22. News and development history

22.1. Changes since the last release

New tools:
  • Added orbit_plot.py (Section 16.41), a tool for plotting camera orientations along an orbit (contributed by Shashank Bhushan).

New camera support:
  • Added the ability to use the CSM camera model with ASTER images (Section 8.17.5).

jitter_solve (Section 16.35):
  • Added the option --weight-image, to weigh observations based on geographic location of triangulated points (Section 12.3.2).

  • Can use anchor points with frame cameras.

  • Added --num-anchor-points-per-tile. This helps when different images have different sizes but want to ensure the same point density.

  • Added the option --anchor-weight-image that is used to limit where anchor points are placed.

  • Added an example for ASTER cameras (Section 16.35.8).

  • Can handle a linescan and frame camera rig (almost parallel rays) (Section 16.35.12).

  • The roll and yaw constraints no longer assume linescan camera positions and orientations are one-to-one.

bundle_adjust (Section 16.5):
  • Added the ability to refine the camera intrinsics per sensor (Section 12.2.2).

  • When optimizing intrinsics, cameras that do not share distortion can have different distortion types and sizes. (Section 12.3.2).

  • Added the option --propagate-errors to propagate the uncertainties from input cameras to triangulated points (Section

  • Added the option --weight-image, to weigh observations based on geographic location of triangulated points. (Section 12.3.2).

  • For ASTER cameras, use the RPC model to find interest points. This does not affect the final results but is much faster.

  • Enforce that option --mapprojected-data use images that were not mapprojected with bundle-adjusted cameras, and also the input cameras themselves are not bundle-adjusted. It is easier to avoid inconsistencies this way.

parallel_stereo (Section 16.47):
  • Added Kaguya processing example (Section 8.14).

  • When a run finished successfully, combine the data from subdirectories and delete these. See --keep-only for more options.

  • Fixed a failure when processing images that have very large blocks (on the order of several tens of thousands of pixels along some dimension, as shown by gdalinfo). A warning, progress bar, and timing info is displayed.

  • For the asp_sgm and asp_mgm algorithms allow cost-mode to have the value 3 or 4 only, as other values produce bad results.

point2dem (Section 16.52):
  • Added the option --auto-proj-center, to automatically compute the projection center for stereographic and other projections (Section

csm (Section 8.12):
  • Fixed several problems in generation of CSM cameras for MSL Curiosity Nav and Mast images. Much large-scale testing was performed. Updated the example showing how to create stereo from either Nav or Mast stereo pairs (Section 8.12.5).

  • A multi-Martian-day example for MSL added (Section

  • Added support for the radial and tangential distortion model with 3 radial distortion parameters and 2 tangential ones. Tested that it agrees with the OpenCV implementation.

  • Fixed a small bug in radial distortion implementation.

stereo_gui (Section 16.64):
  • Can show scattered data with a colorbar and axes (Fig. 16.28).

  • Renamed --colorize-image to --colorbar.

  • Auto-guess and load pc_align error files (Section 16.49.9).

image_calc (Section 16.31):
  • When adding new keywords to metadata geoheader, do not erase the existing ones (if a keyword already exists, its value will be modified).

pc_align (Section 16.49):
  • Add the option --skip-shared-box-estimation.

historical_helper.py (Section 16.26):
  • Added the ability to set a custom path to the needed convert executable and described how that tool can be installed.

sfs (Section 16.61):
  • Added an example for Kaguya TC (Section 8.14.5).

  • Added the option --albedo-robust-threshold.

image_calc (Section 16.31):
  • Added the ability to create a random image.

  • Using PDAL 2.4.2 instead of libLAS for LAS input/output (in pointlas, point2dem, and pc_align). libLAS is no longer developed.

  • Fixed a couple of runtime errors when using conda packages on OSX.

  • Eliminated a procedure for cleaning the name of an input path that was replacing two slashes with one slash, resulting in inconsistencies.

  • Fix an error in conversion between projections for non-Earth images.

  • The OSX build now gets created and tested via GitHub Actions.

22.2. RELEASE 3.3.0, August 16, 2023

DOI: 10.5281/zenodo.8270270

New tools:
  • Added sfm_merge (Section 16.58), a tool to merge several SfM reconstructions that may or may not have shared images.

  • Added sfm_submap (Section 16.59), a tool to extract a submap from a Structure-from-Motion map in .nvm format, as produced by theia_sfm (Section 16.66) or refined with rig_calibrator (Section 16.55).

  • Added a couple of small Python scripts for handling ROS bags (Section 16.56). No ROS binaries are shipped.

  • Added support for the Pleiades NEO exact linescan model (Section 8.21.3).

  • Added sat_sim (Section 16.57), a tool to create simulated satellite images camera models for pinhole or linescan sensors.

  • Added sfm_view (Section 16.60), a tool for viewing orbital Pinhole camera models. It is a modification of the umve program.

csm (Section 8.12):
  • Added initial support for using CSM camera models with MSL Curiosity (Section 8.12.5).

parallel_stereo (Section 16.47):
  • Can propagate horizontal ground plane standard deviations (stddev) specified for each camera through triangulation, obtaining the horizontal and vertical stddev for each triangulated point. For DigitalGlobe RPC and Pleiades linescan cameras the input horizontal stddev can be read from camera files. A formula to go from known CE90 to input horizontal stddev is provided (Section 14).

  • Can propagate the covariances of satellite positions and quaternions to the triangulated point cloud for Maxar (DigitalGlobe) linescan cameras (Section 14).

  • Documented the pre-processing options --stddev-mask-kernel and --stddev-mask-thresh (Section 17.1). Fixed a bug in writing out debug images for this option.

  • The cameras files used in mapprojection can be switched to other ones for the purpose of doing stereo or reusing a stereo run (Section

  • Added the option --matches-per-tile, to attempt to guarantee that each 1024 x 1024 tile has about this many number of matches.

  • Bugfix for stereo with mapprojected Pleiades images. If the mapprojection is done with the exact (non-RPC) cameras, stereo must load the exact cameras when undoing the mapprojection.

bundle_adjust (Section 16.5):
  • Validated that given about a thousand input images acquired with three views and much overlap, one can have very tight control of the cameras in bundle adjustment, while improving the self-consistency of the camera configuration (Section 8.23.1).

  • Validated the option --reference-dem for bundle adjustment. This works even when rays obtained during triangulation are parallel.

  • Added the option --matches-per-tile, to attempt to guarantee that each 1024 x 1024 tile has about this many number of matches (example in Section 16.5.1).

  • Bugfix for slow performance and memory usage for a large number of images.

jitter_solve (Section 16.35):
  • Added the options --roll-weight and --yaw-weight to control the amount of change in these angles relative to the along-track direction. To be used with synthetic cameras created with sat_sim (Section 16.57).

  • Added a section discussing solving for jitter with synthetic camera models (Section 16.35.9).

  • The solver can mix and match linescan and pinhole (frame) camera images if the inputs are all in the CSM format (Section 16.35.11).

  • Added a section on how to prepare interest point matches (Section 16.35.4).

  • Validated the option --reference-dem for bundle adjustment. This works even when rays obtained during triangulation are parallel.

  • Ensure jitter can be solved for the case of one frame and one linescan camera with no baseline, when triangulation becomes tricky (Section 16.35.12).

  • Bugfix for reverse scan direction.

  • Added an example for Pleiades cameras (Section 16.35.7), comparing two ways of setting ground constraints.

sfs (Section 16.61):
  • Created an SfS DEM of size 14336 x 11008 pixels, at 1 m pixel with 420 LRO NAC images with various illuminations and orientations. Reliably bundle-adjusted 814 LRO NAC images in which the shadows were observed to make a full 360 degree loop, with a total of 614k triangulated points. Updated the documentation reflecting latest best practices (Section 13.9).

  • Create more detail in the reconstructed DEM in borderline lit regions. Option: --allow-borderline-data (Section 13.8.2).

  • Added the options --image-list and --camera-list for when the number of images becomes too large to set on the command line.

rig_calibrator (Section 16.55):
  • Added a detailed tutorial describing how this tool was used to create a full 360-degree textured mesh of the JEM ISS module (Section 9.2) using data acquired with two rigs (6 sensors).

  • Added an example for the MSL Curiosity rover (Section 9.3).

  • Allow multiple rigs to be jointly optimized (the rig constraint applies within individual rigs and not between them).

  • Added the option --extra_list to insert additional images close in time to some of the images already on the rig (for the same or different rig sensor). Helps easily grow a map and complete a rig.

  • Added the option of keeping a subset of the camera poses fixed (for when those have been validated in a smaller map).

  • Images for any of the rig sensors (not just the reference one) can be used in registration (but all must be for same sensor).

  • Added the --save_pinhole_cameras option to save the optimized cameras in ASP’s Pinhole format (with no distortion for now).

  • Absorb --rig_transforms_to_float into --camera_poses_to_float.

  • Save alongside an .nvm file a separate file having the values of optical center point that are subtracted from each interest point (for plotting in stereo_gui).

  • Merge the interest point tracks created as part of rig calibration with the matches read from disk beforehand.

  • Fix for too many valid interest point matches being filtered out.

voxblox_mesh (Section 16.68):
  • Added median filtering of input point clouds (option --median_filter).

  • Added weighing of depth points given their distance from the sensor (option --distance_weight).

multi_stereo (Section 16.39):
  • Left and right images can be acquired with different sensors.

  • Use --left and --right to specify which stereo pairs to run.

texrecon (Section 16.65):
  • Can create a texture with images from multiple sensors.

point2dem (Section 16.52):
  • Added the option --propagate-errors to grid the stddev values computed during stereo triangulation.

  • Added the option --input-is-projected to specify that the input coordinates are already in the projected coordinate system.

stereo_gui (Section 16.64):
  • Can read, write, edit, and overlay on top of images polygons in plain text format in addition to the .shp format. Per-polygon colors are supported.

  • Can read nvm files whose features are shifted relative to the optical center, if an .nvm file is accompanied by an _offsets.txt file.

  • Added the option --preview to load one image at a time, and quickly cycle through them with the ‘n’ and ‘p’ keys.

  • Added the option --view-several-side-by-side to view several images side-by-side with a dialog to choose which images to show (also accessible from the View menu).

  • Added the option --font-size, with the default of 9.

  • Added the option --lowest-resolution-subimage-num-pixels to control the behavior of the pyramid of subimages.

  • Noticeable speedup in loading images.

  • Bug fix in loading .nvm files (an SfM format).

image_align (Section 16.30):
  • Can find the 3D alignment around planet center that transforms the second georeferenced image to the first one. This transform can be used to apply the alignment to cameras and point clouds (Section 16.30.3).

dem_mosaic (Section 16.19):
  • Added the option --fill-search-radius to fill nodata pixels in a DEM with nearby valid values. This is different from --hole-fill-length. See an example in Section 16.19.2.

wv_correct (Section 16.69):
  • Maxar (DigitalGlobe) WorldView-2 images with a processing (generation) date (not acquisition date), of May 26, 2022 or newer have much-reduced CCD artifacts, and for those this tool will in fact make the solution worse, not better. This does not apply to WorldView-1, 3, or GeoEye-1.

  • ASP builds after 2023-06-21 (so, version later than 3.2.0), will detect the above-mentioned scenario, and will not apply any correction in that case (a copy of the original image will be written instead and a warning will be printed). This applies to both PAN and multi-spectral images.

corr_eval (Section 16.16):
  • Remove an excessive check. The refined/filtered disparity can be such that left image pixel plus disparity may be outside the right image. Don’t fail in that case, but just skip the pixel, resulting in empty correlation for that pixel.

cam_test (Section 16.9):
  • Added the option --datum. Useful for Pinhole cameras as those don’t know their datum.

  • Added a warning if the camera center is below the datum.

  • Upgraded to ISIS 8.0.0 and USGSCSM 1.7.

  • Throw an error for WorldView products that are not Stereo1B or Basic1B. That because ASP does not support orthorectified Maxar products.

  • Changed the “pretend” height of the RPC cameras from 10 km above ground to around 100 km. RPC camera models do not store this number and it does not make a difference normally, as only ray directions to the ground matter. Yet, .adjust files created with an earlier version of ASP for RPC cameras should be re-created as those use the camera center.

  • The latest version of the Xerces-C XML parser became 10 times slower than before, which may affect the speed of processing for XML-based camera models.

  • Added back the tool view_reconstruction, for examining Theia’s SfM solution (Section 10).

  • The theia_sfm tool can write the optical offsets for a given nvm file which can be used in plotting such files in stereo_gui.

  • Added to hiedr2mosaic.py (Section 16.24) the option --web to invoke spiceinit with web=True. Contributed by Giacomo Nodjoumi.

  • Bugfix for reading .jp2 files. Needed to ship the JPEG2000 driver and set GDAL_DRIVER_PATH.

  • Fixed a failure in mapproject with a small DEM.

  • Bugfix for exporting the TheiaSfM matches in camera_solve.

  • The documentation of the examples chapter was broken up into individual pages (Section 8).

22.3. RELEASE 3.2.0, December 30, 2022

DOI: 10.5281/zenodo.7497499

Added functionality for creation of large-scale meshes and fused textures for small planetary bodies and indoor environments. Added logic for rig calibration. See individual tools below.

New tools:
  • rig_calibrator: Calibrates a rig of N image and/or depth+image cameras. Can also co-register and refine intrinsics of camera images acquired with N sensors with no rig constraint (Section 16.55).

  • multi_stereo: Runs multiple stereo pairs and produces a fused mesh. Uses parallel_stereo, pc_filter, and voxblox_mesh (Section 16.39).

  • voxblox_mesh: Fuses point clouds into a seamless oriented mesh, with each input point given a weight according to its reliability. Based on the third-party VoxBlox software (Section 16.68).

  • texrecon: Creates seamless textured meshes. Based on the third-party MVS-Texturing project (Section 16.65).

  • pc_filter: Filters outliers in point clouds created with pinhole cameras and weighs inliers based on many criteria (Section 16.50).

  • Added CGAL-based tools for mesh smoothing, hole-filling, remeshing, and removal of small connected components (Section 16.13).

  • jitter_solve: A tool for solving for jitter in CSM camera models (Section 16.35). It gives promising results for CTX, Pleiades, and DigitalGlobe data. Examples are provided.

Removed tools:
  • datum_convert: This was an attempt at having a tool applying a transform between datums. It is suggested to use GDAL/PROJ instead. Note that a datum transform may require fetching transformation grids, and without them PROJ will quietly return incorrect results.

New sensors:
  • Support the Pleiades exact sensor (for 1A/1B). See Section 8.21. Implemented as a wrapper around the CSM linescan camera model.

parallel_stereo (Section 16.47):
  • Added the options --match-files-prefix and --clean-match-files-prefix for reusing interest point matches from a previous bundle_adjust or parallel_stereo run. The “clean” interest point matches created by bundle_adjust may have fewer outliers than what stereo can create.

  • Added the option --keep-only to convert all VRT files to TIF (e.g., D.tif), then wipe all files and subdirectories except those specified by given suffixes.

  • Added the triangulation option --max-valid-triangulation-error.

  • The option --prev-run-prefix can be used to start a run with bathymetry modeling at the triangulation stage while reusing the previous stages of a run without such modeling (the needed aligned bathy masks are created, if needed, at the triangulation stage, if not done, as usual, at the preprocessing stage).

  • For SGM and MGM use by default 8 threads and number of processes equal to number of cores divided by number of threads. Less likely to run out of memory that way.

  • Added examples of using PBS and SLURM with ASP (Section 8.16).

  • Added an example of processing SkySat Stereo data (Section 8.23.1).

  • Documented better the option --num-matches-from-disp-triplets for creating dense and uniformly distributed interest point matches. Useful for modeling lens distortion.

parallel_bundle_adjust (Section 16.45):
  • Do not create subdirectories or symlinks, as that results in a massive number of small files. (Unless --save-vwip is used, see below.)

  • Do not save by default .vwip files as those take space and are only needed to find .match files. Use the new option --save-vwip to save them. Note that these depend on individual image pairs, so parallel_bundle_adjust saves them in subdirectories.

bundle_adjust (Section 16.5):
  • Save the convergence angle percentiles for each pair of images having matches. Useful for understating the configuration of cameras.

  • Added the option --tri-weight (default is 0) to keep triangulated points close to their initial values. This looks more promising than other weighing used so far at preventing the cameras from moving when optimizing them. This assumes input cameras are not grossly inaccurate. This adds a robust cost function with the threshold given by --tri-robust-threshold.

  • Added the options --image-list, --camera-list, --mapprojected-data-list, for when the inputs are too many to specify on the command line.

  • Added the option --fixed-image-list to specify a file having a list of image names whose cameras should be fixed during optimization.

  • Pinhole cameras are no longer automatically reinitialized or transformed based on GCP, but only refined given GCP. So, option --disable-pinhole-gcp-init is the default. Use one of the options --init-camera-using-gcp (Section 10.4), --transform-cameras-with-shared-gcp, --transform-cameras-using-gcp (Section 10.2.3) for manipulating cameras using GCP.

  • Bugfix in initializing pinhole cameras based on GCP for off-nadir cameras.

  • When doing multiple passes (which is the default) at each pass resume not only with clean matches but also with the cameras optimized so far, rather than going to the originals.

  • Can do multiple passes with --heights-from-dem. One should be generous with outlier removal thresholds if not sure of the input DEM accuracy (option --remove-outliers-params).

  • Remove outliers based on spatial distribution of triangulated points.

  • Bugfix when the number of interest points is 4 million or more. The algorithm would just stall. It is now replaced by an OpenMVG algorithm.

  • Fold --remove-outliers-by-disparity-params into --remove-outliers-params.

  • Bugfix in residuals_stats.txt; the mean was correct but the median was wrong.

  • Let the default --heights-from-dem-weight be 1.0, and the default --heights-from-dem-robust-threshold be 0.5. These normally need tuning.

  • Added the option --mapproj-dem. If specified, evaluate the disagreement of interest point matches after mapprojecting onto this DEM, per interest point match pair, per matching image pair, and per image. Useful at evaluating registration without mapprojecting the images (Section 16.5.8).

  • Added report files having the camera positions and orientations before and after optimization (for Pinhole cameras only, Section

  • Added options --proj-win and --proj-str for restricting interest points to given area (useful when having many images with footprints beyond area of interest).

  • With --match-first-to-last, write match files from earlier to later images, rather than vice-versa. This was a bug, as the matches were not being picked up correctly later.

  • For pinhole cameras, can read .adjust files via --input-adjustments-prefix, then apply them to existing .tsai files via --inline-adjustments. Until now one could do either one or the other. Also works with --initial-transform.

  • Added a section describing how bundle adjustment is implemented (Section 16.5.5).

point2dem (Section 16.52):
  • Added the Tukey outlier removal method option applied to triangulation errors (error_thresh = 75th_pct + 1.5 * (75th_pct - 25th_pct)). Also print out these percentages even for the regular outlier removal.

bathymetry (Section 8.27):
  • Added scale_bathy_mask.py, for creating a PAN-sized image or mask from an multispectral-sized image or mask, both for WorldView data.

mapproject (Section 16.38):
  • Exposed and documented the --query-projection option.

stereo_gui (Section 16.64):
  • Can plot, overlay on top of images, and colorize scattered points stored in a CSV file (Section 16.64.6). Many colormap styles are supported. See Section 16.14 for the list.

  • Can show side-by-side colorized images with colorbars and coordinate axes (Section 16.64.5).

  • Given a bundle_adjust output prefix, can select via checkboxes any two images to show side-by-side, and automatically load their match file or clean match file (options: --pairwise-matches and --pairwise-clean-matches, also accessible from the top menu).

  • Visualize pairwise matches read from an nvm file, as created by rig_calibrator --save_nvm_no_shift.

  • Zoom to given proj win from the View menu. Useful for reproducibility. Also accessible with the command-line option --zoom-proj-win.

  • Bug fix for slow overlaying of images with different datums.

  • When all images have a georeference, start in georeference mode.

corr_eval (Section 16.16):
  • Bugfix for excessive memory usage with positive --prefilter-mode.

  • Added a note saying that the user should ensure that this tool uses the same --corr-kernel and --prefilter-mode as parallel_stereo.

  • Added the option --sample-rate.

cam_gen (Section 16.8):
  • Can read Planet’s pinhole.json files. Then no further changes are made to the produced pinhole camera.

  • Fix a bug in output camera center determination, when an input camera is provided.

  • Bugfix in initializing pinhole cameras based on GCP for off-nadir cameras given image corners and no prior camera.

  • Added the options --cam-height and --cam-weight to try to keep the camera at a given height above ground.

  • Added the option --cam-ctr-weight, to help fix the camera center during refinement.

  • If --optical-center is not set for pinhole cameras, use the image center (half of image dimensions) times the pixel pitch. The optical bar camera always uses the image center.

pc_align (Section 16.49):
  • Fix a bug with loading very large DEMs; it was failing because of a 32-bit integer overflow.

colormap (Section 16.14):
  • Upgrade to C++-14, Python 3.9, latest libLAS, OpenCV 4, PCL 1.11, Boost 1.72, ISIS 7.1.0, GDAL 3.5, and PROJ 9.1.0. The latter has a whole new API, intended to handle properly transformations among datums.

  • The lronaccal tool in ISIS 7.1.0 appears buggy. Try using an earlier ISIS version if this is needed.

  • Replaced in some locations ASP’s homegrown coordinate transformation logic with what is in PROJ.

  • Added the option of using the CSM camera with DigitalGlobe WorldView images in bundle adjustment, stereo, and mapprojection (use with --t dg). Option name is --dg-use-csm and must be set consistently for all tools. This speeds up ground-to-image computation by a factor of about 20 (which helps with mapprojection and bundle adjustment). The result of projecting into the camera changes by less than 0.015 pixels from before if using this option. That is due to the fact that different methods are used for position and orientation interpolation. The cam_test option --dg-vs-csm can be used for evaluating this discrepancy. Each of these methods is consistent with itself to within 2e-8 when it comes to projecting from camera to ground and back.

  • Increased the cache size to 1 GB per process for each ASP tool. Added the option --cache-size-mb, to set this. Made the warning message refer to this option when the limit is hit. Documented this for all tools.

  • Using -t pinhole now invokes the same logic as -t nadirpinhole --no-datum, which is same code path used by other sessions. This wipes an old alternative approach. Eliminated much other duplicated and mutated code for various sessions at the preprocessing stage.

  • Bugfix for D.tif VRTs as created by parallel_stereo.

  • Allow whitespaces in stereo.default before option names.

  • Fix a crash in ISIS for international users by setting for all ASP programs the environmental variables LC_ALL and LANG to en_US.UTF-8.

  • parallel_stereo will accept (but ignore) Unicode in stereo.default.

  • Eliminate internal fudging of --ip-uniqueness-threshold, and make it equal to 0.8 for both stereo and bundle_adjust. This was shown to increase the number of interest points in situations when not enough were found.

  • The historical_helper.py program expects a local installation of ImageMagick and the convert tool (available on most systems normally).

22.4. RELEASE 3.1.0, May 18, 2022

DOI: 10.5281/zenodo.6562267

New camera additions:
  • Added support for the USGSCSM Frame, SAR, and PushFrame sensors (until now just the Linescan sensor was supported), together with documentation and examples (for Dawn, MiniRF, and WAC, respectively).

  • Added support for ISIS SAR cameras, together with an example in the doc.

  • Added support for the PeruSat-1 linescan camera model (so far just the RPC model was supported for this satellite).

New tool additions:
  • Added the program corr_eval, for evaluating the quality of produced correlation with several metrics. See also the new option --save-left-right-disparity-difference in parallel_stereo.

  • Added the program otsu_threshold for computing an image threshold. It can be used for separating land from water (in WorldView multispectral NIR bands), and shadowed from lit areas (in Lunar images).

  • The program parallel_stereo can function as purely an image correlation tool, without assuming any camera information, via the option --correlator-mode.

  • Added the program image_align. Used to align two images or DEMs based on interest point matches or disparity, with given alignment transform type (translation, rigid, similarity, affine, and homography).

  • Using ISIS 6.0.0.

  • Save the camera state on multiple lines. On reading both the single-line and multiple-line formats are accepted.

  • Bundle adjustment, mapproject, and SfS with the CSM model can be 7-15 times faster than done with the corresponding ISIS mode for linescan cameras (the latter as reimplemented in ASP itself). It is strongly suggested to use CSM for large-scale processing.

  • Bugfix in CSM linescan implementation for some LRO NAC sensors. Also replaced the fixed-point method with the secant method in the ground-to-image logic for CSM linescan cameras, which is faster.

  • Many fixes for reliability of stereo with local epipolar alignment.

  • Added the option --resume-at-corr. Start at the correlation stage and skip recomputing the valid low-res and full-res disparities for that stage.

  • Bugfix: Eliminate edge artifacts in stereo refinement (for subpixel modes 1, 2, and 3).

  • Print in stereo_pprc the estimated convergence angle for the given stereo pair (for alignment methods affineepipolar, local_epipolar, and homography).

  • Added the option --prev-run-prefix, which makes parallel_stereo start at the triangulation stage while using previous stages from this other run. The new run can have different cameras, different session (rpc vs dg, isis vs csm), different bundle adjustment prefix, and different bathy planes (if applicable).

  • Added option --save-left-right-disparity-difference to save the discrepancy between left-to-right and right-to-left disparities, which may help with filtering unreliable disparities.

  • Interest point matching with mapprojected images now happens at full resolution, which results in a more reliable process when there are clouds or if fine features are washed out at low resolution.

  • Expanded the doc to address a big gotcha: if left and right mapprojected images have somewhat different resolutions, then an immense disparity search range can result.

  • Added the option --max-disp-spread to limit the spread of the disparity to this value (useful with clouds in images).

  • Added the option --ip-filter-using-dem to filter as outliers interest point matches whose triangulated height differs by more than given value from the height at the same location for the given DEM.

  • Added a doc section on handling of images with clouds.

  • Disable by default velocity aberration and atmospheric refraction corrections. These are not accurate enough and cause issues with convergence of bundle adjustment. Can be enabled with --enable-correct-velocity-aberration and --enable-correct-atmospheric-refraction. These improve results however with Digital Globe cameras if not doing bundle adjustment. (Note that these are still hard-coded as enabled for optical bar camera models. This would require further study.)

  • Added ready-made ASTER and LRO NAC examples with sample images, cameras, commands, and outputs, all available for download. Contributions of more examples are welcome. See https://github.com/NeoGeographyToolkit/StereoPipelineSolvedExamples.

  • Bugfix for ASTER cameras; this was fully broken.

  • ASP’s SGM and MGM algorithms will always use the cross-check for disparity by default, to improve the quality, even if that takes more time. It can be turned off with --xcorr-threshold -1.

  • Filter outliers in low-resolution disparity D_sub.tif. Can be turned off by setting the percentage in --outlier-removal-params to 100.

  • Filtering of interest points based on percentiles (using also --outlier-removal-params).

  • Folded --remove-outliers-by-disparity-params into --outlier-removal-params.

  • Bugfix in disparity search range handling when it is large.

  • For Linux, in each tile’s directory write the elapsed runtime and memory usage to <tile prefix>-<prog name>-resource-usage.txt.

  • Removed the --local-homography option, as it is superseded by --alignment-method local_epipolar, which blends the local results.

  • The stereo tool is deprecated, and can be used only with the ASP_BM classical block-matching algorithm when invoked without local epipolar alignment. Use parallel_stereo instead.

  • Added the experimental --gotcha-disparity-refinement option, under NASA proposal 19-PDART19_2-0094.

  • Added the option --apply-initial-transform-only to apply an initial transform to cameras while skipping image matching and other steps, making the process much faster.

  • Added the option --auto-overlap-params to automatically compute which camera images overlap, if a DEM and camera footprint expansion percentage are given.

  • Added the option --max-pairwise-matches to put an upper limit on the number of matches, as a large number can slow down bundle adjustment. The default is 10000, likely a large overestimate (but this includes potential outliers). If creating interest points from disparity using --num-matches-from-disp-triplets, similar values should be used for both of these options.

  • Stop printing warnings about failed triangulation if their number goes over 100.

  • Rename verbose final_residuals_no_loss_function_pointmap_point_log.csv to final_residuals_pointmap.csv and final_residuals_no_loss_function_raw_pixels.txt to final_residuals_raw_pixels.txt, etc.

  • Document the useful initial and final residuals_stats.txt files.

  • Added new options for reusing a previous run: --match-files-prefix and --clean-match-files-prefix.

  • SfS was made to work with any camera model supported by ASP, including for Earth. For non-ISIS and non-CSM cameras, the option --sun-positions should be used.

  • Exhaustively tested with the CSM model. It is very recommended to use that one instead of ISIS .cub cameras, to get a very large speedup and multithreading.

  • Added a new --gradient-weight parameter, constraining the first-order derivatives. Can be used in combination with the --smoothness-weight parameter which constrains the second-order derivatives. The goal is to avoid a noisy solution without losing detail.

  • Much work on expanding the documentation.

  • If the input image file has an embedded RPC camera model, append it to the output mapprojected file. (Which makes stereo with mapprojected images work correctly in this case.)

  • Always start a process for each tile. The default tile size is set to 5120 for non-ISIS cameras and to 1024 for ISIS. Use a large value of --tile-size to use fewer processes.

  • Can have different water surfaces in left and right images, so the triangulating rays bend at different heights.

  • bathy_plane_calc can use a mask of pixels above water to find the water-land interface, and also a set of actual lon, lat, height measurements.

  • Added documentation for how to find water level heights at given times and coordinates using National Ocean Service’s tidal zoning map.

  • Add alignment method similarity-point-to-plane. It works better than similarity-point-to-point at finding a scale between the clouds when they have a large shift.

  • Bugfix for alignment methods point-to-point and similarity-point-to-point.

  • Use RANSAC with --initial-transform-from-hillshading, for increased robustness to outliers. Replaced --initial-transform-outlier-removal-params (based on percentiles) with --initial-transform-ransac-params.

  • Add the option --tap, to have the output grid be at integer multiples of the grid size (like the default behavior of point2dem and mapproject, and gdalwarp when invoked with -tap). If not set, the input grids determine the output grid. (The produced DEM will then extend for an additional 0.5 x grid_size beyond grid centers on its perimeter.)

  • Do not allow more than one of these operations in a given dem_mosaic invocation: fill holes, blur, or erode. These won’t work when also having more than one input DEM, reprojection is desired, or priority blending length is used. This is done to avoid confusion about order of operations, and the fact that different input DEMs can have different grid sizes and hence the input parameters have different effects on each.

  • Bugfix for hole-filling and blurring. Tile artifacts got removed.

  • Can cycle through given images from the View menu, or with the ‘n’ and ‘p’ keys, when all images are in the same window.

  • Can save a shapefile having points, segments, or polygons. (These are distinct classes for a shapefile; the shapefile format requires that these not be mixed in the same file.)

  • Noticeable speedup when changing display mode (e.g., from side-by-side to overlayed).

  • Bugfix when overlaying shapefiles with different georeferences.

  • Polygon layers can be set to desired colors from the left pane, when overlaid.

  • On startup, draw rectangular regions corresponding to values of --left-image-crop-win and --right-image-crop-win, if these are passed in as command line arguments together with two images.

  • Quietly accept parallel_stereo options and pass them on if this tool is invoked from the GUI.

  • Add the option --no-georef to remove any georeference information in the output image (useful with subsequent GDAL-based processing).

  • Added the option --longitude-offset to help to deal with the fact that ASP-produced DEMs and orthoimages may have the longitudes in [0, 360] while users may prefer [-180, 180].

  • Bugfix: The --input-nodata value, if set, now overrides the value set in the metadata (the previous value then becomes valid).

  • Added the tool parse_match_file.py to convert a binary match file to text and vice-versa.

  • Add the tool cam_test to compare two different camera models for the same image.

  • Stereo and bundle adjustment with RPC cameras now query the RPC model for the datum.

  • The cam2rpc program saves its datum which is read when needed by the RPC model loader.

  • Add the option --triangulation-error-factor to point2las to allow point cloud triangulation errors multiplied by this factor and rounded/clamped appropriately to be stored in the 2-byte intensity field in the LAS file.

  • Make symlinks relative in parallel_bundle_adjust for portability.

  • The mapprojected image saves as metadata the adjustments it was created with.

  • Save the low-resolution triangulated point cloud (PC_sub.tif) in stereo_corr (based on filtered D_sub.tif).

  • The ipmatch program can take as input just images, with the .vwip files looked up by extension.

  • Bugfix in handling projections specified via an EPSG code.

  • Bugfix when some environmental variables or the path to ASP itself have spaces. (It happens under Microsoft WSL.)

  • Bugfix for the “too many open files” error for large images.

  • Add the build date to the --version option in the ASP tools and to the log files.

  • Bugfix in the original author’s MGM implementation, accepted by the author.

22.5. RELEASE 3.0.0, July 27, 2021

DOI: 10.5281/zenodo.5140581

New functionality:
  • Added new stereo algorithms: MGM (original author implementation), OpenCV SGBM, LIBELAS, MSMW, MSMW2, and OpenCV BM to complement the existing ASP block matching, SGM, and MGM algorithms. See https://stereopipeline.readthedocs.io/en/latest/next_steps.html for usage. These will be further refined in subsequent releases.

  • Added the ability to perform piecewise local epipolar alignment for the input images, to be followed by a 1D disparity search (for non-mapprojected images), as suggested by the Satellite Stereo Pipeline (S2P) approach. This is still somewhat experimental.

  • Added the ability for a user to plug into ASP any desired stereo program working on image clips to which epipolar alignment has been applied (as is customary in the computer vision community) without rebuilding ASP.

  • Added support for shallow-water bathymetry, so creation of terrain models with correct depth determination for well-resolved areas under shallow water. To be used with dg, rpc, and nadirpinhole cameras.

  • Added two supporting tools for this: bathy_plane_calc and bathy_threshold_calc.py.

  • Added CCD artifact corrections for a few WV02 band 3 multispectral images. Apart from the systematic artifacts corrected by this logic, these images have a high-frequency unique pattern, and also jitter, which are not corrected for. Also added tools and documentation to easily tabulate more multispectral bands and TDI.

  • Upgraded to ISIS 5.0.1.

  • Ship a full Python 3.6 runtime, as expected by ISIS.

  • Upgraded to USGSCSM 1.5.2 (ASP’s own build of it has an additional bugfix for LRO NAC not present in the conda-forge package).

  • Validated the CSM model for CTX, HiRISE, and LRO NAC cameras.

  • Added documentation for how to create CSM models from .cub cameras.

  • Export the state of a CSM camera after bundle adjustment and pc_align (only for linescan cameras supported by ISIS).

  • Will now throw an error if --threads is passed in, whose behavior was not defined.

  • Bugifx for Python 3.

  • Added the option --heights-from-dem-robust-threshold.

  • Added the option --save-intermediate-cameras to save the cameras at each iteration.

  • Added the option --match-first-to-last to match the first several images to several last images by extending the logic of --overlap-limit past the last image to the earliest ones.

  • Remove outliers by using a percentile times a factor, in a way analogous to point2dem.

  • Improve the accuracy of the RPC approximation distortion and undistortion.

  • Added the option --shadow-threshold to be able to specify a single shadow threshold for all images. Also added --custom-shadow-threshold-list.

  • Added the option --robust-threshold for situations when the measured image intensity is unreliable.

  • Added the option --estimate-height-errors to estimate the uncertainty in height at each computed SfS DEM pixel. It can be customized via --height-error-params.

  • Added an auxiliary tool named sfs_blend to replace SfS pixels with ones from the original LOLA DEM in permanently shadowed regions.

  • Added the ability to find the contour of a georeferenced image at a given threshold. (It can be later edited, saved to disk, etc.)

  • Bugifxes for polygon drawing logic.

  • Much more responsive for overlaying many images.

  • Support the sign function (can help in creating masks).

  • Bugifx for --initial-transform-from-hillshading with outlier removal.

  • Add the --initial-transform-outlier-removal-params to control outlier removal when finding matches between DEMs to align using features detected in hillshaded images or selected manually.

  • Added --initial-rotation-angle, to initialize the alignment transform as the rotation with this angle (in degrees) around the axis going from the planet center to the centroid of the point cloud.

  • Moved the daily build to the release area on GitHub, at https://github.com/NeoGeographyToolkit/StereoPipeline/releases

  • Upgraded to GDAL 2.4 and PROJ4 5.2.0. (ISIS constrains updating to newer versions of these.)

  • Added the option --ip-per-image to bundle adjustment and stereo, to detect roughly how many interest points should be found per image (only a small fraction of them may eventually match across images).

  • The --min-triangulation-angle in stereo must be always positive if set by the user. Can be set to something very small if desired. This is a bug fix for this rarely used option (before, when set to 0 it would just reset itself to some internal non-small value).

  • Bugifx for the VisionWorkbench implementation of the Levenberg-Marquardt algorithm, it was giving up prematurely in challenging situations.

  • Bugifx for affine epipolar alignment. Use the OpenCV function for finding the alignment matrix instead of the ASP one as OpenCV can filter outliers which cause issues on rare occasions.

  • Bugfix: Do not allow a full run to take place in a directory where a clip was run, as that will produce incorrect results.

22.6. RELEASE 2.7.0, July 27, 2020

New functionality
  • Can first create interest point matches among mapprojected images (automatically or manually) and use those to create matches among the unprojected images when the latter are so dissimilar in perspective that the direct approach fails. See --mapprojected-data.

  • Bug fix when zooming all images to same region when the region is such that all images are seen fully.

  • Added a new very challenging example at the South Pole with drastic illumination changes and using a non-stereo DEM as initial guess.

  • Fixed a bug with craters missing under low light.

  • Fixed a bug with computation of exposures in terrain with many shadows.

  • Print the Sun azimuth angle for all images (useful for sorting them by illumination conditions).

  • When hijitreg finds no match points between two CCDs, the program now emits a warning message to STDOUT with a suggestion to perhaps fiddle with hijitreg manually, and rather than fail with a mysterious exception warning, now gracefully falls back to assuming that there is no jitter correction between the two CCDs that had no matches.

  • Use outlier filtering when computing the bounding box of a DEM. The same option --remove-outliers-params controls this just as for removing outliers by triangulation error.

  • Fixed a bug when finding the extent of the mapprojected image when the DEM to project onto spans the whole planet.

  • Only meshes in .obj format are created. This format can be opened in Meshlab, Blender, or some other mesh viewer.

  • The osgviewer program is no longer shipped.

  • Fixed a bug with invalid points not being filtered.

  • Fixed a bug with insufficient precision (now it can be set by the user and defaults to 17 digits).

  • Added the option --texture-step-size to control the sampling rate for the texture, in addition to the -s option that controls the sampling rate for the point cloud.

  • Updated to C++ 11.

  • Added phase subpixel correlation accuracy parameter.

22.7. RELEASE 2.6.2, June 15, 2019

DOI: https://doi.org/10.5281/zenodo.3247734

New satellites
  • Added support for SkySat, together with a detailed example, including how to jointly align and optimize cameras in respect to a reference DEM, while optionally refining the intrinsics. This approach may be helpful for other images obtained with frame cameras and uncertain positioning information.

  • Added support for CORONA KH-4B, KH-7, and KH-9 declassified images and their panoramic (optical bar) camera models, as well as using and optimizing camera models with RPC distortion (only RPC is supported for KH-7 because it is a linescan camera). An example is in the documentation.

New tools
  • Added parallel_bundle_adjust which computes image statistics and IP matching in a parallel manner similar to parallel_stereo.

  • Added the cam_gen tool to create a correctly oriented pinhole camera model given camera intrinsics, lon-lat coordinates of the corners (or some other pixels), and optionally a ground truth DEM. It can also parse SkySat’s video/frame_index metafile to get this data. It can also take as input any camera supported by ASP via --input-camera and create a most-similar pinhole camera model with given intrinsics.

  • Added the coverage_fraction tool to provide a coverage estimate of the results of a stereo call.

  • Added the image_mosaic tool which merges together images based on interest point matches. Can be used to stitch together Corona scanned images.

  • Added a new tool, n_align, to jointly align n clouds (re-implemented from Matlab, works well for small clouds that are close to each other).

  • Added the option to run a non-SGM subpixel option after running SGM/MGM.

  • Added the phase correlation subpixel option. This is a Fourier transform based method.

  • Added a new approach to finding an initial transform between clouds, when they are DEMs, that may be more robust to large scale or translation changes, or to noise. It is based on hillshading the DEMs and finding interest point matches among them, which are then used to find the transform. Can be invoked with --initial-transform-from-hillshading <transform type>. Supported transforms are: ‘similarity’ (rotation + translation + scale), ‘rigid’ (rotation + translation) and ‘translation’.

  • Added the expression of the Euler angles in the North-East-Down coordinate system around the center of gravity of the source cloud.

  • Bug fix: intersection of bounding boxes of the clouds takes into account the initial transform applied to the source points.

  • Added a new alignment algorithm, based on https://github.com/IntelVCL/FastGlobalRegistration It can be invoked with --alignment-method fgr. It can perform better than ICP when the clouds are close enough to each other but there is a large number of outliers, when it can function with very large --max-displacement. It does worse if the clouds need a big shift to align.

  • Two passes of bundle adjustment (with outlier filtering after

  • first pass) is now the default.

  • The flag --skip-rough-homography is on by default as it usually gives more reliable results. Use --enable-rough-homography to turn this option back on (useful when the footprint on the ground and difference in perspective are large).

  • The flag --disable-tri-ip-filter is also the default as input cameras may not be reliable enough for this filter. Can be enabled back with --enable-tri-ip-filter.

  • Added the --intrinsics-limits option to manually specify intrinsic parameter limits.

  • Added the --num-random-passes option to allow repeat solving attempts with randomly distorted initial parameters.

  • Added option to automatically guess overlapping images from Worldview style XML camera files.

  • Removed the non-Ceres bundle adjustment options.

  • Added the option to share or not share selected intrinsic parameters between pinhole cameras when optimizing intrinsics.

  • Improvements in solving simultaneously for both intrinsics and extrinsics of n camera images if underlying ground truth terrain in the form of a DEM or LIDAR point cloud is present. After this bundle adjustment, pairwise stereo and DEM creation, the DEMs are well-aligned to the ground truth.

  • Added the flag --reference-terrain-weight which, when increased, helps align better camera images to a given reference terrain.

  • Added the option --heights-from-dem. It is very helpful in determining an unknown focal length and distortion parameters for pinhole cameras. It can be used together with ---heights-from-dem-weight.

  • Bug fix in outlier filtering for n images.

  • Updated Ceres version from 1.11 to 1.14. When optimizing with multiple threads, results now vary slightly from run to run. Results from single threaded runs are deterministic.

  • Added a new --parameter-tolerance option. Stop when the relative error in the variables being optimized is less than this.

  • Documented the ability to create a roughly positioned pinhole camera model from an image if its intrinsics and the longitude and latitude (and optionally height) of its corners (or some other pixels) are known.

  • When multiple passes happen with outliers removed, match files are not over-written, but a new clean copy of them gets saved.

  • Renamed --create-pinhole-cameras to --inline-adjustments, and distortion_params to other_intrinsics. This is needed since for the panoramic model there will be other intrinsic parameters as well.

  • Added the option --forced-triangulation-distance for when one really needs to triangulate with poor cameras. Can be used with a very small but positive value of --min-triangulation-angle.

  • Added the option --transform-cameras-using-gcp. If there are at least two images with each having at least 3 GCP (each GCP need not show in more than one image), use this to convert cameras from an abstract coordinate system to world coordinates.

  • Increased the default --num-ransac-iterations to 1000 from 100 so that the solver tries harder to find a fit. Increased default --ip-inlier-factor from 1/15 to 0.2 to help with getting more interest points for steep terrain with the pinhole session.

  • Increased the default --ip-uniqueness-threshold from 0.7 to 0.8 to allow for more interest points.

  • Option to filter interest points by elevation limit and lon-lat limit after each pass of bundle adjustment except the last.

  • Added normalized median absolute deviation (NMAD) output option.

  • Added the option --force-projwin to create a mosaic filling precisely the desired box specified via --t_projwin.

  • Added the ability to manually reposition interest points.

  • Can now show non-synchronous .match files (that is, each IP need not be present in all images).

  • Added basic functionality for drawing/editing/merging polygons on

  • top of georeferenced images or DEMs. The polygons can be saved as shape files, and then used to cut out portions of images with GDAL.

  • Added the option --nodata-value. Pixels with value less than or equal to this are shown as transparent.

  • Added the ability to view .vwip files (specify one per image).

  • Can view (but not edit) GCP files, via --gcp-file (creating GCP is supported in a separate mode, per the doc).

  • The option --dem-file specifies a DEM to use when creating manually picked GCP and --gcp-file specifies the name of the GCP file to use upon saving such GCP.

  • Added the --nearest-neighbor option to use that interpolation method instead of bicubic. This is better for labeled images which should not be interpolated.

  • Can create RPC distortion models of any degree, which can be further optimized in bundle_adjust. Old RPC distortion files are still supported throughout ASP, but not functionality which optimizes them. They can be approximately converted to new type RPC distortion files with this tool if optimization is desired.

  • Compiled against USGS ISIS version 3.6.0.

  • Expanded the documentation explaining how to align cameras to a DEM manually (or initialize such cameras) by selecting matching points between the images and the DEM.

  • The stereo tools and bundle_adjust will now cache image statistics and interest points to files on disk.

  • In stereo and bundle_adjust, when images or cameras are newer than the match files, the latter get recomputed unless the tools are invoked with --force-reuse-match-files.

  • Added a fix to make stereo work with the ZY3 satellite.

  • For stereo and bundle_adjust, added the --no-datum option to find interest points without assuming a reliable datum exists, such as for irregularly shaped bodies. Added the related option --skip-rough-homography to not use the datum in rough homography computation. Added the option --ip-num-ransac-iterations for finer control of interest point matching. Added --ip-triangulation-max-error to control the triangulation error.

  • The cam2rpc tool accepts --t_srs and --semi-major-axis as alternatives to --datum and --dem-file.

  • Add option --theia-overrides to camera_solve to make it easier to customize its behavior via flags.

  • Added an explanation for how the pinhole model works.

22.8. RELEASE 2.6.1, August 13, 2018

New satellites
  • Support Cartosat-1 and Perusat-1 RPC cameras.

New tools
  • Added convert_pinhole_model, to convert between various existing such models.

  • Added camera_footprint as a helpful utility to show where images will project on to the ground.

  • Documented and improved the ipfind and ipmatch tools. ipfind is used to detect interest points in input images, either to generate .vwip files for other tools or to experiment with different IP finding settings. ipmatch matches the IPs contained in .vwip files to create .match files.

New camera models
  • Added simple atmospheric refraction correction to the DG and SPOT5 camera models. This can be enabled using the “–enable-correct-atmospheric-refraction” option.

  • Added support for pinhole camera models where the lens distortion is given by an RPC model (rational polynomial coefficients), of degrees 4, 5, and 6. Such a model may be more expressive than existing ones, and its coefficients can now be optimized using bundle adjustment. An initial model can be created with convert_pinhole_model.

  • Added new options for post-SGM subpixel stereo. Previously only a parabola method was used.

  • Added option to perform cross-correlation checks on multiple resolution levels while using SGM/MGM.

  • Added option --corr-search-limit to constrain the automatically computed correlation search range.

  • Added --corr-memory-limit-mb option to limit the memory usage of the SGM/MGM algorithms.

  • Improved search range estimation in nadir epipolar alignment cases. Added --elevation-limit option to help constrain this search range.

  • Added hybrid SGM/MGM stereo option.

  • Improvements to SGM search range estimation.

  • Added --min-num-ip option.

  • Added the ability to optimize pinhole camera intrinsic parameters, with and without having a LIDAR or DEM ground truth to be used as reference (the latter is recommended though).

  • The tool is a lot more sensitive now to --camera-weight, existing results may change a lot.

  • Added the parameters --rotation-weight and --translation-weight to penalize large rotation and translation changes.

  • Added the option --fixed-camera-indices to keep some cameras fixed while optimizing others.

  • Can read the adjustments from a previous invocation of this program via --input-adjustments-prefix.

  • Can read each of pc_align’s output transforms and apply it to the input cameras via --initial-transform, to be able to bring the cameras in the same coordinate system as the aligned terrain (the initial transform can have a rotation, translation, and scale). If --input-adjustments-prefix is specified as well, the input adjustments are read first, and the pc_align transform is applied on top.

  • Renamed --local-pinhole to --create-pinhole-cameras.

  • Added the parameter --nodata-value to ignore pixels at and below a threshold.

  • Added the ability to transfer interest points manually picked in mapprojected images to the the original unprojected images via --mapprojected-data.

  • Added the flag --use-lon-lat-height-gcp-error. Then, if using GCP, the three standard deviations are interpreted as applying not to x, y, z but to latitude, longitude, and height above datum (in this order). Hence, if the latitude and longitude are known accurately, while the height less so, the third standard deviation can be set to something much larger.

  • Added the ability to do multiple passes of bundle adjustment, removing outliers at each pass based on reprojection error and disparity (difference of pixel value between images). This works for any number of cameras. Match files are updated with outliers removed. Controlled via --num-passes, --remove-outliers-params and --remove-outliers-by-disparity-params.

  • Added the option --save-cnet-as-csv, to save the control network containing all interest points in the format used by ground control points, so it can be inspected.

  • If --datum is specified, bundle_adjust will save to disk the reprojection errors before and after optimization.

  • Can view SPOT5 .BIL files.

  • Add the ability to help the tool with an initial translation specified as a North-East-Down vector, to be used to correct known gross offsets before proceeding with alignment. The option is --initial-ned-translation.

  • When pc_align is initialized via --initial-transform or --initial-ned-translation, the translation vector is now computed starting from the source points before any of these initial transforms are applied, rather than after. The end point of this vector is still the source points after alignment to the reference. This is consistent with the alignment transform output by the tool, which also is from the source points before any initial alignment and to the reference points.

  • The translation vector was expressed incorrectly in the North-East-Down coordinate system, that is now fixed.

  • If the -o option value is specified as filename.tif, all mosaic will be written to this exact file, rather than creating tiles.

  • Added the ability to apply a filter to the cloud points in each circular neighborhood before gridding. In addition to the current weighted average option, it supports min, max, mean, median, stddev, count, nmad, and percentile filters. The --search-radius-factor parameter can control the neighborhood size.

  • Sped up hole-filling in ortho image generation. If this creates more holes than before, it is suggested to relax all outlier filtering, including via --remove-outliers-params, median filtering, and erosion.

  • Added the option --orthoimage-hole-fill-extra-len to make hole-filling more aggressive by first extrapolating the cloud.

  • Rewrote the tool to depend on the Proj.4 HTDPGrids grid shift system. This fixed some situations where the tool was not working such as WGS84/NAD83 conversions and also added support for datum realizations (versions).

  • Vertical datum conversion is only supported in simple cases like D_MARS to MOLA.

  • Even with HTDPGrids, datum support with the Proj.4 library is poor and will hopefully be improved with future releases. Until then try to get external verification of results obtained with the datum_convert tool.

  • Supports WV2 TDI = 32 in reverse scan direction.

  • We now compile against USGS ISIS version 3.5.2.

  • The tools mapproject, dem_mosaic, dg_mosaic, and wv_correct support the --ot option, to round the output pixels to several types of integer, reducing storage, but perhaps decreasing accuracy.

  • The tools mapproject and image_calc support the --mo option to add metadata to the geoheader in the format ‘VAR1=VAL1 VAR2=VAL2’, etc.

  • Handle properly in bundle_adjust, orbitviz, and stereo with mapprojected images the case when, for RPC cameras, these coefficients are stored in _RPC.TXT files.

  • Support for web-based PROJ.4 strings, e.g., point2dem --t_srs http://spatialreference.org/ref/iau2000/49900/

  • Added --max-output-size option to point2dem to prevent against creation of too large DEMs.

  • Added image download option in hiedr2mosaic.py.

  • Bug fix in cam2map4stereo.py when the longitude crosses 180 degrees.

  • Added support for running sparse_disp with your own Python installation.

  • Bug fix for image cropping with epipolar aligned images.

  • The sfs tool supports the integrability constraint weight from Horn 1990.

  • The software works with both Python versions >= 2.6 and 3.

22.9. RELEASE 2.6.0, May 15, 2017

New stereo algorithms
  • ASP now supports the Semi Global Matching (SGM) and More Global Matching (MGM) stereo algorithms. They do particularly well for Earth imagery, better than the present approaches. They can be invoked with --stereo-algorithm 1 and 2 respectively.

New tools
  • Added cam2rpc, a tool to create an RPC model from any ASP-supported camera. Such cameras can be used with ASP for Earth and planetary data (stereo’s --datum option must be set), or passed to third-party stereo tools S2P and SETSM.

  • Added correct_icebridge_l3_dem for IceBridge.

  • Added fetch_icebridge_data for IceBridge.

  • By default, use as many processes as there are cores, and one thread per processes.

  • Large speedup in epipolar alignment.

  • Improved epipolar alignment quality with standard pinhole cameras.

  • Added the options --ip-inlier-threshold and --ip-uniqueness-threshold for finer-grained control over interest point generation.

  • Fix a bug with interest point matching the camera model is RPC and the RPC approximation domain does not intersect the datum.

  • Added new option --stereo-algorithm. Choices 1 and 2 replaces the standard integer correlator with a new semi-global matching (SGM) correlator or an MGM correlator respectively. SGM/MGM is slow and memory intensive but it can produce better results for some challenging input images, especially for IceBridge. See the manual for more details.

  • Added the option --min-triangulation-angle to not triangulate when rays have an angle less than this.

  • Zooming in one image can trigger all other side-by-side images to zoom to same region.

  • Clicking on a pixel prints image pixel indices, value, and image name. Selecting a region with Control+Mouse prints its bounds in pixels, and, if georeferenced, in projected and degree units.

  • Added a 1D profile tool for DEMs.

  • Can visualize the pixel locations for a GCP file (by interpreting them as interest points).

  • Can save a screenshot of the current view.

  • If all images are in the same window, can show a given image above or below all others. Also can zoom to bring any image in full view (from the list of images on the left).

  • Options to set the azimuth and elevation when showing hillshaded images.

  • Added the option --dem-blur-sigma to blur the output DEM.

  • Use by default --weights-exponent 2 to improve the blending, and increase this to 3 if --priority-blending-length is specified.

  • Added the options --tile-list, --block-max, and --nodata-threshold.

  • Display the number of valid pixels written.

  • Do not write empty tiles.

  • One of the two input files can be in CSV format.

  • Save on output the mean values for MEANSUNEL, MEANSUNAZ, and a few more.

  • Added the parameter --gaussian-sigma-factor to control the Gaussian kernel width when creating a DEM (to be used together with --search-radius-factor).

  • Improvements, speedups, bug fixes, more documentation, usage recipes, much decreased memory usage, together with a lot of testing and validation for the Moon.

  • Can run on multiple input DEM clips (which can be chosen as representative for the desired large input DEM region and images) to solve for adjusted camera positions throughout this region.

  • Added parallel_sfs, to run sfs as multiple processes over multiple machines.

  • Can optimize the intrinsic parameters for pinhole cameras. The focal length, optical center, and distortion parameters can be fixed or varied independently of each other. To be used with --local-pinhole, --solve-intrinsics, --intrinsics-to-float.

  • Added the option --overlap-list. It can be used to specify which image pairs are expected to overlap and hence to be used to compute matches.

  • Added the option --initial-transform to initialize the adjustments based on a 4x4 rotation + translation transform, such as coming from pc_align.

  • Added the options --ip-inlier-threshold and --ip-uniqueness-threshold for finer-grained control over interest point generation.

  • Can solve for a rotation + translation or for rotation + translation + scale using least squares instead of ICP, if the first cloud is a DEM. It is suggested that the input clouds be very close or otherwise the --initial-transform option be used, for the method to converge. The option is: --alignment-method [ least-squares | similarity-least-squares ]

  • Built with ISIS 3.5.0.

  • Minimum supported OS versions are OSX 10.11, RHEL 6, SUSE 12, and Ubuntu 14.

  • Ship with GDAL’s gdalwarp and gdaldem.

  • Added integration with Zenodo so that this and all future ASP

    releases will have a DOI. More info in the asp_book.pdf

22.10. RELEASE 2.5.3, August 24, 2016


  • Added the ability to process ASTER L1A VNIR images via the tool aster2asp that creates image files and both RPC and rigorous linescan camera models that can then be passed to stereo. The RPC model seems to work just as well as the rigorous one and is much faster.

  • Added the ability to process SPOT5 images with stereo, bundle_adjust, and mapproject using a rigorous linescan camera model.

  • Added the add_spot_rpc tool to create RPC models for SPOT5 which allows them to be mapprojected with the RPC model.

  • Can solve for a scale change in addition to a rotation and translation to best align two clouds, hence for a similarity transform, using option: --alignment-method similarity-point-to-point.

  • Added ability to mapproject color images.

  • Added option to mapproject on to a flat datum.

  • Added option to accept multiple input camera models.


  • Fix a bug with mosaicking of DEMs over very large extent.

  • Fix a bug with 360 degree longitude offset.

  • Added the option --use-centerline-weights. It will compute blending weights based on a DEM centerline algorithm. Produces smoother weights if the input DEMs don’t have holes or complicated boundary.

  • Added a new colormap scheme, ‘cubehelix’, that works better for most color-blind people.

  • Use transparent pixels for displaying no-data values instead of black.

  • Can delete or hillshade individual images when overlayed.

  • Add control to hide/show all images when in overlay mode.

  • Make ASP handle gracefully georeferenced images with some pixels having projected coordinates outside of the range expected by PROJ.4.

  • Removed the deprecated orthoproject tool. Now mapproject should be used.

  • Fixed a bug in pc_align which caused the --max-displacement argument to be misread in some situations.

  • Removed some extraneous code slowing down the datum_convert tool.

  • Fixed a bug in point2dem handling the Albers Conic Equal Area projection.

  • Added standard thread/bigtiff/LZW options to image_calc.

22.11. RELEASE 2.5.2, Feb 29, 2016


Added a constellation of features and tools to support solving for the positions of input images lacking position information. Can be used for aerial imagery with inaccurate or incomplete pose information, images from low cost drones, historical images lacking metadata, and images taken with handheld cameras.

  • New tool which adds support for aerial imagery etc as described above.

  • Uses the THEIA library (http://www.theia-sfm.org/index.html) to compute camera positions and orientations where no metadata is available.

  • Ground control points and estimated camera positions can be used to find absolute camera positions.

  • Added section to documentation describing ways to use ASP to process imagery from NASA’s IceBridge program.

  • A convenience camera calibration tool that is a wrapper around the OpenCV checkerboard calibration program with outputs in formats for camera_solve and ASP.

  • Added several options to support solving for pinhole camera models in local coordinates using GCPs or estimated camera positions.

  • Improved filtering options for which images are IP-matched.

  • Significantly improved the accuracy of the plotted camera locations.

  • Added option to load results from camera_solve.

  • Now corrects TDI 8 (Reverse) of WV01 and TDI 8 (Forward and Reverse) and TDI 32 (Forward) of WV02. Other correction behavior is unchanged.

  • Added the ability to filter large disparities from D_sub that can greatly slow down a run. The options are --rm-quantile-percentile and --rm-quantile-multiple.

  • A new tool to test out pinhole model lens distortion parameters.

Lens distortion models:
  • Switched from binary .pinhole file format to updated version of the old plain text .tsai file format.

  • Added support for Photometrix camera calibration parameters.

  • New appendix to the documentation describing the .tsai file format and supported lens distortion models.


  • Suppressed pesky aux.xml warning sometimes printed by GDAL.

  • Removed the long-deprecated orthoproject tool.

  • Added icebridge_kmz_to_csv and lvis2kml utilities.

  • Write correct bounding box in the header.

  • Respect projections that are not lon-lat.

  • Increased speed of erode option.

  • Mention DERT, a tool for exploring large DEMs.

  • Added new section describing camera_solve tool in detail.

22.12. RELEASE 2.5.1, November 13, 2015


  • Added jitter correction for Digital Globe linescan imagery.

  • Bug fix for stereo with map-projected images using the RPC session (e.g, for map-projected Pleiades imagery).

  • Added OpenCV-based SIFT and ORB interest point finding options.

  • Much improved convergence for Digital Globe cameras.

  • Added OpenCV-based SIFT and ORB interest point finding options.

point2dem, point2las, and pc_align
  • The datum (-r <planet> or --semi-major-axis) is optional now. The planet will be inferred automatically (together with the projection) from the input images if present. This can be useful for bodies that are not Moon, Mars, or Earth. The datum and projection can still be overridden with --reference-spheroid (or --datum) and --t_srs.

  • Introduce --priority-blending-length, measured in input pixels. If positive, keep unmodified values from the earliest available DEM at the current location except a band this wide near its boundary where blending will happen. Meant to be used with smaller high-resolution “foreground” DEMs and larger lower-resolution “background” DEMs that should be specified later in the list. Changing --weights-exponent can improve transition.

  • Added the ability to compute a manual rotation + translation + scale transform based on user-selected point correspondences from reference to source cloud in stereo_gui.

  • Added the ability to generate ground control point (GCP) files for bundle_adjust by picking features. In addition to the images to be bundle-adjusted, one should provide a georeferenced image to find the GCP lon-lat, and a reference DEM to find the GCP heights.


  • If the input images are map-projected (georeferenced) and alignment method is none, all image outputs of stereo are georeferenced as well, such as GoodPixelMap, D_sub, disparity, etc. As such, all these data can be overlayed in stereo_gui.

  • The output point cloud saves datum info from input images (even when the inputs are not georeferenced).

  • Increased reliability of interest point detection.

  • Decreased the default timeout to 900 seconds. This still needs tuning and a permanent solution is necessary.

point2dem, point2las, and pc_align
  • Accept --datum (-r) MOLA, as a shortcut for the sphere with

    radius 3,396,000 meters.

  • Fix an issue with minor jumps across tiles.

  • Introduce --save-dem-weight <index>. Saves the weight image that tracks how much the input DEM with given index contributed to the output mosaic at each pixel (smallest index is 0).

  • Introduce --save-index-map. For each output pixel, save the index of the input DEM it came from (applicable only for --first, --last, --min, and --max). A text file with the index assigned to each input DEM is saved as well.

  • Rename --blending-length to --extra-crop-length, for clarity.

  • Added the switch --fix-seams to use interest point matching to fix seams in the output mosaic due to inconsistencies between image and camera data. Such artifacts may show up in older (2009 or earlier) Digital Globe images.

  • Added the option --match-file to view interest point matches.

  • Added the options --delete-temporary-files-on-exit and --create-image-pyramids-only.

  • Can read the georeference of map-projected ISIS cubes.

  • Respect --t_projwin to the letter.

  • Can create simultaneously DEMs at multiple resolutions (by passing multiple values in quotes to --dem-spacing).

  • Fix minor discrepancies in the minor semi-axis for the WGS84, NAD83 and WGS72 datums. Now using GDAL/OGR for that.

  • Save the LAS file with a datum if the input PC had one.

  • Fix calculation bug when no-data is present.

  • Upgraded to the latest libpointmatcher. This may result in minor alignment changes as the core algorithm got modified.

  • Save all PC clouds with datum and projection info, if present. Add comment lines with the datum and projection to CSV files.

  • Bug fix when the two DEMs have longitudes offset by 360 degrees.

  • Default style is binary-red-blue. Works better than jet when data goes out of range.

  • Can merge clouds with 1 band. That is, can merge not only PC.tif files but also L.tif files, with the goal of using these two merged datasets to create a merged orthoimage with point2dem.

  • Can create a mesh from a DEM and an orthoimage (DRG file).

22.13. RELEASE 2.5.0, August 31, 2015

Improved speed, coverage, and accuracy for areas with steep slopes for ISIS, RPC and Pinhole cameras by implementing stereo using images map-projected onto an existing DEM. This mapprojection is multi-process and hence much faster than cam2map. This functionality was previously available only for Digital Globe images.

New tools:
  • Added stereo_gui, an image viewer and GUI front-end. Features:

    • View extremely large images using a pyramid approach.

    • If invoked with the same interface as stereo, can run stereo on selected clips.

    • Load images with int, float, and RGB pixels, including ISIS cubes, DEMs, NTF, TIF, and other formats.

    • Can overlay georeferenced images and can toggle individual images on and off (like Google Earth).

    • Show images side-by-side, as tiles on grid, or on top of each other.

    • Create and view hillshaded DEMs.

    • View/add/delete interest points.

    • Create shadow thresholds by clicking on shadow pixels (needed for sfs).

    • Based on Michael Broxton’s vwv tool.

  • Added sfs, a tool to refine DEMs using shape-from-shading. Can optimize the DEM, albedo per pixel, image exposures and camera positions and orientations using a multi-resolution pyramid approach. Can handle shadows. Tested with LRO NAC lunar images at low latitudes and toward poles. It works only with ISIS images.

  • Added image_calc, a tool for performing simple per-pixel arithmetic operations on one or more images.

  • Added pc_merge, a tool for concatenating ASP-produced point clouds.

  • Added pansharp, a tool to apply a pansharp algorithm to a matched grayscale image and a low resolution color image.

  • Added datum_convert, a tool to transform a DEM to a different datum (e.g., NAD27 to WGS84).

  • Added geodiff, a tool for taking the (absolute) difference of two DEMs.

  • Documented the colormap tool. Added a new colormap option based on the paper “Diverging Color Maps for Scientific Visualization” (http://www.sandia.gov/~kmorel/documents/ColorMaps/).

  • Added gdalinfo, gdal_translate, and gdalbuildvrt to the bin directory. These executables are compiled with JPEG2000 and BigTIFF support, and can handle NTF images.

  • Added a documentation section on ‘tips and tricks’, summarizing in one place practices for getting the most out of ASP.

  • Increase the default correlation timeout to 1800 seconds.

  • Fix failure in interest point matching in certain circumstances.

  • Use bundle-adjusted models (if provided) at all stages of stereo, not just at triangulation.

  • Added --right-image-crop-win in addition to --left-image-crop-win. If both are specified, stereo crops both images to desired regions before running stereo (this is different from when only --left-image-crop-win is specified, as then no actual cropping happens, the domain of computation is just restricted to the desired area).

  • Bug fix, remove outliers during search range determination.

  • Added the option --ip-per-tile, to search for more interest points if the default is insufficient.

  • If the input images are georeferenced, the good pixel map will be written with a georeference.

  • Fixed a slight discrepancy in the value of the semi-minor axis in the WGS84 and NAD83 datum implementations.

  • Added the option --median-filter-params <window size> <threshold> to remove spikes using a median filter.

  • Added the option --erode-length <num> to erode pixels from point cloud boundary (after outliers are removed, but before filling in holes).

  • Improved hole-filling, and removed the --hole-fill-mode and --hole-fill-num-smooth-iter, as there’s only one algorithm now.

  • Improved performance when large holes are to be filled.

  • Can create a DEM from point clouds stored in CSV files containing easting, northing, and height above datum (the PROJ.4 string needed to interpret these numbers should be set with --csv-proj4).

  • Fixed a bug in creating DEMs from CSV files when different projections are used on input and output.

  • Expose to user gnomonic and oblique stereographic projections, as well as false easting and false northing (where applicable). This is a shortcut from using explicitly --t_srs for the PROJ.4 string.

  • The default no-data value is set to the smallest float.

  • Can ingest CSV files containing easting, northing, and height above datum (the PROJ.4 string needed to interpret these numbers should be set with --csv-proj4).

  • If the reference point cloud is a DEM, the initial and final errors in the statistics, as well as gross outlier removal, are done using a new distance function. Instead of finding the distance from a 3D point to the closest point in the cloud, the 3D point is projected onto DEM’s datum, its longitude and latitude are found, the height in the DEM is interpolated, and and the obtained point on the DEM is declared to be the closest point. This is more accurate than the original implementation for coarse DEMs. The old approach is available using the --no-dem-distances flag.

  • Fix a bug with a 360 degree longitude offset.

  • Added the ability to specify a custom projection (PROJ.4 string) for output LAS files.

  • Write GeoTIFF files with blocks of size 256 x 256 as those may be faster to process with GDAL tools.

  • Bug fix when the tool is used to re-project.

  • Added the option --weights-blur-sigma <num> to allow the blending weights to be blurred by a Gaussian to increase their smoothness.

  • Added the option --weight-exponent <num>, to allow weights to increase faster than linearly.

  • Added --stddev option to compute standard deviation.

  • Added the ability to fill holes in the output mosaic.

  • Added new parameters, --ip-per-tile and --min-triangulation-angle.

  • Bug fix in handling situations when a point cannot get projected into the camera.

  • Bug fix in the camera adjustment logic. Any .adjust files may need to be regenerated.

  • Bug fixes.

  • Create temporary files in current directory, to avoid access issues to system directories.

  • Can run on multiple machines.

  • Use multiple processes for ISIS images, for a huge speedup.

  • Bug fix, the mapprojected image should not go much beyond the DEM it is mapprojected onto (where it would have no valid pixels).

  • Default penalty weight produces a more accurate fit when creating an RPC model from a DG model.

  • Handle the situation when two images to be mosaicked start at the same output row number.

  • Added --target-resolution option to specify the output resolution in meters.

  • Upgraded to ISIS 3.4.10.

  • Oldest supported OSX version is 10.8.

  • Added documentation for image2qtree and hillshade.

22.14. RELEASE 2.4.2, October 6, 2014

ASP can perform multi-view triangulation (using both the stereo and parallel_stereo tools). The first image is set as reference, disparities are computed from it to the other ones, and joint triangulation is performed.

Added a new tool, dem_mosaic, for mosaicking a large number of DEMs, with erosion at boundary, smooth blending, and tiled output. Instead of blending, the tool can do the first, last, min, max, mean, median, or count of encountered DEM values.

  • Support for multi-band (multi-spectral) images. Use --band <num> to pick a band to mosaic.

  • Bug fix in interest point matching in certain circumstances.

  • Set the correlation timeout to 600 seconds. This is generous and ensures runs don’t stall.

  • Take as input n clouds and optionally n texture files, create a single DEM/orthoimage.

  • Take as input LAS and CSV files in addition to ASP’s PC format.

  • Fix a bug in the interplay of hole-filling and outlier removal for orthoimage creation.

  • Ensure that the DEM grid is always at integer multiples of the grid size. This way, two DEMs with overlapping grids of the same size will be exactly on top of each other, minimizing interpolation error in subsequent mosaicking.

  • Outlier removal is on by default. Can be disabled by setting the percentage in --remove-outliers-params to 100.

  • Use multiple-threads for non-ISIS sessions.

  • Added the parameter --overlap-limit <num> to limit the number of subsequent images to search for matches to the current image.

  • Added the parameter --camera-weight <val>, to set the weight to give to the constraint that the camera positions/orientations stay close to the original values (only for the Ceres solver).

  • Support the EGM2008 geoid. The geoid surface across all Earth is computed with an error of less than 1.5 cm compared to the values generated by harmonic synthesis. A 2.5 x 2.5 minute grid is used.

  • Converted the EGM geoids shipped with ASP to INT16 and JPEG2000, resulting in size reduction of more than 10x.

  • Corrects TDI of 16, 48, 56, and 64 (forward and reverse scan directions) for WV01, TDI of 8 (forward only) for WV01, and TDI of 16, 48, 64 (forward and reverse scan directions) for WV02. Returns uncorrected images in other cases.

  • Fix a crash for very large clouds.

  • Use a progress bar when loading data.

  • Support LAS files on input and output.

  • Bug fix when saving LAS files in respect to a datum.

  • Move the non-ISIS-specific tutorial sections onto its own chapter, to be read by both ISIS and Earth users. Updates and cleanup.

22.15. RELEASE 2.4.1, 12 July, 2014

Added a new tool, bundle_adjust, which uses Google’s ceres-solver to solve for adjusted camera positions and orientations. Works for n images and cameras, for all camera types supported by ASP.

  • Improved corrections for WV01 images of TDI 16.

  • Performance bugfix when the integer disparity is noisy.

  • Fix for large memory usage when removing small islands from disparity with --erode-max-size.

  • Bug fixes for MER cameras.

stereo_tri and mapproject
  • Added the option --bundle-adjust-prefix to read adjusted camera models obtained by previously running bundle_adjust with this output prefix.

  • LAS files can be saved in geo-referenced format in respect to a specified datum (option --reference-spheroid).

  • Bug fix, longitude could be off by 360 degrees.

  • Robustness to large jumps in point cloud values.

  • Ability to read and write CSV files having UTM data (easting, northing, height above datum).

  • Read DEMs in the ISIS cube format.

22.16. RELEASE 2.4.0, 28 April, 2014

Added wv_correct, a tool for correcting artifacts in Digital Globe WorldView-1 and WorldView-2 images with TDI of 16.

Added logging to a file for stereo, pc_align, point2dem, point2mesh, point2las, and dem_geoid.

Added a tutorial for processing Digital Globe Earth imagery and expanded the MOC tutorial.

Bug fixes in mosaicking of Digital Globe images.

  • Use dynamic load balancing for improved performance.

  • Automatically determine the optimal number of processes and threads for each stage of stereo.

  • Added the --skip-image-normalization option (for non-ISIS images and alignment-method none), it can help with reducing the size of data on disk and performance.

  • Added new affine subpixel refinement mode, --subpixel-mode 3. This mode sacrifices the error resistance of Bayes EM mode in exchange for reduced computation time. For some data sets this can perform as well as Bayes EM in about one fifth the time.

  • Hole-filling is disabled by default in stereo_fltr. It is suggested to use instead point2dem’s analogous functionality. It can be re-enabled using --enable-fill-holes.

  • Added the option --erode-max-size to remove isolated blobs.

  • Relaxed filtering of disparities, retaining more valid disparities. Can be adjusted with --filter-mode and related parameters.

  • Added ability to save triangulation error for a DEM as a 3D North-East-Down vector rather than just its magnitude.

  • When acting on map-projected images, handle the case when the DEM used for map-projection does not completely encompass the images.

  • Read and write CSV files in a wide variety of formats, using the --csv-format option.

  • Display the translation component of the rigid alignment transform in the local North-East-Down coordinate system, as well as the centroid of source points used in alignment.

  • Save to disk the convergence history (iteration information).

  • Added the ability to explicitly specify the datum semi-axes.

  • Bug fix for saving transformed clouds for Moon and Mars.

  • More efficient processing of reference and source points by loading only points in each cloud within a neighborhood of the long/lat bounding box of the other cloud.

  • Make it possible to generate ortho and error images using point2dem with the transformed clouds output by pc_align.

  • Replaced the core algorithm. Instead of sampling the point cloud surface, which is prone to aliasing, the DEM height at a given grid point is obtained as a weighted average of heights of all points in the cloud within search radius of the grid point, with the weights given by a Gaussian. The cutoff of the Gaussian can be controlled using the --search-radius-factor option. The old algorithm is still available (but obsoleted) using the --use-surface-sampling option. The new algorithm makes the --fsaa option redundant.

  • Added the ability to remove outliers by triangulation error, either automatically (–remove-outliers) or manually, with given error threshold (–max-valid-triangulation-error).

  • Added two algorithms to fill holes in the output DEM and orthoimage (–hole-fill-mode).

  • The way the default DEM spacing is computed was modified, to make dependent only on the local distribution of points in the cloud and robust to outliers.

  • Can handle highly noisy input point clouds without spikes in memory usage and processing time.

  • Improved memory usage and performance for large point clouds.

  • Bug fix, the DEM was shifted by 1 pixel from true location.

22.17. RELEASE 2.3.0, 19 November, 2013


  • Added pc_align, a tool for aligning point clouds, using the libpointmacher library (https://github.com/ethz-asl/libpointmatcher). Sparse and dense point clouds are supported, as well as DEMs. Two ICP methods are supported, point-to-plane and point-to-point. Memory and processing usage are proportional to the desired number of input points to use rather than to the overall input data sizes.

  • Added lronac2mosaic.py, a tool for merging the LE and RE images from the LRONAC camera into a single map-projected image. The output images can be fed into the stereo tool to generate DEMs.

  • rpc_maprpoject and orthoproject are combined into a single tool for projecting a camera image onto a DEM for any camera model supported by Stereo Pipeline. The old orthoproject is kept for backward compatibility for a while.


  • Stereo Pipeline (almost) daily and fully verified builds for all platforms are available for the adventurous user (http://byss.arc.nasa.gov/stereopipeline/daily_build/, which was later moved to https://github.com/NeoGeographyToolkit/StereoPipeline/releases). When requesting support, please provide the output of stereo --version.

  • The size of Stereo Pipeline output data has been reduced, by up to 40%, particularly point clouds and DEMs are between 30% to 70% smaller. Better encoding is used, output data is rounded (up to 1 mm), and point clouds are offset and saved as float instead of double.

  • Timeout option added for stereo correlation, preventing unreasonably long correlation times for certain image tiles.

  • Subpixel mosaicking in dg_mosaic uses bilinear interpolation instead of nearest neighbor avoiding artifacts in certain situations.

  • dg_mosaic can generate a combined RPC model in addition to the combined DG model. It accepts flags for specifying input and output nodata values.

  • point2dem with the --fsaa option for reducing aliasing at low-resolution DEM generation has been improved as to remove the erosion of of valid data close to no-data values.

  • Bug fixes for parallel_stereo, point2dem, etc.

22.18. RELEASE 2.2.2, 17 MAY 2013

(incremented from 2.2.1 after one more bugfix)


  • stereo_mpi renamed to parallel_stereo and made to work on any machines with shared storage, rather than just on supercomputers using Intel’s MPI library. Bug fixes for homography and affine epipolar alignment modes, etc.

  • Bug fix for dem_geoid path to geoids, more robust datum identification.

22.19. RELEASE 2.2.0, 6 MAY 2013


  • ISIS headers removed from IsisIO’s headers.

  • Removed unneeded mutex inside inpaint algorithm.

  • Interest point matching and description are parallel now.

  • Stereo pprc uses separable convolution for anti-aliasing.

  • IsisIO made compliant with ISIS 3.4.3’s API.

  • Blob consolidation (for inpainting) is now parallel.

  • Yamaha RMAX code dropped.


  • RPC mode can now read Astrium data.

  • DG added additional safety checks for XML values.

  • DG, ISIS, and RPC now have affineepipolar alignment option.

  • All sessions had their API changed. We now use Transform objects instead of LUTs to reverse mapprojections and alignments.


  • Added dem_geoid, dg_mosaic, and stereo_mpi.

  • Added new interest point matching method to stereo.

  • Added new DEM seed mode for stereo.

  • Point2dem sped up by reducing over rasterization of triangles.

  • Added the --use-local-homography option to stereo_corr. Homography transform is applied per tile.

  • Fix point2dem where for certain projections we were setting K=0.

  • Stereo can now operate using command-line arguments only, without stereo.default.

22.20. RELEASE 2.1.0, 8 JANUARY 2013


  • Added documentation for processing GeoEye, Digital Globe, and Dawn FC data.

  • Fixed implementation of internal RANSAC function.

  • DEMError has been renamed IntersectionErr. 3D IntersectionErr is now recordable in local North East Down format.


  • Added RPC processing session.

  • DG sessions now use bicubic interpolation for mapprojection arithmetic.

  • Fixed bug in case where DG XML file had single TLC entry.

  • DG sessions now applies velocity aberration corrections.


  • Have point2dem use correct nodata value when writing DRGs.

  • Fix segfault issue in point2dem due to triangle clipping.

  • Hiedr2mosaic python script now supports missing CCD files and start/stop resume on noproj step for bundle adjustment.

  • Max pyramid level used for stereo correlation is configurable with corr-max-levels option.

  • Stereo accepts left-image-crop-win option for processing of specific image coordinates.

  • Stereo_pprc accepts nodata-threshold and nodata-percentage options for masking (possibly shadows).

  • Stereo command should now correctly call secondary executables so that their dependencies are loaded.

22.21. RELEASE 2.0.0, 20 JUNE 2012


  • Modified ASP according to API changes in ISIS 3.4.0.

  • Added new interest point matching code. Provides better initial guess for search range.

  • Complete changed stereo.default format. See stereo.default.example for an example.

  • Complete rewrote integer correlator for improved speed and less memory use.

  • Relicense code to be Apache 2 licensed instead of NOSA.


  • Add normalization options to PINHOLE session.

  • Added Digital Globe (DG) session. This supports the linearized linescan camera model that is described in the supporting XML file.

  • Deleted KEYPOINT session. PINHOLE essentially does all of that.


  • Added DEMError output example for MOC.

  • Added jigsaw example for MOC.

  • Added HiRISE example dataset.


  • Dropped release of isis_adjust and bundlevis.

  • Fix int32 overflow in arithmetic for subsampling in preprocessing.

  • Remove Python 2.4 incompatible call in cam2map4stereo.py.

  • Speed up point2dem texture access by remove unnecessary mutex.

  • Add earth mode and fix non spherical support in point2dem.

  • Added lronac4staged.py.

  • Implemented D_sub or seeded integer correlation in stereo_corr.

  • Fourth channel of output PC file is now triangulation error.

  • Added --t_srs option to point2dem.

  • Added rpc_mapproject tool. This provides an optional mapprojection step that can be used for DG session.

  • Allow IAU2000:* projection options to be used by point2dem.

  • No-Data is now colored black in GoodPixelMap.

  • Make noproj step in hiedr2mosaic parallel.

22.22. RELEASE 1.0.5, 27 OCT 2011

Fixed ASP to work with ISIS 3.3.0’s new API changes and library dependencies.

Enabled parallel writing in Pinhole Session.


  • Fix possible infinite loop in stereo_corr’s search range.

  • Shutoff rotation invariance in automatic search range for better quality results. This is possible because the input images are already aligned.

  • Sub image produced by stereo_pprc are now limited to around 8MB.

  • Fix disparity_debug to work with integer disparities as well.

  • All ASP tools should now have a ‘–version’ option.

  • Bug fix point2dem where rasterizer was accessing outside of allocated memory.

  • Speed up mask generation in stereo_pprc by avoiding mutex.

  • Speed up hole filling in stereo_fltr by avoiding mutex.

22.23. RELEASE 1.0.4, 23 MAY 2011

Added support for CAHVORE in pinhole sessions.


  • Hide GDAL warnings caused by our file integrity checks.

  • Mostly added standardized options for settings threads and BigTIFF.

  • Have orthoproject return same type as input plus alpha channel.

  • Improved edge_masking, speeds up stereo_fltr and stereo_pprc.

  • Have cam2map4stereo.py explicitly use ISIS’s getkey command.

  • Fix and optimized point2dem. Remove caching and improved rendering times. This should fix BigTIFF problems that have been reported.

  • Improve triangulation times slightly when using mapprojected linescan cameras.


  • Added orthoproject, image2qtree, colormap, hillshade examples to MOC.

  • Added K10 example dataset.

  • Added MER example dataset.

  • Added a non-mapprojected MOC example.

  • Added CTX example dataset.


  • Append notes from Michael about run times.


  • Added threaded writing to colormap and hillshade.

  • Fix hillshade problems with int16 DEMs.

22.24. RELEASE, 16 MARCH 2011

Updated documentation and support text files to insure compatibility with our third party software.

22.25. RELEASE 1.0.3, 11 MARCH 2011


Make quaternion interaction compliant with VW changes.


Correct reading of TSAI camera format.


  • Reduce memory footprint of ISIS_Adjust.

  • MOC Example rewritten.

  • Improve dash script that loads libraries on startup of application.


  • KD-Tree search replace with FLANN, a fast approximate nearest neighbors. This improves speed of ipmatch, and ip alignment option in stereo.

  • Removed exception catch in Bayesian affine sub-pixel.

  • Fixed type deduction problem on 32 bit systems.

  • Pyramid Correlator code cleaned up. Minimal speed improvement.

  • Fixed Camera Relation Network’s memory leak.

  • Fix image2qtree normalization and manual geo-positioning.

  • Correct random seed call with faster solution.

  • Default raster tile size changed to 256.

  • Fix deadlocking in loading of “.vwrc”, Vision Workbench’s settings file.


OSX seems to do excessive locking during multi-threaded rendering. This problem is non-existent in RHEL5 and is still a mystery.

22.26. RELEASE 1.0.2, 9 DECEMBER 2010


  • IsisCameraModel support operator<< style printing.

  • Correct camera pose return to be consistent with VW.

  • Change IsisCameraModel to use shared_ptr to block memory leak.


  • Executables should catch VW and Standard errors and print human readable responses.

  • Stereo is now a python script that call multiple executables.

  • Change correlation progress bar to track total completion.

  • Bundle_Adjust and ISIS_Adjust switch from Euler’s to quaternions.

  • Bundlevis dropped CAHVOR support. Added progress bar. Converted statistics with CDFAccumulator.

  • Point2dem remove excessive rotation call

  • Enforce tile rasterization size to 1024 during integer correlation.

  • Select tools should now write their nodata value in the TIFF metadata.


Still unreleased, and still under development.

22.27. RELEASE 1.0.1, 24 MAY 2010


  • Control Network Loader removed and sent to VW’s Bundle Adjustment Module.

  • Build system can now use Google PerfTools.

  • Kakadu was made optional in build system (ISIS 3.2.x uses this).


  • Optimized IsisCameraModel to use IsisInterface. Custom code can be loaded up for individual camera types so we don’t have to run through ISIS’s entire camera model. This allows us not to call GroundMap when the camera is not mapprojected.

  • Added a series of tests for the IsisCameraModel that perform unit tests with MOC and Galileo.

  • Added custom project code for Linescan cameras so not to rely on ISIS’s LineScanCameraGroundMap. This code is a bit more precise.


Added new optional module called MPI that builds on top of Boost MPI. This is experimental development code and is not used for anything in binary release yet.


Added new optional module call the Photometry Toolkit. This is experimental development code and is not use for anything released in the binary yet. This code focuses on future research of massive mosaics (+100GB) and the ability to perform basic photometric corrections.


Pinhole session modified to read CMOD files as well.


  • Made orthoproject more robust against odd input georeferences.

  • orthoproject’s auto scale and crop works again.

  • Point2mesh’s texture is written to a different file.

  • Added aligndem and geodiff, experimental DEM alignment utilities.

  • Added a quick experimental DEM profile utility called dem_profile.

  • stereo now detects correlation settings automatically using OBALoG and SGrad1 interest point functions.

  • Added cam2map4stereo.py

  • Remove excessive serial number calculations in isis_adjust.

  • Update isis_adjust to VW’s new Bundle Adjustment module for a 2x improvement.

  • Stereo should now use LZW compression by default.

  • Point2dem and Stereo have added option to use directory other than /tmp for intermediate files.

  • Point2dem now uses MOLA datum instead of its previous truncated value.

  • Added safety check to stereo to make sure user is not supplying the same camera.

  • Added point2las, a utility for converting a point cloud to the LAS format.


Switched from CXXTests to GTest framework.

22.28. RELEASE 1.0.0, 23 OCTOBER, 2009


  • OrthoRasterizer.h is subject to change for further VW integration

  • MedianFilter.h is untested/unused

  • BundleAdjustUtils.* is subject to deletion for integration with ControlNetworkLoader.*


  • ISIS Session is the only fully supported session at this time

  • Pinhole Session works but has not been tested for this release

  • Keypoint/RMAX Session status are unknown


Subject to deletion in 1.0.1


  • Point2dem can crash rarely. Still investigating.

  • rmax* utilities are not working