7. Tips and tricks¶
Here we summarize, in one place, some insights in how to get the most from ASP, particularly the highest quality results in the smallest amount of time.
Ask for help or if you have questions. We’re always glad to share what we know, implement suggestions, and fix issues (Section 1.4).
Use the GUI (Section 16.64) to get comfortable with ASP on a small region and to tune parameters (Section 16.64). A solution specific to ISIS images is to crop your stereo pair (using the ISIS
cropcommand being careful to retain SPICE information, or
cam2map4stereo.pyfor map-projected images) to a small region of interest.
The highest quality results with ASP can be obtained with map-projected images (Section 6.1.7).
Run stereo on multiple machines (Section 16.47).
Improve the quality of the inputs to get better outputs. Bundle-adjustment can be used to find out the camera positions more accurately (Section 12.2). CCD artifact correction can be used to remove artifacts from WorldView images (Section 5.5).
Align the output point cloud to some known absolute reference with
Remove noise from the output point cloud. During stereo triangulation, points that are further or closer than given distances from planet center or left camera center can be removed as outliers (Section 17.5). During DEM generation (Section 16.52), points with large triangulation error can be removed using
--remove-outliers-params. Spikes can be removed with
--median-filter-params. Points close to the boundary, that tend to be less accurate, can be eroded (
During stereo filtering, islands can be removed with
Remove noise from the low-resolution disparity (D_sub) that can greatly slow down a run using
--rm-quantile-multiple. Some care is needed with these to not remove too much information.
Fill holes in output orthoimages for nicer display (also in DEMs), during DEM and orthoimage generation with
point2dem(Section 16.52). Holes in an existing DEM can also be filled using
To get good results if the images lack large-scale features (such as for ice plains) use a different way to get the low-resolution disparity (Section 5.6).
If a run takes unreasonably long, decreasing the timeout parameter may be in order (Section 6.1.10).
Manually set the search range if the automated approach fails (Section 11.2.2).
To increase speed, the image pair can be subsampled. For ISIS images, the ISIS
reducecommand can be used, while for DigitalGlobe/Maxar data one can invoke the
dg_mosaictool (Section 16.20, though note that this tool may introduce aliasing). With subsampling, you are trading resolution for speed, so this probably only makes sense for debugging or “previewing” 3D terrain. That said, subsampling will tend to increase the signal to noise ratio, so it may also be helpful for obtaining 3D terrain out of noisy, low quality images.
Photometric calibration (using the ISIS tools) can be used to improve the input images and hence get higher quality stereo results.
If your images have missing or inaccurate camera pose information, and they were acquired with frame (pinhole cameras), such data can be solved for using structure-from-motion and bundle adjustment (Section 10).
Shape-from-shading (Section 16.61) can be used to further increase the level of detail of a DEM obtained from stereo, though this is a computationally expensive process and its results are not easy to validate.
We’ll be happy to add here more suggestions from community’s accumulated wisdom on using ASP.