Gaps in DEM Generation using ASP with SPOT 5

[Daryl Ernesto Ayala Saavedra]

Dear ASP Support Community,

I am currently working on a project to generate a Digital Elevation Model (DEM) for glacier mass balance studies in the Peruvian Andes. I am using the Ames Stereo Pipeline (ASP) with SPOT5 stereo imagery.

I have encountered a significant issue with the generated DEM: it contains numerous data gaps, particularly in areas corresponding to glaciers. These gaps are characterized by pixel values of 255 in the source SPOT 5 images (already map-projected), which is the maximum value for these images. However, I am also observing gaps in non-glaciated areas, suggesting that the issue is not solely related to the presence of ice/snow.

I have attached an image for reference, which shows the "front" and "back" SPOT 5 map-projected images alongside the generated DEM.

I would greatly appreciate any guidance on how to mitigate these gaps and improve the quality of my DEM. I am particularly interested in understanding if there are specific parameters or processing steps within ASP that I should adjust to address this issue.

Here is the code I am using for the ASP processing:

bundle_adjust -t spot5 FRONT/SCENE01/IMAGERY_FRONT.TIF BACK/SCENE01/IMAGERY_BACK.TIF FRONT/SCENE01/METADATA_FRONT.DIM BACK/SCENE01/METADATA_BACK.DIM -o ba_run/out

add_spot_rpc FRONT/SCENE01/METADATA_FRONT.DIM -o FRONT/SCENE01/METADATA_FRONT.DIM

add_spot_rpc BACK/SCENE01/METADATA_BACK.DIM -o BACK/SCENE01/METADATA_BACK.DIM

mapproject DEM/Dem_ref.tif FRONT/SCENE01/IMAGERY_FRONT.TIF FRONT/SCENE01/METADATA_FRONT.DIM FRONT_map_proj.tif -t rpc

mapproject DEM/Dem_ref.tif BACK/SCENE01/IMAGERY_BACK.TIF BACK/SCENE01/METADATA_BACK.DIM BACK_map_proj.tif -t rpc

parallel_stereo -t spot5maprpc FRONT_map_proj.tif BACK_map_proj.tif FRONT/SCENE01/METADATA_FRONT.DIM BACK/SCENE01/METADATA_BACK.DIM st_run/out DEM/Dem_ref.tif

point2dem st_run/out-PC.tif --orthoimage st_run/out-L.tif -o st_run/DEM_PC.tif

Extra info:

• The raster used for the shown image is NASADEM (30m). I tried with a Copernicus COP30 resampled to 200 meters of resolution, but I have the same problem with gaps.

• Version of ASP used: 3.4.0 Stable

Thank you for your time and assistance.

Sincerely,

Daryl Ayala

[Oleg Alexandrov]

You can try to see if --stereo-algorithm asp_mgm works better than the default. We have a section on this in the doc at: https://stereopipeline.readthedocs.io/en/latest/next_steps.html

It is also good to overlay and inspect the two mapprojected images where you later see gaps, and if they agree or there is occlusion or shadow, for example.

In areas where the images are uniformly white stereo is indeed expected to fail.

Sometimes it also helps to take your resulting DEM that has gaps (say after using the asp_mgm algorithm), fill those gaps with dem_mosaic, blur it, so no artifacts exist, then mapproject onto it the raw original images. Maybe they will mapproject better onto this DEM than onto COP30  or NASA DEM, and then stereo will succeed better. The dem_mosaic page has examples for how to fill small holes, fill large areas, and how to grow a DEM and how to blur it.

Also, you use bundle adjustment, but I think you don't use --bundle-adjust-prefix in mapproject and stereo, so in effect you work with original non-bundle-adjusted cameras. I am not sure if bundle adjustment is important, but if it is, it should be used at all steps. 

These experiments can be done on a clip cut from the DEM to mapproject onto, rather than on full image, for speed.

--
You received this message because you are subscribed to the Google Groups "Ames Stereo Pipeline Support" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ames-stereo-pipeline...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/ames-stereo-pipeline-support/0532fca6-81e5-43af-beda-fcfd9faf2c18n%40googlegroups.com.