Mapproject Artifact on Cliff Face from Tri-Stereo Pléiades

[islam koa]

Hello Everyone,

I'm working with tri-stereo Pléiades imagery and attempting to generate map-projected images over a steep, nearly vertical cliff. I'm using the following mapproject command with Ames Stereo Pipeline:
mapproject --threads 18 -t rpc \

  --tr $rawRes_meters \
  ${dem_name} \
  merged_tiles/A.tif \
  merged_tiles/Test/A.tif

The reference DEM I'm using is derived from LiDAR but has been resampled to 50 m resolution.

The issue: the output map-projected image shows blurring or smearing at the cliff face, even though this distortion is not present in the original input image.

My Questions:

1. Could this artifact be due to the low resolution (50 m) of the reference DEM compared to the sharp terrain relief?

2. Is there a recommended way to avoid this kind of distortion when projecting over steep terrain?

3. Would using a higher-resolution DEM or a flat surface (via --datum and --datum-offset) help improve accuracy in such areas?

Thank you in advance for your help or any suggestions!

Best regards,
Islam Koa

[Oleg Alexandrov]

You can try to see if your LIDAR DEM you use for mapprojection is reasonably smooth. A blur may be suggested with dem_mosaic --dem-blur-sigma 5 or so, and then you can retry the mapprojection. That can also make the slopes shallower.

You can also examine the left and right raw images at that location. If you see occlusion, or some features very deformed in one of them, that may mean that there is no good enough info. (You have some pictures above but it is not clear to me if the left one is raw).

>Could this artifact be due to the low resolution (50 m) of the reference DEM compared to the sharp terrain relief?

I would think the lower the resolution and the smoother and shallower the ref DEM is, the less likely you will see such problems. Unless the issue is inherent to the input raw images.

1. Would using a higher-resolution DEM or a flat surface (via --datum and --datum-offset) help improve accuracy in such areas?

A high resolution DEM may work better if it is clear from the left and right raw images that despite the steepness the sloped area is imaged properly, and if the high-resolution DEM is very accurate at that location.

I've only seen such problems in a location in a steep canyon with occlusion.

--
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/d46bd253-bc1b-4474-9a71-dc7dbaf6c695n%40googlegroups.com.

[islam koa]

Dear Oleg and ASP Support Team,

Thank you very much for your reply and helpful suggestions.

To clarify my observations and follow up:

• In the nadir image, I do not see any noticeable occlusion or deformation — it appears clear even over the cliff face.

• In the forward-looking image (the one I previously shared), there is significant occlusion and smearing in the steep terrain area.

• The backward image shows slight deformation in the same steep zone but remains much clearer than the forward image.

• I also tested the --dem-blur-sigma option using dem_mosaic with a sigma of 5, but the artifacts in the forward image remain.

• When I switch to using high-resolution LiDAR (1 m) instead of the 50 m DEM for mapprojection, I still observe some deformation in the forward image, though it is slightly reduced.

Based on these tests and your advice, I am beginning to suspect that the occlusion and distortion in the forward image are due more to the viewing geometry than to the DEM resolution. Would you consider this behavior expected in such terrain, particularly for oblique views?

Also, would it be valid in such cases to exclude the forward image from certain steps (e.g., orthorectification or DSM fusion) to avoid introducing these artifacts, especially if the nadir and backward images are clean?

Thank you again for your guidance and support.

Best regards,
Islam Koa

[Oleg Alexandrov]

>Based on these tests and your advice, I am beginning to suspect that the occlusion and distortion in the forward image are due more to the viewing geometry than to the DEM resolution. Would you consider this behavior expected in such terrain, particularly for oblique views?

Yes, this makes sense and is expected.

>Also, would it be valid in such cases to exclude the forward image from certain steps (e.g., orthorectification or DSM fusion) to avoid introducing these artifacts, especially if the nadir and backward images are clean?

Yes, indeed.

Ideally, there should be a way of removing the problematic portion from a resulting DEM, and then merge the obtained DEM with a big hole with the other good DEM. I am not sure how that could be done automatically. Likely correlation would fail in that area anyway, or it may result in some noise only, and one could try to either erode that in the stereo_fltr step, or exclude it in point2dem based on triangulation error. 

To view this discussion visit https://groups.google.com/d/msgid/ames-stereo-pipeline-support/be6385cb-1f3d-4f4e-83ac-b922752ae900n%40googlegroups.com.