Artifacts in DEM when running parallel_stereo on map-projected Chandrayaan-2 TMC-2 images

[Kalyani]

Hello,

I am encountering an issue with Chandrayaan-2 TMC-2 processing where map-projecting images prior to stereo matching introduces noticeable grid-like patches and missing data artifacts into the DEM.

When I run parallel_stereo directly on unprojected images, the DEM outputs normally. However, map-projecting the images first leads to local matching failures and heavy patchiness across the output DEM.

Dataset & Environment Details

• ASP Version: 3.7.0

• Reference DEM (ref.tif): ch2_tmc_ndn_20231101T0125121377_d_dtm_d18.tif (processed with hole filling and dem_mosaic --dem-blur-sigma 5 as per the manual)

• Nadir Image: ch2_tmc_ncn_20231101T0125121377_d_img_d18

• Aft Image: ch2_tmc_nca_20231101T0125121377_d_img_d18

Workflow & Commands Used

proj="+proj=stere +lat_0=-90 +lon_0=0 +k=1 +x_0=0 +y_0=0 +R=1737400 +units=m +no_defs"

# 1. Map Projection
 mapproject --tr 10 --t_srs "$proj" \ ref.tif \ tmc/nadir.cub \ ba/run-nadir.adjusted_state.json \ tmc/nadir.map.tif mapproject --tr 10 --t_srs "$proj" \ ref.tif \ tmc/aft.cub \ ba/run-aft.adjusted_state.json \ tmc/aft.map.tif

# 2. Stereo Matching parallel_stereo
parallel_stereo \ --alignment-method none \ --stereo-algorithm asp_mgm \ --subpixel-mode 9 \ tmc/nadir.map.tif tmc/aft.map.tif \ ba/run-nadir.adjusted_state.json \ ba/run-aft.adjusted_state.json \ stereo/run \ ref.tif

# 3. DEM Generation
point2dem --tr 10 --t_srs "$proj" \ --orthoimage \ stereo/run-L.tif stereo/run-PC.tif

I have attached screenshots showing how the DEM looks without mapproject versus with mapproject for your review. Could you please check what might be causing this discrepancy?

Thank you!

[Oleg Alexandrov]

You should mapproject at 5 m/pixel, which is the native TMC resolution. It is strongly suggested to overlay the left and right mapproj images onto your mapproj DEM and see if they are in agreement. If you see big shifts or warps that means the data are not registered well, which explains your failure, as I think the search range per tile is too huge. You should also see what the search range that is printed is, if more than 100 - 200 pixels, that would be not good.

I added recently a lot more detail to the documentation. The problem with this data is bad initial alignment, so some care is needed. See https://stereopipeline.readthedocs.io/en/latest/examples/chandrayaan2.html#terrain-mapping-camera-2.

You can also try first doing the stereo with unprojected images, as you do, hole-fill and blur that DEM you made till all artifacts go away, and mapproject onto that one. 

So, there is a chance your external DEM ch2_tmc_ndn_20231101T0125121377_d_dtm_d18.tif may not be well registered with the cameras after bundle adjustment.

--
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/824cdbd5-b99a-4d80-9066-bb82b98cb7c6n%40googlegroups.com.

[Oleg Alexandrov]

To add a little, it is strongly suggested to get a very recent build, from the releases pages, as I put some fixes for local epipolar alignment for Chandrayaan, and in addition to mapprojecting at 5 m/pixel one should also change the DEM grid for point2dem to likely 20 m/pixel, or at least not switch that one to 5 m/pixel as DEM grid should be coarser than the mapproj image grid. 

Another thing that may go wrong, since you have so many missing tiles, is perhaps the node ran out of memory and started to drop tiles. So, if happy enough with the printed search range and consistency between left and right maprpoj images and the input DEM, one could also inspect the memory usage and check the logs for those failed tiles for both search range and some premature stop.