Shadow merges with object in point cloud depth sensing



In both images above, the shadows “merge” with the cube that I am trying to depth sense with. If I add more lighting, there will always be a smaller shadow else where and “merge” with the cube. What is the best environment to take a “depth sense”/“point cloud” with or is there a camera parameter I can change to deal with shadows better? (I am using python currently)
MyPointCloud.ply (2.6 MB)
The above download is the .ply file for the point cloud of my cube on a table with the natural office lighting.

Hi @Flocker
you can try to tune the confidence_threshold to remove from the point cloud the values with low confidence:
https://www.stereolabs.com/docs/api/structsl_1_1RuntimeParameters.html#ade2446d6529e8416f2f8ae002eaeb149

Changing the confidence_threshold has helped and hurt my points a little bit, but made it better overall. Another solution on top of this was putting “textured colored” tape on my object and use a little more lighting and the point cloud information ended up with better data.

Indeed when you are working with pure stereo vision technology the more an object is texturized, the better the depth is estimated.

1 Like