I noticed there could be a problem in the point cloud visibility filter. The problem is described https://github.com/OpenDroneMap/ODM/issues/1785. I believe the current algorithm cannot handle well for the case like the red shot in the below image, where closer points will occlude further points and leads to unpredictable results in the algorithm. I think introducing normal into consideration could help resolving this problem, I tried to ignore the points that have large angle compared to the shot view direction (smaller than 100° for example) which works for solving the issue in my dataset.
Hi,
I noticed there could be a problem in the point cloud visibility filter. The problem is described https://github.com/OpenDroneMap/ODM/issues/1785. I believe the current algorithm cannot handle well for the case like the red shot in the below image, where closer points will occlude further points and leads to unpredictable results in the algorithm. I think introducing normal into consideration could help resolving this problem, I tried to ignore the points that have large angle compared to the shot view direction (smaller than 100° for example) which works for solving the issue in my dataset.