Alignment issues with iPad Pro acquisition #1561
-
|
Currently working on an indoor change detection software, and I’m struggling to understand what can possibly cause this misalignment, and how I can eventually fix it. I’m getting two false positives, reporting that both chairs moved. In the second image, with the actual point cloud overlay (blue before, red after), you can see the two chairs in the yellow circled area. Even if the chairs didn’t move, the after (red) frame is severely distorted and misaligned. The acquisition was taken with an iPad Pro, using RTAB-MAP. |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 1 reply
-
|
In the lastest rtabmap iOS version, we don't threshold anymore the depth image with the confidence for the data saved in the database. We keep both data separately and threshold depth confidence on-demand (before rendering or generating optimized mesh for example). The cloud you see is maybe the cloud at lowest confidence, so you can get some edge bleeding or depth interpolation where the lidar didn't hit or didn't have returns. At highest confidence, these points "could" be removed. How do you generate your point cloud? You may load the depth confidence image too, and threshold the depth for highest confidence (>=100) before projecting the depth image to point cloud. |
Beta Was this translation helpful? Give feedback.
-
|
That was helpful, thank you. Cheers! |
Beta Was this translation helpful? Give feedback.
In the lastest rtabmap iOS version, we don't threshold anymore the depth image with the confidence for the data saved in the database. We keep both data separately and threshold depth confidence on-demand (before rendering or generating optimized mesh for example). The cloud you see is maybe the cloud at lowest confidence, so you can get some edge bleeding or depth interpolation where the lidar didn't hit or didn't have returns. At highest confidence, these points "could" be removed.
How do you generate your point cloud? You may load the depth confidence image too, and threshold the depth for highest confidence (>=100) before projecting the depth image to point cloud.