When the camera captures images on a bin with reflective inner walls, ‘ghost’ point clouds appear because of the reflection. How should this be handled?

My actual bin on site:

The ghost point cloud shown in the image does not actually exist. It is similar to the real object point cloud, and there may also be issues with misidentification in 2D deep learning. How should this be handled?

We recommend introducing the “bin detection” section in the Mech-Vision project.
By using dynamic ROI, we can ensure that only the real point cloud inside the bin is extracted.