Object segmentation using depth camera


I am working on an object segmentation function. Basically I want to detect a package shape on a flat surface (from “top view”/“bird’s eye view”). I am using Intel RealSense d435i camera.

I did a proper object segmentation using RGB image processing and works good. But there are some edge cases that can affect on the RGB image processing such as multi-colored packages, or dark parts on the package. The thing is that I want to use the stereo camera to create a better segmentation, and use it as a mask. This mask can be used to merge/shape the RGB processing segmentation.

This camera also has an IR matrix projector that I already implemented a segmentation with that using clipping distance, and is not bad at all, but there’s a lack of precision. Stereo cameras are more common to get from other brands, so I would like to ask if there’s any well-known method, documentation or suggestion on how to do a segmentation of an object using stereo camera (depth image).

Thank you!