Segmentation of object regions using depth information

The paper describes an object segmentation technique using 3D information extracted by stereo vision. The proposed algorithm is based on the assumption that all objects are lying on a common base plane. From the input stereo images, depth data is first extracted on edge points, and the recovered 3D points are projected onto the base plane. Using the fact that projected points from the same object tend to form a cluster on the base plane, each object region is separated. We demonstrate the efficacy of the algorithm using experimental images.