Segmentation of Depth Images into Objects Based on Polyhedral Shape Class Model

Abstract

A novel approach for object detection in depth images based on a polyhedral shape class model is proposed. The proposed segmentation algorithm decides whether a subset of image points represents a physical object on the scene or not by comparing its 3D shape to several shape classes. The algorithm is designed for cluttered scenes with simple convex or hollow convex objects. The proposed algorithm is trained using a set of 3D models of objects belonging to several shape classes, which are expected to appear in the scene. The presented method is experimentally evaluated using a publicly available benchmark dataset and compared to three state-of-the art approaches.