Integrating Shape and Color Cues for Textured 3D Object Recognition

3D object recognition is a fundamental research topic. However, shape only feature descriptors for 3D object recognition have been the main focus of research. With the availability of low cost range plus color sensors, color based descriptors have attracted increasing attention lately. In this paper we present novel 3D object recognition algorithms which use not only shape but also color cues. We first extend our previously proposed Shape only Rotational Projection Statistics (hereby denoted S-RoPS) to obtain a Color only RoPS (C-RoPS) feature descriptor. The C-RoPS descriptor is based on the color space instead of the 3D shape coordinates. We then use feature level and decision level fusion approaches to combine the shape and color information. Experiments were performed on two popular datasets. The results show that decision level fusion achieves better results than either modality when they are used independently. The performance of C-RoPS was further tested using various color spaces e. g., RGB, HSV, YCbCr and CIELAB.