Scene segmentation based on color and geometry
Liu Qian-la · Journal of Beijing Information Science & Technology University · 2014
Scene segmentation based on Kinect is studied according to the desired segmentation accuracy of the home service robot. A scene segmentation scheme is designed by fusing depth information and color information acquired by Kinect. First,the depth image and color image are described with a one- dimensional vector and a three- dimensional vector respectively. Then the color vector and depth vectors are normalized respectively;a four- dimensional vector description of the scene is formed from their combination by an equilibrium coefficient. Finally the spectral clustering algorithm based on Nystrm method is applied to the four- dimensional vector and the scene segmentation is achieved. Experimental results show the result of the proposed algorithm outperforms scene segmentation algorithms based on geometry or color data alone,and the scene segmentation accuracy requirements of the home service robot can be achieved.