Range image registration by neural network

Range image contain 3D coordinates of points on the object surface which are sampled by range sensors. In general, a sensor can capture only parts of the object of occlusion and sensing limitations. A 3D model needs to be registered using multiple range images to integrate a complete one. Instead of registering range datasets with the traditional iterative closest point (ICP) method, we use a new algorithm combined with image processing and artificial neural network (ANN). Experimental results show the validity and efficiency of our method.