4D feature of point cloud based on robust normal estimation

This paper proposes the point feature histogram based on the correct normal vector estimation. The four dimensional features of each point in point cloud is computed by synthesizing the normal vector information of neighbour field of point cloud. All of four features are binned into histogram. The different type geometric primitives (such as plane, sphere, cylinder etc.) are generated to analyze the points´ signature, and algorithm complexity is reduced by approximating factor parameter. The experiment result proves that point feature histogram has the discriminative power.