A three dimensional point cloud registration method based on backpropagation neural network and random sphere cover set

Abstract

Due to the large number of points in the point cloud, the complexity of registration is quite high. To solve this problem, a registration method based on backpropagation (BP) neural network and random sphere cover set (RSCS) is proposed in this study. For the two point clouds to be registered, each is simplified based on the BP neural network. In order to avoid losing a large number of key points in the simplification process, a fixed RSCS algorithm is used for each point cloud to replace the key points with the super-point (SP) sets, and then the SP sets are combined with the simplified data. The iterative closest point (ICP) algorithm is used for fine registration. The point cloud is simplified by BP neural network and fixed RSCS, which reduce the number of points for the subsequent fine registration. Therefore, the time and space complexity can be effectively reduced. Experimental results show that the proposed method effectively improves the computational efficiency while maintaining almost the same precision details, which is of great significance for the registration of point clouds with a large number of points.