End-to-end point cloud registration method based on feature learning

Abstract

The invention discloses an end-to-end point cloud registration method based on feature learning. The method comprises steps of constructing a local geometric feature of each point by using a neighborhood point of the point in the point cloud, and constructing a mixed feature of the point by using the space coordinate, the normal information and the local geometric feature of each point; establishing an end-to-end point cloud registration deep learning network for simultaneously processing the template point cloud and the source point cloud; designing a translation loss function and a rotation loss function, and completing the training learning of the point cloud registration network under the common supervision of the two loss functions. The method is advantaged in that the end-to-end point cloud registration method based on feature learning is not sensitive to the initial position of rigid body transformation, the probability that the algorithm falls into a local optimal solution is reduced, and precision and efficiency of point cloud registration can be effectively improved.

Description

technical field [0001] The invention relates to a physical point cloud processing method in the fields of computer artificial intelligence and three-dimensional point cloud registration, in particular to an end-to-end point cloud registration method based on feature learning. Background technique [0002] The point cloud registration task is mainly to find the rigid body transformation between two unknown corresponding point clouds, which is widely used in reverse engineering, dimension measurement, robotics and other fields. The disorder of point clouds and the complex initial correspondence between different point clouds increase the difficulty of point cloud registration. The Iterative Closest Point (ICP) algorithm and its variants are widely used effective point cloud registration methods, but this method is very sensitive to the initial corresponding position of the point cloud pair, and it is easy to fall into local optimum. In addition, the way of continuous iteratio...

Images