Parallel 3D ICP Based on Conditionally Constrained Corresponding Points and Applications

The iterative closest point (ICP) tends to fall into local optimality due to inaccurate initial poses during the registration of the multi-view 3D cloud. Therefore, this paper proposes a parallel ICP algorithm with conditional constraint corresponding points. To achieve a fine registration of the point cloud, the corresponding point set is first filtered by adding the normal and color information. Then the OpenMP is introduced to accelerate the program in parallel for ICP. To verify the effectiveness of our algorithm, in the V-Rep simulation environment, the multi-view point cloud data of the scene is obtained by the RGB-D cameras from different angles point cloud. The results show that our algorithm can fuse the multi-view point cloud, improve the accuracy and real-time performance of ICP. Furthermore, in a large-scale calculation, the average single iteration time is less than 0.1s, and the RMSE (root mean square error) is about 0.1, which meets the need of target recognition and sorting in a three-dimensional industrial scene.