SAC-IA Algorithm Based on Parallel KD-Tree Search and Improved Feature Point Selection

Abstract

In recent years, with the rapid development of autonomous driving and intelligent robots, point cloud stitching technology has attracted people's attention. At present, the sampling consistency initial registration algorithm is often used for coarse registration of point clouds, but this algorithm has problems such as large randomness in feature point selection and low search efficiency. this paper proposes a feature point selection method considering the relationship between coordinate distance and spatial similarity, which prevents the occurrence of local optimization by setting the distance threshold, and uses the FPFH to calculate the spatial feature similarity relationship between the source point cloud and the target point cloud point pair. The randomly selected feature pairs with similar spatial characteristics are saved to ensure that the selected point pairs are overlapping as much as possible, which is conducive to the solution of rotation matrix and translation vector. By comparing the registration results of the traditional algorithm, it is concluded that the error of the proposed algorithm is reduced by 15.70%. In addition, the KD-Tree parallel technology based on OpenMP is also used to greatly improve the search efficiency of the corresponding point.