A LiDAR point cloud registration method combining linear feature extraction and TrICP algorithm

We propose a LiDAR point cloud registration method that combines linear feature extraction and trimmed iterative closest point (TrICP) algorithm to address the high initial position requirement of the iterative closest point (ICP) algorithm. It adopts a registration strategy of combining coarse registration and accurate registration. First, the point cloud to be registered is segmented with dynamic thresholds; then, within the segmented area, linear features are extracted by fitting linear parameters using the random sample consensus (RANSAC) algorithm; after that, based on extracted linear features, the density clustering method is used to solve the optimal transformation matrix and performs coarse registration; finally, the point cloud processed by above steps is accurately registered with TrICP. The point cloud is acquired in four environmentally different laboratories by 3i-T1 LiDAR. Large position difference registration experiments and large angle difference registration experiments are separately conducted. The experimental results show that the proposed method has greatly improved registration accuracy compared with the existing algorithms. Therefore, the proposed method can complete accurate registration for point clouds with large differences in initial poses.

Automated summary

This study proposes a LiDAR point cloud registration method that combines linear feature extraction and the trimmed iterative closest point algorithm to address the high initial position requirement of the iterative closest point algorithm. By adopting a registration strategy of combining coarse registration and accurate registration, the proposed method achieves accurate registration for point clouds with large differences in initial poses.