PCR-DAT: a new point cloud registration method for lidar inertial odometry via distance and Gauss distributed

We propose a novel point cloud alignment algorithm, namely PCR-DAT, for radar inertial ranging and localization. In environments with complex feature variations, the distribution trend of features is always changing, and the traditional alignment algorithms often fall into local optimums when dealing with regional point clouds with a combination of rich and sparse feature points, thus affecting the accuracy and stability of point cloud alignment. This paper addresses this issue by constructing a cost function composed of distance factors obtained from lidar measurements, normal distribution factors, and IMU pre-integration measurement factors. The core idea involves analyzing and classifying features in the target environment, defining different residual factors based on feature categories. Sparse features correspond to distance factors, while rich features correspond to distribution factors. Subsequently, a nonlinear optimization process is employed to estimate the robot’s pose. We evaluate the accuracy and robustness of the algorithm in various scenarios, including experiments on the KITTI dataset and field data collected during UGV movement. The results demonstrate that the DAT point cloud registration algorithm effectively addresses the pose prediction problem in the presence of feature degradation.