Loop Closure Detection Method Based on Point Cloud Height Distribution and Contour Features

The loop closure detection module, which reduces accumulated error by constructing loops based on whether the robot has previously visited the current scene, is a crucial component in simultaneous localization and mapping (SLAM). This paper proposes a loop closure detection method based on LiDAR point cloud height distribution and contour features. The point cloud is first downsampled using an adaptive voxel grid filter for efficiency. The point cloud coordinate system is redefined with centroid alignment and principal component analysis (PCA). Height distribution and contour features are encoded into descriptors, and Hamming distance and contour similarities between frames are computed. These similarities are weighted and combined to identify loop closures. Extensive experiments on public datasets validate the effectiveness and robustness of the proposed method.