Laser Radar Data Registration Algorithm Based on DBSCAN Clustering

At present, the core of lidar data registration algorithms depends on search correspondence, which has become factor limiting performance this kind algorithm. For point-based algorithms, coincidence rate is too low, and for line-based method searching correspondence complex unstable. In paper, a laser radar algorithm based DBSCAN (Density-Based Spatial Clustering Applications with Noise) clustering proposed, avoids establishment corresponding relationship. Firstly, ring band filter designed to process outliers noise in point cloud. Then, adaptive threshold used extract line segment features cloud be registered, density obtain key clusters rotation angle translation matrix. order evaluate similarity two frames after registration, kernel estimation proposed describe registered cloud, K-L divergence find optimal value clusters. The experimental results show that direct between points or lines scenes many clouds, can effectively improve robustness suppress influence relative error result actual within 10%, accuracy better than ICP