A Robust Point Cloud Registration Method in Urban Dynamic Environment

Considering the enormous moving objects, serious occlusions, and low precision of GPS (global positioning system) in urban dynamic environment, a fast and robust registration method which can adapt to different initial position errors without detecting dynamic objects is proposed. Firstly, the region growing method is used for object segmentation of obstacle point cloud without ground data; and by setting constraint conditions, segmentation result is optimized to generate object gravity centers of one point cloud. Then, an algorithm framework with multi-nested loop of RANSAC (random sample consensus) in which registration results is updated iteratively is put forward to achieve rough registration of centroid sets and remove outliers. ICP (iterative closest point) is finally used for precise registration. The contrast experiments with the traditional RANSAC algorithm show that the proposed method can achieve accurate and reliable point cloud registration in complex dynamic scenes with large initial position errors, and the registration success rate and registration speed are significantly higher than those of traditional methods.