VEHICLE RECOGNITION IN AERIAL LIDAR POINT CLOUD BASED ON DYNAMIC TIME WARPING

Abstract. A two-step vehicle recognition method from an aerial Lidar point cloud is proposed in this paper. First, the Lidar point cloud is segmented using the region-growing algorithm with vehicle size limitation. Then the vehicle is recognized according to the profile shape based on dynamic time warping. The proposed method can detect vehicles parking under trees in an urban scene, and classifies the vehicles into different classes. The vehicle location, orientation, parking direction and size can also be determined. The experimental result based on a real urban Lidar point cloud shows that the proposed method can correctly recognize 95.1 % of vehicles.