A method of objects classification for intelligent vehicles based on number of projected points

Objects classification is an important issue in environment perception of intelligent vehicles. Feature extraction is the basis for the classification. Due to the large amount of point clouds data measured by 3D LiDAR sensor and the real time performance demand of intelligent vehicles, the efficiency is a great challenge in data processing. This work proposes a new method of objects classification based on the feature NoPP (number of projected points). This idea is inspired by the phenomenon, that the nearer the object is to the LiDAR sensor, the more scan points the object has. Firstly, the measurement principle of the 3D LiDAR sensor is introduced. Then, the relationship of NoPP, the measurement distance and object size is geometrically analyzed. Based on the derived function relationship, the calculation process achieves the high efficiency. Lastly, the experimental results based on the Velodyne HDL-32E measurement point clouds are presented and the effectiveness of the proposed method is verified.