A Common Traffic Object Recognition Method Based on Roadside LiDAR

Environment perception and object recognition are the key to realize cooperative vehicle infrastructure system. This paper proposes an object recognition method based on roadside LiDAR, which realizes pedestrian, bicycle and vehicle recognition through feature extraction and convolutional neural networks. First, the DBSCAN clustering algorithm is used to segment the point clouds from LiDAR. In order to decrease space complexity, the traditional features are optimized by feature selection. Then, a lightweight VGGNet network is built as the recognition network, and the self-attention mechanism is added to improve the representation ability of features and recognition accuracy. Experimental results show that the proposed method has a recognition rate of 87% for common traffic objects, which has good robustness and real-time performance.