A Forward Obstacle Detection Approach for Trains Based on 4D Radar

With the expansion of urban scale, rail transport has become an indispensable component of the city's public transportation. Achieving intelligent assisted driving in rail transport can reduce incidents due to driver factors. Sensors enabling accurate perception of the surrounding environment are the prerequisite for intelligent assisted driving. 4D Radar generates point clouds that include both 3D position and velocity information. Especially, 4D Radar works well in bad weather conditions compared to Lidar and camera. This paper proposes an obstacle detection method based on 4D Radar in the train-travelling environment, which dynamically adjusts the obstacle detection range using DBSCAN. Moreover, the network is improved for the characteristic of sparse 4D Radar point cloud, and enhancing the detection capability of the network through the introduction of an attention module. Experimental results show that the method achieves satisfactory results in real scenarios and has great potential.