3D Object Detection Algorithm in Adverse Weather Conditions Based on LiDAR-Radar Fusion

In adverse weather conditions such as rainy or foggy weather, the point cloud data from lidar may not reliably describe the environment. In such situations, the robust information from millimeter-wave radar, which is less susceptible to environmental interference, can compensate for the limitations of lidar. Effectively integrating the information from both radars to enhance the robustness of target detection under adverse weather conditions poses a challenge. To address this issue, a road scene multi-object detection algorithm based on the fusion of lidar and millimeter-wave radar in adverse weather conditions is proposed, leveraging improvements based on the CenterPoint framework. The algorithm incorporates a transformer-enhanced network and fuses information from both radars at the feature level. Experimental evaluations conducted on the Fog-nuscenes dataset, which underwent simulated foggy weather conditions, demonstrate that the proposed network achieves a mAP of $49.52 \%$ Compared to the original CenterPoint network, this represents a significant improvement of $10.75 \%$.