Radar Voxel Fusion for 3D Object Detection

Automotive traffic scenes are complex due to the variety of possible scenarios, objects, and weather conditions that need be handled. In contrast more constrained environments, such as automated underground trains, automotive perception systems cannot tailored a narrow field specific tasks but must handle an ever-changing environment with unforeseen events. As currently no single sensor is able reliably perceive all relevant activity in surroundings, data fusion applied much information possible. Data different sensors modalities on low abstraction level enables compensation weaknesses misdetections among before information-rich compressed thereby lost after sensor-individual object detection. This paper develops low-level network for 3D detection, which fuses lidar, camera, radar data. The trained evaluated nuScenes set. On test set, increases resulting AP (Average Precision) detection score by about 5.1% comparison baseline lidar network. proves especially beneficial inclement rain night scenes. Fusing additional camera contributes positively only conjunction fusion, shows interdependencies important result. Additionally, proposes novel loss discontinuity simple yaw representation Our updated orientation estimation performance input configurations. code this research has been made available GitHub.