Real-time dynamic object detection using multi-sensor fusion for autonomous driving

Autonomous driving requires reliable perception of the vehicle’s surroundings to safely operate in adversarial weather conditions and to avoid potential collision with dynamic obstacles. This work studies three dynamic object detectors and fusion of the objects detected by each detector for autonomous driving. A dynamic object detector was designed to fuse state-of-the-art camera based detection, radio detection and ranging (RADAR) fused with camera detection and segmented clusters from light detection and ranging (LiDAR) sensor point cloud data from the three detectors. Reliable perception can be achieved by sensing the environment with multiple sensors to create a fused model of the environment including detected dynamic objects. State-of-the-art object detection precision for autonomous driving has advanced with recent breakthroughs in neural networks. A public dataset was used to evaluate the designed detector’s precision and computational performance. The designed detector was also compared with a detector that was used as a base for the implementation. The designed detector’s fusion performance proved to be reliable for autonomous driving with the identified requirements. The computational performance requirements were not met on the system used for testing