Research on Environmental Perception Based on YOLOv8 and LiDAR Data Fusion

To enhance the sensing capability of driverless cars in complex dynamic scenarios, a fusion method of radar and depth camera is emphasized and a technical framework for integrating multi-source sensors is proposed. A YOLOv8 neural network detection model combined with a self-attention network structure is first used to achieve correct recognition and depth estimation of vehicles and pedestrians, then further distance data of dynamic environment is acquired by M10P LiDAR, and finally information from D435i camera and radar information are fused by matrix projection. In the experiments, the conventional YOLOv8 model was compared to the YOLOv8 model enhanced with self-attention, the camera and LIDAR measurements were evaluated against the fusion results. The experimental results demonstrate that the new model significantly improves the confidence and stability of target detection, while the fused distance measurements are more accurate than those from individual sensors.