Fig. 1. Detection flow of YOLOv5

Fig. 2. Network structure of 3D target detection method based on improved YOLOv5 algorithm

Fig. 3. Overall network framework structure of Complex-YOLO algorithm

Fig. 4. Rotating target box labeling. (a) Definition of angle of rotation; (b) schematic of angle coding

Fig. 5. Structure of small target detection layer

Fig. 6. Visualization of feature map. (a) RGB-Map; (b) feature map of small target layer; (c) feature map of shallow layer; (d) feature map of middle layer; (e) feature map of deep layer

Fig. 7. CBAM Attention Module Structure

Fig. 8. AP and detection speed comparison between different algorithms on KITTI dataset

Fig. 9. Comparison of detection performance between Complex-YOLO, PointPillars algorithm, and proposed algorithm

Fig. 10. Visualization results of proposed method