Real-time stereo vehicle detection and collision warning system based on YOLOv8 and SGBM

This study proposes a robust, real-time stereo vision system integrating YOLOv8n for fast, accurate vehicle detection and SGBM for reliable depth estimation, addressing traditional method limitations. By fusing YOLOv8n's detections with SGBM's disparity maps, 3D vehicle positions are calculated for collision risk assessment. Trained and evaluated on KITTI, the system achieves exceptional detection accuracy and depth precision with over 100 FPS, significantly outperforming conventional approaches and offering a viable solution for low-cost, high-efficiency driver assistance.