LiDAR 2D and Camera Fusion for ADAS: A Practical Approach With YOLOv9 and ROS2 Framework

This study introduces a cost-effective sensor fusion system that integrates 2D LiDAR and camera data for Advanced Driver Assistance Systems (ADAS), implemented within the Robot Operating System 2 (ROS2) framework. The objective is to enable accurate and real-time obstacle detection using lightweight hardware. The system leverages the YOLOv9 object detection model, optimized with TensorRT for high-performance inference, and includes a projection pipeline that aligns LiDAR point clouds with camera images to enhance spatial localization. By combining geometric precision from LiDAR with visual semantics from the camera, the architecture achieves effective multi-modal perception. Unlike more expensive 3D LiDAR-based solutions, the use of 2D LiDAR reduces costs while maintaining competitive accuracy. Experimental evaluations in static real-world environments show robust performance, with a distance estimation error below 3% and an angular error under 0.3°. The proposed system supports modular deployment and real-time operation, making it suitable for low-cost embedded ADAS applications.