A Fast Inertial-LiDAR Odometry with YOLO Recognition for RGB-Colored and Object-labeled Mapping

This paper presents a rapid inertial-LiDAR odometry system integrated with YOLO object detection for constructing RGB-colored and object-labeled point cloud maps. By combining excellent target detection capabilities of YOLOv5 with precise mapping framework of FAST-LIO2, the system is adapted effectively for environment mapping with fast and accurate object detection. It features two core modules: a YOLO-based object detection module that processes RGB images to identify desired objects, and a SLAM-based mapping module that incorporates detection results to improve the efficiency and precision of point cloud map construction. Additionally, a novel image-to-point cloud conversion technique is developed for 2D-3D mapping, resulting in detailed colored point cloud maps. Experimental validation demonstrates the system's adaptability and high performance across diverse settings, enhancing environmental recognition without sacrificing accuracy or efficiency. The system also performs robustly on non-hard synchronized devices.