Iterative Closest Point of Object-based Point Clouds for Multi-View Extension

Importance of LiDAR-based vehicle detection techniques for ensuring safe driving is increasing. However, LiDAR data from a single vehicle may have occlusions caused by obstacles, presenting a challenge. Eliminating these occluded areas through point cloud registration obtained from two vehicles can enhance autonomous driving systems. However, matching every point for point cloud registration requires a significant computational process, making real-time processing challenging. This paper proposes a novel technique that tackles this issue by introducing the iterative closest point (ICP) of object-based point clouds, enabling fast registration of point clouds scanned by two vehicles. The proposed approach demonstrates the ability to reduce the number of points significantly, resulting in a faster registration process compared to the conventional ICP approach.