Towards Practical Fixed-Point LiDAR Object Tracking: Improving Accuracy and Processing Speed

This study aims to prevent accidents in parking lots caused by sudden pedestrian or vehicle movements emerging from blind spots, through the development of an object tracking algorithm that utilizes fixed-point LiDAR. Unlike conventional RGB camera-based systems, which are vulnerable to low-light conditions and raise privacy concerns, LiDAR provides reliable distance measurements regardless of ambient lighting and enables object detection and tracking without capturing personally identifiable information. To this end, we propose a handcrafted method that combines DBSCAN, Random Forest, and ByteTrack, with an emphasis on practical applicability through adaptability to diverse environments, improved detection accuracy, and high processing speed. We evaluated the method across six different scenes, demonstrating that it achieves both reliable detection performance and high processing speed. Notably, when the number of points per object was below 4000, the algorithm consistently achieved processing speeds exceeding 10 FPS. These results indicate that the proposed approach is a viable and effective solution for real-time object tracking using fixed-point LiDAR in a variety of real-world scenarios, including nighttime environments.