A Proof-of-Concept Study for LiDAR-Based Traffic Measure Estimation: Traffic Density Estimation and Vehicle Trajectory Extraction

Abstract

This study explores the potential of vehicle-mounted LiDAR sensors for road traffic monitoring, focusing on traffic density estimation and vehicle trajectory extraction. Traditional traffic monitoring systems, dependent on fixed infrastructure like cameras or loop detectors, face challenges such as high maintenance costs and limited spatial coverage. LiDAR technology, capable of dynamically collecting high-resolution 3D point cloud data, provides a scalable and efficient alternative. Using the KITTI dataset, this study evaluates three models—PointPillars, PartA2-Net, and PV-RCNN—for vehicle detection and traffic metric estimation. PV-RCNN demonstrated the highest mean Average Precision (mAP) in 3D object detection, excelling in both accuracy and reliability. The results highlight LiDAR's ability to provide detailed and continuous traffic data, addressing the limitations of traditional methods. This approach offers a promising solution for real-time traffic monitoring, with potential applications in adaptive traffic management as automated vehicle technologies advance.