LiDAR Vehicle Trajectory Reconstruction with Arterial Shockwave Detection and Space–Time Analysis

Abstract

Roadside LiDAR (light detection and ranging) sensors have become increasingly significant in collecting high-resolution trajectory data of vehicles, pedestrians, and other road users, supporting the deployment of connected and automated vehicle applications. These sensors operate effectively under various illumination and weather conditions, providing accurate three-dimensional positioning. However, the raw trajectory data often suffers from quality issues, such as missing segments, misaligned directions, inappropriate appearances, and sudden disappearances. These issues arise from field-of-view constraints, road user or infrastructure occlusions, and sensor or analytic engine limitations. To address these challenges, this paper presents a LiDAR trajectory reconstruction method that utilizes spatial-temporal analysis and shockwave detection to improve incomplete and noisy vehicle trajectories. The method includes a lane-based centerline map-matching algorithm to determine vehicle lane assignments and create lane-by-lane spatial-temporal trajectory diagrams. Fragmented trajectories are matched by analyzing their upstream–downstream continuation relationships. A shockwave detection technique is introduced, which assesses the convexity and concavity of each vehicle trajectory and identifies peaks in curvature. Finally, trajectories are stitched based on estimated vehicle states within regions separated by shockwaves. The proposed model was evaluated using roadside LiDAR trajectory data collected from two intersections during a workday in February 2024 at the DataCity Smart Mobility Testing Ground in New Brunswick, New Jersey. The reconstruction results show promising performance in handling LiDAR coverage blind spots and accurately reconstructing shockwave patterns, as verified by field observations.