Occlusion-Aware Trajectory Discontinuity Correction for Roadside LiDAR Using Time-Space Analysis.

Area of Science:

• Transportation Engineering
• Computer Vision
• Data Science

Background:

• Roadside sensing technologies like LiDAR offer high-resolution vehicle trajectory data.
• LiDAR is valuable for traffic monitoring and safety analysis but suffers from noise and occlusion.
• Sensor noise and occlusion lead to trajectory errors, impacting traffic data reliability.

Purpose of the Study:

• To propose a post-processing method for detecting and correcting occlusion-induced trajectory discontinuities in LiDAR data.
• To enhance the reliability of vehicle counts, traffic state estimation, and conflict analysis.
• To reconstruct continuous vehicle paths and recover realistic traffic patterns from fragmented data.

Main Methods:

• A post-processing method based on time-space analysis is proposed.
• The method exploits spatiotemporal consistency of vehicle movements to identify and correct trajectory errors.
• Validation performed on real-world LiDAR data from an urban intersection.

Main Results:

• The proposed method achieves high performance with an average precision of 0.989 and F1-score of 0.948.
• Significantly outperforms benchmark methods like IMM, GNN-RM, and HMM + Viterbi.
• Vehicle count accuracy improved from 85.5% to 97.4% under occlusion conditions.

Conclusions:

• The time-space analysis method effectively detects and corrects LiDAR trajectory discontinuities caused by occlusion.
• The approach enhances the accuracy of traffic monitoring and surrogate safety analysis.
• This method offers a reliable solution for improving traffic data quality from roadside LiDAR systems.