Anchor-free real-time vehicle tracking using LiDAR data.

Area of Science:

• Intelligent Transportation Systems (ITS)
• Computer Vision
• Robotics

Background:

• Accurate vehicle trajectory data is crucial for developing intelligent transportation systems (ITS) and supporting connected and autonomous vehicles (CAVs).
• LiDAR sensors provide real-time microscopic vehicle trajectories, but precise tracking from raw point clouds is a significant challenge.
• Existing tracking methods often rely on bounding boxes, which can be computationally intensive and less accurate in certain scenarios.

Purpose of the Study:

• To propose a novel real-time vehicle tracking mechanism for roadside LiDAR systems.
• To enhance the accuracy and computational efficiency of vehicle trajectory generation.
• To adapt the Simple Online and Real-Time Tracking (SORT) algorithm for a track-by-point approach using 2D LiDAR data.

Main Methods:

• Developed a new tracking mechanism integrating a 2D LiDAR data structure with the SORT algorithm.
• Replaced traditional bounding box detection with a track-by-point approach using vehicle center points.
• Introduced a novel index that incorporates LiDAR distance measurements for precise vehicle spatial localization.

Main Results:

• The proposed method demonstrated higher tracking accuracy compared to the traditional bounding box approach.
• The new tracking mechanism achieved faster computation speeds, enhancing real-time application feasibility.
• Evaluation using data from a 32-channel portable LiDAR at signalized intersections validated the method's effectiveness.

Conclusions:

• The developed track-by-point tracking method significantly improves real-time vehicle trajectory generation from roadside LiDAR data.
• This approach offers a more accurate and computationally efficient alternative to conventional bounding box methods for ITS applications.
• The findings indicate a promising advancement for the operational capabilities of connected and autonomous vehicles.