Tracking Multi-Vehicles With Reference Points Switches at the Intersection Using a Roadside LiDAR Sensor

Abstract

Obtaining real-time trajectories with high accuracy and high resolution of connected and unconnected vehicles with roadside LiDAR (Light Detecting and Ranging) sensors can highly benefit connected vehicle (CV) systems. In order to track vehicles with a roadside LiDAR sensor, the vehicle-corner point with the shortest distance to the sensor is usually used as the reference point since only parts of the vehicle can be scanned, but trajectory interruptions and speed errors caused by neglecting reference point switch were observed. With the movement of a vehicle, the reference point may switch from one corner point to another, especially at the intersection where it may switch more than once when turning left/right. Moreover, the vehicles with various lengths will increase the impacts of reference point switch on tracking. This paper proposes a tracking method considering reference point switch and occlusions for multi-vehicles at the intersection based on a roadside LiDAR sensor. In particular, reference point switch patterns are deeply explored and trajectory prediction for occlusions based on Kalman filtering method is developed. Then they are integrated into a global nearest neighbor (GNN) method for pairing reference points between frames. At last, the proposed method was validated by LiDAR data collected by a 32-channel 360-degree LiDAR sensor at the intersection of McCarran Blvd and Evans Ave in the city of Reno, Nevada, U.S. The new approach correctly captured switches of vehicle reference points and solved short-term occlusion issues to avoid trajectory interruptions, speed errors, and corner point misrecognitions.

Keywords

• Roadside LiDAR sensor
• vehicle tracking
• reference point switch
• vehicle occlusion