Vehicle trajectory extraction at the exit areas of urban freeways based on a novel composite algorithms framework

Abstract

The exit areas of urban freeways always experience serious traffic safety and congestion problems. As a basic task, vehicle trajectory data are difficult to extract by traditional manual counting method because of the complicated weaving flow and large traffic volume at the exit areas of urban freeways. This paper presents a novel vehicle trajectory extraction composite framework combining YOLOv4 vehicle detection algorithm, SORT vehicle tracking algorithm and KD-Tree trajectory data reconstruction algorithm (YSKT algorithms framework). An unmanned aerial vehicle (UAV) was used to collect traffic videos of urban freeways exit areas, and the YSKT algorithms framework was adopted to extract vehicle trajectory data from the collected traffic videos. According to the test results of 4 traffic video samples, around 95% of complete vehicle trajectories in the videos could be extracted. Furthermore, basic traffic flow characteristic parameters, traffic efficiency parameters and traffic safety parameters were calculated and analyzed according to the extracted vehicle trajectory data, which was expected to help researchers analyze traffic problems in this kind of road segment in future studies.

Keywords:

• KD-Tree
• SORT
• vehicle trajectory extraction
• YOLOv4

Acknowledgments

This research was funded by National Natural Science Foundation of China (Grant No. 51778141 and 52072069), Jiangsu Creative PHD student sponsored project (KYCX20_0138) and Henan science and technology project (Grant No.182102310733). Their assistance is gratefully acknowledged.

Disclosure statement

No potential conflict of interest was reported by the author(s).

References