Extracting and Denoising Vehicle Trajectory Automatically from Aerial Roadway Surveillance Videos

In recent years, unmanned aerial vehicle (UAV) has become an increasingly popular tool for traffic monitoring and data collection on highways due to its low cost, high resolution, good flexibility, and large coverage. Extracting high-resolution vehicle trajectory data, which provides wide support for both microscopic and macroscopic traffic flow analysis, from aerial videos taken by UAV flying over road section becomes a critical research task. In this study, we propose a novel methodological framework for automatic and accurate vehicle trajectory and length extraction from aerial videos. We first employ an ensemble detector to detect vehicles on the target region. Then the kernelized correlation filter (KCF) is applied to track vehicles in a fast and accurate way. The vehicle positions are mapped from the physical coordinates to the Frenet coordinate to obtain the vehicle trajectories along the road. The data quality control is applied in the procedure and a Wavelet Transform is used to denoise the biased vehicle positions in the trajectory data. Our model was tested on two aerial videos on freeway segments. The experimental results show that the proposed method extracts vehicle trajectory at a high accuracy (i.e., measurement error of Mean Squared Deviation (MSD) is 28.854 pixels, Root-mean-square deviation (RMSE) is 2.187 pixels, the Pearson product-moment correlation coefficient (Pearson's r) is 0.999), which provides us a reliable trajectory for analyzing traffic flow. This study fills gaps in UAV-based automatic vehicle trajectory extraction, and has the potential to benefit a variety of future research.

  • Supplemental Notes:
    • This paper was sponsored by TRB committee ABJ50 Standing Committee on Information Systems and Technology. A Novel Framework for Automatic Vehicle Trajectory Extraction and Denoising from Aerial Videos: This is an alternate title.
  • Corporate Authors:

    Transportation Research Board

    ,    
  • Authors:
    • Chen, Xinqiang
    • Li, Zhibin
    • Yang, Yongsheng
    • Wu, Huafeng
    • Ke, Ruimin
    • Zhou, Wenzhu
  • Conference:
  • Date: 2019

Language

  • English

Media Info

  • Media Type: Digital/other

Subject/Index Terms

Filing Info

  • Accession Number: 01697662
  • Record Type: Publication
  • Report/Paper Numbers: 19-03147
  • Files: TRIS, TRB, ATRI
  • Created Date: Dec 7 2018 9:34AM