Nighttime Vehicle Detection, Counting, and Classification Using Kinect Depth Virtual Loops

Due to nighttime poor lighting conditions, traditional vehicle detection and counting methods based on video cameras usually suffer from the problems of low detection accuracy and poor robustness. This paper presented a novel method for nighttime vehicle counting based on Kinect depth virtual loops. First, a pre-processing method is proposed to derive the Motion Depth Map (MDM) and the Hole Depth Map (HDM) from depth map. Virtual loops are set on these two maps to compute 1-D motion signals from the integral images in the loops. The motion signals from the MDM and the HDM are fused to formulate the description of vehicle motions, from which vehicles are detected and counted. With the virtual loops, a method was also proposed to identify vehicle change-lane behaviors to further enhance vehicle counting accuracy. Furthermore, the authors also compute vehicle geometric features from virtual loops for vehicle type classification. The proposed method has been validated with the actual nighttime traffic data collected by Kinect installed on the second WHUT Bridge, Youyi Avenue, Wuhan City. It can detect and count vehicles accurately in nighttime environments with 99.8% and 99.3% true positive rates on one-lane and two-lane scenarios, respectively. In average, it took 7ms to process one depth frame. The proposed method can also classify vehicle types with 99.80% recognition rates. In addition, the experimental results also demonstrate that it also successfully identified all the change-lane events.

• Supplemental Notes:
  • This paper was sponsored by TRB committee ABJ35 Standing Committee on Highway Traffic Monitoring.
• Authors:
  • Hu, Zhaozheng
  • Zhang, Rufeng
  • Mu, Mengchao
• Conference:
  • Transportation Research Board 97th Annual Meeting
  • Location: Washington DC, United States
  • Date: 2018-1-7 to 2018-1-11
• Date: 2018

Language

• English

Media Info

• Media Type: Digital/other
• Features: Figures; Photos; References; Tables;
• Pagination: 13p

Subject/Index Terms

• TRT Terms: Automatic vehicle classification; Automatic vehicle detection and identification systems; Data collection; Depth; Lane changing; Night; Traffic counting; Traffic flow
• Geographic Terms: Wuhan (China)
• Subject Areas: Data and Information Technology; Highways;

Filing Info

• Accession Number: 01664083
• Record Type: Publication
• Report/Paper Numbers: 18-02724
• Files: TRIS, TRB, ATRI
• Created Date: Mar 23 2018 10:32AM