This paper focuses on creating a framework for real-time behavior interpretation from traffic videos. It relates that video-based surveillance systems have many different applications in traffic monitoring, providing more information compared to other sensors. The paper presents a rule-based framework for behavior and activity detection in traffic videos. The videos comes from stationary object feeds, which segment moving targets from the image feed and track them in real time. A novel Bayesian network approach classifies the different segmented objects using image features and image sequence- based tracking, resulting in robust classification. Traffic and classification information can help analyze behavior. The paper states that behavior falls into two observable categories: interaction between cars, and interaction between a car and a stationary object. The system looks for both with implicit knowledge of the cameras field of vision and common interactions available within the system. It can recognize behavior from video input and provide readout on such behavior. The system demonstrates successful behavior recognition result for pedestrian-vehicle interaction and vehicle-check post interactions.


  • English

Media Info

Subject/Index Terms

Filing Info

  • Accession Number: 00989341
  • Record Type: Publication
  • Source Agency: UC Berkeley Transportation Library
  • Files: BTRIS, TRIS
  • Created Date: May 3 2005 12:00AM