Abnormal traffic incident detection is an important issue in intelligent traffic surveillance. The traditional methods based on video object detection and tracking usually fail in complex urban traffic conditions. This research aims to propose a robust detection method without using the detection and tracking techniques, which has the ability to detect the traffic abnormal incidents under crowded traffic scenes. In this paper, the optical flow features are adopted to extract traffic information from surveillance videos instead of tracking analysis. We adopt the optical flow as the observation of each video frame. The observations of video frames are encoded with Hidden Markov Models (HMM) to estimate the traffic state in the complex environment and used to train HMMs that model the variations of optical flow in the different traffic states. By calculating the new observations' likelihood based on these HMMs, we can automatically identify abnormal traffic incidents. Since not based on the vehicle trajectories, our method also has a good performance on crowed scenes where tracking-based approaches often fail. The experimental results on traffic videos demonstrate the efficiency, effectiveness and robustness of our method.
Abnormal Traffic Incident Detection Based on Hidden Markov Models
Third International Conference on Transportation Engineering (ICTE) ; 2011 ; Chengdu, China
ICTE 2011 ; 3098-3103
2011-07-13
Conference paper
Electronic Resource
English
TRAFFIC LIGHTS DETECTION AND STATE ESTIMATION USING HIDDEN MARKOV MODELS
British Library Conference Proceedings | 2014
|Multiuser detection based on hidden Markov models
British Library Online Contents | 2003
|Anomaly Detection Using Hidden Markov Models
NTRS | 1993
|