¹Department of Information Technology, MIT Campus, Anna University, Chennai, India-600 044.
²Department of Electronics Engg., MIT Campus, Anna University, Chennai, India-600 044.

---

Abstract—Traffic video surveillance has received significant attention in recent years. Anomalous Event Detection is gaining popularity among vision community. Existing methods on Intelligent Traffic Surveillance (ITS) systems are inefficient in detecting abnormal events, as they employ high level object features. This paper proposes an alternate solution named Optical Flow based Frequent Pattern Mining (OFFPM) based on low level pixel features. OFFPM employs temporal pattern analysis for detecting abnormal events from the sequences of video streams. Optical flow method is used to determine the spatial information and direction of moving pixels. In addition to the spatial and direction information, temporal information is also determined for every window of sequences. OFFPM relies mainly on extracting spatial and temporal information of motion pixels and mining frequent temporal patterns in a sequence of video frames. Frequent pattern mining is applied to discover regular patterns of normal events. The irregular patterns of events are classified as anomalies. Experiments are conducted on the Queen Mary University of London (QMUL) traffic junction & Roundabout dataset and results show the proposed method accurately detects abnormal traffic events.

Index Terms—Motion pixels, optical flow, regular and irregular patterns, temporal pattern analysis.

[PDF]

Cite: Deepika R., Arvinth PrasathV. S., Indhumathi M., Ashok Kumar P. M., Vaidehi V., "Anomalous Event Detection in Traffic Video Surveillance Based on Temporal Pattern Analysis," Journal of Computers vol. 12, no. 2, pp. 190-199, 2017.