An Unsupervised Abnormal Crowd Behavior Detection Technique using Farneback Algorithm
M.R. Rejitha, Sudhish N. George · 2019
Anomalous behavior of crowd in surveillance videos have gained tremendous interest in the past few years. Although a number of human detection and activity recognition systems are available today, there is a need for new method to address the key challenges in different scenarios. Intelligent video surveillance systems observe human activities within very crowded surroundings and have a vital role in public safety scenarios. `Abnormal behavior' in crowd scenes implies an event which is unexpected in nature. In this paper, a temporal abnormality situation is considered, in which a sudden random movement of people happens after an alarming situation such as, bomb explosion, terrorist attack, etc. arises. In this paper, using Farneback optical flow algorithm and an activity map based image segmentation, an unsupervised anomalous behavior detection technique among crowd scenes is depicted. Depending on the drastic variation in two metric parameters such as entropy and Temporal Occupancy Variation (TOV), the abnormality is detected. The proposed algorithm is compared with different state-of-the-art methods and shows a better performance based on various quantitative metric values.