Comparative Performance Analysis of Optical Flow Algorithms for Anomaly Detection

Abstract

Presently there is major interest in visual surveillance systems for crowd anomaly detection. Horn-Schunck, Lucas–Kanade and Farneback optical flow methods are used for the estimation of motion in the scene. The motion vectors; magnitude and orientation are analyzed in this work to detect anomaly in crowd. It helps classify the crowd as normal or abnormal, walking or running, vehicle entering in crowd. Various databases are evaluated to check the validity of the feature extraction process. In order to cluster the behaviour as normal or abnormal, Artificial Neural Network is used as a classifier. The results from Farneback optical flow estimation algorithm are promising for crowd behaviour understanding and anomaly detection.