Anomaly detection in crowd using optical flow and textural feature

Abstract

This paper aims at solving real-world surveillance problems using computer vision and motion estimation techniques. It focuses on detecting abnormal crowd behaviour and locating it in dynamic crowd condition. In this paper, a combined approach is the proposed using the crowd motion analysis and texture-based analysis. Lucas–Kanade optical flow method is used for the estimation of motion in the scene. Also, texture-based feature and entropy give the statistical measure of randomness which is used for localization of crowd. The University of Minnesota (UMN) database has been used for testing.