Detection Of Anomalous Behaviour In Crowds Using Newton Pratt'S Curve Fitting Technique

Analysis of Crowd behaviour and determining the occurrence of an anomaly is a very complex problem to solve efficiently. Crowd behaviour is analysed by detecting and tracking the objects in a crowd using particle advection schemes. The general flow pattern of a crowd clearly shows that crowds have a standard behavioural flow. Anomalies can be detected in crowds when there occurs a deviation from such standard behaviours. This paper proposes a method to identify crowd behaviours in uniform visual scenes and help detect anomalies using a light and simple algorithm. The algorithm involves splitting a given video sequence into fixed size windows, obtaining clear flow patterns for each window using Lucas-Kanade algorithm, thresholding and determining the best fit circle for the flow pattern of each window using Newton Pratt's' curve-fitting technique. The algorithm detects an anomaly in the crowd behaviour when it realizes an aberrant change in the centre and dimensions of the best-fit circle of a particular window. Although, this algorithm is susceptible to shock effects and can only be used when there exists a single flow pattern in a sequence, its light and simple structure, low execution time and high accuracy make it convenient for use in situations where there exist single crowd behaviours. The methods validity has been tested by running it for various crowd sequences from the UMN dataset, YouTube, and other web sources.