Suspicious behavior assessment for visual surveillance using neural network classifiers

Assessment of abnormal and suspicious behaviors can to some extent be performed by an automated visual surveillance system. In this paper, we present an approach to visual surveillance of a car park against potential offenders by use of neural network classifiers. First, trajectories of people within the car parks are extracted. Then velocity patterns sampled from each trajectory are used for behavior classification. Performance of different neural network classifiers including Modified Probabilistic Neural Network, Multi-Layer Perceptron, and Self Organising Map are compared in terms of accuracy and computational requirements.