Detection of Anomalous Behaviour in an Examination Hall Towards Automated Proctoring

Abstract

The anomalous behaviour is hard to be detected simultaneously in a complex scene such as detecting abnormal movements of examinees in examination rooms. Modelling activities of moving objects and classifying them as normal or anomalous is a major research problem in video analysis. In this paper, we make use of the of neural networks and Gaussian distribution to help solve this problem by building a prototype of a monitoring system that consists of three stages; face detection using haar cascade detector, suspicious state detection using a neural network and lastly anomaly detection based on the Gaussian distribution. The main idea is to decide on whether the student is in a suspicious state or not using a trained neural network and then decide that a student performs an anomalous behaviour based on how many times he was found in a suspicious state in a defined time duration. The complete system has been tested on a proprietary data set achieving 97% accuracy with 3% false negative rate.