Video analysis for malpractice detection in classroom examination

Abstract

Malpractice in examinations is one of the banes in the education system of a nation. This paper aims in developing a robust face detection algorithm that tracks and analyzes multiple faces in a real-time video scene in a classroom examination. This work proposes automated face detection from a preprocessed surveillance video. First, the foreground object is extracted and the face region is detected using the Haar cascades. Second, the activity classification is performed to detect whether it is normal or suspicious, based on the face orientation, hand contact detection using background subtraction, and Gaussian mixture model (GMM). This system detects the commonly occurring suspicious activities like object exchange, peeping into others’ answer sheet, and people exchange during examination. Automated suspicious activity detection would help in decreasing the error rate due to manual monitoring. Promising results were obtained using this approach which show significant improvement over the existing methods, especially in case of illumination and orientation.