Classification of Violations in the Examination Room Based on Deep Detection Algorithm

Abstract

In the field of proctoring, cameras play an irreplaceable role, but the traditional invigilation methods for monitoring students during exams require a large number of invigilators, which can be a waste of manpower, resources, and finances. In recent years, the rapid development of deep learning technology has shown potential for recognizing abnormal behavior in various settings, including examination rooms. Test postural assistant detection algorithm in the examination room studied in this paper aims to use computer technology instead of human power to automatically detect behaviors in surveillance video. In this paper, five algorithms are used to identify any violations of test discipline in the examination room and compare the recognition results. The experimental results show that DenseNet can effectively detect video frames containing violations in the video sequence, and the average detection accuracy is 95.62%, which can accurately identify these behaviors such as looking left and right in the examination room and passing items. The study also presents a self-built database to support model training and testing. The experiments were conducted on self-built dataset of examination room videos. The findings of this study have important implications for improving examination security and reducing the need for large numbers of invigilators.

Ähnliche Arbeiten