Fast Exam Video Summarization Using Targeted Evaluation of Scene Changes Based on User Behavior

As the demand for distance learning has increased, so has the need for electronic exams (e-exams). In e-exams, where video monitoring is used, one of the main tasks of the proctors is to detect cheating in exam videos. Monitoring a large number of videos is challenging for proctors, and summarizing videos to focus on cheating scenes can significantly reduce the difficulty of the monitoring task. Several important points should be taken into account in e-exam video summarization, such as various background, the clothes of each examinee, and the variety of video surveillance devices. One of the most effective methods for video summarization is the use of deep neural networks. However, the existing methods require a large dataset for training which is time consuming, besides, they are unable to accurately summarize the video containing a new background scene. In this paper, a deep learning-based method is proposed to recognize the objects in the scene and monitor any change in the examinee’s body parts with different priorities. In this approach, the abnormal behaviors of the examinee can be automatically detected without the need for a pre-training with multiple background, then the video is automatically summarized. The experimental results demonstrate that the proposed method, while having a high processing speed, can summarize exam videos up to 70%.