Student activities detection of SUST using YOLOv3 on Deep Learning

Abstract

This article describes the main phases of a new learning system in which YOLOv3 is used for deep learning to identify student activities. A student’s exploits in the SUST- (Shaanxi University of Science and Technology) should be perceived to circumvent any unwanted problems. In this project, we have investigated the problem of image-based student activity detection in the SUST. It involves making a prediction by analyzing student poses, behavior, and activities with objects from complex images instead of videos. Comparing with all approaches, we conclusively decided to use an algorithm YOLOv3 (You Only Look Once) which is the latest and more convenient. The algorithm utilizes anchor boxes, bounding boxes, and a variant of Darknet. We have created our own dataset collecting images from SUST and annotated the dataset manually. During the research with this project, we have considered student activities in the SUST into seven sections namely reading, phoning, using a laptop, taking books, smiling, looking, and sleeping. The proposed system provides not only multi-tasking knowledge with classification but also localization of students and the equivalent actions instantaneously. Our intention is to detect the student position automatically, efficiently, confidently, and strictly with the help of extracted image functions. Interestingly, the proposed approach achieved a mean average precision (mAP) 97%. In the future, a combination of real-time data analysis will improve value to this scheme.