Smart Classroom Surveillance System Using YOLOv3 Algorithm

One of the major concerns associated with educational institutions is the attendance survey, monitoring, and surveillance. Owing to the labor-intensive nature of manual attendance system involving the management of attendance records, the current focus is on the emergence of an efficient and accurate attendance system. This paper presents a maiden attempt to propose a smart classroom attendance and surveillance system using YOLOv3 algorithm, a novel deep learning approach. An attempt has been made to avoid the unnecessary wastage of time spent during attendance marking and also to avoid fake attendance. Using YOLOv3 algorithm in the DarkNet framework, a realistic dataset of images with around 14 students and faculty members has been used to train the test model. The dataset has been formed by acquiring the realistic images from the Department of Instrumentation and Control Engineering, Dr. B. R. Ambedkar National Institute of Technology Jalandhar, India. The test results demonstrate the efficiency of YOLOv3 algorithm in effective face recognition, thereby endorsing its capability and usage in smart classroom surveillance system. In addition, the performance of YOLOv3 has been compared with YOLOv3-tiny algorithm to validate its robustness and competence in classroom surveillance tasks. The experimental results demonstrate a maximum accuracy of 99% by YOLOv3 algorithm.