Facial Recognition Attendance & Surveillance System For University Examination (PRESENTIAL)
Since the recent pandemic, the world has adapted to various touchless technologies. However, the current state of tertiary examinations in the Mauritius academic system remains unchanged. The prevailing practice is manual attendance and surveillance which is intrusive, time consuming, unreliable, and high risk of community spread. This research suggests an automated exam attendance and surveillance system that employs a touchless biometric image identification using facial recognition together with object detection, head pose estimation and pose classification techniques to execute real time attendance and surveillance. A system has been built to facilitate these exam procedures. Prior to exams, students are expected to register their face biometrics using the application. Exam attendance requires the placement of a video camera outside the exam venue to detect faces of students. For facial recognition attendance, ArcFace has been used together with various liveness detection techniques. Exam surveillance requires one or more cameras inside the exam venue to get the entire coverage of the room. For cheating detection, RetinaFace and 6DRepNet models are used to detect suspicious head movements and YoloV5 is used to detect unauthorized object such as cell phone. To detect exam events such as toilet breaks, a pose classification technique achieved using BlazePose was used. Additionally, YoloV5 together with DeepSort and Person ReID was used to ensure correct seating of students. At the end of an exam, a report is also generated.
Keywords: Face Recognition, Attendance, Camera Surveillance exams, Computer Vision, Exams Cheating
Completed by Yaj Siburuth
BSc (Hons) Computer Science
Department of Information and Communication Technologies
Faculty of Information, Communication and Digital Technologies
University of Mauritius, 2022
