A scalable real-time computer vision system for student posture detection in smart classrooms

Technological advancements have ushered in a new era of global educational development. Artificial Intelligence (AI) holds the potential to enhance teaching effectiveness and foster educational innovation. By utilizing student posture as a proxy, computer vision technology can accurately gauge levels of student engagement. While previous efforts have focused on refining posture classification models, this study uniquely addresses the comprehensive implementation of a real-time posture detection workflow, encompassing software, hardware, and network aspects. The proposed posture detection system leverages surveillance cameras equipped with cutting-edge computer vision technology, specifically employing the Open Visual Inference & Neural Network Optimization (Open VINO) model for precise student posture detection. Data transmission is facilitated using the Message Queuing Telemetry Transport (MQTT) protocol, effectively establishing a seamless posture detection workflow within the classroom setting. To validate the system, video recordings from a real teaching environment (a fifth-grade class in the Chinese compulsory education system) were analyzed, resulting in posture classifications with impressive accuracies of 0.933 for standing, 0.772 for sitting, and 0.959 for hand-raising. Achieving a frame processing time ranging from 109 to 758 milliseconds, the system efficiently delivers real-time posture data to educators. Consequently, the posture detection system developed in this study possesses the capability to intelligently monitor student postures in the classroom, with the potential to enhance teaching quality in smart classrooms.