LEARNSCOPE: A Training Free and Privacy Preserving Learning Engagement Analysis System Using MediaPipe Based Behavioral Signal Fusion

Abstract

The rapid expansion of online and hybrid educational programs has established a strong demand for automated systems that can track student participation throughout the day. Instructors in physical classrooms use visual cues to determine whether students pay attention to their lessons, but virtual learning environments reduce the availability of these visual signals. Most automated systems that analyze student engagement use supervised deep learning models which need extensive facial and behavioral data to operate, but this method creates problems because it breaches privacy rights, introduces bias, demands extensive processing power, and lacks system transparency. This paper presents LEARNSCOPE, a training free and privacy preserving learning engagement analysis system based on MediaPipe driven computer vision techniques and rule based behavioral signal fusion. The system extracts eye gaze direction and blink rate and head pose and facial activity proxies and presence consistency from webcam input on the user’s device without storing raw video data or performing face recognition. The system combines features through an interpretable heuristic scoring model which uses smoothing methods to create stable engagement states throughout time. The system uses an edge centric architecture which provides quick processing times together with secure privacy protection methods. The system demonstrates its ability to track student engagement patterns through its analytics tools which assist instructors, all while operating without the need for data labeling or time-consuming machine learning training processes. The proposed approach demonstrates that practical, explainable, and ethical engagement monitoring is feasible for real world educational environments.