Detection of Student Focus Level using Body Posture Analysis with Blazepose

Abstract

University student engagement monitoring is crucial for effective learning outcomes, yet traditional methods rely on subjective educator observations and time-consuming questionnaires. This research develops and evaluates a real-time university student focus monitoring system based on body posture analysis using the BlazePose pose estimation model with a rule-based classification approach. The system detects 33 body keypoints from video input using MediaPipe Pose, utilizing 7 specific keypoints including nose, left and right shoulders, left and right hips, and left and right knees to calculate head and body angles for classifying postures into three categories: Focused, Less Focused, and Not Focused. Testing was conducted on 10 video datasets with various sitting posture scenarios. The system successfully operates in real-time with a frame rate of 25-30 FPS and achieves an overall average accuracy of 81.05%, exceeding the minimum performance target of 80%. However, consistency remains challenging with only 60% of test videos reaching the target threshold. System performance proves excellent under ideal conditions such as adequate lighting and contrasting backgrounds, but decreases significantly under challenging conditions including poor lighting, noise, and ambiguous postures. The system successfully demonstrates proof of concept, yet requires further optimization to enhance robustness against environmental variations for practical classroom implementation.