FocusCam: A Real-Time Deep Learning System for Detecting User Distraction Using Webcam Video

Maintaining our focus during online learning or any computer based works has become a big problem for us due to distractions like using our phones, yawning from fatigue and short attention spans. This results in a significant drop in productivity and learning efficiency in such environments where direct supervision is not possible. To overcome this problem we propose FocusCam a lightweight real-time distraction monitoring system that checks the webcam feed to determine if a person is focused or distracted. Our model makes use of CNN models for Yawn detection, gaze detection and finally phone usage detection module YOLO during online learning sessions. the high quality datasets that were available such as YawDD and MPIIGaze have been preprocessed to emulate webcam like footage and achieved a accuracy of 97 % in predicting yawning and 80 % accuracy for gaze detection. These results demonstrate that combining multiple models results us in an effective tool to measure the attention. Future improvements can be made by integrating temporal modeling by the help of LSTM architectures.