Design of a Light-Weight Video-Based Blink Detection System with an Eye Monitoring Interface

Blinking plays a crucial role in maintaining eye health by cleaning the ocular surface and moisturizing the eyeball. Blink pattern can also serve as an indicator for ocular diseases, attention levels and the degree of fatigue. This work proposes a video-based blink detection system utilizing a lightweight eye states classification network and a blink recognition algorithm based on probability thresholds. The network reports a classification accuracy of $\mathbf{9 8. 1 4 \%}$ on MRL Eye Dataset and $\mathbf{9 6. 7 0 \%}$ on a customized dataset, and the overall system achieves an F1-score of 0.940. The processing speed of the system is up to 109.1 frames per second. Additionally, an eye monitoring interface is also implemented on the basis of the blink detection system, enabling real-time monitoring and recording of blink patterns, which has the potential to migrate to other application scenarios.