A Machine Learning Approach to Eye Blink Detection in Low-Light Videos

Abstract

Inadequate lighting conditions can harm the accuracy of blink detection systems, which play a crucial role in fatigue detection technology, transportation and security applications. While some video capture devices are now equipped with flashlight technology to enhance lighting, users occasionally need to remember to activate this feature, resulting in slightly darker videos. Consequently, there is a pressing need to improve the performance of blink detection systems to detect eye accurately blinks in low light videos. This research proposes developing a machine learning-based blink detection system to see flashes in low-light videos. The Confusion matrix was conducted to evaluate the effectiveness of the proposed blink detection system. These tests involved 31 videos ranging from 5 to 10 seconds in duration. Involving male and female test subjects aged between 20 and 22. The accuracy of the proposed blink detection system was measured using the confusion matrix method. The results indicate that by leveraging a machine learning approach, the blink detection system achieved a remarkable accuracy of 100% in detecting blinks within low-light videos. However, this research necessitates further development to account for more complex and diverse real-life situations. Future studies could focus on developing more sophisticated algorithms and expanding the test subjects to improve the performance of the blink detection system in low light conditions. Such advancements would contribute to the practical application of the system in a broader range of scenarios, ultimately enhancing its effectiveness in fatigue detection technology.