A Review Of Video-Based And Machine Learning Approaches To Human Eye Blink Detection In Video

Abstract

Automation of detection of human eye blink in video has a broad array of applications, including detection of disease, anti-spoofing software, and helping individuals with physical disabilities interact with computers. The present work provides a review of several papers within the past two decades which propose methods for automated blink detection, highlighting the evolution of the field alongside developments in machine learning techniques. Then, the strengths and shortcomings of several popular approaches are evaluated in the context of eye blink detection. Namely, I focus on appearance-based and motion-based computation, support vector machines, convolutional neural networks, and long short-term memory networks. Finally, I report the beginnings of a reproduction of the methods outlined in one of the papers reviewed.