Real-time full face closed eyes detection

Abstract

Intelligent systems based on machine-learning techniques are becoming a common way of solving problems in many different areas. In this thesis, our goal is to apply machine learning to a computer vision problem. We propose a new solution to detect closed eyes in full-face images. Accurate detection of closed eyes can be used in many problems such as driver drowsiness detection, human-computer interaction, and computer user monitoring. Most algorithms used to detect blinks or closed eyes follow a similar workflow. They require detection of a region of interest that is then used with different algorithms and techniques to determine if the eyes are closed. These methods are usually slow or require specific hardware that is not easily obtainable, making the solutions less viable and not portable. To overcome these issues, we created a convolutional neural network capable of detecting closed eyes in full-face images. We tested our network in two different experiments using images from different datasets. Our method was able to detect closed eyes with high accuracy. Different from other algorithms, our network is fast enough to perform accurate real-time detection in small GPU boards.