Convolutional Neural Networks for eye detection in remote gaze estimation systems

An eye detection algorithm based on Convolutional Neural Networks (CNN) architecture was developed. The algorithm was designed to detect eyes in video images from a remote gaze estimation system that is part of a gaze-controlled human-computer interface. The CNN for eye detection has two stages of convolutional and sub-sampling layers followed by a fully connected feed forward neural network with a total of 1227 trainable parameters. Experiments with 3 subjects showed that for the full range of expected head movements, the CNN achieved a detection rate of 100%, for images with fully opened eyes, and a false alarm rate of 2.65 X 10-(4) %. The CNN failed to detect eyes that were either partially or completely covered by the eyelids. The CNN for eye detection did not require pre-processing or normalization and was shown to be robust to changes in scale, rotation and illumination of the eyes.