A novel blink detection system for user monitoring

As the amount of time that the average person spends interacting with computing devices increases, it becomes important for the devices to be able to adjust to the user´s state. Blinks are known to be indicative of the emotional state of a person and detecting blink patterns of a computer user can lead to devices that comprehend the user´s state and adjust their characteristics accordingly. In this paper we propose a method to reliably detect blinks by analysing the motion vectors in the face and eye regions. It consists of three main steps: detecting the face and eye regions in the image, calculating the motion field in the face region and finally, exploiting the level of similarity between the vectors in the eye and face regions to distinguish images containing blinks from those that do not. Our algorithm is able to detect blinks even in images where the head is tilted or is moving - conditions that are very important for practical applications. We have achieved an accuracy of 96.9% in our experiments.