Development of Real-Time Eye Tracking Algorithm

Abstract

This project develops a real-time eye tracking algorithm to improve the accuracy of iris detection and gaze classification. Eye tracking has been utilized in many applications to detect driver awareness, predict human behaviour, and assist paralyzed individuals. Detecting the centre of iris is a crucial step in eye tracking development and Conventional Circular Hough Transform (CCHT) is widely used for the iris detection. However, the accuracy of the CCHT method decreases when the head of an object is not orthogonally positioned to the camera under ambient light changes. To overcome this problem, the facial landmark detector is implemented to detect the eyes as the Region of Interest (ROI), track the eyes and identify the gaze by classifying the eye positions to left, right and middle. The eye tracking algorithm has an added feature to detect eye blinking for drowsiness detection. To gain better accuracy, the classifier uses a scan method to classify the eye position based on the levels of pixel intensity. The eye tracking algorithm is implemented in OpenCV using Python Software for ease of portability. The results show that the average accuracy of 100% and 90% are achieved in the iris detection and gaze position classification, respectively.