Research On Eye-gaze Tracking System Based On One Web-camera

In order to solve the problems such as high cost,complex system composition and calibration algorithm of the eye-gaze tracking systems.An eye-gaze tracking system based on one low cost web camera was designed.The hardware of the eye-gaze tracking system is simple,only has one web camera and a computer.The web camera can capture 640×480 or320×240 pixel RGB images.The software of our eye-gaze tracking system consists of three modules,eye positioning module,feature detection and tracking module,dynamic calibration module.In eye positioning module,a fusion algorithm of Adaboost algorithm and the Active Appearance Model algorithm is proposed to detect 66 feature points on the face,then the location of these feature points are used to locate the eye position.It solves the problem which is diffculte to locate eye positions in real-time,thereby improving the real-time and robustness of eye positioning in eye-gaze tracking system.In feature detection and tracking module,the gradient vector method is used to detect the pupil center in the low-pixel eye image.Then the algorithm based on Active Appearance Model is used to locate the position of eye corner and the three-dimensional information of head.It solves the problem that it's difficult to detect the iris center in low pixel images.After detecting the required feature points,the optical flow method is used to track the feature points.In the research of dynamic calibration module,a dynamic calibration algorithm based on artificial neural network is prososed.The application of artificial neural network to the calibration process of eye-gaze tracking system solves the problem that tracking system needs to be calibrated before each operation.Experiments show that the developed low-cost eye-gaze tracking system not only has good robustness but also has high precision,the average error is 1.34° when the head is stationary,and the average error is 3.26° with head movement.The proposed low-cost eyegaze tracking system has the characteristics of simple composion and low cost,experiments show that the performance and the practicality of our eye-gaze tracking system are better than the existing eye-gaze tracking systems.In addition,the proposed dynamic calibration algorithm improves the robustness of the system.