Research And Application Of Eye Detection And Tracking Based On Video Sequence

With the rapid development of the Internet and the widespread use of imaging devices, image and video become the main mode of transmission of information and permeate in our daily life explosively. In order to make the computer analyse and process of the acquired information automatically, computer vision attracts more and more attention. Since the human eye contains a rich variety of information of gender, identity, expression, etc., eye detection and tracking become the core technology of some applications such as face recognition, psychological test and fatigue test, which attract wide attention of many researchers and make eye detection and tracking become one of the important topic in the field of computer vision. The paper does some research on eye detection and tracking based on video sequence, and applies them into the fatigue detection. The main research work are as follows:(1) Do some research on eye detection based on Adaboost algorithm. When training the classifier using Adaboost algorithm, due to multiple misclassifications the weight of some samples in the critical junction of positive and negative samples become bigger with the increasing of the number of iterations, which enable the algorithm to focus shifted to the trigger weight imbalance. Considering that the value of positive samples missed is much larger than the false detection of negative samples, the paper improves Adaboost algorithm and puts forward three layer eye detection for the defect of false positive rate of eye detection is higher.(2) Do some research on eye tracking based on Kalman filter and CamShift algorithm. In the target tracking process, CamShift algorithm lack of prediction module and is easy to lose the target when the color of the target and background color is similar or the target area has large occlusion. Furthermore, the target cannot be recovered once lost. The paper improves CamShift algorithm and puts forward eyes tracking method based on Kalman filter algorithm and improved CamShift for the size of the search window has great influence on eye tracking.(3) The eye changes most obviously when people is in a state of fatigue. On the basis of eye detection and tracking, the paper studies fatigue detection based on eye state. The work is divided into two parts specifically. Primarily, recognizing eye state based on image processing method. First using the Ostu method to make the located eye region binary; Second extracting the eye peripheral contour in binary image using the Freeman chain code method; Then fitting the eye peripheral contour using the least squares ellipse fitting method; Finally, recognizing the eye state by the ratio of elliptical long axis and short axis. Posteriorly, introducing two fatigue criteria that are the PERCLOS and blink frequency. Through statistics of PERCLOS and blink frequency in a continuous period to determine whether it is in the state of fatigue currently.