Human Eye Detection Optimization Algorithm Based On AdaBoost

Eye detection is the key of face recognition and tracking technology.It has important research value and application prospect are reflected in the field of fatigue detection and eye tracking and so on.The eyes are sensitive to the changes of external environment and make the response rapidly,and the information conveyed from eyes are rich and complex,so the study of vision based on the mechanism is still a challenging task.Because of the important research value of eye detection,it has been highly sought after and emerging out a variety of good detection algorithm,according to the requirements in practical application it can be divided into two categories:coarse detection and precise detection.But each algorithm has its advantages and disadvantages.One of the most successful and widely used algorithms is the algorithms that based on AdaBoost Haar feature in the numerical method.The eye detection algorithm based on AdaBoost algorithm is optimized in this paper after analysis and exploration of the AdaBoost algorithm carefuly.Firstly the traditional AdaBoost algorithm is optimized make the degradation phenomenon does not appear in the training samples,than analyzed the training sample selection of AdaBoost classifier and the experiment results,and tested several groups of different positive and negative samples,finally conclud a set of the best combination of positive and negative samples from the experimental results.Based on the above research process,an improved AdaBoost algorithm for eye detection is proposed,use the best positive and negative samples to select binocular and monocular samples and trained the binocular and monocular classifiers separately,than use the two classifier cascade make a double layer named "Eye Classifier",use the "Eye Classifier" as a strong classifier of the whole detection structure.The first layer of the double-layer structure effectively eliminates the false detection of the human eye picture,and improves the accuracy of the second layer.The system shortens the detection time and reduces the false detection rate,which has a great improvement on the accurate positioning of the eyes.The most important problem of human eye detection is the detection accuracy and speed,the future resaech will focus on the advantages of detecting the integration in a variety of methods,find the rapid detection of human eye information from more complex or dynamic background images.