The Research And Implementation Of Algorithms Of Eye Localization

In the modern society,pupil localization has played a crucial role in many applications,such as personal identification,human-computer interaction and facial expression analysis.The higher pursuit of accuracy and precision make the people put forward new requirements and interactive experience.At the same time,the target image is often affected by environment,which makes the localization a difficult task.At first,the industry generally locates the human eye based on geometrical characteristics,but due to the complexity of different parts in the face,which result in the local extrema,and cannot get the good result.In the recent years,the application of machine learning becomes wider and wider,and some related algorithm was also be introduced in the field of computer vision,such as Adaboost and support vector machine(SVM).But these machine learning methods are belong to the classification tree and not suitable for accurate localization,this paper proposes random forest based on adaptive gradient boosting decision tree..The main work of this paper is to improve the regression tree from a large number of diverse normalized eye samples.Because of the weakness of a single regression tree,this paper use a strong classifier based on the integration of multiple regression trees,furthermore,we have improved the performance by a multistage structure like pyramid.Performance results of our methods showed that it can obtain an accuracy of 92 percent on BioID database and gives a frame processing time of less than 1ms because of the low computation cost.