The Algorithm Research Of Face Pose Estimation Based On Multi-layer Random Forests Classification

In recent years,the researches of face image is increasing.Face detection and feature points localization,face tracking and recognition are developing rapidly,many researchers have developed effective methods and techniques which were applied to front face and can get better effect and high accuracy.Face pose estimation will provide more technical support for the further researches of face image.The topic of this paper is based on the specific studies which will use the shooting video sequences to replace the faces of the original video.When the face angles range for?30°to 30°in yaw and?30°to 30°in pitch,using the different classifiers can get the different orientations angle values in the images or videos.This paper used the Active Shape Models(ASM)feature points detection method and combined with random forests classification method to estimate the face pose.Collecting samples by the designed sampling device and these samples were used for random forests classifier training.Using the ASM facial feature points detection templates can get 68 feature points and they will through the normalized processing,then it can choose 7 key points and calculate the distance between the key points and other feature points.These data would be used as the characteristic data for training classifier.Because the number of training data is large and the data contains much redundant information,the Principal Component Analysis(PCA)algorithm can be used to select the better training data and this way can reduce about 90% of the data.Because of the better processing efficiency of random forests,using the different orientations training samples will have the different orientations random forests classifier that can have a better effect.Experiments show that the algorithm of this paper based on multi-layer random forests classification can get higher accuracy and efficiency when the angles range for?30°to 30°in yaw and?30°to 30°in pitch by comparing with other pose estimation algorithms.