关键词: 头部姿态估计, 特征融合, 主成分分析（PCA）, Bagging-SVM

Abstract: Aiming at the problem that the performance of existing common classifiers can’t meet the requirements of head pose estimation accuracy and the influence of illumination changes on head pose estimation accuracy, a method about head pose estimation based on Bagging-SVM integrated classifier is proposed. Firstly, the image preprocessing and the detection of face area based on Adaboost algorithm are used to reduce the influence of background, illumination and other interference factors on head pose feature extraction. Secondly, the histogram of orientation gradient (HOG) and the local binary pattern (LBP) are fused to extract the contour features and texture features of the face respectively. Then, principal component analysis (PCA) is introduced to select the fused head pose features, and the principal component features are extracted for the classifier to train. Finally, several training datasets are obtained by Bagging, each dataset is trained by support vector machine and several weak classifiers are produced. The categories of the test samples are voted by the plurality of weak classifiers. Experimental results on the Pointing ??04 dataset, CAS-PEAL-R1 dataset and self-built dataset show that the proposed algorithm has higher classification accuracy than the commonly used classification methods such as linear discriminant analysis (LDA) and na?ve Bayesian (NB) classifier, and has better robustness to illumination changes.

Key words: head pose estimation, feature fusion, principal component analysis (PCA), Bagging-SVM