A Head Pose Estimation Method Based on Multi-feature Fusion

Since head pose estimation is influenced by illumination variation, expression, noise disturbance and other factors, which results in low rate of recognition, a method of head pose estimation based on multi-feature fusion is proposed in this paper. At first, a pose feature combining the second-order histogram of oriented gradient (HOG) and the uniform pattern of local binary pattern (UP-LBP) is proposed, which is used for head pose estimation in single image. Then, an improved random forest algorithm is adopted for classification of head pose and solving the instability problem of the algorithm. Finally, the improved random forest classifier is used for head pose estimation of the novel pose feature. The experimental results show that, the method proposed in this paper is more capable of classification and with better robustness to illumination variation.