Research On The Algorithm Of Face Pose Estimation

Face pose estimation is a key technology in related areas such as intelligent monitoring systems, human-computer interaction systems, and coincides with the cranial system. For the problems of existing pose estimation methods, in this paper we perform two different methods, one based on the geometric method and the other on manifold learning, reducing the complexity of the algorithm and improving accuracy. The main work includes:(1)A new face pose estimation method based on the geometry is proposed, using some of the features of a human face and human face metrology knowledge to establish face plane model. Second, it builds relationships between the face images and face model by pinhole imaging model. Finally, we calculate face pose with Newton iterative method and the least squares principle. Model established with this method is relatively simple, the use of optimization methods can obtain good results.(2)We give the other face pose estimation method based on Adaboost and LLE. The face images are classified according to the posture, thereby pose estimation problem into a pattern classification problem. This paper uses Kernel and supervised locally linear embedding to reach High-dimensional face image dimensionality reduction. Then in a low-dimensional space, face images can be to classify with Adaboost classifier. The algorithm can reduce the computational complexity and the ability to obtain a higher recognition rate.(3)In this paper, different algorithm is tested using a different face database, Geometry-based method with CMU PIE face databases and another with Facepointing face databases. Some Matlab simulation experiments prove the effectiveness of the two methods.(4) Design and implement a face pose estimation system, that provides two algorithms for face pose estimation.