Abstract: The method of face pose estimation is vulnerable to ‘self-occlusion’ at present. To
solve this problem, an improved ASM algorithm is used to extract facial feature points, and the
geometric statistical knowledge of the face shape is used to estimate the depth of the facial feature
points. Then the sparse face model is established based on the main features of human face. After
estimating the face pose approximately with relevant face feature points, 3D space face pose is
estimated accurately via the algorithm of least-squares method. The experiment results show that
the method has better estimated results for the case of ‘self-occlusion’, and has better estimation
accuracy compared with the same kind of method.

Key words: face pose estimation, sparse model, feature points, least-squares