Aiming at the problem that head posture estimation（ HPE） used to be affected by face pose

facial expression

occlusion

and inaccurate feature extraction

and results in recognition rate is not high

a robust head pose estimation method is proposed. Firstly

local binary features（ LBF） are generated

then

68 key facial feature points extraction model are regressed by training random forest. Finally

a classifier is trained by combining back propagation（ BP） neural network to recognize head pose. Results of experimental on CAS-PEAL-R1 and Pointing’04 dataset show that the accuracy of head pose estimation of the proposed method achieves 98. 57 %

and the mean absolute error is 3. 90°

which solves the problem of inaccurate positioning of facial key feature points and further improves the recognition precision of head pose.