Abstract: When the deflection angle is large, great noises are brought to the face pose measurement results because the saliency of facial feature points decreases. To solve the problem, a multi-innovation disturbance rejection filtering algorithm is proposed, in which the pose changes between the moving face and the standard face model are taken as filtering observations. (1) Multi-innovation is introduced to modify filtering estimation, and the states of the face pose changes are estimated by the time series of multi-group observations. (2) The convergence of filtering is judged in real time, and then the noise covariances in the observation and the process are estimated by multi-innovation in time to adjust the Kalman gain matrix. (3) The pose coordination model is established and then the camera motion parameters are calculated according to the changes about face pose after filtering, to realize the coordination between the camera and face pose. On the basis of the hardware structure of the test device, the proposed algorithm is compared with the adaptive Kalman filter (AKF) algorithm. The experimental result shows that the pose estimation error of the proposed algorithm is less than 10 mm and the time of camera coordination is about 25 ms for coordination on face pose. Compared with the AKF algorithm, the pose accuracy is improved by 23% and the coordination efficiency is improved by 30%. The influences of the noises from facial pose calculation in pose coordination can be effectively suppressed by the proposed algorithm. It can improve the stability of the system of coordination on face pose and ensure the real-time response.