Summary: Aiming at the problem of inaccurate estimation results caused by occlusion of feature points in visual target pose estimation system, a distributed fusion estimation method using adaptive unscented Kalman filter (AUKF) as local filter is proposed in this paper. The process noise is adapted by introducing the improved Sage-Husa noise estimator. According to the recognition quantity of feature points, the occlusion is divided into partial occlusion and severe occlusion. The partial occlusion subsystem is estimated by local filtering after repairing the missing observation points according to prior information. The severe occlusion subsystem does not participate in the fusion, and it is initialized with the current global estimation result. The threshold of distinguishing occlusion is obtained by simulation, and the experimental results show that the proposed method can improve the estimation accuracy and robustness of the system under occlusion.