Abstract: To address the problem of jitter and missing human joint data due to Kinect V2’s own error and joint occlusion, a method is proposed to integrate human kinematic features with an improved adaptive Kalman filtering algorithm. We introduced the filter convergence criterion and skeletal distortion coefficient into the adaptive Kalman filtering algorithm to reduce the computational effort of the algorithm and accelerate the convergence of the adaptive parameters, and combined the human skeletal length invariance and motion continuity to obtain the a priori coordinate measurements of the occluded joints, and then substituted the improved adaptive Kalman filtering algorithm to obtain the relocation coordinates of the occluded joints. The experimental results show that the method can meet the user’s real-time requirements and effectively improve the accuracy of human joint data.