In the field of human-computer interaction,the position information of human hands is often directly used in the interpretation of interactive instructions and the calculation of interactive results,therefore,high-precision real-time hand position detection is an important basis for non-contact and natural human-computer interaction.In order to solve the problem of fluctuation and error in the 3D coordinates acquired by Kinect 2.0,a hand position detection algorithm based on mean processing of correlation points is proposed in this paper.Based on the depth information,the hand image is segmented using the human wrist as the segmentation threshold,and the human hand position information is processed by spatial and temporal average of the related points to improve the accuracy of position detection.The experimental results show that the algorithm based on correlation points mean processing is effective and the detection error is less than 5 mm,which can meet the basic requirements of human-computer interaction.