Close-range Human Motion Analysis Based On Joint Points By Kinect

The analysis of the human motion state has always been a research hotspot in the field of artificial intelligence,and its application is also extremely extensive.The traditional analysis of the human motion state generally uses a color camera as an input device to obtain the motion information of the human body through analysis of color images.With the appearance of more and more depth camera devices represented by Kinect,the input device can acquire depth information while acquiring a color image,thereby greatly improving the detection and segmentation accuracy of the human body region.Therefore,human motion analysis based on depth information has important research value.In this paper,Kinect 2.0 is mainly used as a hardware device to analyze human motion in close range.The main work is as follows:1.For the problem that there is a lot of noise in the depth image obtained by Kinect,the bilateral filtering algorithm is improved,and the preprocessing of the Kinect depth image is achieved,and good results are obtained.2.There is a problem of offset and jitter in the joint point information acquired by Kinect.The region circle limitation method is improved,and the true position of the offset joint point is estimated.The Kalman filter algorithm is used to repair the joint point jitter..3.Based on joint point-based motion recognition,we propose a close-range motion recognition method based on 25 joint angles of the human body combined with dynamic time warping.Tests were performed on MSRAction3D datasets and self-produced datasets,achieving an average recognition rate of 84.7%and 87.5%,respectively.In the experimental stage,the joint points in the 567 motion sequences in the MSRAction3D data set were identified and repaired,and an average recognition rate of 84.7%was obtained.For the "High arm wave" action,40 sets of Kinect raw data and 40 sets of data after depth image preprocessing and joint repair were recorded,respectively,and the recognition rates of these two sets of actions were 82.5%and 87.5%,respectively.The experimental results show that the preprocessing of the Kinect depth image and the repair of joint points can help to improve the stability and accuracy of the motion recognition.