Research On Human Joint Angle Recognition Based On Dual Azure Kinect

With the rapid development of computer vision science,in today’s society,many people’s bodies are often in a state of physical sub-health,in order to achieve disease prevention and timely rehabilitation treatment,the medical field of computer vision technology means is increasingly prominent demand.Since medical diagnosis and sports rehabilitation in the medical field have strict requirements on the measurement accuracy of human bone joints,the recognition technology of human bone key points in machine vision has become one of the hot research fields.As a single depth camera has problems such as insufficient precision and easy data point jitter in the identification of key points of human bones,and Range of Motion(ROM)is a common and important evaluation index in clinical medical diagnosis and rehabilitation,Therefore,this paper proposed a human bone key point recognition method based on double Azure Kinect to carry out human ROM calculation,and compared the recognition accuracy with the single Kinect human bone key point recognition method based on depth image,which is of great significance for future medical diagnosis and sports rehabilitation.The work of this paper is as follows:First of all,aiming at the deviation of the image acquired by the camera,this paper uses Zhang Zhengyou calibration method to complete the calibration of the Kinect camera,calculates the internal parameters and distortion parameters of the camera,and corrects the image.Secondly,in view of the low efficiency of the traditional human bone key point detection technology,this paper uses a single Kinect camera to obtain the depth image of the human body,and uses the human bone tracking algorithm to identify the human bone key point.Then the key point data is smoothly preprocessed,and Canvas tool is used to overlay the key point data obtained by Kinect and color images.Finally,through the experimental analysis,it is found that the single camera identification method has a certain applicability in the identification of key points of human bones,but there will be poor fitting phenomenon in the real-time tracking of the node position,the situation is not ideal.Then,aiming at the problem of unstable recognition of human skeleton key points by single camera,this paper proposes a point cloud-based dual Kinect human skeleton key points recognition method.Firstly,real-time synchronization between two cameras was carried out,and depth and color information of human body was obtained by Kinect camera to generate point cloud.Then,point cloud denoising and matching are carried out,and point cloud network is used to identify key point information of human skeleton.Finally,through the experimental analysis,it is found that the dual-camera recognition method has better accuracy and real-time performance in the recognition ability of human bone key points.Finally,the principle of ROM measurement is introduced,and the optimal distance and height experiment of depth camera for human body measurement is completed.The optimal distance X=2.1 and the optimal height Y=1.2 of Kinect camera distance measurement target are found.Then,the ROM values of shoulder joint and elbow joint under different posture angles were measured and calculated respectively.Finally,the quantitative analysis showed that the recognition performance of the dual camera was superior to that of the single camera,which is of great significance for the future research on the recognition of the key points of human bone.