Research On Markerless Human Gait Capture Method And System Based On Depth Cameras

Human gait capture has extensive and important applications in many fields such as medical rehabilitation,physical training and interactive control.This paper proposes an markerless human gait capture system based on depth camera to effectively improve the accuracy and stability of the markerless human gait capture system in view of the complexity,high cost and inconvenient operation of the current marked gait capture system,and the fact that the accuracy and stability of the markerless human gait capture system cannot meet the requirements of relevant fields,so as to truly apply the human gait capture analysis to relevant fields and improve the rehabilitation and training effects of users.This paper mainly studies the accuracy and stability of joint localization of markerless gait capture system based on depth camera.At present,the stability and accuracy of the markerless motion capture method,which use machine learning to process and locate the joints in deep images,can only meet the requirements of attitude recognition and mild interactive control,and still cannot meet the requirements of specific gait parameters acquisition.In view of these conditions,a gait capture method was proposed to reconstruct the personalized lower limb 3D model with two depth cameras,and to improve the accuracy and stability of joint positioning through the registration of the model and the actual point cloud.Specific research work is as follows:(1)Aiming at the problem of limited view angle and complicated operation in external parameter calibration of multi-depth camera,a new calibration method based on digital model registration is proposed.The calibration target of this method adopts plane hollow-out geometric pattern,which is easy to make and can calibrate multiple depth cameras at the same time.The digital model of the calibration target was taken as the intermediate registration target,and the real point cloud of the calibration target was registered.Then,according to the feature point pairs of the digital model,the external parameters of the camera were solved,which effectively reduced the reduction of the calibration accuracy in the direct registration process of the real point cloud due to the absence of the corner point and boundary of the point cloud.The validity of this method is verified by test,which provides a convenient,fast and accurate scheme for depth camera calibration.(2)In order to solve the problem that the posture of the human body should remain unchanged and the human body should be reconstructed by means of the turntable,a method of segmental three-dimensional reconstruction of the human body is proposed.In this method,the point cloud of human body is divided into several independent parts of human body that can be regarded as rigid body according to the joint,and then KinectFusion algorithm is used to reconstruct each part of human body onto the basic frame to form a complete three-dimensional model,to avoid the failure of reconstruction caused by the body posture mismatch between the front and rear frames during the rotation of the body.According to the actual verification,the human body only needs to rotate in front of the depth camera to complete the 3D model reconstruction.(3)Aiming at the problem of joint location instability,jitter and dislocation in markerless human gait capture system based on machine learning,a segmental model registration method for human joint location is proposed.This method can be used for joint localization by embedding a reliable human skeleton in the lower limb model.Then the model and the real human point cloud were segmented to locate the location of the joint,which improved the accuracy and stability of the location of the joint.At the same time,compared with the traditional model-based markerless gait capture,the problem of human posture estimation and model deformation can be avoided by using segmented model registration to locate joints,which effectively reduces the computation amount and complexity,and ensures the accuracy of joint positioning results.Finally,a series of experiments were conducted to verify the improvement of accuracy and stability of joint localization by the markerless gait capture method proposed in this paper.