Research On Pose Estimation And Gait Detection Of Exoskeleton Based On Deep Learning

With the development of assisted walking exoskeleton robot technology,accurate,fast and stable control of it has attracted more and more attention.In order to meet the requirements of cooperative control of assisted exoskeleton robot,it is necessary to accurately obtain the gait information of the assisted exoskeleton robot.The purpose of this paper is to use the deep learning method to realize the pose estimation of the exoskeleton device wearer in the video,and then to obtain the gait detection and gait prediction parameters based on the pose estimation results,and preliminary verification is carried out in the control experiment of the exoskeleton robot.The video-based gait information acquisition method is convenient for the deployment of experimental equipment and does not affect the movement of the experimenter.Therefore,the use of video-based methods for gait detection of exoskeleton equipment wearers has certain research significance.In this paper,a deep learning method is proposed to get the pose estimation of the exoskeleton in the video,that is,to obtain the coordinate positions of the joint points in each frame,and then obtain the gait detection information and predict the gait parameters according to the pose estimation result.The method applies human pose estimation to exoskeleton gait detection.Specifically,this paper mainly focuses on the following researches:1)Research on human pose estimation and gait detection for normal people without exoskeleton equipment,that is,using deep learning methods to estimate human poses in the video data obtained by cameras,and then obtain correlations based on the human pose estimation results.The gait parameters obtained by this method were compared with those obtained by the sensor-based method.The mean value of the relative mean deviation of each joint was 8.61%.The comparison results showed that the gait parameters obtained by the two methods were similar.2)The pose estimation method based on deep learning is applied to the exoskeleton robot,and the pose estimation results of the exoskeleton robot are presented.Based on the pose estimation results of the exoskeleton robot,the gait of the exoskeleton is detected,and the joint angle parameters and other gait spatiotemporal parameters of the exoskeleton robot during motion are obtained,and are compared with the gait parameters obtained by the exoskeleton platform acquisition system.The mean value of the relative mean deviation of the spatiotemporal parameters of each gait was3.63%.The comparison results show that the gait parameters obtained by the two methods are relatively similar,which verifies that the exoskeleton gait parameters obtained based on the pose estimation method have certain accuracy.3)Dynamic analysis of the exoskeleton robot,and dynamic simulation of the exoskeleton robot using ADAMS simulation software to provide theoretical guidance for subsequent experiments.The gait parameters of the exoskeleton robot during steady motion are predicted,and the gait parameter prediction results of a fixed time interval after the current moment are obtained.Using the gait parameter prediction method,the exoskeleton robot control experiment was carried out.The experimental results show that the gait prediction method can help the exoskeleton device to reduce the user’s leg resistance.The state prediction method is helpful for the cooperative control of exoskeleton robots.