Research And Application Of Gait Analysis System Based On Multi-source Information Fusion

In recent years,human gait research based on computer vision technology has made great progress in the fields of rehabilitation medical diagnosis,athlete training,and identity recognition.These research results have extremely important scientific significance and application value.However,in practical applications,the existing gait analysis system has different problems such as high price,cumbersome operation,and insufficient test parameters.This article "Gait Analysis System Based on Multi-source Information Fusion" is a project implemented by the author at the Beijing Zhongke Lingrui Innovation Base of the Chinese Academy of Sciences.Currently,the Chinese Academy of Sciences' "Gaitboter Portable 3D Gait Motion Capture and Analysis System" project(hereinafter referred to as“Gaitboter System”)can obtain three-dimensional gait models through gyroscopes,high-definition microphones and pressure sensors,but it lacks gait space parameters such as step length and step width.In response to this problem,this paper proposes a gait analysis method that integrates multi-source information such as gyroscopes,pressure and vision sensors.The image data of the gait process is collected through the camera,and the depth network model and the derivation method of spatial distance are used to obtain the step length.Gait data such as gait,step width,etc.,with low price and simple operation to supplement the parameter integrity of the Gaitboter system,and provide a low-cost,high-precision,comprehensive parameter solution for the direction of gait behavior detection.The main research work includes:(1)Establish the geometric model of the camera imaging based on the conversion of the coordinate system,and use the visual equipment to collect the gait video information,and carry out the camera calibration experiment based on the Zhang's calibration method under the ROS system to obtain the parameter matrix of the camera.The key to the vision-based gait parameter calculation method is to extract the coordinates of the key points of the human body in the image and calculate the distance of the key points at different times.For this reason,two optimization algorithms are proposed.(2)Use a monocular camera to record gait data,predict the position of the left and right foot shoes through two target detection models,use the center point of the shoe prediction frame as the key point of the foot,and integrate multi-source information to provide the human body support phase time period of the foot For the image frame at a still moment,the distance of a specific key point at different moments is calculated on the basis of camera calibration,and the step length and step width parameters are obtained by the monocular estimation ranging algorithm.(3)In order to improve the accuracy of experimental data and reduce computational complexity,a depth ranging algorithm based on Kinect depth camera is proposed.The human body pose estimation algorithm Open Pose is used to extract 25 bone key points of the human body,and the corresponding image frame is provided by the fusion algorithm in(2),and the Euclidean distance between the depth values of the key points is directly calculated to obtain the required parameters.In the test experiment,by comparing the calculation and analysis of the gait video with the manual calibration method,taking the walking distance as a reference,the calculation errors of the step length,step width and other parameters are all within 5%,which achieves the system design goal.