Human Motion Detection And Gait Analysis Based On Depth Image

With the progress of the society and the increasing attention for people drawing on their health,also with the diverse demand for social medical resources,the low-cost diagnostic equipment,which can be easily used in community and family,has been an important complement for the hospital care.Human gait can reflect some physical,psychological and medical condition,so the gait analysis is an effective way to evaluate the human health situation.However,existing gait analysis system is expensive to buy and complex to operate.For the low-cost gait analysis system based on the depth image,in the paper,four aspects of foreground extraction,target tracking,joint positioning and gait analysis have been researched based on the method of unmarked point of human motion detection and gait analysis.The main work and innovation of the paper has been arranged as follows:(1)According to the characteristics of Kinect depth image,ViBe algorithm is introduced and improved to detect human motion on depth image.Considering the motion near the ground is difficult to detect due to the continuity of pixel values of depth image,a modeling method based on adaptive hierarchical image processing and different neighborhood patterns is proposed.And a reference model for removing the “ghost” phenomenon is raised.A foreground detection process is added in pixel classification,which can remove the “ghost” phenomenon through comparing the current pixel with the reference model.In terms of model update,a foreground pixels based background model update strategy is adopted to solve the “shadow” phenomenon problem.(2)For the specific target tracking of the multiple moving targets,Camsift algorithm is introduced and improved.The algorithm increases the adaptive,limited range of pixel histogram model update strategy to solve the problem of targets tracked hopping and occlusion tracking failure.To solve the problem of locating box migration,Kalman algorithm is introduced.When the positioning results' offsetting goes too far,replace detecting position with the predictive position and update the histogram model by reducing the sampling range.Finally,by target location obtained to optimize ViBe algorithm's processing result,achieve the extraction of the target human body.(3)In order to solve the problem of human lower limb joints positioning,a multi joint skeleton model is established.Firstly,the presence of occlusion phenomenon of the knee joint and ankle joint is judged depending on whether the front and rear legs contains continuous background pixels.Then an adaptive threshold method is used to distinguish the front leg from the rear leg.Finally,the scaled method and edge positioning method are used to extract lower extremity joint.(4)In order to extract the maximum curvature angle indicators from the location results of the joint.Firstly the least square method is introduced.The discrete data set,knee joint difference and knee joint angle changes are fitted using the least square method.By analyzing these trajectories,the gait parameters which contains gait period,walking speed and so on can be obtained.Finally,the evaluation index is given.In the positioning experiment of the human lower limb joint,the positioning results of the algorithm in this paper has been compared with the manual calibration and the mean value of the positioning error is within 1% of the height of the human body.The accuracy of health evaluation index is verified by comparison between normal and abnormal gait.