Human Motion Evaluation Method Based On Depth Camera

With the progress of society and the improvement of tangible and cultural level,people pay more and more attention to the needs of daily entertainment and fitness.A human motion evaluation system,which can be conveniently used in the home and in the community,can greatly enhance the fun and experience for the entertainment and fitness groups.Existing methods of human motion analysis are mostly based on professional systems of wearable devices,which are expensive and inconvenient to use.Aiming at the problem of regular continuous human motion evaluation,this paper studies the following four aspects: human target tracking,human skeleton model correction,human continuous motion feature extraction and human motion scoring method.The main work and innovation of this paper are as follows:(1)According to the characteristics of Kinect depth image,the traditional Meanshift object tracking method combined with Kalman filter was introduced and improved to track the moving human body.Firstly,the reasons of "near-object shift" and "far-object shift" when used the traditional Kalman-Meanshift method to track the human body in the depth image which was obstructed by obstacle were analyzed.Aiming at these two kinds of defects,the method of tracking template updating based on the change of depth value and the retracking method based on human target detection and obstacle screening were respectively proposed to solve the phenomenon that the tracking frame shifts when the obstacle obstructed human object,and increase the tracking accuracy of the human object in depth image.(2)Through the joints points' linear motion model,a new method based on Kalman filter predicting to correct the joint point was proposed.Firstly,the problem that the Kinect skeleton model shifts momentarily at some joint points when human limbs crossed each other,blocked or the human body made large movements was analyzed.The linear motion model was established for the easily shifting joint.The predicting position of the output of Kalman filter is used to judge and correct the shift of the joint point so as to get more accurate Kinect joint position.(3)In order to make a more accurate evaluation of human motion,a feature descriptor based on space geometric among human joint points was proposed.Firstly,the relative area of the nine constructed triangles and the angle between the verticaland normal directions of the plane where the triangles were located were taken as features,and then the features are adaptively weighted by the rate of change of the features to obtain the final weighted feature descriptor.Experiments show that the proposed method of feature extraction can describe human motion more accurately than other feature methods.(4)Based on the thoughts of template contrasting,a method of human motion evaluation based on improved DTW algorithm was proposed.Firstly,an improved DTW method considering first-order and second-order differentials was introduced,which can enhance the matching accuracy of feature serials.Then,DTW distance between standard motion and contrast motion and professional man-made scores was used for distance-score correspondence,so as to give the scores of contrast motion.Experiments showed that the human motion evaluation method based on the improved DTW algorithm can give a good score evaluation of the human body motion,and the evaluation accuracy is better than the traditional DTW method.