Research On The Recognition Method Of Human Posture Based On Depth Camera

The recognition technology of human posture is one of the important research directions in the field of computer vision,and is widely used in fields such as humancomputer interaction,security monitoring,and medical rehabilitation training.However,due to the time-varying nature of human movements and the complexity of application scenarios,the accuracy and robustness of human pose recognition algorithms are still not ideal.How to accurately capture and recognize human poses in real-time has become a current research hotspot.This article is based on the Kinect V2 platform and conducts research from three aspects: joint data filtering,motion recognition algorithms,and motion evaluation analysis.The main research content of this article is as follows:Firstly,aiming at the problem of large estimation error of human joint data collected by Kinect V2 camera in complex environment,a joint relocation method combining the improved adaptive Kalman filter algorithm with human kinematics features is proposed.By introducing filter convergence criteria and bone distortion coefficients,the computational complexity of the algorithm is reduced and the convergence speed of adaptive parameters is accelerated;Combining the invariance of human bone length with motion continuity to obtain prior coordinate measurements of occluded joints,and introducing filtering algorithms to obtain more accurate repositioning coordinates of occluded joints.The experimental results show that this method can effectively improve the accuracy of Kinect V2 in capturing human joint data and meet the real-time needs of users.Then,aiming at the problem that existing motion recognition methods have poor ability to distinguish similar actions executed at different speeds,a human posture recognition method based on dynamic time warping algorithm is proposed.Firstly,on the basis of capturing joint position sequences in Kinect V2,the relative position features and motion trend features of joint points are extracted,and weights are set in the similarity measurement function to combine the two features;Then,the early termination technique is introduced to search the optimal path,which reduces the computation of the distance matrix between sequences,and reduces the search process of the invalid optimal path;Finally,ranking the importance of joint points reduces the amount of data that needs to be processed.The algorithm performance tests were conducted on both public and self built datasets,and the experimental results showed that the proposed method has good accuracy and real-time performance.Finally,an indoor fitness assistance system based on Kinect V2 was developed.Firstly,collect human joint position data through Kinect V2 cameras and establish a standard action library;Secondly,match the real-time actions of users with those in the standard template library to identify the types of user actions;Once again,based on the differences between user actions and standard action libraries,provide improvement suggestions for user action amplitude and action speed;Finally,a dangerous action alarm system was established to monitor the safety status of users.The experimental results show that the system can accurately identify and analyze user actions in real-time,with both intelligence and security.