Capture And Recognition Of Human Posture And Motion Based On Video Image

In recent years,human posture detection techniques that detect the position of human joints in video images to identify human posture,as well as motion capture and analysis of limb movements based on real-time tracking of human body parts based on joint point information,have important application value and practical significance in fields such as human-computer interaction,sports training,and safety monitoring.However,there are still some shortcomings in the implementation of human posture detection,human motion capture,and human motion recognition based on video images in terms of environment,data volume,accuracy,and real-time performance.Existing technologies need to be further improved in terms of algorithms and data acquisition.Based on the above,this thesis has done the following work:(1)This thesis propose a 3D human pose domain adaptive detection method based on Mediapipe by velocity threshold method of joints,human tilt statistics method,limb proportional simulation method,and numerical filtering method of one-euro filtering and mean filtering.In the process of 3D human pose detection,Mediapipe solved the problems of inaccurate recognition of 2D coordinates of human pose due to changes in lighting environment,inaccurate detection of depth Z-value in 3D human pose due to body movements and human inclination,pulse interference due to rapid changes in nodal velocity over a short period of time,and uniform noise due to unstable values over a long period of time,respectively.Experimental results show that this method has the advantages of strong adaptability to different human posture detection,good robustness to environmental changes,and high detection accuracy for 3D human posture.(2)This thesis proposes a method for realizing human motion capture through joint local coordinate system transformation based on quaternion method.The input of the system is the coordinate of human joint points adaptively detected in the 3D human pose domain based on Mediapipe.According to the characteristics of the world coordinate system and local coordinate system in the Unity 3D model,the orientation function Look Rotation and the norm function Normal Rotation that can express the joint local coordinate system transformation are established using quaternion,and limit the angle range of the resulting joint Euler angle,the problem of low accuracy of virtual characters tracking real characters in traditional inverse kinematics is solved.This method has the advantages of more accurate spatial rotation of child nodes compared to parent nodes,and more consistent movement of various joints of the human body with human kinematics.A responsive UDP based on communication method is proposed,which solves the problems of high latency and instability in communication between the server and the client.It has the advantages of strong real-time transmission and better security.(3)Aiming at students’ static and dynamic movement recognition in classroom and physical exercise scenarios,a human motion recognition method based on static and dynamic classifiers is proposed.The static classifier inputs the 3D coordinates of the human joint points,and uses the connection between the joint points as the feature of the static action,and realizes the classification and recognition of the static action through twice KNN;The input of the dynamic classifier is the joint rotation angle,and two thresholds are used to delimit the time window of the dynamic action.During this time,the integrity of the action is judged to achieve action recognition and counting.Finally,the abnormal values and noise data in the classification results are smoothed through EMA.The experimental results show that this method has the advantages of high recognition accuracy and good real-time performance.