Research On Pose Matching Based On Openpose And Its Application In Physical Education

For a long time,the situation of primary and middle school students' physical learning and exercise is not optimistic,especially the basic action learning exercise after school.Due to the lack of convenient action evaluation way,it is impossible to evaluate the situation of numerous students' physical learning after school.With the rise of online teaching,the effective realization of basic sports evaluation has practical significance for improving the evaluation of physical education.Through the research on the posture matching algorithm based on Openpose,this thesis proposes an algorithm that uses the spatial characteristics of human skeleton points to achieve posture matching of basic sports actions,The algorithm can complete the scoring of the action standard degree and identify the deviating limbs in the substandard actions to realize the evaluation of basic sports learning..This thesis proposes the human body's differential posture features and extraction algorithms for basic sports actions.According to the shape characteristics of the human skeleton space,the algorithm extracts the information of the different posture characteristics of the human body.The algorithm utilizes the skeleton point coordinate data obtained through Openpose and standardized processing.According to the spatial geometric relationship of the human body morphology,the human body weight center information and the human body limb angle information are calculated as basic human body posture features,Combining foot direction,shoulder joint vector and hip joint vector to extract vector information of upper and lower body limbs as human body orientation information,and the three kinds of feature information are merged into the human body's different posture features,which improves the discrimination of the posture features in the basic sports actions from similar actions,and is used for basic sports posture matching.This thesis proposes a DTW posture matching algorithm based on human body's different posture features to achieve posture matching and perform action evaluation.The main idea is to calculate the distance between the human body posture features in the video frame to be matched and the template video frame,the search condition of the regular path in the DTW algorithm is changed and quantify it as a score,and the dynamic matching of the frame sequence is realized through the monotonic maximum matching degree search principle,which solves the problem that the length of the action sequence to be matched and the template action sequence are inconsistent and cannot be matched;The algorithm define the limb angle deviation factor and calculate the deviation degree of each limb angle,and the limbs with deviations greater than the set standard are exported,the corresponding limb that affects the overall action to reach the standard is recognized,and finally the evaluation and result feedback of the basic sports actions are realized.Based on the above research,this thesis designs an intelligent the basic sports action teaching evaluation system,the system matches the video of the action to be matched with the standard action of the template library,and obtains and feeds back the evaluation result of the action.The teacher obtains the action learning situation of numerous students through the system,and the action learners adjust their actions through the feedback of the system,thereby achieving convenient interaction in the teaching of basic sports movements.