A Comparative Study Of Typical Movements Joint Angles Of Elementary Changquan Routine No.3 Based On Two-Dimensional And Three-Dimensional Motion Capture

Propose:With the continuous advancement of current motion capture and motion analysis equipment,many sports programs use the most advanced analysis equipment to record and quantify motion patterns,thereby providing more intuitive evaluations than visual observation alone.The integration of sports programs with three-dimensional and two-dimensional skeleton recognition technology provides professional and reliable data in competition and training.At present,3D skeleton recognition technology has the advantages of high accuracy and fast sampling rate,but due to its high price,labor and time costs,and dependence on strict laboratory conditions,there are many restrictions in its application.2D video analysis technology uses readily available cameras to capture the movements of performers and loads the videos into a computer system for comparison with standard videos.It is real-time and efficient and is being adopted by more and more sports programs.There is little research on the evaluation of kinematic effectiveness using 2D posture estimation throughout the entire movement process,and it does not include martial arts routine movements.The integration of martial arts routine movements into feature recognition is a relatively new application.Due to the complexity of the movements and the coordination of hand-eye-body-step during practice,which causes the body’s own movements to be obscured,it increases the difficulty of capture and recognition.At present,the research on 3D and 2D motion capture in martial arts routines（Changquan）is still in the exploratory stage.Therefore,this study will explore the performance of posture estimation in identifying the accuracy of Elementary Changquan Routine No.3,determine the relationship between them,and derive the corresponding difference intervals.Compared with 3D,The recognition accuracy of the video capture effect in 2D compared to 3D motion capture system,and the degree of matching of2D joint parameters with 3D joint parameters.Through comprehensive analysis and balanced use of two-dimensional and three-dimensional equipment,it is expected that two-dimensional can be widely applied and achieve good analysis results.Methods:A total of 12 experimental subjects were recruited for this study,and the screening criteria were level 2 and above.Taking the Elementary Changquan Routine No.3 as an example,the in-plane joint movement angles of six movements,namely,empty stance and block palm,empty stance and low side punch,cross-legged stance and windmill smash,cross-legged stance and downward punch,crouch stance and palm blocks over head,crouch stance,windmill and chop punch,were selected as the test movements for this study.The study was divided into two parts.In the first part subjects performed two-dimensional and three-dimensional movement tests to obtain parameters of the spine,elbow,Shoulder,hip,knee and other joints.The second part is the comparison of two-dimensional and three-dimensional measurement results,and the Pearson correlation coefficient and paired-samples T-test are used for analysis to compare the differences between the two systems.The third part is the validation of the two-dimensional recognition effect,using linear fitting and binomial method to fit and correct the two-dimensional measurement results with the three-dimensional system measurement results as the gold standard.Result:Analyzing the recognition accuracy and correlation of Elementary Changquan Routine No.3 pose angle using 2D posture estimation and 3D motion capture.（1）The recognition efficiency of nine joints of empty stance and block palm was 78%,upper limb joints 44%>lower limb joints 33%>unrelated joints23%.Upper extremity joints had a moderate correlation（r=0.642-0.771）,and lower extremity joints had a moderate-to-strong correlation（r=0.718-0.959）.（2）In the empty stance and low side punch movement,the recognition efficiency of nine joints is 66%,the proportion of upper limb joints,lower limb joints and unrelated joints is equal.The correlation was moderate to strong（r=0.598-0.866）.（3）In cross-legged stance and windmill smash,the recognition efficiency of nine joints was 78%,upper limb joints 56%>lower limb joints 22%=unrelated joints 22%.There was a strong correlation in the upper extremity joints（r=0.897-0.953）,and only the left hip had a moderate correlation in the lower extremity joints（r=0.641）.（4）The recognition efficiency of the nine joints in cross-legged stance and downward punch was 67%,the upper limb joints 45%>unrelated joints 33%>lower limb joints 22%.Upper extremity joints had a moderate to strong correlation（R=0.580-0.844）,and lower extremity joint angles had a general correlation（R=0.583）.（5）The identification efficiency of the nine joints in crouch stance and palm blocks and crouch stance,windmill and chop punch over head was 89%,and that of the upper limb joints was 56%>that of the lower limb joints was 33%>that of the unrelated joints was11%.For example,there was a moderate to strong correlation（r=0.692-0.908）in the lower extremity joints and a moderate to strong correlation（r=0.634-0.796）in the lower extremity joints.There was a strong correlation（r=0.728-0.915）between the upper limb joints and the lower limb joints（r=0.632-0.811）.（6）Compared with three-dimensional capture,the joint angles of the original recognition in two-dimensional capture are slightly higher,but after fitting,all joint angles have a high degree of matching and there is no significant difference.The R_max²=0.921,P=0.995 for the linear fit and R_max²=0.672,P=0.954 for the curve fit.the fitted data models still differed only in the spine joint of the false step planting fist,the right arm joint and the spine joint and the right hip joint of the resting swinging fist.Conclusions:（1）The study observed the joint angle recognition of Elementary Changquan Routine No.3 movements by three-dimensional and two-dimensional motion capture technology.In this experiment,the recognition rate of individual movements from high to low was crouch stance and palm blocks over head,crouch stance,windmill and chop punch,empty stance and block palm,cross-legged stance and windmill smash,cross-legged stance and downward punch,empty stance and low side punch.（2）The recognition accuracy of the elbow joint in all movements reached91.6%,and the recognition accuracy of the arm and hip joints was 80%-90%,while the recognition accuracy of the knee joint was poor（below 70%）,It is because the poor visibility of this joint leads to lower recognition accuracy,while all other joint angles show more consistent results with 3D capture recognition and analysis.In general scenarios,2D video analysis is an effective alternative to three-dimensional motion capture.（3）Using 3D recognition as the gold standard,in the fitting of 2D motion recognition accuracy,linear fitting and curve fitting have good calibration effects.After calibration,the mean,standard deviation,and paired sample T-test of the two-dimensional recognition joint angle have all improved.