Reliability and Accuracy Analysis of Depth Camera from AP analysis during Deadlifts

The purpose of this study is to analyze the kinematic variables of joints during the deadlift using a depth camera and to verify the reliability and accuracy of sagittal plane angles. 93 male and female students(male-age: 24.41±2.15 years, height: 174.93±4.70 ㎝, weight: 71.88±6.97, BMI: 23.42±1.99 ㎏/㎡; female–age: 22.2±2.34 years, height: 163.61±4.98 ㎝, weight: 57.17±6.39 ㎏, BMI: 21.26±1.79 ㎏/㎡) participated in this study. Total 2,232 data of deadlift were collected to analyze the reliability and accuracy of the two camera data. Joint angle was calculated from position data, the analysis method used the Pearson correlation coefficient for reliability analysis, Root Mean Square Error(RMSE) and Normalized Root Mean Square(NRMSE) for accuracy analysis and max/min difference of angle between two cameras. Significant differences in sagittal plane joint angles were observed between the depth and infrared cameras, with the infrared system reporting larger values for the hip (F = -26.209 to -27.922, P = 0.000) and knee joints (F = 25.919 to 27.541, P = 0.000). Strong correlations were found for the shoulder (r = 0.84–0.96), hip (r = 0.79–0.97), and knee joints (r = 0.91–0.94), whereas the elbow showed poor correlations (r = -0.05–0.30). RMSE values ranged from 6.83° to 35.45°, with the highest errors in the hip joint. NRMSE was lowest in the knee (16.16–28.86%) and highest in the shoulder (14.46–53.50%), indicating joint-dependent measurement reliability. In static movements, continuous jitter occurs due to the characteristics of the depth camera, and analyzing the data using time-series curves or patterns would provide a more accurate assessment.