Validation of video analysis of marker-less barbell auto-tracking in weightlifting

We determined the marker-less barbell auto-tracking accuracy using the Kanade-Lucas-Tomasi (KLT) algorithm in a digital video for two-dimensional analysis (2D-AT). The position coordinates of the barbell's right end during multiple loads (60%-90% of one-repetition maximum) of snatch motion in eight participants were recorded using a three-dimensional motion capture system. Simultaneously, the snatch motion was recorded by a digital camera from the right side. Based on the digital video, the center of the barbell's right end was auto-tracked using the KLT algorithm. Six barbell kinematic variables (forward, backward, and vertical displacement, peak forward, backward, and vertical velocity) were calculated. Intraclass correlation coefficient (ICC) analysis was performed to establish the agreement level between the methods. The Bland-Altman plots and regression were used to examine the agreement between the methods. ICCs of 0.999-0.971 revealed a strong agreement level between the methods. The Bland-Altman plot showed small bias (-0.001-0.001 m, -0.034 --0.005 m/s). We considered that 2D-AT could obtain barbell position coordinates with sufficient accuracy to discriminate the difference in the lifter's level and a successful or unsuccessful lift.