Measure athleticism from video

In 20 recreationally active men, jump heights from an iPhone app using high-speed video differed from a force plate by 1.1 cm on average, almost perfect agreement (intraclass correlation coefficient 0.997).² In 12 highly trained sprinters, split times over 40 meters from an iPhone app using high-speed video were nearly identical to the timing gates (r = 0.989 to 0.999).³ An iPhone app's change-of-direction times for 20 adolescent athletes closely matched timing gates (r = 0.964).⁴

Markerless systems track joints from video, without the body markers that lab motion capture uses. Across 20 studies of jumps, markerless knee angles differed from lab motion capture by 4.4 degrees and ankle angles by 4.9 degrees (root-mean-square error).⁵ Markerless jump heights came out 2.9 cm lower than lab measurements on average.⁵ In basketball players' jumps, a markerless system measured knee and ankle bend accurately, but its hip angles differed from lab motion capture.⁶

In three athletes filmed at 25 frames per second, how far each joint moved matched lab motion capture closely (r = 0.916 to 0.994), but accelerations did not (r = 0.232 to 0.677).⁷ Detecting key events such as foot contact required video at 100 frames per second or more.⁸ An error of 20 milliseconds in timing foot contact changed the knee angle measured at that moment by up to 20 degrees.⁸

Each pilot films jump and sprint tests at 100 frames per second or more. A ConductVision engineer measures sample frames by hand to check the jump and sprint numbers, and timing gates add a second check where the academy has them.