Analysis of speed accuracy using video analysis software

Abstract

Determining an athlete’s speed from broadcast video is a common practice in sport. Many software packages that perform data extraction from video files are expensive; however, open source software is also available, but lacks published validation for speed measurements. The purpose of this research was to examine the error of speed measurements extracted from video during an ice hockey skating exercise using open source software. The subject completed four exercises, at two speeds recorded by broadcast cameras set at five angles. The speeds from the broadcast cameras were compared to speeds calculated from a high-speed camera placed orthogonally to the exercise. Speeds from the broadcast cameras correlated well with the high-speed video for motion more than 12 m away from the broadcast camera. When comparing all the measured speeds, no significant difference was found between the speeds calculated by the high-speed camera (slow: 4.46 m/s ± 0.2; fast: 7.2 m/s ± 0.7) and the speed calculated from the broadcast cameras (slow: 4.50 m/s ± 0.4; fast: 7.34 m/s ± 0.6) (p > 0.05). The open source method was found to be less accurate when the athlete was close to (within 12 m of camera position) or moving directly toward the broadcast cameras.