PITTSBURGH, NORTHBROOK, ILL., and PASADENA, CALIF.—
Disney researchers have figured out of a way to search sports footage using the equivalent of hand-drawn play diagrams.
“We have developed a novel query paradigm and retrieval system, which we call Chalkboarding, that allows the user to issue queries by drawing a play of interest—similar to how coaches draw up plays,” the research abstract
The technology was created in conjunction with STATS, Queensland University of Technology and Caltech as an alternative to keyword searching. It leverages the copious data collected by the STATS SportsVu six-camera system developed to track and record the positions of players, referees and the ball—at a rate of 25 times per second. The system also logs passes, shots, fouls and other game elements.
The method is more effective than keyword searching, according to Patrick Lucey, director of data science at STATS, who started working on Chalkboarding while a scientist at Disney Research.
Keyword-searching “three-point shot” in a database of 600 games, for example, might return 20,000 candidate plays, Lucey is quoted as saying. Sorting through 20,000 plays manually would take nearly a full month at 24/7 if the plays were around two minutes each.
“Moreover, obtaining fine-gained information such as the location and motion of the players would require an enormous amount of keywords to capture all of the specifics—a prohibitive task,” Lucey said.
Chalkboarding also can be done over a broadcast feed of a game, using an extra step to calibrate the camera angle.
In a lab implementation of the technology, 10 users versed in basketball play jargon performed eight specific searches using both keywords and Chalkboarding, which yielded a substantially higher rate of precision, as explained in the research team’s abstract.
The Chalkboarding team also included Long Sha of Queensland University of Technology, Peter Carr and Iain Matthews of Disney Research, Yisong Yue of Caltech and Charlie Rohlf of STATS.