In team sports, single players can sometimes get lost on the crowded field. But a new system can follow multiple players at once, even when they’re buried under a pile of bodies in a football tackle or crouching behind another player. The new tool, developed at École Polytechnique Fédérale de Lausanne’s Computer Vision Laboratory (CV Lab), can track athletes’ movements continuously from the time they enter the field by overlaying their numbers and jersey colors on their computer screen images. These superimposed images make it easier for spectators, referees and coaches to distinguish individuals out on the field, and they can be produced without special chips, extra gear or additional markers. Innovative multicamera algorithms can automatically track multiple basketball players. Courtesy of CV Lab, EPFL. The system consists of a computer and eight video cameras: Two are set up on each side of the field, two look down onto the field from overhead, and two zoom in on players. Software running on the computer uses three algorithms to detect, track and identify the players. The first algorithm divides the field into a grid of squares measuring 25 sq cm each, then removes the background in all images simultaneously. It deduces the probability of the presence of a player in each small square. The other two algorithms connect the series of results obtained from the first to establish individual trajectories. Each algorithm uses global optimization methods, which result in a system that can track people in real time in a reliable manner. The tracking software deduces player positions even in the presence of obstacles, such as other players. “Our technology is nonintrusive, in the sense that people do not need to wear additional markers or devices in order to be tracked,” said Horesh Ben Shitrit, a doctoral candidate in the CV Lab and one of the developers of the algorithms. “We manage to track people for a long period of time and preserve their identities.” Although the scientists are concentrating their tracking on sports players, Shitrit said the technology, with some adaptations, also could track pedestrians to monitor traffic in an area, or follow the movement of clients in a store for marketing purposes. Currently, the team also is involved in a new project for tracking medical staff in operating rooms to optimize the work flow during surgery, he said. Next, the researchers plan to use their trajectories, in addition to other visual cues, to analyze the behavior and performance of teams and their players. “We would like to find patterns in the game’s style of different teams and evaluate different techniques of individual players,” Shitrit said.