A multiple camera methodology for automatic localization and tracking of futsal players

• The paper deals with automatic futsal player localization and tracking.
• A method for estimating the player positions as prob. distributions is presented.
• A solution for dealing with multiple cameras is presented.
• Particle filters and appearance models are used to enhance the observation model.
• Results show tracking errors below 70 cm per player.

There is a growing scientific interest in the study of tactical and physical attributes in futsal. These analyses use the players’ motion, generally obtained with visual tracking, a task still not fully automated and that requires laborious guidance. In this regard, this work introduces an automatic procedure for estimating the positions of futsal players as probability distributions using multiple cameras and particle filters, reducing the need for human intervention during the process. In our framework, each player position is defined as a non-parametric distribution, which we track with the aid of particle filters. At every frame, we create the observation model by combining information from multiple cameras: a multimodal probability distribution function in court plane describing the likely players’ positions. In order to decrease human intervention, we address the confusion between players during tracking using an appearance model to change and update the observation function. The experiments carried out reveal tracking errors below 70 cm and enforce the method’s potential for aiding sports teams in different technical aspects.