Teamwork Activity Recognition

Recognizing teamwork in embodied agents has important applications in areas such as surveillance, training, automated annotation and commentary, as well as improving the ability of robotic agents to participate in human teams. Most of the time, the teamwork of embodied agents have a natural expression in coordinated movement. Naturally, teamwork can have many other aspects, in addition to or replacing coordinated movement; examples are voice commands and wireless communication. However, the focus of our work, is on recognizing team actions which are expressed in a coordinated movement pattern of agents with agents playing specific roles in the team. Examples of teams displaying such patterns are sport teams such as football, basketball or soccer, dance groups, pickpockets in action, groups of animals such as wolf-packs, bodyguards defending a celebrity, military units in the battlefield or a MOUT environment, tank units, air squads, terrorists in a busy marketplace and many others.

This web page presents contributions to the theory and practice of teamwork activity recognition. A particular focus of our work is to improve our ability to collect and label representative samples, thus making the team activity recognition more efficient. A second focus of our work is improving the robustness of the recognition process in the presence of noisy and distorted data.

On this web page we introduce the software tools and datasets we have developed in our teamwork recognition system. We also present the main challenges and applications of teamwork activity recognition systems.