A three-layer framework is proposed for cooperative distributed face tracking. In this framework, multiple distributed cameras cooperate both for stable face tracking and for effective registrations. The framework consists of an independent task layer, a local cooperation layer and a global cooperation layer. In the independent task layer, each camera asynchronously works as an agent to track and to register a face. In the local cooperation layer, a set of agents forms an agency and it cooperatively tracks a particular person with estimating the 3D face location. A concept of asynchronous cooperative tracking is introduced for all the agents to independently track the target person and to utilize the integrated 3D information for the effective tracking. In the global cooperation layer, the assignments of agents to persons are dynamically changed in order to effectively track and register multiple persons. We call this mechanism a dynamic agency configuration. Each agent is transfered ...