Concluding Remarks

This paper described an overview of our five years project on Cooperative Distributed Vision.

After one year research, we have obtained various promising results:

1. From a scientific viewpoint, we proposed a functional dependency model to explore the mechanism of the integration of perception, action, and communication. While this model is not well developed yet, we can get some basic characteristics of vacuous and embodied Active Vision Agents:
   1. The essence of action is the internal state transition of an AVA. The world state transition caused by the physical action should be modeled by the side-effect of the action.
   2. While in vacuous AVAs no communication can be realized without the message exchange, the communication between embodied AVAs is best characterized by multi-channel communication links formed by versatile combinations of perception, action and message exchange processes.
   3. By analyzing the linear dynamic system model used for controlling the soccer robot, the following points become evident:
      • The reciprocal process consisting of the action-driven perception and the perception-driven state-change followed by the action-selection are the essential scheme of the model.
      • To perceive complicated dynamic situations, the internal state should have a certain amount of memory.
   To make the proposed model really meaningful, we should augment it by introducing temporal features, i.e. design a dynamic model.
2. From a technological viewpoint, we have the following results:
   1. We developed the Multi Focus Camera for real time 3D range sensing, whose prototype showed its practical utilities as a compact wide-range real time range finder.
   2. We proposed the Appearance Sphere Camera as a model of active imaging method for observation stations. Sophisticated camera calibration methods have been developed to make an off-the-shelf video camera work as the APS camera. Experimental results proved its practical utilities in generating a wide panoramic image by an active pan-tilt-zoom video camera.
   3. Using the APS camera, we are developing various algorithms for real time object detection and tracking. Experimental results have shown that they can be used in the real world. We plan to employ parallel processors to increase the processing speed and multiple observation stations to realize the cooperative object detection and tracking in a wide spread area.
   4. We proposed a behavior recognition method based on the event driven selective attention mechanism. While the current system is not so sophisticated, versatile behavior recognition will be realized by incorporating multiple observation stations.
   5. We developed cooperative soccer robots and showed that they can learn some primitive cooperative behaviors.
   Intensive works on hardware/software developments, design of real time video processing/generation algorithms, communication protocols for cooperation, and visual learning are being conducted to set technological foundations of CDV systems.

We plan to organize a series of workshops annually to promote the CDV project. In addition to scientific and technological idea exchange and discussion, collaborative liaisons with industries will be established to apply our research results to practical fields. All these project activities are listed on the project homepage (URL: http://vision.kuee.kyoto-u.ac.jp/CDVPRJ/).

Finally, I would like to express my sincere thanks to Miss Hiromi Taguchi for her help to prepare figures in the paper.