Background and Basic Idea

Roughly speaking, the history of the computer vision research can be summarized as follows (Fig. 3):

-1970s:

Image Processing: Image Image
A given input image is transformed into an output image to enhance its quality and to detect image features.

1980s:

Computer Vision: Image Scene
Recover 3D scene information from observed 2D image(s) based on geometric and photometric models of the imaging process.

1990s:

The following two disciplines are being studied:

1. Active Vision: Computer Vision Physical Action Intelligent Tasks
   Integrate visual perception and physical action to perform intelligent tasks.
2. Image Media Processing:
   Computer Vision Computer Graphics Versatile Visualization
   Integrate image analysis and synthesis to realize versatile visualization.

  
Figure 3: History of computer vision research

The key idea of CDV is to introduce network communication capabilities into active vision and image media processing systems. That is, a CDV system incorporates a group of network-connected cameras (with processors) to augment and improve capabilities of computer vision and visualization systems in the following points:

Dynamic Wide Area Image Capturing and Scene Understanding

: A group of network-connected cameras are distributed over a wide spread area to capture its multiple different views simultaneously and then understand dynamic situations and events in the global scene.

Robustness

: Multiple information from different cameras is integrated to increase the accuracy and reliability of image analysis.

Flexibility

: Cooperative camera group organizations are dynamically formed to cope with various situations in the real world.

Smooth Degradation

: Cameras sharing views compensate each other in case of failures.

With the introduction of network communication capabilities, CDV systems are endowed with three functions of Visual Perception, Action, and Communication. To study the integration of these three functions seems reasonable and timely;

• While active vision[1] integrates visual perception and physical action, no mechanism of communication between multiple cameras is taken into account.
• In DVMT (Distributed Vehicle Monitoring Testbed)[2], on the other hand, perception and communication are integrated to understand dynamic situations in a wide spread area. However, it lacks action; no active sensing mechanism is supported.

We believe the concept of CDV offers a fundamental scheme of computer vision systems in the 21st century.