Hardware and Network Organization of a Prototype CDV System

Fig. 8 illustrates the overall hardware and network organization of a prototype CDV system under development. It consists of the following components:

  
Figure 8: Hardware and Network Organization of a Prototype CDV System

Observation Station

As an practical implementation of a vacuous AVA, we are developing the observation station with the following equipments.

sensor

: As an active pan-tilt-zoom video camera, we have designed and developed the Appearance Sphere Camera. (APS camera, in short. Its characteristics are described in Section 4.2.2.) Since the developed camera is too large to be embedded in the real world (see Fig. 12 ), we use SONY EVI-G20, a compact computer controllable pan-tilt-zoom color video camera, in the prototype system. Sophisticated calibration softwares have been devised to make EVI-G20 work as an APS camera. The calibration has been proved to be enough accurate for practical use[10]. In addition to video cameras, we have developed the Multi Focus Camera for real time 3D depth sensing. (See Fig. 10. Its characteristics are described in Section 4.2.1.) The future system will employ this camera as its visual sensor to get mutually registered optical and range images in real time.

processor

: Currently, we are developing a real time vision processor by augmenting high performance UNIX PC with IMAP-VISION board(s)[4]. This board has 256 SIMD parallel processors with real time video capturing and data transfer capabilities. Using this vision processor, real time 3D scene understanding will be realized.

Ultra High Speed Network

We have developed a PC cluster system, where 9 UNIX PCs with real time video capture boards are connected by Myrinet, an ultra high speed network (1.28Gbps). This PC cluster system can be used for real time multi viewpoint stereo to generate dynamic 3D images. Ordinary 10Mbps and 100Mbps Ether Net and 155Mbps ATM Net are used to connect observation stations in the current system. Wireless LAN will be introduced to link mobile robots.

Parallel Distributed Algorithm and Software

Various different types of parallel distributed algorithms and softwares should be developed to realize a CDV system: real time video image processing, 3D object recognition, 3D image rendering, message exchange for cooperation, and robot control. While the prototype system uses ordinary UNIX softwares, we plan to develop an original parallel distributed software system on the PC cluster.

Mobile Robot Platform

As an practical implementation of an embodied AVA, we have developed a mobile robot with vision, a radio-controlled miniature car with a color video camera (see Fig. 21 in Section 4.5). We plan to develop more capable mobile robot platforms which can cooperate with observation stations.