Multimedia Processing
and Communications Systems
Computing Japan visits the Multimedia Systems Laboratories of Fujitsu Laboratories, Ltd.
For this month's R&D Focus, Computing Japan visited Fujitsu's Multimedia Systems Laboratories in Kawasaki. Senior Editor Steven Myers reports on selected research projects of the MSL's Media Processing Lab and NetMedia Lab.by Steven Myers
The multimedia systems laboratories, located near the head office of Fujitsu Laboratories, Ltd., in Kawasaki, is one of five major research centers that make up Fujitsu Laboratories, a wholly owned subsidiary of the famous parent company. Fujitsu Laboratories was created in 1962 by merging R&D sections that previously had been managed by separate technical divisions within Fujitsu. Today, the labs employ almost 1,600 persons and conduct research and development in fields ranging from high-end multimedia systems to low-level electron devices and materials.
The Multimedia Systems Laboratories, three other labs in Japan, and one lab in California round out Fujitsu Laboratories. The others are the Personal Systems Labs (Akashi), Electronic Systems Labs (Atsugi), Electron Devices and Materials Labs (Atsugi), and Fujitsu Laboratories of America, Inc. (Santa Clara, California). In addition to these five laboratories, Fujitsu also manages the Institute for Social Information Science (Numazu), which conducts research in areas such as biological information science and computer mathematics.
The Multimedia Systems Laboratories
Computing Japan's visit to the Fujitsu Kawasaki facilities was arranged by Mike Beirne, who is in the public relations department of Fujitsu Ltd. at the company's headquarters in Chiyoda-ku, Tokyo. Our thorough tour of the expansive Multimedia Systems Labs, which included numerous demonstrations, was capped off by a visit to the "museum" located in the head office building of Fujitsu Laboratories. The museum contains exhibits of many of the famous products and technologies developed by Fujitsu over the past fifty years.The Multimedia Systems Laboratories' facilities are enormous; it was difficult to grasp the scope and scale of the work conducted there based on a single visit. The facilities are broken down into a dozen or so smaller labs, including the Media Integration Lab (natural language processing, optical character recognition, and multimedia databases), NetMedia Lab (software agents and Internet applications), Media Processing Lab (video servers and image processing), and System LSI Lab (LSI chips for image processing). All of the labs work to develop technologies in their respective areas, with an overall goal of producing systems for multimedia processing and communications.
In visiting research labs, it is not an uncommon situation that much is talked about, but little is shown. This was definitely not the case at the Multimedia Systems Labs. The ability of Fujitsu's researchers to explain their work and their willingness to demonstrate the systems they have built were impressive. At each lab, we were first given a thorough introduction to the research goals for the projects underway, then shown an impressive array of demonstrations of the experimental systems built by the project teams. In addition to supplying us with papers and technical reports, all of the researchers gave high-quality presentations and were quick to answer any questions about their work.
Described here are just three of the many projects we saw demonstrated at the Multimedia Systems Laboratories.
AIMS
Drawing upon its vast experience in developing technologies for VLSI chips, storage devices, ATM (asynchronous transfer mode) equipment, and software systems, a group at the Media Processing Lab is currently at work building a scaleable multimedia server. This server eventually will be used for advanced services such as VOD (video-on-demand), teleshopping, and remote learning. The immediate goal for this team is to develop an interactive visual information system that can distribute high-quality MPEG-2 videos to clients over an ATM network.The AIMS (ATM-based Interactive Multimedia Server) system utilizes an architecture that provides a direct connection between the disks controlled by the server and the ATM interfaces. With this massively parallel architecture in place, 20 programs at speeds of 6MB per second each can be delivered through an ATM port. The system is fully scaleable; the size and speed can be increased or reduced simply by adding or removing of unit modules. (Each unit has the specifications shown in figure 1.) It is noteworthy that the system can handle up to 350 simultaneous streams per unit at a speed of 3.3MB per second.
The AIMS system uses an MPEG-2 decoder card implemented with a newly developed Fujitsu video decoder VLSI chip, the MB86365. This new card conforms to PCI bus specifications and is fully compatible with the MPEG-1 and -2 standards. (Figure 2 depicts a block diagram of the decoder.) Using a SPARClite RISC processor, the card provides flexible support for the MPEG system layer syntax. Although the AIMS server is currently being used only for experimental purposes, the team expects that the system will have a strong impact on broadcasting and communications in the near future.
ISHTAR
Another team at the Media Processing Lab is developing a computer vision system that can recognize the motion of objects. ISHTAR (Image Sequence Hardware for Temporal Analysis in Real-time) uses an optical flow algorithm for calculating the velocity of a moving object. Although the algorithm is relatively simple to implement, it has a prohibitively high computational cost on traditional systems. The ISHTAR team has therefore designed and built its own linear array processor, one that directly processes TV images in real time and thus allows for a more reasonable implementation of the optical flow algorithm.The actual system consists of a video I/O board and several DSP (digital signal processing) boards. The video board digitizes NTSC analog signals from a video camera or VCR, separates the image and synchronous data, and transfers them to the DSP boards. Each DSP board contains two TMS320C40 DSP chips and can be connected in tandem with other DSP boards to build a linear array system.
One of the leaders of the ISHTAR project, Shigeru Sasaki, explained that in several experiments the system has proven capable of calculating velocities for image sequences in real time. In one of these experiments, the team used a version of ISHTAR consisting of a video board and four DSP boards to successfully do real-time calculations of the velocity of shifting and rotating objects. Sasaki went on to explain that Fujitsu is looking at applying the technology developed by the ISHTAR team to a variety of fields, most notably traffic surveillance.
DUET
The DUET team, headed by Ichiro Iida of the NetMedia Lab, aims to develop an agent-based personal communications system that integrates several diverse network technologies into a single architecture that is invisible to users. Under such an architecture, users would be able to employ any type of telephone, PDA (personal digital assistant), or PC to communicate with other persons on the network, regardless of the device types used by the receiver. The details of each physical network are encapsulated in a "service control network layer."Cooperation among network nodes in the service control network is managed by a group of Personal Agents (PAs). Each user has his or her own PA, which maintains a set of information about that user, including location and preferred method of interface. To send information, a user need only specify the ID number for another user. The system then contacts the PA corresponding to the receiver, retrieves the information for the receiver, and converts the data to be sent into a suitable format.
Rather than having the user manually register his or her ID with the exact devices to be used, the goal is to have this task be performed dynamically by the network. One solution proposed by the group is to have each user carry a small wireless personal device that transmits the user's ID by radio signals. The network will then determine the person's location by polling the devices in each area, and automatically register the user to the appropriate devices.
Iida demonstrated some of the features of this architecture for us using a prototype system developed by the team that allows users to move around and perform different communication activities using PDAs and PCs. In one example, an Apple Newton with a wireless connection was used to send e-mail and make telephone calls to other users on the network. Iida showed us the information stored by the system for the users, and demonstrated how the information is updated automatically when a user changes his/her physical location. Iida notes that this type of architecture is especially suited for advanced communications networks that must integrate real-time voice and non-real-time messaging services.
Contact information
Fujitsu Laboratories Ltd.
Phone 044-777-1111
Fax 044-754-2580
