Japanese Researchers Develop 'Real-Time' 3D Image Rendering System

A Japanese research group prototyped a system that renders 3D images by using pictures taken by 64 cameras. It takes only 0.5 seconds for the system to capture images and display a 3D image. The group consists of researchers from Hitachi Ltd and associate professor Takeshi Naemura of the Graduate School of Information Science and Technology, the University of Tokyo. The rendered images look three-dimensional to the naked eye. And the operation is "almost real-time," the group said.

In addition, the prototype system has two more major features. One is the capability to adjust displayed images and their 3D effect. The other is the increased mobility of the system. The prototyped 3D image display system used Hitachi's 3D display, which enables viewers to see 3D images with the naked eye, and Tokyo University's array of 64 cameras (8 units lengthways by 8 units sideways). Hitachi had already prototyped the 3D display before. The specifications of the display used are almost the same as the previous model's, excluding the increased luminance, Hitachi said.

The prototype naked-eye 3D image rendering system. On the left is the 3D display and on the right is the camera array. 8 x 8 units of cameras are arrayed on a 700 x 700mm board. The display can output 320 x 240 resolution images. The camera array weighs about 10kg, excluding every component other than the 64 cameras, while the total system, including the camera-mounting board and the rack, etc, weighs roughly 100kg, the group said. Although the workstation cannot be seen in this photo, it was placed on the bottom shelf of the rack under the 3D display.


The display employs "IV (Integral Videography) technology," one of so-called multi-eye image displaying technologies, which use images shot from multiple angles. With an array of micro-lenses mounted on an LCD panel, it displays a 3D image by emitting light in 60 directions from each micro-lens. Accordingly, the display needs 60 angle shots from 60 directions. As a result, the display processes a number of images equivalent to 60 angle shots from the images shot by the 64-camera array. Image processing uses "free angle live image synthesis technology" developed by the University of Tokyo. This technology processes an image by integrating a number of images shot by multiple cameras.


Structure for Integral Videography


This technology enables to display images of the subject and adjust 3D effect. If a person is standing in front of someone and you want to focus on the person in front, for example, the display can show the front person clearly and the other one in the back vaguely. If you want to focus on the person in the back, the display can show the person in front vaguely and the other one sharply.

When adjusting 3D effect, the display arranges the distance between angle positions in image synthesis process. 3D effect will strengthen if the distance is increased, and weaken if the distance is decreased. These capabilities to display the subject and adjust 3D effect are aimed at enabling the display render optimal images by content.

Integrated into a single workstation
The development group emphasizes the prototype system's mobility as its feature attraction. That is because the system can be formed with no more than a multi-purpose workstation, a camera array and a 3D display. The group also made the system movable by placing the display, workstation and camera array in wheeled racks. The development group made it possible to take the 3D image system outside, assuming an application for observing animals at the zoo, for example.

The 64 cameras in the camera array shoot images, compress them into "Motion JPEG" format and input them to the workstation via a 1Gbps Ethernet cable at about 80Mbps. Operation has been accelerated through the parallel processing of those images transmitted from the 64 cameras and synthesized by the GPU in the workstation. The workstation incorporated NVIDIA Corp's "GeForce 8800 ULTRA" GPU, 3GHz "Xeon" dual core microprocessor and 3-Gbyte RAM. When synthesizing images, the GPU allows playback of 256 x 192 resolution images at 7fps, including image data of 60 angles per pixel. One 3D display pixel corresponds to one micro-lens.

The system's commercialization has not been determined yet. Accordingly, the group has not specified expected pricing for the system. But the price of the workstation and the camera array were roughly ¥500,000 (US$4,799, price may vary depending on country) and ¥2 million, respectively, according to the group.

The group is planning to demonstrate the prototype system at the "3D Image Conference," which will take place at the University of Tokyo from July 10 to 11, 2008.

By Tadashi Nezu, Nikkei Electronics