Coverage from the PMA 2008 (January 31 - February 2). Get a first-hand look at JVC’s and Samsung’s new 1080p camcorders, the wireless SD Eye-Fi card, Delkin’s Image Router and Panasonic’s new camcorders. » View the complete coverage
Manchester (UK) and Vancouver (British Columbia) - Researchers at the University of Manchester and Dolby Canada have developed a way to create "accurate enough" 3D texture models using only a high resolution flash camera and special software. The result is a very fast and inexpensive way to produce 3D surface texture maps which are accurate enough to "fool the eye" most of the time.
Current 3D texture capture technologies employ expensive, time consuming, and often complex and dangerous pieces of hardware (such as lasers) all working in coordination. Whereas these processes produce hyper-accurate models of the 3D surface, the reality is that for most applications such a degree of accuracy is far more than is needed. What is often desired by consumers is something close, but not perfect, especially when it comes with a lower price tag and much faster turnaround.
For example, geologists like to create 3D models of their digs. To do this today, they would have to create a grid all around their dig, employ the specialized hardware to every part of the grid, and then assemble all of the data. This method also requires that sometimes portions of the 3D texture mapping data be hand-corrected for capture errors.
The new method
The newly developed system makes the whole process much easier. While it does reduce overall accuracy notably, it's not often enough to be significantly noticeable visibly. This means it's more than sufficient for games and models where only visible closeness is needed.
While the same kind of grid system is still required, the process involves taking two images with a regular digital flash camera. Since the flash is slightly offset from the camera's optics, it creates brighter pixels and shadows that differ from the non-flash image. The researchers developed software algorithms to determine how much of the color difference is the result of the shadow, and how much is the result of the brightness of the flash. They applied that logic to a program which apparently can reasonably accurately extrapolate 3D data.
The new method does not require any special hardware, only the newly developed software operating on the two images. To test the results, the research teams asked 20 volunteers to compare the final products captured from the expensive techniques to the double-pic flash technique. None of the 20 could notice any differences of substance, the researchers said.