Archive notice: This article was originally published on February 24, 2010. Links and embedded videos are preserved as part of the historical record.
As every CG artist knows, creating realistic reflections on objects in computer-animated films is a challenge. Postproduction in particular also takes a great deal of time. New software from the Max Planck Institute for Informatics in Saarbrücken is intended to speed up and simplify this process. The same laboratories have produced software that makes it possible to integrate three-dimensional representations of real objects into virtual film scenes in no time at all.
When a brilliantly gleaming car races through a commercial, glides through urban canyons or across dazzling desert sand, today it is only rarely the real vehicle made of steel, but usually a CG representation. Anyone wishing to showcase their latest vehicle model leaves nothing to chance: no still camera and no film camera can portray a car as ideally as a computer or make the light flow so beautifully across the paintwork. Although computer animation has long been standard in film, it still reaches its limits - for example, when it comes to conjuring perfect reflections on the paintwork. Thorsten Thormählen, a computer scientist at the Max Planck Institute for Informatics in Saarbrücken, is a specialist in such borderline cases and has developed methods that allow deceptively realistic computer graphics to be created much more easily and quickly than before.
Among the techniques he will present at CeBIT (hall 9, stand B 43) in Hanover from March 2 to 6 is “Interactive Reflection Editing” - a method that allows light reflections on the surface of virtual objects to be altered in no time. Until now, this has been extremely tricky: a virtual object only looks convincing if the sun and its surroundings are reflected in its surface. Graphic artists therefore devote a great deal of effort to creating realistic reflections in a car’s paintwork, windows or headlights. It becomes problematic when the reflections have to be revised once again at the end. The graphic artist then has to erase the light reflections from the graphic piece by piece and draw them in again - a time-consuming undertaking.
“Interactive Reflection Editing”, developed by Thormählen together with his colleagues Tobias Ritschel and Makoto Okabe, automates this work. It is sufficient to click on the light reflections in the graphic and use the mouse to move, compress or stretch them until the reflection is in the desired position. Disruptive points of light can also be pushed across the virtual paintwork. What sounds so simple is the result of a powerful mathematical process, says Thormählen. “When they are moved, the old and new positions of many hundreds of pixels have to be calculated simultaneously in fractions of a second.” The skill lay in developing a calculation rule capable of handling the volume of data without overloading the computer’s computing capacity, so that the image does not stutter during editing. According to the Max Planck Institute, Thormählen’s development has already attracted attention at a Siggraph Asia in Japan.
The same applies to another demanding project on which Thormählen has worked - and whose solution he is also presenting in Hanover: transferring a real object into a computer-animated film scene. Anyone wanting to transfer the representation of a real car into a computer today uses an outrageously expensive laser scanner that creates a digital copy of the real object. Only a few film producers can afford such scanners. Thormählen has therefore developed a method that is both faster and inexpensive. All he needs is a standard video camera. It is sufficient to walk once around the object – a real car, for example – and feed this 360-degree video sequence into the computer. The software developed by Thormählen extracts the object’s 3D information from the moving image and uses it to generate a 3D object that can then be integrated into any film scene. “In future, computer-game enthusiasts could film their own car, integrate it into their favourite computer game and race it,” the researcher explains.
Thormählen calls his video extraction process “3D Scene Analysis”. “One trick is to identify prominent reference points in the moving scene and calculate their position. These can then be used to determine the position and shape of the actual object, such as the car.” That sounds complicated, but is significantly simpler than older methods, in which markers first have to be attached to the car and the background in order subsequently to calculate the car’s position and shape. “Our process determines its own reference points in the video and is therefore considerably more user-friendly.”
Thormählen is certain that computer games are only one of many possible applications for his “3D Scene Analysis” techniques. “It is also conceivable that the position of endoscopic medical devices during keyhole surgery could be shown in a video image, which could make it easier for the doctor to orient himself inside the patient’s body.” And in sports training, for example, a virtual 3D figure could be calculated from the video of a rotating discus thrower in order to study and improve the sequence of movements.