Beefing Up Animation Software
Researchers believe the secret to making animated characters look more realistic is in the muscles.
Computer animations have become increasingly realistic over the years, but a few lingering nuances still trigger our brains to recognize that a character is not flesh and blood. It’s the subtle details – a facial expression or muscle movement – that usually gives it away. But now researchers from the Bournemouth University in the U.K., have developed easy-to-use tools that make skin and the muscles underneath it move more realistically.
When animators build a character, they usually start with its external appearance, rather than internal features such as skeletal structure and musculature, says lead researcher Jian Zhang, professor of computer graphics at the University of Bournemouth. Details such as the way a muscle looks when flexed are added after the basic look of the character is complete.
This approach, called skin-based, is popular in the animation industry because it’s the most intuitive for animators to use, he says. And, indeed, drawing an arm with a bulging biceps is easy enough – if the character isn’t moving.
However, the skin-based approach isn’t the most accurate way to capture the nuanced movement of skin as muscles contract below: that requires information about the shape of the muscles themselves. And ill-fitting skin can make an animated character look fake.
A more accurate method, Zhang says, is to build a character’s muscles first and then add the skin on top of them. It’s more natural looking, he says, but it requires detailed knowledge of anatomy, and it’s counterintuitive for an artist who has the external appearance of a character in mind. For those reasons, he says, the muscle-based approach has been mostly an academic curiosity.
The researchers’ algorithm combines the realism of the muscle-based approach with the intuitive appeal of the skin-based model. It allows an animator to start with a character’s outward appearance and modify the movement of major muscle groups – without having to build the musculature from scratch.
The problem is more complex than it may sound, says Zhang. As soon as an arm is flexed, for instance, the animation algorithm needs to use information about the shape of the muscle under the skin to make the muscle contract and the skin deform accurately. When the only information about the muscle comes from the shape of the skin around it, it’s challenging to make flexing look realistic.
Zhang approached the problem by looking at the muscle groups that most commonly make skin move, such as those in the shoulder, neck, arms, thighs, and calves. He then used software to analyze the shapes of these muscles and simplify them into ellipses. “The muscles are not biologically correct,” Zhang says. “They are estimated shapes.” But when the muscles are simplified, he says, an animator doesn’t need detailed knowledge of their shapes to make their movements look real. In addition, using simplified shapes cuts down on the computational power the process requires.
In informal subjective surveys, people found that characters created using the new algorithm were more realistic than those created using the traditional skin-based method. However, the algorithm does not produce better results in all cases; it works best for characters with more defined muscles, Zhang says. In characters with fewer muscles and smoother bodies, it doesn’t predict the skin changes very well.
The researchers’ algorithm is a “refinement” of the tools already available to animators, says Michel Besner, a senior director of product management at Autodesk, a manufacturer of design software and technology. And it could make their jobs easier. “When we ask an artist to do more quality [animations] it forces them to become doctors,” he says.
The Bournemouth researchers are tackling an important problem in animation, agrees Karan Singh, professor of computer science at the University of Toronto, who calls the group’s algorithm “an incremental contribution on what exists.” The research is similar to the work that he and colleagues have done, says Singh, but the new approach has even broader applications than his work.
Zhang’s algorithm is still in the early stages and needs to be integrated into software with an easy user interface. The researchers’ goals are not necessarily to produce a product immediately, though. They plan to further reduce the amount of computation needed to animate a muscle, and improve the realism in the skin of less muscular characters.
Keep up with the latest in intelligent machines at EmTech Digital.
The Countdown has begun.
March 25-26, 2019
San Francisco, CA