A new motion capture technology will allow you to create digital dog avatars after filming them with a single camera.
Digitizing a dog is typically like digitizing a person. The process involves multiple cameras and a canine version of those ball-coated motion capture suits. A team of computer scientists from the University of Bath made all that, according to research shared on the GitHub team. It is an uncommon development, but one that could significantly improve fields from animation to veterinary care.
The software may be used for a wide variety of purposes, from assisting vets in diagnosing sickness and tracking their canine patients' rehabilitation, to entertainment applications such as making it easier to incorporate digital dog images into movies and videogames.
Virtual runway
RGBD images have been used for the first time tracking dogs' movement using a single camera. The process, according to scientists, is much more affordable than traditional multi-camera motion capture systems.
The team started out with the usual getup: several cameras and a suit for motion capture. However, all the extra hardware became unnecessary after training a computer model on the gait and poses of 14 dog breeds. A single RGBD camera - which also records how distant objects are - is sufficient for the model to accurately detect and predict and virtually recreate a dog's moves.
"For the entertainment industry, our research can help produce more authentic movement of virtual animals in films and video games," Bath researcher Sinéad Kearney said in a press release. "Dog owners could also use it to make a 3D digital representation of their pet on their computer, which is a lot of fun!"
Time capsule
While there is much research on automatic analysis of human motion without markers, Professor Darren Cosker, Director of CAMERA, said the animal kingdom is often overlooked.
The technology might also be a valuable tool for veterinarians who track the health of a dog aside from fun and games.
"This technology allows us to study the movement of animals," Kearney added, "which is useful for applications such as detecting lameness in a dog and measuring its recovery over time."
The team also started testing their method with some promising results on computer-generated images of other four-legged animals, including horses, cats, lions, and gorillas. They plan to expand their animal dataset in the future to make the findings more accurate. Scientists will also make the data set accessible to others for non-commercial use.
"Our research is a step towards building accurate 3-D models of animal motion along with technologies that allow us to measure their [movement very easily]," Cosker said. He added that the breakthrough has many exciting applications across a range of areas, from veterinary science to video games."
The team presented their research at the CVPR (Computer Vision and Pattern Recognition) conference on June 17 to 18 at one of the world's leading AI conferences.
