BodyTrak: Estimating Full-Body Poses Using Wrist-Mounted Camera

Researchers from Cornell University presented BodyTrak – "an intelligent sensing technology that can estimate full-body poses on a wristband." It just needs one miniature RGB camera to capture the body silhouettes, learned by a deep-learning model to estimate the 3D positions of 14 joints on arms, legs, torso, and head.

BodyTrak was tested by 9 participants, who performed 12 daily activities, such as walking, sitting, or exercising, in varying scenarios with a different number of camera settings. The researches say that the system can infer the full body pose with an average error of 6.9 cm using only one camera pointing towards the body.

The authors of the paper believe that this technology can improve individual fitness performance analysis, inform users about their posture, and even help care for elderly individuals in emergency situations by recognizing uncommon body postures.

Learn more about BodyTrak here and don't forget to join our Reddit page and our Telegram channel, follow us on Instagram and Twitter, where we share breakdowns, the latest news, awesome artworks, and more.