Robust Action Recognition using Depth-cameras for Rehabilitation and Bio-feedback Applications

点击翻译按钮获取中文摘要

With the emergence of inexpensive consumer level hybrid cameras (depth+color) such as the Kinect device (already at the second generation and that sold over 20 million units) and the Intel RealSense camera now standard in several laptop and tablet models opened the door to a whole new universe of consumer applications and a new era of home computing and gaming. Since they can acquire both geometric information as well as color information one of the benefits of such hybrid cameras is that they can, for the first time, facilitate real-time marker-less motion capture and action recognition. In fact these devices already are shipped with built in rudimentary motion capture and action recognition. Unfortunately, while inexpensive, these devices lack the accuracy and robustness required in medical applications such as rehabilitation software. In this project we will develop in partnership with Jintronix, new accurate as well as robust action recognition and motion capture methods based on consumer level depth-cameras.