Gait Analysis Using Stereo Camera in Daily Environment

Gait is one of the important features to assess physical or mental conditions of the elderly which is directly related to their health status. The changes or abnormalities in gait may reflect health risks. Screening tests like TUG (Timed Up and Go) and POMA (Performance Oriented Mobility Assessment) include gait assessment. However, analyzing gait is a complex process usually handled by professionals at clinical facilities. With the development of IoT (Internet of Things) sensors and computer vision technologies, more automatic and accurate approaches are demanded to measure gait at nursing facilities or at-home in the daily environment. In this paper, we propose an automatic and privacy-considered way to analyze gait using a non-contact sensor, a stereo camera. This approach applies a cutting-edge deep learning technology to detect a human subject in 2D images and then combining 3D sensing data to measure gait features, such as step length and walking speed. Compared to Kinect or a single 2D camera, our approach is not only accurate for various walking patterns but also robust to camera setting environment. Experiments at a daycare facility in Tianjin, China showed that our approach is suitable for assessing TUG or POMA tests in the daily environment.