Gait Analysis and Rehabilitation Using Web-Based Pose Estimation

Gait abnormalities are one of the most common health conditions in the elderly population, with almost one in three people over 60 experiencing symptoms that disrupt their movement [1]. These symptoms can cause disability [2] and present an increased fall risk [3] [4]. Detecting these abnormalities early is, therefore, crucial as it reduces the likelihood of injuries and accidents. Current treatments for gait abnormalities depend on the condition, but many treatment plans commonly incorporate some form of physiotherapy. Clinicians typically deliver physiotherapy in the form of gait assessments and targeted exercises or therapies. Recent research has also shown that virtual reality (VR) treadmill walking, using motion capture technology, can be an effective method of treating certain gait abnormalities [5] [6] [7]. This thesis covers the development of a web-based VR treadmill walking system to make VR physiotherapy cheaper and more accessible. The system uses convolutional neural networks to assess the patient’s gait from an RGB webcam feed and provides them with live feedback on their body position within a VR environment. The system’s gait assessment capabilities are validated by comparing it to a gold standard – the OptiTrack motion capture system. The results demonstrate that the system’s percentage error (ϵ˜%) was much less for temporal gait metrics (0.24 < ϵ˜< 12.40) than it was for spatial ones (70.90 < ϵ˜% < 79.72). Four out of five spatial metrics also had a “very strong correlation” (0.74 < r < 0.86) when compared to the OptiTrack’s metrics, meaning the accuracy could be increased using a gain factor. These findings establish the basis for a similar study with a larger sample size. They also raise the possibility that this system could analyse gait in the clinic and the home without specialist motion capture equipment or facilities.