Home-Based Physical Therapy with an Interactive Computer Vision System

Yiwen Gu, Shreya Pandit, Elham Saraee, Timothy Nordahl, Terry Ellis, Margrit Betke; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2019, pp. 0-0

In this paper, we present ExerciseCheck. ExerciseCheck is an interactive computer vision system that is sufficiently modular to work with different sources of human pose estimates, i.e., estimates from deep or traditional models that interpret RGB or RGB-D camera input. In a pilot study, we first compare the pose estimates produced by four deep models based on RGB input with those of the MS Kinect based on RGB-D data. The results indicate a performance gap that required us to choose the MS Kinect when we tested ExerciseCheck with Parkinson's disease patients in their homes. ExerciseCheck is capable of customizing exercises, capturing exercise information, evaluating patient performance, providing therapeutic feedback to the patient and the therapist, checking the progress of the user over the course of the physical therapy, and supporting the patient throughout this period. We conclude that ExerciseCheck is a user-friendly computer vision application that can assist patients by providing motivation and guidance to en-sure correct execution of the required exercises. Our re-sults also suggest that while there has been considerable progress in the field of pose estimation using deep learning, current deep learning models are not fully ready to replace RGB-D sensors, especially when the exercises involved are complex, and the patient population being accounted for has to be carefully tracked for its "active range of motion."