Omuz Egzersizlerinin RGB-D Verisi Kullanılarak Gerçek Zamanlı Kestirimi İçin Sanal Egzersiz Sistemi

Abstract

Shoulder pain and discomfort are common and serious problems. Shoulder treatment benefits from a structured and repetitive program. In traditional physical rehabilitation programs, patients frequently perform exercises with intermittent feedback following the demonstrations from the physiotherapist. However, with at-home rehabilitation, the patient does not receive feedback after the initial demonstrations. This may lead to interruption in the treatment process, improper treatment and even self-inflicted injuries.This work propose Virtual Training Environment for Shoulder Exercises (ViTES) as a promising new tool to achieve sustained therapy practice and patient motivation for shoulder rehabilitation. ViTES can train users and assess their exercise performance concurrently with real-time recognition from incoming RGB-D data stream. To create the learning model that we use with ViTES, we also created V-Shoulder Dataset. The dataset consists of 739 exercise samples of 7 different shoulder treatment exercises in total and was created using Kinect RGB-D sensor. We validated the usability and the efficacy of ViTES by a two-part user study. In the first part, where the users evaluated ViTES via a short questionnaire, it was seen that all users regarded the system positively and found it easy to use. In the second part, we compared the similarities of the exercise movements performed by the users as automatically assessed by ViTES with respect to the model learned from the V-Shoulder Dataset. The results show that ViTES has a remarkable potential to be a beneficial tool in complementing the traditional physiotherapy process.