Existing home-based training systems use games as the primary trend to enhance the level of motivation [1,2,3,4,5]. In addition, the training efficiency, in terms of repetitions, may be lower when utilizing the home-based training system without feedback [6]. If the fun environment is enriched and the correct movement is guided by visual feedback during training exercises, this approach would be a feasible training approach. Cloud computing technology has advanced recently, cloud-based networks can easily connect the training from home to the center, and therapists can adjust the training protocol remotely and access the training data. Two markerless systems for motion tracking 3D depth sensor technique and RGB camera system. 3D depth-sensing cameras, employing technologies such as stereo vision, time of flight, or structured light, are now capable of identifying 3D body segments. Notable examples include Kinect [7,8,9,10,11], ZED [12, 13], Intel RealSense [14]. Another type of system using markerless’s AI-driven motion capture technology with RGB cameras (e.g., Theia3D) [15] to construct the 3D skeletal models for tracking joint movements and balance control. In this work, we raise the following questions: Is it possible to have an interactive guidance system for home-based users during training, and what is the effectiveness of motor recovery after 800 min of home-based training using computer-guided visual feedback?
The home-based training approach [16,17,18] provides training for chronic stroke and benefits stroke survivors with an increased amount of training time and more feasibility in the training schedule, especially in patients with severe levels. The depth sensor-based training systems (i.e., UINCARE Home + (UINCARE Corp., South Korea) [19, 20], MindMotion® GO (Switzerland) [21], EvolvRehab (Spain) [22], LongGood TeleRehabilitation System (Taiwan) [23]) provide the visual feedback with specific symbols or targets to guide the user in achieving the tasks. However, these training approaches display virtual objects in a video game or demonstration videos without providing feedback on the user’s real-time motions. In this study, we developed a home-based guidance training system for stroke survivors, enabling to provide the RGB-depth sensor to capture 25 artificial anatomical landmarks to reconstruct the body skeleton and real-time visual feedback on users' body segment movements and joint angles during training.
The purpose of this study was to investigate the improvement in motor performance for stroke survivors with moderate to severe motor impairment after 800 min of training using the home-based guidance training system with interactive visual feedback. With the technology of depth sensors, home-based training with computer-guided motion guidance in real-time may help users improve motor performance with minimal therapist assistance, which can greatly enhance the flexibility to facilitate their training time schedule.