A Digital Upper Limb Rehabilitation System with Markerless Tracking: Design and Preliminary Accuracy Assessment

Abstract

This paper presents the design and preliminary evaluation of a noninvasive rehabilitation game system for upper limb motor recovery in post-stroke patients. The system integrates the Orbbec Femto Bolt depth sensor with the Azure Kinect SDK and a Unity-based game application to enhance patients’ motivation and support therapist-guided rehabilitation. Shoulder joint angles are detected and calculated in real-time, enabling interactive gameplay control and on-screen visualization of both the virtual character movements and joint angle. Accuracy testing across clinically relevant motion ranges demonstrated a mean absolute error of 0.45 degrees and an average accuracy of 98.9 percents. These preliminary results suggest that the system holds strong potential for clinical application. Moreover, the Orbbec Femto Bolt has been shown to be compatible with the Azure Kinect SDK and Unity-based applications. However, further enhancements are recommended, particularly in refining the skeleton tracking algorithm to address limitations inherent to the Azure Kinect SDK. Additional studies involving real patients and therapist input are also essential to evaluate practical usability and therapeutic impact.