Advances in Computer Vision for Home-Based Stroke Rehabilitation

This chapter examines the application of computer vision (CV) and deep learning (DL) techniques in the design and development of stroke rehabilitation systems. We begin by reviewing motion-capture-based solutions and discussing their inherent challenges and limitations. Next, we explore the requirements for next-generation home-based rehabilitation systems using simple RGB camera sensors, focusing on two main challenges: (1) the scarcity of high-quality data in healthcare settings, and (2) the limited explainability and usability of CV techniques due to their black-box nature. Addressing the crucial interaction between healthcare providers and rehabilitation systems, we present a series of experiments in collaboration with multiple hospitals to demonstrate the efficacy of a cyber–human intelligent system design in overcoming these challenges. Furthermore, we outline essential design principles for building low-cost, minimally intrusive rehabilitation systems that can be deployed in patients’ homes. Finally, we discuss the potential of 3D CV advances in designing the next generation of rehabilitation systems and review future opportunities in this domain.