IMU-based motion analysis with deep learning and transferring movements to the virtual reality environment

Virtual reality (VR) technology has become a powerful instrument of transformation across a broad spectrum ranging from healthcare and education to rehabilitation. Realistic and interactive environments accelerate skill acquisition for healthcare professionals, while for patients they provide safe, measurable exercise experiences that enhance motivation. In this context, VR contributes to improving motor abilities in individuals with stroke and musculoskeletal disorders, enhancing balance and coordination, and strengthening adherence to therapy. This thesis analyzes VR's clinical potential through an integrated system that monitors users' exercises in real time and automatically determines the type of movement at the end of each session. The system combines exercise scenarios developed on the Meta Quest 3 with the wireless transmission of data obtained from a wearable IMU sensor. During the session, the user completes the selected exercise; the streaming data are organized, and at the end of the exercise the model produces a prediction regarding which movement was performed. While gamified exercises support user motivation, the resulting outputs provide a framework for customizable rehabilitation exercises that can be adapted to home and clinical settings.