HAR Multi-View by Vision RGB-D for Home Neurological Rehabilitation

CONTEXT Neurological rehabilitation (post-stroke, Parkinson's, multiple sclerosis) requires regular monitoring of therapeutic exercises. Home monitoring via RGB-D cameras (color + depth) allows for objective monitoring without having to travel to a care center. This internship aims to develop a solution for recognizing and evaluating rehabilitation exercises at home. INTERNSHIP OBJECTIVES • Identify 12 to 15 typical neurological rehabilitation exercises • Evaluate the quality of movement execution (scoring) • Multi-view robustness (different camera angles) • Clinical validation with real data TASKS REQUIRED • Acquisition and annotation of RGB-D data • Feature extraction by view (2D/3D pose, skeleton) • Multi-view fusion and temporal modeling • Exercise classification and quality scoring • Clinical validation on a cohort of patients DELIVERABLES • Annotated multi-view RGB-D dataset • Operational recognition and scoring model • Monitoring interface for the patient and therapist • Scientific report