Multimodal Vision System for Mobility Assessment in Parkinson's Elderly

Traditional methods for evaluating mobility in older adults with Parkinson's Disease (PD), such as the TUG test, are often subjective or rely on uncomfortable wearable sensors, leading to a gap in objective and detailed assessment of crucial metrics like tremors, balance, and obstacle avoidance. To address this, we developed a new non-invasive, vision-based system that uses high-resolution cameras to track a patient's movements in real-time. The system uniquely analyzes three key factors: it calculates tremor frequency using Fast Fourier Transform (FFT), measures obstacle avoidance skills using AI (YOLOv8), and determines a real-time Fall Risk Score by modeling balance dynamics based on the inverted pendulum mechanism. This comprehensive approach provides objective feedback and lays the groundwork for future AI-driven disease severity classification.