Vision model learns to detect freezing of gait in Parkinson’s without wearables
Supervised Cross-Modal Feature Alignment for Zero-Wearable Freezing of Gait Detection in Parkinsonism
Computer Vision and Pattern Recognition
Summary
People with Parkinson’s disease often experience freezing of gait (FoG), which is hard to detect without special sensors worn on the body. The authors designed a way for a camera-based system to learn from detailed wearable sensor data and clinical information during training, so it can recognize FoG later without needing any sensors attached. This method helps the camera system handle tricky movements like turning in place, where visual tracking is usually unreliable. Their system achieved over 85% accuracy in detecting FoG using just video at test time.
What this means in practice
- •For clinical monitoring teams: Use video cameras alone to accurately detect freezing of gait in Parkinson’s patients without attaching wearable sensors.
- •For home health technology developers: Develop unobtrusive, camera-based systems for continuous Parkinson’s symptom monitoring that avoid the need for wearable devices.$Commercial implications: Enables sale of non-contact monitoring products to healthcare providers seeking continuous gait monitoring for Parkinson’s patients.
Authors
Aryan Singh, Chandan Biswas
Abstract
Objective assessment of Freezing of Gait (FoG) in Parkinson's disease (PD) relies predominantly on wearable Inertial Measurement Units (IMUs). While IMUs provide optimal kinematic precision, mandatory sensor attachment restricts continuous clinical deployment. Conversely, unobtrusive vision-based alternatives suffer substantial classification errors during turning-in-place tasks, where geometric self-occlusion degrades deterministic skeletal coordinates and obscures the high-frequency precursors required for FoG detection. To resolve these physical observation limits, we propose a supervised cross-modal subspace distillation framework. During optimisation, pre-trained kinematic data from IMU sensors and contextual clinical metadata act as oracles to guide a deployable visual architecture. By incorporating joint velocity and acceleration derivatives, utilising a confidence-based gating mechanism, the visual model mitigates some of the tracking errors during occlusion events. Empirical evaluations confirm this latent alignment transfers the predictive fidelity of hardware sensors directly into the visual representation, yielding $85.5\%$ accuracy, and $82.4\%$ balanced accuracy. All the while maintaining a vision only model at inference.