Design of hand rehabilitation movement evaluation system based on lightweight critical point detection

Abstract

Unlike traditional hand rehabilitation training, which relies on on-site guidance by therapists, this system can meet the quantitative assessment needs of home rehabilitation. This problem was addressed by designing a lightweight, deep learning-based system for hand key point detection and rehabilitation action evaluation. Through improvements totheMobileNetV3 network architecture and the combination of knowledge distillation and deep separable convolutional technology, the system achieves real-time, accurate localization of hand keypoints. To recognize key point sequences, a spatio-temporal convolutional network is constructed, and a quantitative assessment is established based on kinematic parameters. Deploying the system on mobile devices provides patients with convenient home rehabilitation training while verifying the application value of the lightweight model in medical rehabilitation.