Deep learning-based gesture recognition for control of mobile body-weight support platform

It is important for stroke patients with hemiplegia to carry out lower limb rehabilitation training. At present, various body-weight support (BWS) platforms for lower limb rehabilitation training have been developed. Most of them are BWS treadmill training (BWSTT) which cannot follow the patients' movement. In this paper, a BWS locomotion training (BWSLT) platform based on deep learning-based gesture recognition for control of mobile BWS platform has been proposed and developed. Image database is established with 600 images that has classified into five categories indicating forward, backward, left, right and stop respectively to control BWS platform. A CNN-architecture network modified from AlexNet are trained and tested using this database. The experimental results showed that the recognition accuracy of gestures via proposed recognition method based on deep learning can reach 99.98% which is precise enough to control the mobile BWS platform.