An FPGA-Based Upper-Limb Rehabilitation Device for Gesture Recognition and Motion Evaluation Using Multi-Task Recurrent Neural Networks

This work presents a mobile FPGA-smart rehabilitation system that can be used at home. The prototype is a rehabilitation table instrumented with a capacitive sensor array (CSA) to track the upper-extremity motions of the user through proximity or touch. In addition, inertial measurement units (IMUs) are placed on the affected upper limb and combined with the CSA data with our sensor fusion signal processing architecture. Motions are classified and evaluated using multi-task convolutional recurrent neural networks. The prototype achieves real-time execution time with above 99% accuracy using 32-bit fixed-point format implementation for simultaneously recognizing dynamic motions and identifying unnatural characteristics in upper limb motions based on sensor values. The C-LSTM/C-GRU fusion classification network is implemented on a 200 MHz Zynq ZCU104 FPGA using an HLS-based design optimized with pipelining and parallelism techniques.