FPGA-Prototyped Application-Specific RTL Architecture for Real-Time Sign Language Translation Using Dual-Glove Sensor Fusion

Abstract

This paper proposes an FPGA-prototyped, application specific Register Transfer Level (RTL) architecture that translates hand gestures captured by a dual-glove wearable sensor system into real-time text and audio output. Each glove incorporates five flex sensors and one accelerometer, providing ten flex and two motion inputs for a total of twelve; the binary flex states provide 2^10 = 1,024 possible finger configurations, while the five discretized motion states of each of the accelerometers provide 5^2 = 25 combined motion states, yielding a theoretical sensor-state space of 25,600 combinations. The proposed hardware pipeline executes sensor interface, finger-state encoding, motion-state classification, sensor fusion, command decoding, and message generation completely in dedicated RTL, as opposed to using software on a general purpose or embedded processor. The final output of the system is a real-time, human-readable translation of the signer's hand gesture, displayed as text and rendered as audio speech -- generated directly from raw glove sensor data with no intermediate processor or ML inference stage in the recognition pipeline. The architecture is functionally verified through RTL simulation, and demonstrated on an FPGA prototype, laying a hardware-validated foundation for a subsequent ASIC implementation in which area, power, and timing can be quantitatively optimized.