Magic Glove - Sign Language Recognition
A digital-circuit-lab glove that recognizes sign language — flex sensors and gyro feed an FPGA-accelerated CNN to translate gestures into text
Magic Glove
Magic Glove is the final project for the digital circuit laboratory course. The goal: a wearable glove that recognizes sign language gestures in real time and translates them into text on screen.
Hardware
The glove is equipped with flex sensors on each finger and a gyroscope on the back of the hand, capturing both finger bend angles and hand orientation. An Arduino reads the sensor data and transmits it over Bluetooth to a PC, which relays the signal to the FPGA for processing.
FPGA Recognition Pipeline
The FPGA runs a CNN (Convolutional Neural Network) implemented in Verilog for fast inference on the incoming sensor data. To improve accuracy beyond raw classification, the pipeline includes two additional stages:
- Dynamic Time Warping — smooths out timing variations in how different people sign the same gesture
- Viterbi algorithm — searches a built-in dictionary to correct nonsensical letter sequences into the closest valid word
The result is displayed on screen with minimal delay.