Character Recognition in Air-Writing based on Network of Radars for Human-Machine Interface

Abstract

Radar technology can detect hand gestures without touching, making it an intuitive way to interact with computers. Air-writing means writing in the air with hand movements. We’ve created an air-writing system using millimeter-wave radars. Our method has two steps: first, we figure out where your hand is and track its movement. Then, we use this data in two ways: one uses a special kind of neural network to understand the hand’s path and recognize characters, and the other turns the hand's path and recognize characters, and the other turns the path into an image and uses another neural network to figure out which letters were drawn. The first method works really well, with a 98.33% accuracy rate for character recognition, similar to the second method. We tested this with real data from three radar sensors at 60 GHz. This setup and method show promise for human-computer interaction without touching.