Real-time Air-written English Numeral StringRecognition using Deep Learning

Air writing is gesturing to draw 3D letters without tools, which could allow devices to work in new ways. Single letters are easy to recognize; however, continuous writing is difficult because there are no clear breaks between letters. Stray hand moves also add noise that confuses re cognition systems. Eliminating noise and accurately sorting letters are crucial for better accuracy. To help solve these issues, we aim to use Recurrent Neural Networks (RNNs), which can handle time-base d data like gestures. Numbers often share parts too, making this another tricky case. Digits look alike, so telling them apart is hard. Our approach uses common ways people write digits to figure out which digit came first. It take s x-y coordinates as input that any camera can capture, so anyone can use it. We tested our system with English numerals from combined MNIST, Pen-digits, and our ISI-Air Dataset of air-written English numerals. Under standard conditions, it correctly identified 98.75% single -digit English numerals and 84 % multiple digit sequence.