Air Writing using Deep Learning

Abstract

Air typing is writing in free air using hand or finger OPs, without using physical medium_HCI research and intelligence. Deep learning methods, which can learn abstract features hierarchically in a manner inspired by human cognition, have gained huge attention in air typing applications due to their ability to learn high-level patterns and features from low-level signals. This paper highlights some of the deep learning concepts like CNN, Index content cloud typing, classification typing, CNN, RNN, LSTM that would be utilized for air typing. The biggest problem with air typing is that there is no zero-knowledge environment — this leads to an extremely high variance in speed, direction, and style of typing. Deep learning models have great potential to address these issues by directly learning spatiotemporal patterns from the raw input data, provided that they are trained on large enough datasets of sufficient diversity. We talk about the basic methods of how to take raw data (from devices, cameras etc.) and input them to neural networks. We also examine recent advances in aerial typing including gesture recognition and natural language processing to achieve greater accuracy.

Keywords: Aerial Typing Recognition, Convolutional Neural Network (CNN), Hand Tracking, Recurrent Neural Network (RNN), and Short Term Memory Network (LSTM)