RNN and CNN for Way-Finding and Obstacle Avoidance for Visually Impaired

Abstract

Way-finding is crucial for visually impaired as well as sighted persons. Already navigated way is useful for the visually impaired if reused. In this research, we present an assistive technology solution of reusable way-finding with obstacle avoidance for the visually impaired. We trained a recurrent neural network (RNN) model to predict the navigation activities. These activities are used as the building blocks of reusable way. A fine-tuned convolution neural network (CNN) model is used to detect obstacle. Both models are incorporated in a smart phone application to construct, share, and reuse a navigation way. The evaluation shows that using the application the visually impaired were able to navigate 95% times accurately without external help.