A Transfer Learning Approach for Smart Home Application Based on Evolutionary Algorithms

Abstract

Building new systems used for indoor sign recognition and indoor wayfinding assistance navigation, especially for blind and visually impaired persons, presents a very important task. Deep learning-based algorithms have revolutionized the computer vision and the artificial intelligence fields. Deep convolutional neural networks (DCNNs) are on the top of state-of-the-art algorithms which makes them very suitable to build new assistive technologies based on these architectures. Especially, the authors will develop a new indoor wayfinding assistance system using aging evolutionary algorithms AmoebaNet-A. The proposed system will be able to recognize a set of landmark signs highly recommended to assist blind and sighted persons to explore their surrounding environments. The experimental results have shown the high recognition performance results obtained by the developed work. The authors obtained a mean recognition rate for the four classes coming up to 93.46%.