A Deep Learning based Real Time Assistive Framework for Visually Impaired

Visually impaired people face great difficulties in interacting with unfamiliar surroundings and making use of public amenities. Commuting and using public services without any external help is very challenging to the visually impaired. Over time there have been numerous developments to aid visually impaired but mostly these have been limited the use of sensors to measure distances and send warning signals. The proposed work presents a deep learning technique to aid the visually impaired in real time by identifying nearby public amenities that are used by all in everyday life like Restrooms, ATMs, Metro stations and Pharmacies. This method used faster region-convolutional neural networks (Faster R-CNN) with resnet50 to identify the symbols of public amenities that are the same around the world. This algorithm was trained with a set of over 450 images and was tested on a varying database and achieved an accuracy of 92.13 percent. The experimental results to identify public amenities like Restrooms, ATMs, Metro stations and Pharmacies are robust, efficient, low in cost and helps in providing sight to the visually impaired up to a large extent.