Walking Assistant for Vision Impaired by Using Deep Learning

Abstract

In this paper, we address the people in the globe living with a visual impairment or living in isolation. Although it is already very difficult to live with sight loss, still we try to provide some support for them to access the environment freely. The lack of limited accessibility to activities and information, the societal stigma and the deficiency of joblessness are all factors frequently leading blind or low-vision individuals in isolation. We have used CNN technique and developed a method for processing an image and tells about the obstacle and depth of potholes to make the walking easy for visually impaired person. CNN has almost dominated computer vision and shows powerful performance on object classification. We create our own network on the basis of RCNN algorithm in which we consider 128 filters having size of (2 × 2). This network consists one input layer, nine middle layers, and five final layers and train the same network with CINIC-10 dataset having initial learn rate is 0.001, Learn rate drop factor is 0.1; learn rate drop period is 8, and maximum epochs are 40. After validation of our trained network, it gives the accuracy of 64.7%, and it wonderfully makes out results.