A Novel Approach for Object Detection Using Optimized Convolutional Neural Network to Assist Visually Impaired People

Abstract

Human race is blessed with the five basic senses such as touch, taste, smell, hearing and the most important of them all 'vision or eyesight'. It is very difficult to survive without any one of them. Unfortunately a mass population across the globe suffers from the ill effects of vision, hampering their daily life. Detecting objects and providing navigational instructions in an indoor environment can considerably improve the day-to-day quality of life of visually impaired people. The motive of this research work is to propose a solution approach for assisting visually impaired population by identifying obstacles in front of them considering indoor environment. This approach focuses on feature extraction and object detection using Convolutional Neural Network (CNN) from a real time video. For this a head mounted image acquisition device may be used to detect the objects from the scene ahead and information of the detected objects is provided to the visually impaired (VI) person through the audio modality. As a first step towards the overall conceptual process, an object detection system is presented in this article, which processes the live video stream captured through the acquisition device. The video is processed frame-by-frame, treating each frame as a separate image and then using the proposed feature extraction and object detection algorithm to identify the objects.