Machine Vision Intelligent Travel Aid for the Visually Impaired (ITAVI) in Developing Countries*

Abstract

The visually impaired have little or no effective visual sensory input and have to rely on external assistance for navigation. Several electronic travel aids have been developed to aid independent navigation of the visually impaired, however they are not without limitations and dependence on third-parties. This paper describes the design and implementation of an Intelligent Travel Aid for the Visually Impaired, it combines the detection and recognition of objects in real-time with audio feedback to provide aid to the visually impaired users. This assistive device uses machine vision for object recognition detection, a camera for capturing object images respectively and a speaker all collectively form the core of the system. The system notifies users of obstacles and objects via synthesized speech. Using a quantized MobileNet based Single Shot multibox object detection model pre-trained on the Common Objects in Context dataset, the device was able to detect objects/obstacles, as well as determine the relative position and approximate distance. The device, when tested, was found to achieve real time performance of up to 70.56 frames per second for detections. Audio feedback was also achieved using the eSpeak Text to Speech engine to provide real time voice instructions to the user. All algorithms were implemented using Python language. The device is user friendly, allowing the visually impaired to enjoy easier navigation. However, other features such as extra object classes as well as language variety could be added in order to boost the robustness of the device.