Inclusive mobility solution for visually impaired people using Google Cloud Vision

Abstract

Mobility in cities is of particular and growing importance nowadays due to the demographic increase and the existence of people with reduced mobility, as is the case of visually impaired people. Of the various situations where mobility represents a challenge, obtaining the notion of positioning, at times when the person loses track of where he is and becomes disoriented, can be extremely useful and a way to contribute to greater autonomy for this segment of people. This paper proposes a visual positioning system using the Google Cloud Vision API. The architecture includes a mobile application that captures an image via the mobile phone and sends it to a backend server that makes use of Google Cloud Vision to recognize the image, which may consist of text, logos or landmarks. In a first phase, the solution was evaluated individually and, in a second phase, on a route chosen in the city of Braga, in Portugal. Logo recognition achieved an accuracy of 98% and proved to be sensitive to image resolution. The frontal text recognition obtained an accuracy of 100% while the lateral recognition and at a 3 meters distance obtained lower values, with worse results in images with more text and of reduced dimensions. Landmark recognition always returned the correct result, although the average accuracy is 82%. The processing time was around 3 seconds in tests done with Wi-Fi network and about 2 seconds in field tests made with mobile network. The obtained results prove the adequacy of using this solution to be adapted in a real scenario.