Proposed real-time obstacle detection system for visually impaired assistance based on deep learning

Abstract

The mobility in an unfamiliar environment independently is one of the biggest challenges for visually impaired and blind people. It becomes difficult for those with this kind of disability to keep a track of their routine environments, and so many of them tend to rely on their sighted friend or family member for assistance. In this paper, the authors are proposing a real time based-vision system using object detection, object tracking and object-camera distance estimation for the visually impaired people assistance. The main purpose is to come up with a low cost system which can detect and track the surrounding obstacles and estimate their risk on the users, in order to facilitate theirmobilityin any kind environment. Firstly, the You Only Look Once approach is the main object detector, by applying YOLO the obstacles will be detected in a frame t and their bounding boxes will be tracked in the coming frames. Secondly, DisNet which is a deep learning based method is applied for the distance estimation.

Development of an Intelligent Navigation and Spatial Orientation System for People With Visual Impairments

“…Полягає у можливості забезпечити цим рішенням виконання усіх або більшості вимог до системи, делегуючи його на підготовлені нейронні мережі в одному або декількох екземплярах. Серед переваг незалежність від платформи розгортання, простота заміни та масштабування, проте недоліками є велика затребуваність ресурсів для реалізації специфічного рішення, високе споживання ресурсів на виконання [6][7][8]. На рис.…”