Obstacle detection and distance estimation using monocular camera in navigation services for visually impaired people

Abstract

In this paper, we propose a method for obstacle detection and distance estimation using monocular camera mounted on a mobile robot. The proposed system aims to support visually impaired people navigating in indoor environment. The obstacles include static and dynamic objects on that encumber human mobility. For static objects, supporting information such as type of object, positions, and corresponding images in relevant scenes are stored in database (DB). To detect them, the images captured during robot’s movements are compared with the corresponding images through a localization algorithm proposed in \cite{nguyen2016developing}. Then the existing objects in DB will be identified and distances from them to current robot’s position is estimated. For dynamic objects, such as movements of people in scenes, we use HOG-SVM algorithm \cite{dalal2005histograms}. To estimate distance from camera to detected obstacles, we utilize a disparity map which is built from consecutive frames. The experiments are evaluated in the hall of building floor of 60 meters under different lighting conditions. The results confirm that the proposed method could exactly detect and estimate both static and dynamic objects. This shows the feasibility to help visually impaired people avoiding obstacles in navigation.