Outdoor Situation Recognition Using Support Vector Machine for the Blind and the Visually Impaired

Abstract

Traffic intersections are most dangerous situations for the pedestrian, in particular the blind or the visually impaired person. In this paper, we present a novel method for automatically recognizing the situations where a user stands on, to help safe mobility of the visually impaired in their travels. Here, the situation means the place type where a user is standing on, which is classified as sidewalk, roadway and intersection. The proposed method is performed by three steps: ROIs extraction, feature extraction and classification. The ROIs corresponding to the boundaries between sidewalks and roadways are first extracted using Canny edge detector and Hough transform. From those regions, features are extracted using Fourier transform, and they fed into two SVMs. One SVM is trained to learn the textural properties of sidewalk and the other is for intersection. On online stage, these two SVMs are hierarchically performed; the current situation is first categorized as sidewalks and others, then it is re-categorized as intersections and others. The proposed method was tested with about 500 outdoor images, then it showed the accuracy of 93.9 %.