Wearable Scene Classification System for Visually Impaired Individuals

Advancements in technology are made to make human life easier. According to survey made by the World Health Organization (WHO), millions of people are suffering from visual impairement. Individuals with visual impairment face many troubles in their daily life tasks. In this research, our objective is to design a cheap, secure, wearable, and portable system for visually impaired individuals. The proposed methodology includes the acquisition of indoor samples using Raspberry Pi Night Vision Camera and then preprocessing using the Retinex Filtering Technique and Low Light Image Enhancement Technique. Local Binary Pattern Features are uprooted from the preprocessed images which are then trained using a different set of classifiers including Fine Tree (FT), Quadratic Discriminant (QD), SVM-Quadratic (SVM-Q), Weighted-KNN (W-KNN) and Ensemble. Using 10-fold cross-validation, Quadratic Discriminant gives maximum accuracy of 92.0%.