Android-based object recognition for the visually impaired

Abstract

The large number of blind and visually impaired individuals in the society has motivated research groups to search for smart solutions that use vision-based technologies to improve their quality of life. This paper describes an Android-based application for object recognition developed to help the blind understand their environment better. This application is based on extracting local features of the object of interest, which are then matched to the corresponding features of objects saved in a knowledge base previously created. The local features are tested against more than one classification method and the results are analyzed. Deploying the application to a Samsung Galaxy Tab, the system is evaluated using a dataset especially developed for this purpose. The dataset contains more than 600 images of twelve objects under several distortions and viewing condition changes. Results of the analysis show that the system achieves real-time performance with high accuracy under most viewing conditions.