SMART APPLICATION FOR THE VISUALLY- IMPAIRED

Abstract

In this work, an application that would allow recognizing objects from images recorded by the camera of a mobile device was developed. An android phone camera was used to take images of some objects and then store them in the android database and the name of each object was stored in an audio mode. The SIFT (Scale-Invariant Feature Transform) was applied for the development of the application. To improve the performance of the application, one of the fastest corner detection algorithms, the Features from Accelerated Segment Test (FAST) algorithm was implemented. Since the algorithm was implemented on a smartphone, OpenCV for Android SDK was used. The cascaded filters approach was used by SIFT to detect scale-invariant characteristic points, where the difference of Gaussians (DoG) was calculated on rescaled images progressively. A blob detector based on the Hessian matrix to find points of interest was used by SURF. To measure local change around the points, the determinant of the Hessian matrix was used, and points were chosen based on where this determinant is maximal. The determinant of the Hessian was used by SURF to select scale. The auditory presentation of object recognition results to the blind user was done through a pre-recorded message. 97% accuracy is recorded in the performance of the system.