Stereo image based object localization framework for visually impaired people using edge orientation histogram and co-occurrence matrices

A new framework that uses internet-based images for detecting objects and estimating real world location of the objects via stereo images is proposed. This framework provides a self-learning ability for detecting desired objects in the scene without pre-prepared classifiers by harvesting sample images of the objects from the internet. Histogram and co-occurrence matrices of edge orientation are used as features. The objects are recognized based on likelihood scores and distance in the feature space between every window in the scene and k-nearest prototypes. A local feature matching is used to match the feature points in stereo pair. Disparities from stereo images are used to estimate real world distance and direction of the objects. The experiments on 120 pairs of stereo images from three object classes show the satisfying results in comparison to baseline methods.