A Visual Navigation Method for Robot Based on a GOR and GPU Algorithm

Abstract

A GOR (general object recognition) method is proposed. It refers to the statistical model of BOW (bag of words), and makes use of SIFT (scale-invariant feature transform) detection algorithm to describe feature vectors. Especially, in order to increase redundancy of image information, the statistical information of spatial relationships among object parts are used to describe the spatial relationships of all the feature points in an image, including relative distances and angles, which augments the feature vectors in the original BOW model. The unsupervised discriminated classifier-support vector machine(SVM) is used to recognize objects. At the same time, GPU (graphic processing unit) acceleration technology is used to guarantee the real-time feature extraction and description of SIFT algorithm. Then, based on the hand drawn map, this method is successfully applied to indoor robot navigation. Experiments show that the mobile robot navigation technology based on this method is robust and effective.