Research On Road Detection Algorithm Based On Generalized Hough Transform

With the rapid development of remote sensing technology and computer technology,the acquisition of remote sensing images become very easy.Object feature recognition and detection in remote sensing images become more and more important.Straight line is one of the basic features of objects in images.It is of great significance for computer vision and pattern recognition to study the detection algorithm of lines in images.Road information in remote sensing images is a very important geographic information in remote sensing image analysis.Hence,the research of the road detection has become the focus in the field of remote sensing at home and abroad.Hough transform is the more typical algorithms of straight line feature extraction,which has got a good application in many application systems.Hough transform is essentially the clustering of pixels that have some relationship in digital image space,which has good robustness and can solve the problem of linear feature extraction.The main contributions of this article are as follows:1.Dictionary learning based Hough transform method is presented.It based on the mapping relation of Hough transform and the Radon transform formula uses the dictionary learning method to approximate the Radon transform,treats a Radon transform as a linear transform,which then facilitates parallel implementation of the Radon transform for multiple images,and obtains better performances than Radon transform in the term of computational efficiency.We conduct extensive experiments on the popular RSSCN7 database,and our algorithm is superior to the traditional Hough transform,Radon transform.2.Regularized multispectral dictionary learning based Radon transform is proposed.it introduces the concept of multiview which avoids the loss of image information and improves the accuracy of line detection,introduces prior knowledge by adding the regularization.Prior knowledge adds the person's prior knowledge of the model to the learning model and forcibly learns the desired characteristics of the person,such as smoothness,sparseness and so on.We conducted a large number of comparative experiments in the database RSSCN7,the proposed algorithm has a good detection effect.