Automated Road Junction Recognition Based On Model Restriction

With the development of the photogrammetry and remote sensing, people can obtain abundant spacial information. The information acquired from aerial and remote sensing images becomes an important means to upgrade special information and extensively applied to the national economy production and military target spying. But how to automatically acquire the information we need from the image data becomes an important problem in the information-oriented society. Nowadays object recognition plays an increasing role in the military and civil fields. Automatic object recognition from aerial and remote sensing images has obtained considerable development in the last few years but we still can 't find an universally applicable method due to the complexity of the scene and the deficiency in the human vision mechanism. The road information in remote sensing images is important thematic information of geography and the road junctions are important characters of the road network. So the recognition of road junctions from the aerial and remote sensing images is an important research subject in the image interpretation and object recognition.At the beginning, this paper introduces the basic knowledge of object recognition, and then centers on the methods of automatic recognition of road junctions. This paper summarizes the existing relevant methods of recognizing road junctions in detail. And then presents two new methods for small and medium scale aerial and remote sensing images separately:1 . The first method is based on shape restriction. For small scale aerial and remote sensing images, the roads are brighter lines with small width, and the road junctions are just the points of intersection of lines. Through designing proper shape model of road junctions, and combining with spectral feature, we just need to provide appropriate parameters and the road junctions can be detected.2 The second method is based on differential geometry. The surface of image can be looked as a curved surface, and we can simulate a conicoid function in local area. Based on this function we can estimate the gradient and the curvature of the curved surface points, and set proper thresholds of gradient and curvature, we can detect the centre lines of roads with correlative process including removing small areas, area growth, thinning, etc. Lastly through seting proper thresholds of the gradient and the curvature again, the original road junctions can be detected and then verified by the road lines detected formerly. The experiment results prove that the two methods can make better results.