Semi-automatic road tracking by template matching and distance transform

Semi-automatic extraction of road networks is greatly needed to accelerate the acquisition and update of geodata. However, the road surfaces are seriously disturbed by occlusion of vehicles or shadows on high resolution remotely sensed imagery in urban areas, which makes most of road trackers, using least-squares template matching, inefficient. Fortunately, the scale of many disturbing features such as vehicles, zebras, lane markings is smaller than one of ribbon road surfaces in urban areas. As a matter of fact, Euclidean distance transform can dilate the pure road surface and erode the small disturbing features if a coarsely template matching by thresholding the differences of gray values is firstly performed. Consequently, the Euclidean distance transformation makes the template matching more robust in tracking road networks in urban areas. In this paper, a novel semi-automatic scheme based on template matching and Euclidean distance transformation is presented to extract ribbon roads in urban areas. A scene of QuickBird image over Tai'an area was used for test. The results show our improved method can reliably and robustly extract road networks in urban areas. The main contribution of this paper is that the method of distance transformation besides least squares can be used in template matching to track road networks with much complexity has been strongly proved.