Research On The Algorithm Of Road Edge Detection In High Resolution Remote Sensing Images Using Dynamic Programming

As the remote sensing technology develops, space shuttle and satellite system of different sizes have provided sufficient remote sensing images whose spatial resolution, spectral resolution and time resolution have improved greatly for the enrichment of information. The question of how to make the best use of the remote sensing data has drawn crucial attention to scholars of the field of global information science.Because of the importance and the wide application of the road networks data in traffic control, city programming and emergency handling, the road data extraction has been one of the important parts of the remote sensing data extraction. It also has become one of the important research subjects of image interpretation and object recognition. This thesis, focusing on the automatic road extraction, includes the following parts:1. Aiming at the disadvantages of the traditional phase-grouping method when itacquires its line-support regions, this thesis puts forward an improved method of combination phase-grouping. By analyzing and experiments, it selects an adequate gradient computing stencil and then carries out an attended operation on the traditional supporting area. This method has better solved the problems of the deficiency of integrality and the serious breakage between line-support regions, both of which are caused by the traditional phase-grouping method;2. Within a smaller phase-grouping area, combing the dynamic programmingalgorithm and the improved method of combination phase-grouping, the dynamic programming method is applied to search the edges. The combination of these two methods makes the method of road edge detection not only bear the ability of detecting the not necessarily of high contrast edge but also have the advantages of dynamic programming which could detect the arbitrary shapes. Meanwhile it greatly reduces the big amount of operand when adopting the dynamic programming algorithm separately, so the edge searching speed is accelerated.3. When searching the boundary point by using the one dimension dynamic programming algorithm, this thesis adopts the method of dummy starting point and ending point, which has successfully avoided the difficulty and insufficient accuracy when fixing the starting point and ending point.4. By making use of the properties that the edge lines are parallel, start with a longer edge line, detect the candidate position of the external edge and then fix the external position of a certain section. This method could better extract the road section under the situation that only one complete road edge exists.5. After computing the center line of each road section, combine the center lineswhose endpoints approach each other, and whose directions differentiateslightly. Repeat this process, the whole road network could be detected.This thesis chooses two high resolution remote sensing images of different areasof two cities to evaluate the precision of the algorithm from the aspects of vision andquantity respectively. As for the vision aspects, the detected road edge is very close tothe one in the original image. It also could better detect the not necessarily of highcontrast edges. As for the quantity aspects, the average precision is 88.83%.Experiments show that this method can be applied to the road edge detection ofcomplicated images and it has the stronger ability of detecting the not necessarily ofhigh contrast edges.