Research On The Technology Of Recognition Of Bridges Over River In Complex Scenarios

With the development of the computer vision technology,automatic recognition of target in remote sensing image has become both hot topic and emphasis on research.Bridge as a typical and important artificial construction,is the throat of transportation route.Valid recognition of the bridge in collecting images has a wide range of requirements in the civil and military.This paper based on the automatic recognition of bridge on optical remote sensing images in complex scenarios,study on the problem of the automatic extraction of rivers and the orientation of the bridge.Rivers have great differences in different images.The surfaces of rivers are calm and the distributions are uniform while the waves and the water turbidities make the water texture rich and the distributions uneven.Also the background region is diverse.A variety of natural scenery,crops or artificial constructions,such as forest lands,cultivated lands and residents lands,make the background contain various kinds of textures.On this basis,we construct a set of complete water bridge automatic recognition system.The main research content is as follows.(1)Although,the river on bridge images in complex scenarios,presents different forms,for example,its gray value is high or low and the distribution is uniform or uneven,the color similarity between river areas is high while color is different between river areas and background areas.According to the characteristics of color similarity,we propose a unsupervised segmentation method that combined K-means clustering with Harris corner points automatically extracted river area.(2)For more complex bridge images,in river areas,parts of the colors are different.So K-means clustering and Harris corner combining method hardly extract the complete river areas,but can extract part of sample in the river.Using this part of extracted river sample to represent the characteristics of color and texture in the river about the image,and learning form the characteristics,we use self-supervision method to classify all the pixels in the image to segment the complete river areas.(3)After morphological operation and eliminating disturbance areas,we obtain a complete river profile.Then,we dilate and erode rivers binary image while plugging truncation position to obtain connecting rivers binary image.Subtracted with the original river binary image,the truncation areas that are suspected bridges are extracted.According to the characteristic of river skeleton intersects bridges,we can eliminate some false bridges.Thus,based on the characteristics of the bridge inflection points,we verify the existence of the real bridge to complete the real bridge to be obtained.Finally,we mark validated bridges in the position of the original image to implement the positioning of the bridge.(4)The automatic recognition of the bridge we propose is developed with Visual Studio 2010 as the development platform and combined the OpenCV source visual library.Experiments show that the system can automatically recognize the high-altitude water bridge objects in the image and has certain applicability.