Research On Automatic Detection Method Of Building In Aerial Stereo Image Based On Elevation Constraints

In recent years,the rapid development of cities has put forward higher demands for the current situation and accuracy of urban basic geographical information data.As buildings are the main features of urban areas,the timely updating of geometric information,such as the spatial position and shape and size,is particularly important.How to quickly acquire and update relevant information of buildings has become an important research issue in the field of mapping geographic information.Due to its advantages of relatively mature technology,large imaging scale,high ground resolution,and simple acquisition methods,aviation remote sensing has become one of the major approach to acquiring and updating geographic data.However,because of the existence of "homologous spectrum" and "isospectral foreign matter" in high-resolution images,the traditional primitive-based image processing methods cannot meet the requirements of reliable information extraction.The object-oriented method takes a series of homogenous regions as the research object,makes full use of the characteristics of remote sensing images such as geometry,texture,and spectrum to classify and extract features,which largely guarantees the classification accuracy.However,when object-oriented methods are used to extract buildings,due to the similarity of spectral characteristics of roads and building in aerial image,some of the roads are mistakenly classified as buildings,which will cause many errors and affect the accuracy of building extraction.Height is one of the effective features to distinguish between buildings and roads.Effective use of this information will greatly improve the results of building extraction and the accuracy of building detection.Therefore,based on object-oriented extraction of buildings,this paper makes full use of the elevation information implied in aerial imagery stereo pairs,and proposes a multi-level automatic detection method for buildings based on elevation constraints.Firstly,object-oriented method is used to obtain the initial extraction results of the buildings.Then,the initial extraction result of the building is mathematically morphologically calculated,and then the noise spots are removed by the method of removing the small spots based on the edge detection,and the building refined extraction result map is obtained.Finally,after image matching by using probability relaxation matching method,the elevation of the edge points of all the spots in the refined extraction result is calculated through the beam adjustment method.The threshold value is set based on the elevation to delete the erroneous spots in the refinement result,and get the result of building optimization extraction.This article uses two different sources of aerial stereoscopic imagery for experimentation.The first type of sensor is the UltraCam digital aerial camera,photographed in June 2011,covering the area of Xianlin University City,Nanjing,Jiangsu Province.The second type of data,from the ISPRS test data set,the sensor type is UltraCam D digital aerial camera,covering Toronto,Canada.Comparing building optimization extraction result with building refined extraction result,the accuracy of quantitative extraction is improved from 47.83%to 91.43%,and the best experimental results can reach 100%;in the second set of data,the accuracy of quantitative extraction increases from 81.82%to 93.75%.The experimental results show that the multi-level automatic detection method based on elevation constraints can effectively remove the error information in the results and improve the accuracy of building detection.