Building Heigh Information Extraction From Shadow Derived From High Resolution Satellite Image Based On Scene Classification

With the rapid development of digital city construction,extracting urban building height information from remote sensing images is of great significance to the planning,construction and development of digital cities.Extraction of 3D information of buildings,including plane and heigh information,from high resolution remote sensing images has become a hot research topic in recent years.This research analyzed the shadows of buildings from the aspects of spectral and geometric characteristics,classified the scenes of buildings based on their spatial locations,distribution density,as well as geometic information,and proposed a building shadow detection method which combines multiple features fusion of high-resolution remote sensing images and rule-based object-oriented.A multi scene building height inversion model was established based on the proposed building shadow detection method.A case study of extraction of building height information from high resolution remote sensing images in Xian city was conducted to verify the proposed theory.Wordview-3 high resolution images and non-parameter Google Earth high-resolution remote sensing images were used.The results show that:(1)Building scenes were classified based on the distribution of buildings and their shadows derived from high resolution remote sensing images in terms of location,distribution,and geometry of buildings through field investigation.Within different building sccces,the shadows of buildings were detected and the lengths of the shadow feature line were extracted.A technology framework of urban buildings inversion in different scenarios was proposed considering geometric relationship of shadows of the buildings,sun azimuth,sun altitude,satellite azimuth,satellite altitude,scene complexity of the building and the local terrain.(2)Typical building shadows were detected by a combination method of image multiple features fusion approach and rule-based object-oriented technology.This method can not only effectively weaken the interference on detection results caused by vegetation,water body and dark ground objects,but also can remove non-buildings' shadows.Hence,patches of obtained building shadows were comparatively complete and no small patches were included,Overall,Results had high shadow detection accuracy with 94.7%in simple scenes buildings and 54.1%in complex terrain buildings.In addition,the required information can be kept intact.On this basis,the characteristic line's length of building shadows was acquired by means of the fishnet method and it was compared with the length obtained by manual.Compared with the value measured by manual,the value acquired by the fishnet line was overall high but within an absolute error between 0?1m.Therefore,this method can be used to obtain the length of characteristic line.(3)The height inversion of building heights was gained by the height inversion model established in this paper.The length of the shadow characteristic line was obtained by means of the fishing net method and the manual measurement method.The model's results showed that buildings with the precision consisted with the requirement took up 84.6%,and 92.3%respectively of all;based on the results derived from scale factor inversion,the building's proportion which met the practical precision requirement consisted 73.7%and 89.5%respectively of all simple Wordview-3 image scenes.;Correspondingly,100%and 91.7%respectively of all buildings in complex terrain meet the accuracy requirement;,The total length of the shadow L and the visible shadow L2 of the Google Earth image were regarded as characteristic lines.As a result,The proportion accounts for 100%and 80%,respectively in simple scenes while 92%in complex terrain.A comprehensive analysis was conducted by regarding image spatial resolution,shadow feature line category,shadow feature line length extraction method,scene complexity,height inversion method as variables.Results showed as follows:different shadow characteristic lines should be selected appropriately depending on different building heights of remote sensing images inversion technology;In order to eliminate the influence caused by height calculation on shadow length,the scale factor method can be used to calculate the building height when the scene complexity is high or the image does not contain parameter information;when there are few interference features and the scene complexity is low,both methods can be applied in this situation.The model method which has high automatic character has an advantage in improving the work efficiency and extensive use in inverting building heights.In addition,there is no need to distinguish buildings in the same scene by height.(4)Objective error sources in the process of inverting building height cannot be effectively eliminated.It only can be weakened by manual intervention;while subjective errors are mostly derived from shadow detection and length extraction,whose boundary of shadow are could be processed smoothly by opening and closing operations with a result of less splintery patches;combined with image features,an appropriate method is selected to eliminate outliers in the length extraction process.