Photogrammetric Record,16(94):643–650(October1999)
ORTHOPHOTOGRAPH PRODUCTION IN
URBAN AREAS
By D.S KARLATOS
University College London
(Paper read at the Thompson Symposium of the Photogrammetric Society held at the University of Durham on19th April,1998)
Abstract
This paper investigates possible solutions to the problem of urban orthophotography.Extensive work has been carried out to attain the best
possible results,using all the existing software in the Department of
Geomatic Engineering,University College London.Two of the most prom-
滴水观音有毒吗
侦探类小说ising methods,which yield the best results,are analyd further.New
routines were written in C language,as additions to the existing software,
to allow for the special problems that ari in urban areas.Thefinal results
overcome the problem of double mapping(or occlusions)and will poten-
tially enable even greater success to be achieved.A simple accuracy
asssment of the results has been made.Difficulties and current limitations
are analyd and prented.
K EY WORDS:orthophotography,urban imagery
I NTRODUCTION
O RTHOPHOTOGRAPHY,which is a well-known alternative to line mapping,has evolved rapidly in recent years due to developments in computing and technology.It has many advantages over traditional line mapping and many more applications.The fact that orthophotographs are an excellent
backdrop for a GIS is a strong incentive for further investigation.Due to the concentration of people in large towns,the GIS community is being directed towards urban applications and therefore urban orthophotography is needed as a backdrop.Another interesting application which involves urban orthophotography is three dimensional city visualization.
Orthophotographs may be a robust solution for rural areas,but problems ari when they are applied to urban areas,particularly at large scales.The main difficulties and limitations are the double mapping effect,displacement,and concealment of information by high buildings.The aims of the project described in this paper were to investigate ways to overcome the problems and to asss the degree of success that could be achieved.Using existing software at University College London(UCL), it was intended that project results should be balanced in terms of accuracy and visual acceptance,whilst maintaining a maximum level of automation.
643 Photogrammetric Record,16(94),1999
S KARLATOS.Orthophotograph production in urban areas
幼儿小故事100篇F IG.1.The area of study on the left photograph.
Data and Software Available
The existing software included Phodis ST,OrthoMax,Erdas Imagine,Arc/Info and HIPS,a command line image processing package developed in hou.
Two aerial photographs were available,taken over the Bloomsbury region of central London at1Ϻ5500scale.This part of London contains many high buildings, such as the British Telecommunic
ations tower.The area of study,unfavourably positioned towards the edge of the photography,is200ϫ154m in extent and includes buildings with heights of between3m and30m(Fig.1).The calibration report of the camera was available;the focal length of the lens of the camera ud for photography was153mm.The photographs were scanned at42 m with a Sharp JX-600scanner.
The ground control points(GCPs)available were road junctions in the area, digitized from Ordnance Survey1Ϻ1250scale maps,last revid in the early1950s. The points might not have been the perfect choice,but the project was initially conceived as a feasibility study on orthophotograph production.
Lester(1995)had manually collected points over an area of200ϫ154m in order to form a digital elevation model(DEM).The spacing of the points was2m and they were obtained using the Kern(Leica)DSR analytical plotter at English Heritage. The DEM is shown in Fig.2.
M ETHODOLOGY
Two principal methods were ud in the project:merging andfine DEM.A simple accuracy asssment was carried out on the best orthophotograph from each method.
644Photogrammetric Record,16(94),1999
S KARLATOS.Orthophotograph production in urban areas
F IG.2.The manually collected DEM with2m spacing.The black pixels are“void”pixels which were not
measured due to occlusions.
The Merging Method
This method wasfirst introduced by Jenn et al.(1994).Two DEMs are created and then ud to produce two different orthophotographs.One DEM is created from the building points(using points on the roof tops)and the other from the remaining points,called the ground DEM.The reason for using tw
o DEMs is to compensate for the large altitude differences between the ground and roof top points,which produce steep slopes in the DEM surface,resulting in an orthophotograph with many errors. Each DEM is ud to produce the corresponding orthophotograph.Theoretically,in each orthophotograph the corresponding features will be positioned correctly.Hence by using the correct parts from each orthophotograph,it is possible to create a single correct othophotograph.The vectors of the buildings are also required for data input; the were collected manually.
The Fine DEM Method
Thefine DEM method,introduced by Ecker(1992),us a single very good DEM which models the buildings in an optimum manner.The fac¸ades of the buildings should be as steep as possible,which can only be achieved with a den DEM.It can then be ensured that the building fac¸ade appearance is minimal and equal at most to the DEM step.With this method,the vector information for the buildings is still required,collected manually.
There are two alternative procedures for employing thefine DEM method:one step and two step.In the one step procedure,densification and the vector information are ud in a single step to produce thefine DEM.In the two step procedure,thefirst step makes the ground DEM more den and the cond step models the buildings by “raising”the points which are inside the building polygons.
During thefine DEM process,it was possible to mark the DEM pixels hidden behind buildings where there was no information available and therefore it was possible to mark the corresponding pixels in the orthophotograph.
645 Photogrammetric Record,16(94),1999
喝饮料的英文S KARLATOS.Orthophotograph production in urban areas
炒苍术
F IG.3.The merged orthophotograph using3DIM.The vector information is overlaid for accuracy asssment
and also indicates the areas which were“cut”from the building orthophotograph.
P ROCEDURES AND R ESULTS
我爱家乡Different software packages were ud at different stages of the two methods (for example,OrthoMax,Phodis ST and HIPS were ud for orthophotograph production;3DIM,Arc/Info,Phodis ST and OrthoMax were ud to create DEMs). In this paper it is not possible to give details of all the work carried out,so the most reprentative results for each method are described.
It should be mentioned that becau multiple systems were ud,different model orientations were made and therefore degradation of the overall accuracy could be expected.For both methods it was necessary to collect the building vector infor-mation,which was achieved using an analytical stereoplotter(DSR with CADMAP software).
The Merging Method
Both the partial orthophotographs and the corresponding DEMs were produced using the3DIM software.The.dxffile,with the clod polygons reprenting the buildings,was employed to produce a mask,which was ud to add the clod polygons with the image rooftops in the ground orthophotograph.The double mapping effect was not avoided(Fig.3).
646Photogrammetric Record,16(94),1999
S KARLATOS.Orthophotograph production in urban areas
F IG.4.The orthophotograph created by thefine DEM method,with0·25m pixel size.
The Fine DEM Method
福字草书This method was implemented using the3DIM software which could be customized more easily than other packages.Three routines were written in C programming language and introduced into the procedure.
Thefirst routine was ud tofilter the initial DEM for single void pixels which produced aesthetically poor effects in thefinal orthophotograph.In practice,the routine was able tofill in even bigger gaps of up to8pixels,bad on the number and formation of the pixels,which were ur defined parameters.
The cond routine was ud for densification of the DEM.The input consisted of DEM points,the vector information and the output DEM spacing.The DEM with incread point density was produced with the following constraints:
(1)the void pixels/points should be maintained;
(2)all points in the building polygons should have the polygon height or the
nearest building point height;and
(3)the remaining ground pixels should be assigned values using either bilinear
interpolation or nearest neighbour method(if the new pixel is next to a
building rooftop and the bilinear interpolation is impossible,the height value
should be assigned from the nearest ground point).
The third routine was the inver camera model,modified to compensate for the void pixels in the DEM where no information should be printed on thefinal orthophotograph.The result is shown in Fig.4.
647 Photogrammetric Record,16(94),1999
S KARLATOS.Orthophotograph production in urban areas冰箱温度如何调节
T ABLE I.Results from check points.
Left image Right image
s.e.X(s.e.Y(s.e.X(s.e.Y(m) Fine DEM0·2560·2520·2090·246 Mer
ging0·3660·3640·3360·325
Accuracy Asssment
In order to examine accuracy,14points were measured on a digital stereoplotter (with Phodis ST)using the same digital imagery and model orientations as employed for the orthophotograph production.Five ground points and eight building corners were lected as check points.The same points were then also measured on the orthophotographs.The results are given in Table I.
C ONCLUSIONS AN
D D ISCUSSION
At the beginning of this project,the main query concerned the possibility of producing orthophotographs of an urban area such as the centre of London.At the end of the project,not only were aesthetically good orthophotographs produced,but the occlusion areas were automatically marked as part of the orthophotograph production algorithm and the asssment proved that thefinal results are sufficiently accurate to fulfil requirements for mapping.It should not be hastily concluded that all problems of urban orthophotography have been solved.In the author’s opinion,further rearch should be focud on automation.
The merging method is simple and easy.It has been proved that the procedure can be fully automated.The main disadvantage of this method is the manual acquisition of the necessary vector information.It does not need a detailed DEM with a large number of points,in comparison with thefine DEM method which needs ample points to exactly describe the three dimensional man-made objects.With the merging method,the necessary vector information may be ud,in addition to the building points,for the production of the building DEM and therefore improve the accuracy.It is also possible to u only the vector information to form the building DEM using a TIN structure and hence reduce the acquisition time if DEM points are being collected manually.On the other hand,the coar DEMs ud may lead to loss of accuracy in comparison with thefine DEM method.If the orthophotographs are going to be ud as backdrops in a GIS,then the merging method is the better method to u,since it is not as demanding as thefine DEM method.Another disadvantage of the merging method,as implemented in this project,was the difficulty of identifying building corners due to the double mapping effect.
Thefine DEM method,as applied in this project with purpo written software, managed to overcome the problems of occluded areas and to map buildings success-fully.It is difficult to implement becau manual collection of information is time consuming and purpo written software is needed.
However,it produces more accurate results and it is possible,with an extension of the basicfine DEM procedure, tofill in missing information from other orthophotographs bad on the black areas of the existing orthophotograph.The camera model is robust and therefore estimation of occluded areas on every photograph is possible(Amhar and Ecker,1996).The vector information available could be ud to reduce manual collection of points. The densification program of the DEM can work even without points in the polygons.
648Photogrammetric Record,16(94),1999