SPOT stereo matching for DTM generation Page 1 SPOT stereo m

This paper presents a matching algorithm for automatic DTM generation from SPOT images that provides dense, accurate and reliable results and attacks the problem of radiometric differences between the images. The proposed algorithm is based on a modified v

SPOT stereo matching for DTM generation

Emmanuel P. Baltsavias, Dirk Stallmann

Institute of Geodesy and PhotogrammetrySwiss Federal Institute of Technology

ETH-Hoenggerberg, CH-8093 Zurich, Switzerland

Tel.: +41-1-3773042, Fax: +41-1-3720438, e-mail: manos@p.igp.ethz.ch

ABSTRACT

ThispaperpresentsamatchingalgorithmforautomaticDTMgenerationfromSPOTimagesthatprovidesdense,accurateandreliableresultsandattackstheproblemofradiometricdifferencesbetweentheimages.Theproposedalgorithmisbasedonamodi edversionoftheMultiphotoGeometricallyConstrainedMatching(MPGC).Itisthe rstalgorithmthatexplicitlyusestheSPOTgeometryinmatching,restrictingthusthesearchspaceinonedimension,andsimultaneouslyprovidingpixelandob-jectcoordinates.Thisleadstoanincreaseinreliability,andtoreductionandeasierdetectionofblunders.ThesensormodellingisbasedonKratky’spolynomialmappingfunctionstotransformbetweentheimagespacesofstereopairs.Withtheirhelpepipo-larlinesthatarepracticallystraightcanbedeterminedandthesearchisconstrainedalongtheselines.Thepolynomialfunctionscan also provide approximate values, which are further re ned by the use of an image pyramid.

Radiometricdifferencesarestronglyreducedbyperformingmatchingnotinthegreylevelbutingradientmagnitudeimag-es.Thus,practicallyonlytheinformationinstripesalongtheedgesisusedformatching.Edgesthatexistinonlyoneimagecanbedetectedbysubtractingquasiregisteredimagesintheupperlevelsofanimagepyramid.Thepointstobematchedareselect-edbyaninterestoperator.Grosserrorscanbedetectedbystatisticalanalysisofcriteriathatareprovidedbythealgorithmandby a robust analysis of the heights within local neighbourhoods.

TheresultsofanextensivetestusingastereoSPOTmodeloverSwitzerlandwillbereported.Matchingwithdifferentop-tions and the qualitative comparison of the results based on thirty thousand check points will be presented.

1. INTRODUCTION

ThemotivationbehindthisresearchwastheaimtoimprovematchingofSPOTimagesbyintegrationofthesensorgeometryandthetreatmentofradiometricdifferences.Asfarastheauthorsknow,noneofthepublishedmatchingmethodsexplicitlyex-ploitstheSPOTgeometrytorestrictthesearchspace.SinceSPOTdoesnothaveaperspectivegeometryin ightdirection,ac-curateepipolarimagescannotbegeneratedwithouttheuseofaDTMandthustheexistingmatchingalgorithmsperforma2-Dsearch.TheauthorsinvestigatedKratky’spolynomialmappingfunctions(PMFs)(Kratky,1989a)whichtransformfromimagetoimage,imagetoobjectandobjecttoimagespace.Forthecomputationofthecoef cientsofthePMFstheresultsofarigorousbundleadjustmentareused.ThePMFsaremuchfasterandalmostequallyaccurateastherigoroustransformations.WhentherayofanimagedpointisprojectedbyusingthePMFsontotheotherimageofastereopair(epipolarline)itiswithaverygoodapproximationastraightline.Ifastraightlineisde nedbytheprojectionofasmallraysegmentwhichiscentredatthecorrectpointpositioninobjectspace,thenthedeviationoftheepipolarlinefromthestraightlinewouldbe0.25pixelforaheighterrorofmorethan7km.Thus,theaforementionedstraightlinecanbeusedasaquasiepipolarline.Moredetailsonthecharacteris-tics of the PMFs can be found inBaltsavias and Stallmann, 1992a.

Attemptstocircumventtheproblemofradiometricdifferenceshaveconcentratedonusingimagesacquiredwithinashorttimeinterval.However,thisisdif culttoachieveanddoesnotsolvetheproblem.Thealong-trackstereowillstronglyreducetheseproblemsbutcannoteliminatethemduetodifferentperspectiveviews,cloudsandocclusions.Fusionandmatchingofmultitemporalandmultisensordata,asinchangedetectionapplications,willretaintheirimportanceevenintheeraofalong-trackstereo.Thus,theauthorsdecidedtoattackthisproblem,whichhasbeenuptonowtreatedonlytoalimitedextent.Theideaistousegradientmagnitudeimages,thuseliminatingradiometricdifferencesinareasoflowtexture.Preliminaryinvestiga-tionshaveshownthatthemajorityoftheedgesremainstable.However,differentedgesexistduetoclouds,shadows,differentperspectiveviews,newedgeswithin eldsduetoagriculturalactivities,humanintervention,waterlevel,snowcoverage,chang-es in the tree canopies etc. (Figure 1). A method should be developed to try to detect the different edges.

To be published in: SPIE, Aerospace and Remote Sensing,Orlando, 1993