E M O R T Making DEMs H O Orthophotos

Getting Started

Making DEMsand

Orthophotos

with

TNTmips®

DEM/ORTHO

page 2

Making DEMs and Orthophotos

Before Getting StartedThis booklet introduces techniques for extracting elevation raster objects (DEMs)from stereo imagery in TNTmips® and creating orthoimages by applying theelevation information to the source imagery. These processes are among themost powerful and complex in TNTmips and should not be casually approachedby the beginner. This booklet shows how the basic steps work. It does notpretend to teach the complex science of digital photogrammetry.

Prerequisite Skills This booklet assumes that you have a fair amount of experi-ence with TNTmips. At the minimum, you should have completed the exercisesin Getting Started: Displaying Geospatial Data, Getting Started: Navigating,Getting Started: Georeferencing, and Getting Started: 3D Perspective Visual-ization. The exercises in those booklets introduce essential concepts and skillsthat are not covered again here.

Sample Data The exercises in this booklet use sample data that is distributedwith the TNT products. If you do not have access to a TNT products CD, youcan download the data from MicroImages’ web site. In particular, this bookletuses objects in the AIRPHOTO, EPIPOLAR, DEM_TIN, and ORTHO Project Files in theDEMORTHO data collection. An option in the TNT installation process makes aread-write copy of these files on your hard drive; you may encounter problems ifyou work directly with the read-only sample data on the CD-ROM.

More Documentation This booklet is intended only as an introduction to DEMextraction and orthoimage creation. Consult the TNT reference manual for moreinformation.

TNTmips and TNTlite™ TNTmips comes in two versions: the professional ver-sion and the free TNTlite version. If the software license key for the professionalversion is not attached to your computer, TNT operates in TNTlite mode, whichlimits object size and enables data sharing only with other copies of TNTlite.

DEM extraction and orthoimage creation are not available in TNTedit or TNTview.The processes are available in TNTlite, but the TNTlite size limits make it hard toprovide real-world sample data that produces meaningful results. The sampledata used in the exercises in this booklet exceed the TNTlite size limits.

Keith Ghormley, 21 September 2000

It may be difficult to identify the important points in some illustrations withouta color copy of this booklet. You can print or read this booklet in color fromMicroImages’ web site. The web site is also your source for the newestGetting Started booklets on other topics. You can download an installationguide, sample data, and the latest version of TNTlite:

http://www.microimages.com

page 3


All airphotos contain systematic distortions causedby camera angle, distance, and surface elevation.As a result, their internal geometry differs fromthe ideal geometry of map products and thereforethey are of limited use in professional geospatialanalysis. In some cases, simple resampling and warp-ing processes can produce corrections that are “goodenough.” But a more rigorous solution is to usedigital photogrammetric techniques to createorthophotos, which have map-like geometry.Orthophotos provide an accurate image base formany projects in geospatial analysis.

The Photogrammetric Modeling process in TNTmipsproduces orthophotos in five steps:· Georeference· Interior Orientation· Relative Orientation· DEM Extraction· Orthorectification

This booklet introduces each step of the process.

Please note that in one sense there is no such thingas “getting started” with digital photogrammetryin TNTmips. Photogrammetric Modeling is no placefor beginners. Ideally, you would be well schooledin the principles of photogrammetry before you satdown at a computer. If you lack a university levelbackground in photogrammetry, you should not ex-pect to follow the examples and exercises in thisbooklet easily.

A series of intermediate results are included in thesample data, so you will be able to continue withthe sequence of exercises, even if you are not ableto produce usable results with the previous exer-cise. Don’t be discouraged. But neither shouldyou proceed with naive hopes of learning an easyway to do digital photogrammetry.

An epipolar image pair hasbeen resampled to have thesame cell size, and rotatedso that all stereo parallax isin the horizontal dimension.

A DEM (Digital ElevationModel) is a raster object thatcontains elevation values fora site.

An orthophoto is a digitalimage that has beenprocessed to correct fordistortions of cameraperspective and surfaceelevation. An orthophotohas map-like geometry and isuseful as an accurate imagebase for geospatial analysis.

The exercise on page 5discusses georeference.Page 6 presents interiororientation. Pages 7-9 coverrelative orientation. Pages10-15 introduce DEMExtraction. Pages 16-18discuss orthoimage creationand evaluation.

A TIN (Triangulated IrregularNetwork) is a set of 3Dnodes connected by edges toform a network of triangles.TINs are fast, efficientstructures for representingelevation surfaces.


page 4


Stereo to DEM to OrthophotoOverlapping airphotos contain stereo informationthat can be digitally processed to extract elevationinformation. The elevation information can in turnbe used to adjust the position of features in theimages to correct for systematic distortions causedby camera perspective and surface elevation. Inuncorrected airphotos, tall features, nearer to thecamera, are too big and appear to lean away fromthe center of the photo. After compensating forcamera tilt, digital orthophoto techniques adjustthe position of features inward and reduce theirsize, based on their elevation and their distancefrom the center of the photo.

Thus, if you have a pair of stereo airphotos, youcan create a DEM and an orthoimage for the over-lap area. Likewise, if you already have a DEMand one airphoto, you can create an orthoimage.And of course, if all you want is a DEM for yourstudy site, you can create one from a pair of stereoairphotos and not bother with creating anorthoimage.

A pair of georeferenced airphotoswith stereo overlap region

a small set of orientation tie points

Left-Right pair containing theextracted overlap region, processed tohave epipolar orientation

a large set of parallax tie points

TIN and DEM extracted from theLeft-Right epipolar pair

Orthoimage created from the DEMand any georeferenced image.

Be sure you make a read-write copy of the sample data on your hard drive. You cannotcomplete these exercises using the read-only sample data directly from the CD-ROM.

Orthorectification

Relative Orientation

Interior Orientation

DEM Extraction

The quality of the DEMproduced by stereoscopicanalysis is limited by thequality of the input stereoimages. Clear, high-contrast images that havemany widely distributedidentifiable features arebest. Images of largefeatureless regions ofuniform landcover (such asflat agricultural fields) willusually not support detailedDEM output.

page 5


Prerequisite GeoreferenceBefore you open the Photogrammetric Modelingprocess, you must prepare your input airphotos byestablishing suitable georeference (Edit /Georeference). If you don’t have georeference con-trol, you can still extract relative elevation, but if youwant real-world elevation values in the DEM, and ifyou want to produce an orthoimage, you must sup-ply accurate map coordinates and elevation values.

As you enter map coordinates, be sure to enter anelevation value for each georeference point. Inparticular, enter elevation control points for the highand low features in the image. If you have a moun-tain top that is 5,000 meters, put a georeferencepoint on that peak even if you must use estimatedmap coordinates (surround the mountain with lowelevation control points of known coordinates, andthen put a point on the peak using the EstimateCoordinates button and enter the known elevationof the peak).

Consider: If you use an x-ydigitizer with a 1:24,000 topomap for georeferencecontrol, then 1 mm on theprinted map = 24 meters onthe ground. So if yourdigitizer click is off by 1 mmand your airphoto cell size isbetween 3 and 4 meters,then your map control isaccurate only to 6 to 8 pixels(worse if you account for theaccuracy standards of theprinted map).

For help with theGeoreferenceprocess, refer to theTNTmips reference manual,and to Getting Started:Georeferencing.

Examine your georeferencepoints for accuracy. If yourresidual values are muchlarger than the cell size ofyour images, then the pointmay be misplaced.

Every georeference point must have an elevation valueassociated with it. In particular, put georeference points onthe extremes of high and low elevation in the scene.

You may open AIRPHOTO / RED145 in the Edit / Georeferenceprocess to examine its georeference control.

page 6


Interior OrientationThe DEM extraction and orthoimage creation pro-cess is found on the TNT menu path Process / Ras-ter / Photogrammetric Modeling.

The first information the DEM and orthophoto pro-cess needs is used to establish the interior orienta-tion of the georeferenced input airphotos. The pro-cess must know· the focal length of the camera lens that was

used to collect the images,· the scanner resolution used to digitize the pho-

tos, and· the cell address of the principal point.

With full airphotos, use Display / Spatial Data toexamine the airphotos to determine the cell ad-dress of each fiducial mark, and then calculate thecell address of the Principal Point. (If the fiducialmarks are on the edges, then the column addressof the Principal Point is halfway between the col-umn addresses of the top and bottom fiducialmarks; the line address of the Principal Point ishalfway between the line addresses of the left andright fiducial marks.)

STEPS� launch the Photogram-

metric Modeling process� choose Interior

Orientation from theMode option button

� for the left image, selectRED145 in the AIRPHOTO

Project File� select RED146 for the

right image� examine the Focal

Length, ScannerResolution, and PrincipalPoints addresses in theCamera Parameters tab

The sample data is cut out from a full airphoto, sothe coordinates of the principal point reflect itslocation outside the extents of the raster.

The Principal Point is the addressof the cell where lines that connectopposite fiducial marks intersect.

Fiducial marks show as pinholeson the edges or in the corners ofairphotos.

In the Camera Parameterstab, record focal length,scanner resolution andPrincipal Point addresses.

Interior Orientation requiresonly that you enter theorientation parameters.There is no associatedprocessing operation withthis step.

page 7


Tie Points for Relative OrientationYou will enter tie points at two stages of the Photo-grammetric Modeling process. You must manuallyenter the first, smaller set of orientation tie pointson the airphotos to provide information for the Rela-tive Orientation step. Later, you can use a combina-tion of automatic and manual methods to generate amuch larger set of parallax tie points on the Left-Right epipolar images.

The goal in placing orientation tie points is to estab-lish the rectangular extents and common orientationof the overlap area in the airphotos. Put tie pointson common features in the corners, edges, and onthe center lines of the overlap area.

STEPS� choose Relative

Orientation in the Modeoption button

� click [Open...] in the TiePoints tab

� select the ORIENTATION tiepoint subobject that isoffered

� turn on the ShowGeoreference Points andShow Color CodedCorrelation toggles in theParameters tab

� click Redraw in theView window

The correlation valueshows how well theimmediate area around a tiepoint matches statistically inthe two images. Add tiepoints only if they have astrong correlation value(aim at first for 80% orhigher) AND clearly arepositioned on the sameimage feature.Green, yellow, and blue labels indicate

strong, weaker, and very weak correlation.

Put orientation tiepoints on commonfeatures in thecorners, edges andcenter lines of thestereo overlap area.

To place a new tie point, (1) click on the left image in the View window, (2) adjust thecrosshair position in the magnifier windows, and (3) click the Add push button.

page 8


Adjusting Orientation Tie PointsSTEPS� turn on Show Overlap

Area in the Parameterstab

� add tie points outside theoverlap box if you want toexpand its extents

� click [Adjust AllTiepoints] on the AutoGenerate tab

� manually adjust tie pointpositions if necessary

� click [Save...] to saverevised tie pointsubobject

The orientation tie points you enter on thegeoreferenced airphotos are used to produce theLEFT and RIGHT epipolar images, which form thefoundation of all your later work. Evaluate theorientation tie points carefully to ensure that youget the best possible set.

First, check the placement of the points and the redbox that marks the overlap area (Parameters tab /Show overlap area toggle)

· If the box excludes part of the image you want toinclude, enlarge the box by putting tie points inthe excluded area you want.

· If the box is skewed and has corners that are notat least approximately right-angled, then exam-

ine each tie point in your list: oneor more are probably miss-matched.

Use the Adjust All Tie Pointspush button in the Auto Gener-ate tab to have the process au-tomatically look for ways to im-prove the correlation by adjust-

ing each orientation tie point. After the processadjusts the tie point positions, look again at the

correlation values. Ifyour tie points still havelow correlation values,edit their position manu-ally and then repeat theautomatic adjustmentprocedure.

Turn on Show Overlap Areain the Parameters tab.

The red overlap areabox should not beskewed and shouldhave right-anglecorners.

page 9


Process for Relative OrientationAfter you have a good set of orientation tie points,you are ready to run the Relative Orientation pro-cess. Check to see that you have tie points aroundthe perimeter of the area you want to process andthat the overlap area box has square corners. Thenclick the Run icon button.

The Relative Orientation process extracts a rasterpair from your source airphotos. The new LEFT-RIGHT pair is resampled to have the same cell sizeand rotated to have epipolar orientation.

The Relative Orientation process closes the inputairphotos and automatically opens the new LEFT-RIGHT epipolar pair in the View window. Whenthe process opens the new epipolar pair, it orientsthem so that cross-eyed stereo viewing is possible.As you add parallax tie points (refer to the nextexercises), you can periodically look at the pair incross-eyed stereo to make sure you are includingthe high and low points of elevation in the scene.

STEPS� visually check the

distribution of your tiepoints

� click the Run iconbutton

� resize the View windowto fit the aspect of theimages, and attempt toview the new Left-Rightepipolar images in cross-eyed stereo

Human eyes do anexcellent job of perceivingstereo information. If yourimagery is too poor forstereo visualization, there islittle chance that thephotogrammetric modelingprocess will be able toidentify and processenough image parallax tocreate a good DEM.

Cross-eyed stereo viewing is difficult for some people, but it can be a quick anduseful technique, once you get the hang of it. If you don’t know how to do cross-eyed3D, try this: (1) hold a pencil a few inches in front of your nose as you look at the Left-Right epipolar stereo pair on the computer screen 20 - 25 inches away. While looking atthe stereo pair, move the pencil forward and back until its unfocused, double image iscentered in front of each image on the screen. (2) Then refocus your eyes on thepencil. The stereo images will appear to have a third image between them (directlybehind the pencil). Keep your focus on the pencil, while you concentrate on the“middle” image. When your eyes adjust, the image should resolve into 3D and you canmove the pencil out of the way. The trick is to hold the features in the “middle” imagetogether against the tendency of your eyes to re-focus and let the stereo view slideapart into the discrete component pair again.

(1) Focus on the image pairwhile you position a pencil.

(2) Focus on the pencil, then on the “center”image behind it.

(For other stereovisualization tech-niques, refer to GettingStarted: 3D Perspec-tive Visualization.)

Don’t give up too easily; ittakes practice. Cross-eyedstereo is similar to the popular“magic eyes” 3D pictures andposters.

page 10


Parallax Tie Points for DEM ExtractionSTEPS� turn off Show Overlap

Area (Parameters tab)� click [Open] in the Tie

Points tab and select theTIEPOINT subobject

� click [Adjust All Tiepoints]on the Auto Generate tab

� add a tie point as near aspossible to eachgeoreference point

� click [Auto Generate TiePoints] in the AutoGenerate tab

In some cases you maydecide to disregardcorrelation values. Forexample, if an open watertank in a remote, featurelessarea shows clearly in bothphotos, it may make a goodtie point, even if one photohas a sun flash on the waterthat gives the feature a verylow correlation value.

The Relative Orientation process created a LEFT-RIGHT epipolar pair of images that includes onlythe extents of the stereo overlap area. These newLEFT and RIGHT raster objects are loaded automati-cally as the input for the DEM extraction process.

You need a much larger set of parallax tie pointsfor DEM extraction than you used for RelativeOrientation. While the orientation tie points onthe airphotos were placed in a regular pattern,around the edges and down the center lines, theparallax tie points on the epipolar pair are to beplaced across the entire image, with special em-phasis on elevation extremes in valleys and onridges.

Open the tie point set that was carried forward fromthe Relative Orientation step (the new copy wasautomatically named TIEPOINT). Manually add tiepoints across the images, keeping an eye on thecorrelation value under the Right Image magni-fier in the Tie Points tab, trying for tie points withan 80 percent or higher correlation to begin with.

18 orientation tie pointsplaced manually.

559 parallax tie pointscreated with two iterations ofguided TIN densification inthe Auto Generate process.

page 11


Iterative Generation of Parallax Tie PointsThe Auto Generate feature allows you to create sev-eral generations of tie points using the TIN densifi-cation algorithm. In TIN densification, the processconstructs a temporary TIN from the initial set of tiepoints. Then the process looks in the middle ofeach triangle on both images and searches locallyfor a new point of high correlation. If it finds a goodpoint, the process adds it to the parallax tie point listand reconstructs the TIN.

Click the Auto Generate Tie Points button in theAuto Generate tab to create a new generation oftie points. Unless you disable the Auto Load newTie Points toggle in the Auto Generate Tab, theprocess automatically loads the new tie points,gives you the option of saving the previous set,and shows the new points in the View window. Inthe next two exercises, you will survey the distri-bution of the new tie points to look for areas thatare sparsely populated, andyou will also look at the Y re-siduals in the tie point list toidentify tie points that are in-correctly placed.

STEPS� turn on Show Color

Coded Residuals in theParameters tab

� if the tie point labels areall blue, click [Adjust AllTie Points] in the AutoGenerate tab

� click [Auto Generate TiePoints] to create asecond generation of tiepoints

The temporary TIN used ineach iteration of thedensification process canbe saved after each runas a TIN object if you pushin the Save As TIN toggle.Then you can view eachTIN in Display / 3D Groupto see how the surfacelooks so far.

You can refresh the tie pointlist and update the Y parallaxvalues by pressing [Adjust AllTie Points].

Click Auto GenerateTie Points to create anew generation of tiepoints by TINdensification.

TIN densification quicklygenerates many goodparallax tie points. Theprocess naturally concen-trates tie points in areaswith highly definedfeatures while “avoiding”areas of uniform, feature-less appearance.

page 12


Evaluating and Adding Tie PointsSTEPS� turn off the Show Tie

Points Labels toggle inthe Parameters tab

� select the Tie Points taband manually add tiepoints in sparse areas

� click [Auto Generate TiePoints] in the AutoGenerate tab

For tie point placement, pickfeatures thathave distinctpatterns oflight-darkboundaries,such as abright rockoutcrop in adarkervegetatedarea, or adark tree in a light field.

Help the processfind more correlationin problem areas bymanually addingparallax tie points.

Increase the triangle sizeto steer the processaway from dense, highlycorrelated areas.

After each Auto Generate iteration, you may noticethat new parallax points are concentrated in certainareas of the image, while other areas have been ne-glected. Not surprisingly, the TIN densification al-gorithm is very good at finding and following areasof strong correlation, and not as good at addingpoints to featureless areas of uniform texture andlow correlation. Areas that have poor correlation,will have reduced detail and accuracy in the outputDEM. (A poor DEM may still be suitable for

Orthoimage creation, even though it maynot be acceptable for other uses.)

You can do two things to steer the pro-cess towards areas that need more paral-lax tie points. First, increase the Mini-mum Triangle Area to keep the processfrom adding more small triangles indensely correlated areas. Second, manu-

ally add tie points in the areas with few points. Yournew tie points will force the process to look again in

problem areas.

If you still aren’t getting enoughpoints in difficult areas, decrease theminimum correlation value (Min.Cross Correlation) so that more tiepoints will qualify. Of course, if youlower the minimum correlation valuetoo far, you may start getting falseparallax points, which will give in-correct results in the DEM output.

page 13


Correcting High Y ResidualsThe essential quality of theepipolar Left-Right pair is thatthe stereo parallax is all locatedin the X dimension. Thus, theleft-right stereo effect is createdbecause the position of commonfeatures differs in only the horizontal direction. (Allvertical displacement was minimized in the RelativeOrientation step, which created the epipolar pair.)As a result, when you examine the parallax values inthe tie point list, the parallax values should be verylow (a cell or less) in the Y dimension. Any tie pointthat has a Y parallax larger than a couple of cells isalmost certainly positioned incorrectly. If you try torun DEM Extraction with a set of tie points that in-cludes such bad tie points, the process warns youof the problem with a Verify dialog box like the oneillustrated at the top of this page. Do not proceedwith the DEM Extraction step while your tie pointlist includes points with high Y-parallax. Select eachbad point from the list, repositionit, click the Update button, andlook at the new parallax value inthe tie point list to be sure thepoint is now correctly placed.

STEPS� scroll through the tie

point list and select ahigh residual tie point

� manually correct itsposition

� click [Update]� look for the next high-

residual tie point

93% correlation usually indicates agood tie point, but this diagonal roadon a featureless background showsthat high correlation does not alwaysmean matching features.

Points with high Y residualsmust be deleted orcorrected by adjusting theirpositions and clicking[Update].

The Filter tab offers an automatic wayto delete tie points with high Y residuals.

page 14


Make TIN and DEMSTEPS� view one of your tie point

TIN’s (or DEM_TIN /TIEPOINT_TIN) in Display /Spatial Data / 3D

� choose DEM Extractionin the Mode optionbutton

� click [Estimate Param-eters]

� click [Save Result AsTIN] in the Options tab

� select 16-bit signedinteger as the OutputData Type in the Optionstab

� click the Run iconbutton and create anew object for theoutput

3D Perspective view of a tie pointTIN with 2230 nodes.

You can view each TINobject in thestandard displayprocess or the 3D Groupdisplay process to see how thecurrent surface looks.

The z values of the TIN nodes arederived directly from the x parallaxvalues of the parallax tie points.

12,000 nodes is enough for orthophotowork. To get sharper surface detail,increase the maximum to 40,000 or more.

Once you have several hundred tie points, you areready to run DEM Extraction. Before doing so,you can use the 3D Group display process to viewthe most recent TIN representation of the elevationsurface (created by the Save As TIN toggle in theAuto Generate tab — see page 11). If you see obvi-ous problems in the surface, go back to pages 12 -13 and correct the related tie points.

Select the DEM extraction process by selecting DEMExtraction in the Mode option button. Click [Esti-mate Parameters] in the Parameters tab to have theprocess derive reasonable processing control val-ues from the current tie point list. Select the Op-tions tab and turn on the Save Result as TIN toggle,and change the Output Data type to 16-bit signed.

Click the Run icon button and when the processfinishes, view the result in theDEM View tab.

NOTE: The Surface Fitting process (Process / Surface Fitting) offers a selection ofadditional methods for creating a DEM raster from the TIN object. For example, youcould choose a high order polynomial fitting method to get a 64-bit floating point DEMwith a continuous, curved surface.

page 15


Evaluate DEMPause at this point in the sequence to evaluate theDEM results. The orthorectification step will usethe DEM to determine how far to move each pixel toproduce the orthoimage. If that is your only consid-eration, then a DEM that shows sharp local detail isno better than a fuzzy low resolution DEM. Moreimportant than sharp local detail is how close thehighest points in the surface are to the true eleva-tion. If a mountain ridge of 1000 feet is representedby elevation values of only 800 feet, then theorthoimage result may be affected (the farther fromthe principal point, the worse the effect). But thepresence or absence of local detail that depicts den-drite minutia of drainages has no measurable effecton the orthoimage. (Nevertheless, you may be inter-ested in local elevation detail for other reasons.)

Use Examine Raster to check known points of eleva-tion at several locations across the DEM, includingvalleys and ridges. Thecell values of the DEMshould be close to theknown control values.

Select the Report tab andreview the extractionsummary statistics.

STEPS� select DEM View tab� visually inspect the

overall DEM� use Examine

Raster in the DEMView tab to comparecell values to knownpoints of elevation

� correct any misplacedtie points (page 13)related to surfaceanomalies and re-runthe DEM extractionprocess (page 14)

The orthoimage producedfrom the DEM created inthese exercises showscorrections of up to 10 - 15pixels. More local detail inthe DEM would not improvethe orthoimage.

This extreme exampleshows a surface anomalycaused by a misplaced tiepoint that has a respect-able correlation of 72%.

The Y-parallax for the misplaced points is 3 pixels (high, but not impossible).Since the X-parallax was much greater than that of surrounding tie points, theprocess concluded there must be a steep surface mound at that point.

The tie point magnifiershows both themisplacement of the tiepoints and the similarityof light/dark imagepatterns that fooled theautomatic tie pointgeneration process.

page 16


OrthorectificationOnce you are satisfied with the DEM, you can ap-ply it to either or both of the Left - Right epipolarimages to create an orthoimage. SelectOrthorectification from the Mode option button.

The orthorectification process is very sensitive tothe georeference and orientation control associatedwith your images. After you select the input airphotoand DEM, the process displays the camera orienta-tion statistics it derives. Even small errors ingeoreference control can result in unusable inputobjects. In particular, look at the Orientation angles:the camera is supposed to be pointing nearly straightdown: so the first two values in the orientationangles should approach 0, 0 (the third angle is notcritical).

Even though the input button is labeled “Left Im-age...” you can select any georeferenced airphotothat shares an overlap area with your DEM. Thus, ifneither of your epipolar images work, try AIRPHOTO /RED146 with DEM_TIN / DEM from the sample data.

STEPS� select Orthorectification

from the Mode optionbutton

� select Left Image andDEM input raster objectsand click [COMPUTE] tosee the OrientationStatistics

� if the processderives a workableorientation, click [Run] tocreate an orthoimage

Problems? Consider: If youuse a 1:24,000 topo map forgeoreference control, then 1mm on the printed map = 24meters on the ground. So ifyour georeferencingdigitizer click is off by 1 mmand your airphoto cell sizeis between 3 and 4 meters,then your original control isis accurate to 6 to 8 pixelsat best.

In this example, the firsttwo computed camerarotation angles deviateradically from the ideal 0,0 (“straight down”)orientation, so theprocess cannot proceed.As a side effect, theCamera Position valuesshow a position below thesurface. Even smallerrors and inaccuracies ingeoreference andorientation control cancause these values tojump out of useful ranges.

page 17


Evaluate OrthoimageNo map product is completely accurate at everypoint, so you should always do some evaluation toget an idea of the accuracy characteristics of yourorthoimage. The kind of reference control you haveavailable will determine the kind of evaluation youcan perform. Ideally, you would compare theorthoimage coordinates reported for the cursor lo-cation in the display process with the coordinatesof a number of accurate survey points that werenot used as input control in the stereoscopic mod-eling process.

If you lack such control values for checking, youmight be able to compare several sample measure-ments made on the airphoto and orthoimage withknown distances from an accurate source. If theorthoimage measurement values are consitentlybetter than the airphoto values, you may reason-ably conclude that the geometry of the orthoimageis better than that of the uncorrected airphoto.

Adopt reasonable expecta-tions and evaluation criteriafor your results consideringthe accuracy and distributionof your georeference control,the quality of your inputimages, and the use youplan to make of the outputraster objects.

Test measurements were made on three images to evaluate the orthoimage.

Not every test measurement compared as well. Nevertheless, all the ORTHOIMAGE

measurements were closer to the values from the topo map than the RED146 values were.

(2) AIRPHOTO / RED146: 1.6340 mi.

(3) ORTHO / ORTHOIMAGE: 1.6100 mi.

(1) the “truth” - a USGS topo map: 1.6103 mi.

page 18


Many users of TNTmips are attracted to the DEMand orthoimage process because it promises to de-liver an accurate, digitally corrected image basefor their project materials. Since orthoimages pre-pared by professional photogrammetrists can beprohibitively costly, the DEM and orthoimage ca-pability in TNTmips looks like a bargain. Others arelooking for ways to create reliable DEMs for areaswhere no other source of DEMs is available.

It sounds too good to be true. Unfortunately, forthe naive user, it often is too good to be true.MicroImages has provided reliable and useful toolsfor making DEMs and orthoimages. However, thebest tools cannot compensate for poor data or poorreference control.

New users contact MicroImages technical supportwith problems that stem from· poor source images: grainy, low-resolution im-

ages, images that contain too little parallax, im-ages that differ greatly in contrast and bright-ness (perhaps collected on different dates orunder different circumstances);

· poor georeference control: control points withresiduals many times greater than the cell sizeof the image; no control points on high-eleva-tion features;

· poor tie point correlation: inability to place use-ful tie points in large featureless areas of thephotos where even the human eye finds too littlestereo information for 3D viewing;

· too few tie points for DEM extraction: blocky,low-resolution DEM results.

If you do not get impressive results on your first try,do not be discouraged. The DEM and orthophotoprocess is by nature very sensitive to the accuracyof input control values. Learn all you can aboutdigital photogrammetry from university-level text-books.

Many of the questions newusers ask do not relate to theTNTmips software, but to thescience of digital photogram-metry. MicroImages cananswer your questions aboutTNTmips, but our softwaresupport engineers cannot beexpected to teach you auniversity course inphotogrammetry.

What Went Wrong?

Inaccuracies in input controlvalues almost alwayscause problems in thephotogrammetric modelingprocess. The processmust have good qualityimagery and accurategeospatial control.

The DEM extractionprocess can produceimpressive results for sometypes of input images. Youcan expect good results ifyour input images havelarge relief displacementsand many high contrastfeatures. You should NOTexpect good results if yourinput images are of flat,featureless terrain thatoffers a uniform appear-ance, such as desert,pasture, or large agricul-tural fields.

page 19


Can the DEM/ortho process produce a contour map with one foot or one meteraccuracy from 3-meter imagery?

No way. The DEM/ortho process uses concepts developed in the science ofPhotogrammetry, the science of making precision measurements from pho-tographs. The key words here are “precision measurements.” 3-meter imagesare simply too coarse to achieve 1-meter results, no matter how you processthem.

What are the main problems that I should anticipate?The most common problem in the application of the DEM/ortho process is inthe expectation that the source images and control data can produce resultsbetter than is photogrametrically possible. The number two problem is inac-curate ground control points and tie-points.

What features are planned for future versions of Photogrammetric Modeling?A visual semi-interactive interface will be added for finding and measuringthe position of fiducial marks on the input airphotos in order to computelocation of the principal point. Another feature will remove distortion fromthe input images based on camera calibration data.

Can I get special help for my project?Yes, within reason. MicroImages software engineers and support specialistsregularly work with clients of all kinds. Please recognize that the DEM/ortho process is complicated and not easy to apply properly. One notableMicroImages client has persevered through several versions of the process toachieve impressive results creating a “DEM” and mosaicked orthoimagesfrom video frames of an underwater archaeological site.

My DEM is fuzzy and doesn’t show much detail. Is it still useful?It depends on the application. Detail in the DEM is not critical in the prepa-ration of acceptable orthophotos, 3D views, and flybys. For those applica-tions, as long as the extremes of elevation in the DEM are close to actualvalues, a fuzzy DEM is good enough. However, accurate detail in the DEMis more important for geomorphology (such as extracting drainage paths andwatersheds), GIS (such as assigning Z coordinates and viewsheds), and engi-neering (such as cut and fill analysis, and gravity flow systems). So if yourapplication requires sharper surface detail, increase the number of parallaxtie points. The more tie points you get, the more local surface detail you willget. (Of course processing takes longer and eventually you will reach a pointof diminishing returns.)

Frequently Asked Questions

Advanced Software for Geospatial AnalysisDEM/ORTHO

Voice: (402)477-9554FAX: (402)477-9559

M icro Im a g es, In c .

MicroImages, Inc. publishes a complete line of professional software foradvanced geospatial data visualization, analysis, and publishing. Contact usor visit our web site for detailed product information.

TNTmips TNTmips is a professional system for fully integrated GIS,image analysis, CAD, TIN, desktop cartography, and geospatial data-base management.

TNTedit TNTedit provides interactive tools to create, georeference,and edit vector, image, CAD, TIN, and relational database project ma-terials. TNTedit can access geospatial data in a wide variety of com-mercial and public formats.

TNTview TNTview has all the same powerful display features forcomplex visualization and interpretation of geospatial materials asTNTmips. TNTview is perfect for those who need flexible access tothe TNT project materials but do not need the technical processing andpreparation features of TNTmips.

TNTatlas TNTatlas lets you publish and distribute your spatial projectmaterials on CD-ROM at low cost. TNTatlas CDs contain multipleversions of the TNTatlas software so that a single CD can be used onany popular computing platform.

TNTserver TNTserver lets you publish TNTatlases on the Internet oron your intranet. Navigate through massive geodata atlases with yourweb browser by using the free, open-source TNTclient Java applet (orany custom applet you create) to communicate with TNTserver.

TNTlite TNTlite is a free version of TNTmips, TNTedit, andTNTview for students and professionals with small projects. You candownload TNTlite for your computer (about 100MB) from MicroImages’web site, or you can order TNTlite on CD-ROM with the current set ofGetting Started tutorial booklets (shipping and reproduction chargesapply).

email: [email protected]: www.microimages.com

11th Floor – Sharp Tower206 South 13th StreetLincoln, Nebraska 68508-2010 USA

Documents

E M O R T Making DEMs H O Orthophotos