Historical Orthoimage : Différence entre versions

De MicMac
Aller à : navigation, rechercher
(DEM processing and orthorectification)
(Absolute orientation)
Ligne 154 : Ligne 154 :
  
 
== Absolute orientation ==
 
== Absolute orientation ==
 
+
First, input a few ground control points (GCPs) on a few images (here, on 3 images only), to give a first approximation of the geo-referencing.
 
<pre>mm3d SaisieAppuisInitQT "OIS-image1.tif" Relative id_appuis.txt MesuresInit.xml
 
<pre>mm3d SaisieAppuisInitQT "OIS-image1.tif" Relative id_appuis.txt MesuresInit.xml
 
mm3d SaisieAppuisInitQT "OIS-image2.tif" Relative id_appuis.txt MesuresInit.xml
 
mm3d SaisieAppuisInitQT "OIS-image2.tif" Relative id_appuis.txt MesuresInit.xml
 
mm3d SaisieAppuisInitQT "OIS-image3.tif" Relative id_appuis.txt MesuresInit.xml</pre>
 
mm3d SaisieAppuisInitQT "OIS-image3.tif" Relative id_appuis.txt MesuresInit.xml</pre>
 +
 +
Use these points to get into the cartographic coordinate of the GCPs.
 
<pre>mm3d GCPBascule "OIS-.*tif" Relative TerrainInit Mesures_appuis.xml  MesuresInit-S2D.xml</pre>
 
<pre>mm3d GCPBascule "OIS-.*tif" Relative TerrainInit Mesures_appuis.xml  MesuresInit-S2D.xml</pre>
<pre>mm3d SaisieAppuisPredicQT "OIS-.*tif" TerrainInit Mesures_appuis.xml MesuresFinales.xml</pre>
+
 
 +
Then input all the GCPs on all the images (or at least quite a few), using pre-pointed approximate GCPs:
 +
<pre>mm3d SaisieAppuisPredicQT "OIS-image2.tif" TerrainInit Mesures_appuis.xml MesuresFinales.xml</pre>
 +
Again, use these points (now more numerous) to get into the cartographic coordinate of the GCPs.
 
<pre>mm3d GCPBascule "OIS-.*tif" TerrainInit TerrainBrut Mesures_appuis.xml  MesuresFinales-S2D.xml</pre>
 
<pre>mm3d GCPBascule "OIS-.*tif" TerrainInit TerrainBrut Mesures_appuis.xml  MesuresFinales-S2D.xml</pre>
 +
Perform a bundle adjustment and a refinement of the camera calibration using the GCPs:
 
<pre>mm3d Campari "OIS-.*tif" TerrainBrut TerrainFinal GCP=[Mesures_appuis.xml,0.1,MesuresFinales-S2D.xml,0.5] AllFree=1</pre>
 
<pre>mm3d Campari "OIS-.*tif" TerrainBrut TerrainFinal GCP=[Mesures_appuis.xml,0.1,MesuresFinales-S2D.xml,0.5] AllFree=1</pre>
 +
 +
Visualize if wanted:
 
<pre>mm3d AperiCloud "OIS-.*tif" TerrainFinal SH=HomolMasqFiltered</pre>
 
<pre>mm3d AperiCloud "OIS-.*tif" TerrainFinal SH=HomolMasqFiltered</pre>
 
<pre>meshlab AperiCloud_TerrainFinal.ply</pre>
 
<pre>meshlab AperiCloud_TerrainFinal.ply</pre>

Version du 31 janvier 2017 à 16:21

Picto-liste.png Tutorials index

Description

This tutorial will present the method to process historical aerial images into DEM and Orthoimages. With this kind of products, you can monitor changes in an arean (urbanization, landscape changes, etc...).

The USGS NAPP program offers a large amount of free scanned images over the continental US (mostly), often with calibration data, though the Earth Explorer.

If you are looking for a special area in France, you can use the Geoportail (IGN) to download your own images and process it.

Download

Presentation

Tutorial

Internal Orientation

MicMac use EXIF metadat in order to determine image format and focal length. However, historical images often don't have such metadata, so we have first to create a xml file called MicMac-LocalChantierDescripteur.xml.

Example of MicMac-LocalChantierDescripteur.xml
Change the values according to your camera.
<Global>
  <ChantierDescripteur >

    <!-- Define a camera model (name and sensor/film size) -->
    <LocCamDataBase>
        <CameraEntry>
              <Name> ZeissRMKATOP15  </Name>
              <SzCaptMm>  226.004 226.008  </SzCaptMm> <!-- MidSideFiducials or "MaxFidX-MinFidX MaxFidY-MinFidY"-->
              <ShortName> Zeiss RMK A Top15* and Zeiss Pleogon A3/4 </ShortName>
         </CameraEntry>
    </LocCamDataBase>

    <!-- Associate images to a camera model -->
    <KeyedNamesAssociations>
            <Calcs>
                 <Arrite>  1 1 </Arrite>
                 <Direct>
                       <PatternTransform> .*  </PatternTransform> <!-- Regular expressions of the group of images with the following camera model -->
                       <CalcName> ZeissRMKATOP15 </CalcName> <!-- Name of the camera for these images -->
                 </Direct>
             </Calcs>
             <Key>   NKS-Assoc-STD-CAM </Key>
    </KeyedNamesAssociations>
	
    <!-- Associate images to a focal length -->
    <KeyedNamesAssociations>
            <Calcs>
                 <Arrite>  1 1 </Arrite>
                 <Direct>
                       <PatternTransform> .*  </PatternTransform> <!-- Regular expressions of the group of images with the following focal length -->
                       <CalcName> 153.664 </CalcName>	<!-- See calibration report -->
                 </Direct>
             </Calcs>
             <Key>   NKS-Assoc-STD-FOC  </Key>
    </KeyedNamesAssociations>
	
  </ChantierDescripteur>
</Global>

Scanned images also need to be normalized so the calibration is the same for all images. In order to achieve that, the fiducial marks coordinates need to be know both in film space (these values should be in the calibration report) and in image space.

FiducialCoord
Fiducial Coordinates from USGS Report No. OSL/2782

To report the film space coordinates to MicMac , you need to create an xml file called MeasuresCamera.xml in a sub folder called Ori-InterneScan. MicMac requires the origin of the system to be the top left corner, so the coordinates from the calibration files (that usually are centered in the center of the image, with the Y axis going upwards) need to be manipulated : Y axis inverted (Yinv=-Y) and then the coordinates translated (X'=X-Xmin and Y'=Yinv-Yinv_min). Be careful with calibration files that might use different names for the fiducial marks than the ones printed on the images, and also the orientation of the images that may be wrong.

Example of MeasuresCamera.xml corresponding to USGS Report No. OSL/2782
Change the values according to your camera.
<?xml version="1.0" ?>
<MesureAppuiFlottant1Im>
     <NameIm>Glob</NameIm>
     <OneMesureAF1I>
          <NamePt>P1</NamePt>
          <PtIm>1.0040  226.9950</PtIm>
     </OneMesureAF1I>
     <OneMesureAF1I>
          <NamePt>P2</NamePt>
          <PtIm>226.9920    0.9960</PtIm>
     </OneMesureAF1I>
     <OneMesureAF1I>
          <NamePt>P3</NamePt>
          <PtIm>0.9960    1.0070</PtIm>
     </OneMesureAF1I>
     <OneMesureAF1I>
          <NamePt>P4</NamePt>
          <PtIm>226.9930  226.9950</PtIm>
     </OneMesureAF1I>
     <OneMesureAF1I>
          <NamePt>P5</NamePt>
          <PtIm>0.9940  114.0040</PtIm>
     </OneMesureAF1I>
     <OneMesureAF1I>
          <NamePt>P6</NamePt>
          <PtIm>226.9980  113.9940</PtIm>
     </OneMesureAF1I>
     <OneMesureAF1I>
          <NamePt>P7</NamePt>
          <PtIm>114.0000    1.0060</PtIm>
     </OneMesureAF1I>
     <OneMesureAF1I>
          <NamePt>P8</NamePt>
          <PtIm>113.9950  227.0140</PtIm>
     </OneMesureAF1I>
</MesureAppuiFlottant1Im>

Then to input the image coordinate of the fiducial marks, you should use the SaisieAppuisInitQT command on each image like this (id_fiducial.txt is a text file with a point name on each line):

mm3d SaisieAppuisInitQT "image.tif" NONE id_fiducial.txt MeasuresIm-image.tif.xml 

The resulting MeasuresIm-image.tif-S2D.xml file (created in the image folder) should be moved in the Ori-InterneScan directory and renamed MeasuresIm-image.tif.xml (note the removal of "-S2D").

If you have images where the fiducial marks are easily recognizable (they look like targets, not just a dot), and if the images are already close to be aligned, you can use Kugelhupf to compute the position of the points starting with the second image (appearance and position of the points are dictated by the first image that you processed manually).

mm3d Kugelhupf .*tif Ori-InterneScan/MeasuresIm-image.tif.xml SearchIncertitude= ??

Then all the images can be re-sampled to fit in the same geometry and can therefore be processed like digital images. The user need to input the scan resolution (in the example line, 0.025 -> 0.025mm=25microns). This process is slow (ca. a minute per image), but is parallelised.

mm3d ReSampFid ".*.tif" 0.025

The user should then move the original images to a sub-folder, or state OIS.*.tif as the regular expression in futur steps.

Relative orientation

First, you need to find tie points between your images:

mm3d Tapioca MulScale "OIS.*tif" 1000 2500

Be aware that you shouldn't use a very high resolution for finding tie points in scanned because of both the usually very large image files and the noise often present in scanned data.

If camera postions are (approximately) known

If you have the position of the camera for each image (set in a txt file similar to a GCP file), you can create a file with the reference of images potentially in contact (sometimes, the info is printed on the images). In that case, run this instead of the aforementioned Tapioca command:

mm3d OriConvert OriTxtInFile GPS_sommets.txt Sommets NameCple=Couples.xml
mm3d Tapioca File Couples.xml 2000

To be able to ignore the fiducial marks and other inscriptions on the images that would yield nonsensical tie points, a mask need to be created.

mm3d SaisieMasqQT ""OIS-image.tif"

Once created, the mask should be renamed filtre.tif.

mm3d HomolFilterMasq "OIS.*tif" GlobalMasq=filtre.tif

Because historical images were typically taken with long focal lenses, only at a nadir point of view and with limited overlap, the calibration is not very stable. A good way to constrain it is by fixing the focal length at the value stated in the calibration report, hence the LibFoc=0 option in Tapas.

mm3d Tapas RadialBasic "OIS.*tif" InCal=CalibInit Out=Relative SH=HomolMasqFiltered LibFoc=0
mm3d AperiCloud "OIS.*tif" Relative SH=HomolMasqFiltered
meshlab AperiCloud_Relative.ply

Absolute orientation

First, input a few ground control points (GCPs) on a few images (here, on 3 images only), to give a first approximation of the geo-referencing.

mm3d SaisieAppuisInitQT "OIS-image1.tif" Relative id_appuis.txt MesuresInit.xml
mm3d SaisieAppuisInitQT "OIS-image2.tif" Relative id_appuis.txt MesuresInit.xml
mm3d SaisieAppuisInitQT "OIS-image3.tif" Relative id_appuis.txt MesuresInit.xml

Use these points to get into the cartographic coordinate of the GCPs.

mm3d GCPBascule "OIS-.*tif" Relative TerrainInit Mesures_appuis.xml  MesuresInit-S2D.xml

Then input all the GCPs on all the images (or at least quite a few), using pre-pointed approximate GCPs:

mm3d SaisieAppuisPredicQT "OIS-image2.tif" TerrainInit Mesures_appuis.xml MesuresFinales.xml

Again, use these points (now more numerous) to get into the cartographic coordinate of the GCPs.

mm3d GCPBascule "OIS-.*tif" TerrainInit TerrainBrut Mesures_appuis.xml  MesuresFinales-S2D.xml

Perform a bundle adjustment and a refinement of the camera calibration using the GCPs:

mm3d Campari "OIS-.*tif" TerrainBrut TerrainFinal GCP=[Mesures_appuis.xml,0.1,MesuresFinales-S2D.xml,0.5] AllFree=1

Visualize if wanted:

mm3d AperiCloud "OIS-.*tif" TerrainFinal SH=HomolMasqFiltered
meshlab AperiCloud_TerrainFinal.ply

DEM processing and orthorectification

Create a pseudo orthoimage (with a "flat" terrain as target) to be able to draw a mask on the approximate area of interest.

mm3d Tarama "OIS-.*tif" TerrainFinal
mm3d SaisieMasqQT TA/TA_LeChantier.tif

Compute the DEM (DEM is the file called MEC-Malt/Z_Num7_DeZoom2_STD-MALT.tif , or similar).

mm3d Malt Ortho "OIS-.*tif" TerrainFinal MasqImGlob=filtre.tif NbVI=2 ZoomF=2 ResolTerrain=0.5 DefCor=0 CostTrans=4

Once the DEM is generated, other products can be generated : A hillshade:

mm3d GrShade  Z_Num8_DeZoom2_STD-MALT.tif ModeOmbre=IgnE Out=Hillshade.tif Mask=MEC-Malt/AutoMask_STD-MALT_Num_7.tif

An image representation of the DEM:

mm3d to8Bits MEC-Malt/Z_Num8_DeZoom2_STD-MALT.tif Out=hypso.tif Coul=1 Dyn=3 Mask=MEC-Malt/AutoMask_STD-MALT_Num_7.tif

An Orthoimage:

mm3d Tawny Ortho-MEC-Malt Out=Mosaique.tif

A point cloud (drapped with the ortho image):

mm3d Nuage2Ply MEC-Malt/NuageImProf_STD-MALT_Etape_7.xml Attr=Ortho-MEC-Malt/Mosaique.tif Out=PointCloud.ply