5 Preparation for the grid generation
5 Preparation for the grid generation
This chapter describes the Match-T set-up which is the basis for all the automatically generated grid cal-culations which are included in the investigations. The Match-T version employed is version 2.2.0.
5.1.1 Image material
The location of the digital image material must be indicated with the full path connection. In the Match-T version 2.2.0, the image material must be stored in *.itex, *.tif or *.bt format. For each model, the pair of images must be indicated, and the location of the models in the block must be given.
In this project, the images have the format *.bt or *.itex, and it is noted that aerial photos have been used.
Each pair of images and their location paths are given.
5.1.2 Geometric data
Under the heading of geometric data, the terrain type and the mean elevation over the terrain, the calibra-tion report of the camera, control points, earth curvature and refraccalibra-tion are indicated. The informacalibra-tion from the calibration report of the camera can be keyed in or transferred per file, if the information lies in Match-T camera format. The co-ordinates of the control points can be keyed in with an indication of whether it is a planimetric control point, a height control point or a fully measured control point, its accu-racy, etc. The control point co-ordinates can also lie in a separate control point file to which a reference is given with full path connection. Finally, it must be indicated whether the earth curvature and the refraction must be taken into consideration.
In this project the terrain type is given as flat, the mean elevation as 75 m, the information from the cali-bration report of the camera is keyed in manually, the control points are keyed in manually and the earth curvature, refraction and lens distortion have been considered.
5.1.3 External data
A generation of elevations can be supported by external data. This is done by including external elevation data as a part of the calculation itself, for instance a grid or as break lines.
However, this possibility has not been used in this project, even though an actual elevation model exists.
That is because the purpose is to find out how well Match-T is able to determine elevations without the influence of other data. The existing elevation data will only be included as a standard of comparison.
After the initial set up, the project was chosen. The further calculation process is divided into four steps:
orientation of the images, determination of primary data, the DEM generation and, finally, the post-processing.
5.1.4 Orientation of the model
In Match-T, first the inner, then the absolute orientation are determined. The relative orientation is not de-termined, cf. Chapter 2 part 2.4.1.
5.1.4.1 Inner orientation
In Match-T, the inner orientation can be determined manually or automatically.
The automatic measuring of the inner orientation was not stable at the start of the project, giving a poorer result than if the orientation had been done manually, therefore, the inner orientation has been done manually.
5.1.4.2 Absolute orientation
Parameters which might have an influence on the absolute orientation are the possibilities of stereo vision and the definition of the control points. The definition of the control points is dependent on among other things, the resolution of the images.
Stereo vision.
The control points in the images are chosen according to those criteria which are used in a stereo sys-tem. In the Match-T version 2.2.0, orientation of the images is done, however, in a mono-system, there is no possibility of orientating the images in stereo. The control points in a mono-system must be physical objects which can be recognised in the images. Not all the chosen control points live up to this demand.
For example, the height control points are chosen on flat areas, for instance in a field or in the middle of a
5 Preparation for the grid generation
courtyard, where it was possible to see in stereo and thus determine the elevation. The determination of the height control points is, therefore, not possible because of the lack of stereo vision in Match-T.
The influence of resolution on the control points.
The control points are chosen in the analogous images. The experience gained in this project shows that control points chosen in analogous images cannot be transferred directly to digital images. The analo-gous aerial images lose some of their clarity by later scanning which is particularly important to the defini-tion and contrast of the points. The contrast between control points and their surroundings must therefore be greater by digital photogrammetry than by analytical photogrammetry. Their definition in 15 µm images is a little blurred, and even more so in the 30 µm images. As for the 60 µm images, the control points are so blurred that they can only be recognised/localised by an operator who ”knows” where the control points are, and therefore can recognise their location in the images.
To avoid that the standard of comparison, of the different image resolutions, vary because of different ori-entations, it has been decided to ”re-use” the same orientation for each image, regardless of resolution.
The orientation of the individual images can be achieved by use of different methods. For example, the orientation of 15 µm images can be transferred to the 30 µm and 60 µm images, as the definition of the control points is best in the 15 µm images. However, incomplete elevation orientation caused by lack of stereo vision will still be a problem.
As it has not always been possible to select physical objects as height control points, and to avoid differ-ent oridiffer-entations, it has been decided in this project to oridiffer-entate all images in a Planicomp, and do a sub-sequent bundle adjustment in the programme package Bingo. As regards the images in scale 1:5,000, the orientation exists as a consequence of the creation of the frame of reference. As regards images in scale 1:15,000 and 1:25,000, this extra process is done to achieve a uniform orientation of the images.
The orientations achieved are then transferred to the respective images, regardless of their resolution. As a result, the orientation of the individual images will be the same, and the orientation will, therefore, not be included as a source of error in the comparison of the individual analyses, or by the evaluation of the method used for automatic generation of elevations in general.
The orientation of the images.
The table below shows the maximum deviation from the absolute orientation in x, y, z, φ, ωand κ for im-ages in the different scales achieved in Bingo. The results can be seen in Appendix A, section A.3.
Table 5.1 shows that the poorest accuracy is the planimetric, whereas the elevation is somewhat better determined which is important for this project, as it is the elevations which are of interest. If the deviation is expressed in ‰ of the flight altitude, the result is 0.09‰ of the flight altitude for images in scale 1:5,000, 0.07‰ for scale 1:15,000 and 0.05‰ for scale 1:25,000.
The results of the bundle adjustments of the images in 1:5,000, 1:15,000 and 1:25,000 can be found in Appendix A, section A.3 and A.5.
Transfer of the orientation to Match-T
In Match-T, it is possible to transfer the orientation parameters from Bingo automatically. The transfer presupposes, however, that the orientation and scanning of the images are done in the same direction which is not the case in this project. In Planicomp, all the images have been orientated towards south. As regards the scanned images, the greater part is scanned with north upwards. Still, images which were
Scale Accuracy in x (m)
Accuracy in y (m)
Accuracy in z (m)
Max. deviation in φ (mgon)
Max. deviation in ω (mgon)
Max. deviation in κ (mgon)
1 : 5,000 0.103 0.108 0.064 7.9 6.7 3.6
1 : 15,000 0.218 0.226 0.152 5.9 5.3 2.6
1 : 25,000 0.277 0.241 0.197 4.6 3.3 1.7
Table 5.1: The accuracy of the absolute orientation in x, y, z,
φ
,ω
andκ
ages in 1:5,000, the first strip was scanned with north upwards, the second strip with south upwards, the third strip again with north upwards and the last strip with south upwards.
In this project, the orientation parameters are therefore keyed in manually.
5.1.5 Pre-processing of primary data
After the orientation of the images, the pre-processing is carried out. First the images are normalised, then an image pyramid and an object pyramid respectively are created.
5.1.5.1 Normalisation
Each image was normalised, that is, an epipolar geometry was built up according to the principles de-scribed in Appendix 1, section A.1.4. These normalised images were then stored on the hard disk.
5.1.5.2 Image pyramid
For each of the normalised images, an image pyramid was built. As standard, the image pyramid was built by the Gaussian function over 5 x 5 pixels. For images with a resolution of 15 µm, 9 levels more than the original image were built. For images with resolution 30 µm, the image pyramid was built with 8 levels, and for the 60 µm images with 7 levels more than the original image. (Cf. Appendix A, section A.1.7).
The standard set-up for Match-T has been chosen for this project.
5.1.5.3 Object pyramid
The object pyramid with interest points is built from the image pyramid. This is done according to the prin-ciples described in Appendix A, section A.1.8.
Primary data, that is, the normalised images and the image and object pyramids are not changed by the later correlation or DEM generation. This process is therefore not influenced by changes such as, for in-stance, a different set up of parameters for the DEM generation. Therefore, primary data is re-used in the subsequent calculations with different mesh sizes. This is done to save calculation time.
5.1.6 The DEM generation
Input arguments for the grid generation include indication of the calculation area defined by the lower left co-ordinate (lower left, II) and the upper right co-ordinate (upper right, ur). In addition, a starting point is indicated. This starting point must be located within the model.
In this project, the co-ordinates for (II) and (ur) are chosen, so that the calculation areas are indicated by means of the corner points in such a way that neighbouring areas adjoin without overlap. This is done to avoid elevations being generated over the same area twice. This is valid for neighbouring models as well as between strips. By avoiding overlapping areas, possible problems arising from such double elevations will not have to be taken into consideration later in the project.
The mesh size is chosen under the parameters for DEM generation. For this project, elevations are gen-erated in a grid with a mesh size of 5 x 5 m, 12.5 x 12.5m and 25 x 25m respectively. As mentioned ear-lier, the primary data is not re-calculated, it is only the DEM generation itself which is determined for each mesh size.
To achieve the optimum standard of comparison, the automatically generated grids must be coincident with the frame of reference which has a mesh size of 25 x 25m and 5 x 5m respectively. Therefore, the automatic elevations are generated in the same grid as the frame of reference. This means that for the automatically generated grids with the mesh size 25 x 25 m, the same co-ordinate endings are fixed as for the frame of reference, which means that the planimetric co-ordinates X an Y end on ***00.00,
***25.00, ***50.00 and ***75.00. For grids with a mesh size of 12.5 x 12.5 m, the co-ordinate endings are
***00.00, ***12.50,***25.00, ***37.50, ***50.00 etc. For this mesh size every other grid point will be coinci-dent with a reference point. For automatically generated grids with a mesh size of 5 x 5 m, the planimetric co-ordinates end in ****5.00 or ****0.00. Here each point will be coincident with a reference point in the reference grid of 5 x 5 m, while there will be coincidence in every 5th reference point in the 25 x 25m grid.
5 Preparation for the grid generation
5.1.6.1 Correlation
The standard value for the correlation window is 5 x 3 pixels. The threshold value of the correlation coeffi-cient is as a standard value 0.75. The correlation coefficoeffi-cient is described in Appendix A, section A.1.1.1.1.
In this project the landscape is considered as flat.
The standard value for the correlation window and the correlation coefficient is selected and fixed. The choice of a flat terrain entails a parallax bound in the x-direction (because of the normalised images) of 3 pixels, see Chapter 2, section 2.4.3.1.
5.1.6.2 The finite-element reconstruction
In the finite-element reconstruction, the values for the curvature, the torsion, the standard deviation etc.
are fixed. The standard value of the curvature and torsion is 3, see the description of curvature and tor-sion in Chapter 2, section 2.4.3.2. The standard value for the standard deviation of the determined terrain points is 0.20m. The iteration process goes through 2 separate phases. The standard value of the first phase is 6 and for the last phase is 4. The last parameter used in the set up in this project is the threshold for the bias, the standard value is 35 %.
All these standard values are fixed. The chosen values are presented in table 5.2:
Function Parameter Value Comments
The image and object pyramid Gaussian function over 5 x 5 pixels 9 levels Standard for 15 µm images The image and object pyramid Gaussian function over 5 x 5 pixels 8 levels Standard for 30 µm images The image and object pyramid Gaussian function over 5 x 5 pixels 7 levels Standard for 60 µm images
DEM generation Window size for convolution 5 x 3 pixels Standard window size DEM generation Parallax bound in row direction 3 pixels Standard value for flat terrain DEM generation Threshold for correlation coefficient 0.75 Standard value
Finite-element Curvature and torsion 3 Standard value
Finite-element Standard deviation for grid points 0.20 m Standard value
Finite-element First iteration process 6 Standard value
Finite-element Second iteration process 4 Standard value
Finite-element Threshold value for bias 35 % Standard value
Table 5.2: The standard set-up for Match-T for all calculations.
The above parameters are fixed in all automatically generated grid calculations. As described in Chapter 4, there are digital images in three different scales and three different resolutions. In addition, calculations are planned with three different mesh sizes, that is, 27 calculations in all.