Deliverable 3: Geometric Correction, Orthorectification and Mosaicking

You might also like

Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 16

4/6/2018 Deliverable 3

Geometric Correction,
Orthorectification and Mosaicking

Hayden Polski B. Sc
GEOSPACE
Hayden Polski, B. Sc (Honours)
96 Cole Cres
Niagara-on-The-Lake, Ontario, L0S 1J0
519-841-4151
Haydenpolski@gmail.com
April 6th, 2018
Janet Finlay B.A, B. Sc
Professor of GISC9216 Digital Image Processing
Niagara College
135 Taylor Road
Niagara-on-The-Lake, Ontario, L0S 1J0

RE: Completion of GISC9216 Digital Image Processing Deliverable 3


I would like to submit the third deliverable for GISC9216 Digital Image Processing, as enclosed
below. I utilized information obtained in lectures and skills learned during in class exercises to
complete the project to the desired specifications.
The deliverable requires the creation of two mosaic maps created from the same air photos. The
air photos of Aberfoyle, Ontario are first geometrically corrected using two different methods. A
polynomial method is done first, correcting the photos with ground control points to an existing
subset image and then mosaicking the resampled and photos together. The original photos are then
resampled again, but fiducial points and a camera model are used instead. Fiducial points and
ground control are used before the same resample, trim and mosaicking are repeated. Both mosaics
are overlaid with building polygons and road polylines and given all appropriate map features. The
accuracy of each method of geocorrection was then assessed and the output maps examined for
errors.
Should you have any questions/issues with the enclosed documents please feel free to contact me
at your convince at (519) 841 4151. Thank you for your time and teachings. I would appreciate
your comments and suggestions.

From
Hayden Polski, B. Sc
H.P./

Attached: Fiducial Map.pdf, Polynomial Map.pdf, GISC9216D3HPOLSKI.docx

96 Cole Crescent, Niagara-on-The-Lake, ON, L0S 1J0


Tel: (519) 841 4151 Email haydenpolski@gmail.com
Project: Aberfoyle Orthoimagery Mosaicking

Client: Janet Finlay Prepared by Hayden Polski B. Sc

Subject: Geometric Correction, Orthorectification and Mosaicking

Date: April 6th, 2018 Project Number: 54796

Introduction
The creation of orthoimagery is not always a straight forward task. Sensors used to
capture images of the earth often take snapshots, capturing an area beneath the focal point. In
order to make use of the captured data for larger purposes, these snapshots often need to be
joined together into a single file. Creating a single file out of multiple images is called
mosaicking.
The captured images often contain a degree of overlay, allowing for like points to be
connected using geographic analysis software. In this case ERDAS IMAGINE 2016 was being
used to combine several air photos that have been taken over Aberfoyle, Ontario. An existing
subset is being used as a base for which each of the air photos can be corrected two, although the
subset is itself missing sections of data. Information concerning structures and roads within the
study area have also been provided by the client for use in accessing the accuracy of the
mosaicking process as well as being used in the creation of maps.

1
96 Cole Crescent, Niagara-on-The-Lake, ON, L0S 1J0
Tel: (519) 841 4151 Email haydenpolski@gmail.com
Study Area
The mosaic processing was done on orthoimagery acquired around Aberfoyle, Ontario and
Puslinch, Ontario. It is centered a section of Ontario Highway 401 running between Cambridge,
Ontario and Toronto, Ontario, as well as a section of Ontario Highway 6 leading to Guelph,
Ontario. Figure 1 shows the study area.

Figure 1: Study Area (Google Earth, 2017)

Methodology
The mosaicking process can be divided into two phases. The first phase used only
polynomial correction before creation a mosaic, while the second phase used a camera model for
orthorectification and fiducial points before creating a mosaic. Both phases were done using

96 Cole Crescent, Niagara-on-The-Lake, ON, L0S 1J0


Tel: (519) 841 4151 Email haydenpolski@gmail.com
ERDAS IMAGINE 2016 using the provided data. The provided data consists of seven files that
are detailed in Table 1.
File Description
Photo_1.tif Portion of the orthoimagery of study area
Photo_2.tif Portion of the orthoimagery of study area
Photo_3.tif Portion of the orthoimagery of study area
Subset_existing.img An almost complete orthoimage of the study area
DEM.img Elevation data of the study area
Buildings.dbf File containing polygons of buildings within study area
Roads.dbf File containing polylines of roads within the study area
Table 1: Provided Data

The polynomial correction was done by first opening each of the files and using ERDAS
to build pyramids. From there, each of the .Tif files was opened one at a time before the Control
Points tool was used. The polynomial geometric model was used and then the Subset_existing.img
file was opened for use as the reference image that the .Tif files are being correct to. Figure 2

Figure 2: Control Point Creation (Left: Photo_1.tif with control points, Right: Subset_existing.img with
control points)

shows the first Photo.tif file with the control points and the existing subset with the same control
points.
As the control points were added, their location was recorded. After the addition of the third control
point, ERDAS began to predict the location of the corresponding control point on the subset image
when a point was added to the photo .Tif. The accuracy of the predicted point improved in accuracy
as additional points were added. Once eight control points were created the Root Mean Square
Error (RMSE) was calculated for each of these points. The RMSE refers to the difference between

96 Cole Crescent, Niagara-on-The-Lake, ON, L0S 1J0


Tel: (519) 841 4151 Email haydenpolski@gmail.com
the control points on the two images. Figure 3 shows the table created during the placement of
control points and the RMSE calculation.
After this was done, the .Tif image was resampled. The resampled image is tilted and lays
overtop the subset_existing.img. This process was then repeated for the Photo_2.tif and

Figure 3: Control Point Table for Photo 1

Photo_3.tif. Creating eight control points for each of them and creating a subset image. Once all
were subjected to this process, and .aoi file was created for each and used to trim away the
boundaries. Figure 4 shows the created .aoi used in the trimming process for the resampled
Photo_1.tif and the produced trimmed image.

Figure 4: Trimming process (Left: .aoi file for resampled Photo_1.tif, Right: Trimmed resample of Photo_1.tif)

The creation of .aoi files and trimming was done for the other two resampled photos. Lastly.
Using the three trimmed photo files, a mosaicking process was run. Using the Mosaic Express
wizard in ERDAS, the three trimmed, resampled .Tif files were loaded. The wizard was set to use
the nearest neighbour setting and the output path was specified before the tool was run. The
resulting output combined the images into one, matching them together based upon their relations
to one another and accuracy of the control points.

96 Cole Crescent, Niagara-on-The-Lake, ON, L0S 1J0


Tel: (519) 841 4151 Email haydenpolski@gmail.com
The second phase of the analysis is very similar to the first, the main difference occurring
at the beginning of the process. The photo.tif were opened and loaded using the control point tool.
But rather than the polynomial geometric mode, a camera model was used instead. The X and Y
principal points were set to 0.000 and the focal length for the camera is 152.468 mm. The elevation
for the study area was taken from the dem.img file. The curvature of the Earth was not accounted

Figure 5: Camera Model

for due to the small size of the study area, and a total of five iterations were done. . Figure 5 shows
the camera model settings used in all the images.
The fiducial points where then added to the camera model film, creating set points along
the X and Y axis. The values for these points were provided in the terms of reference and can be
seen in Figure 6.

Figure 6: Fiducial Points for Photo 1, Image and Film

96 Cole Crescent, Niagara-on-The-Lake, ON, L0S 1J0


Tel: (519) 841 4151 Email haydenpolski@gmail.com
Once the fiducial points were set for the film the camera model was saved and the first
photo.tif loaded. Using the Viewer Fiducial Locator the fiducial points on the image were selected.
These are eight points around the perimeter of the air photo located at the cardinal points and the
corners of the area. Following the creation of the fiducial points on the image, the ground control
points were made using the same method described in phase 1. Figure 7 shows the fiducial points
(around the edge) and the ground control points (scattered throughout the center) for the first
image.

Figure 7: Photo_1.tif with Fiducial points and ground control points

From there the analysis process mirrors the first phase, with each of the images being given
ground control points and resampled, followed by the creation of a. aoi file containing a polygon
that traces around the perimeter and using it to trim the boarder. Then the creation of a mosaic
using the mosaic express wizard.
Once both mosaics were made, they were opened in ArcMAP 10.5. The Roads.dbf and
Buildings.dbf files were opened in the save screen and laid overtop the mosaic. Each mosaic was
given all proper map attributes before being exported.

96 Cole Crescent, Niagara-on-The-Lake, ON, L0S 1J0


Tel: (519) 841 4151 Email haydenpolski@gmail.com
Discussion
Question A
After entering the first several ground control points to the Photo.tif and
Subset_existing.img the Control Points tool began to predict the location of further ground control
points. At first a point was selected on the photo screen and then another had to be manually added
to the subset screen. After the fourth point was added the subset existing point would be
automatically generated. The accuracy of these points on the subset existing screen were initially
somewhat inaccurate, potentially missing the actual corresponding point by a great margin.
However as further points were added, the prediction accuracy improved until the computer
predicted control points were only off by a small degree.
The initial inaccuracy could be caused by the distance between the pre-placed control
points. If several points were scattered in a tight cluster then the programs ability to predict outside
of this area would be lacking accuracy. As further points were added over a wider range the
accuracy of the predictions increased.

Question B
For each entered control point an error was generated. This error is based upon the
displacement between the predicted and actual points. The Control Points tool generates a model
using control points and the RMS error represents the distance of a point from the predicted. Table
2 contains the RMS error for the eight control points as well as the sum of the error and the average.
The average error represents, on average, how far each of the control points is from the model used
for predicting where a ground control point should be located.
Points Photo_1 Photo_2 Photo_3
1 0.034 0.17 0.03
2 0.018 0.06 0.07
3 0.037 0.09 0.03
4 0.062 0.08 0.05
5 0.052 0.15 0.05
6 0.039 0.05 0.02
7 0.04 0.03 0.06
8 0.056 0.03 0.04
Total Error 0.338 0.66 0.35
Average Error 0.0845 0.0825 0.04375
Table 2: Polynomial RMS Error

Question C
All three images have a similar pixel size, but there is slight variations between them.
Despite being taken from the same sensor there is variation between the times at which they were
taken, even if that is only several moments apart. High altitude images, while removing terrain
distortion, are still subject to variations due to factors such as elevation or turbulence. Even if only
taken moments apart it is possible for the sensors elevation to have shifted enough to cause the
pixel size to change, increasing in size if it moves closer to the ground or decreasing in size if it

96 Cole Crescent, Niagara-on-The-Lake, ON, L0S 1J0


Tel: (519) 841 4151 Email haydenpolski@gmail.com
moves further. The error can also cause distortion in the pixels size, resulting in smaller/larger
pixels depending on the error for each of the control points.

Question D
When creation the mosaic map it appears fine when looked at as a whole. However when
examined more closely there are obvious signs of error. Figure 8 shows three examples of error
present on the mosaic. The left shows a poorly joined roadway, with the line representing the actual
road skirting along the side of the road in the orthoimagery. The center shows an intersection and
bridge crossing a highway. The road lines are offset from the orthoimagery by a great deal,
meaning that when the resample and control points were created that there was significant error.
The right image is a house and the polygon representing the hours, which is clearly offset. The
error and shift can be attributed to the error calculated when creating the ground control points.

Figure 8: Errors in Polynomial Mosaic

Question E
When working with the fiducial image, the predictive abilities of the ground control points
tool was somewhat improved. The first several points were off from the predicted location by a
large margin, but this quickly increased in accuracy as further points were added. The rate at which
the prediction accuracy increased with the eight fiducial points was greater that of when the tool
was running with only the ground control points. The addition of the fiducial points increased the
effectiveness of the control point’s tool accuracy.

Question F
The improved accuracy of using fiducial points in orthorectificaion can be seen not just
visually, but when looking at the RMS error. Table 3 contains the RMS errors for the control points
and subset image for when fiducial points were used. When comparing the total and average errors
of each photo in the polynomial correlation (Table 2) and fiducial point correlation (Table 3), the
values for the fiducial point correlation are substantially lower. This can be attributed to the
improved accuracy provided by the fiducial points.

96 Cole Crescent, Niagara-on-The-Lake, ON, L0S 1J0


Tel: (519) 841 4151 Email haydenpolski@gmail.com
Points Photo_1 Photo_2 Photo_3
1 0.016 0.013 0.008
2 0.014 0.018 0.009
3 0.016 0.014 0.007
4 0.018 0.002 0.004
5 0.009 0.008 0.007
6 0.007 0.005 0.003
7 0.019 0.009 0.009
8 0.002 0.007 0.015
Total Error 0.101 0.076 0.062
Average Error 0.012625 0.0095 0.00775
Table 3: Fiducial Point ERM Error

Question G
The pixel sizes for the images corrected with fiducial points vary. The average pixel size
is 0.5008m by 0.5008m. The largest is 0.5028m square and the smallest being 0.4996m square.
This variation in size can be attributed to the error. Due to the error associated with the ground
control points, when a new image is sampled there may be a degree of distortion from the error.

Question H
Like with the polynomial mosaic, the fiducial mosaic map does contain some error.
However when comparing Figure 9 (Fiducial Map Errors) and Figure 8 (Polynomial Map Errors)
it is obvious that the errors in the fiducial map are less pronounced.

Figure 9: Fiducial Points Mosaic Errors

The left section shows the same road as seen in the polynomial error, but the offset across
photos is less intense and the road line is better aligned with the orthoimagery. The center image
shows the same intersection and highway, but once again the road lines match up better than they
did in the polynomial mosaic. Finally, the same house is shown on the right, but the polygon
representing the house is fits almost perfectly in the fiducial point mosaic. The shift between air

96 Cole Crescent, Niagara-on-The-Lake, ON, L0S 1J0


Tel: (519) 841 4151 Email haydenpolski@gmail.com
photos used to create the mosaic is far less pronounced in the fiducial mosaic. The road and
building files align better with the orthoimagery than they had in the polynomial mosaic.

Findings
After creating both a polynomial and fiducial mosaic, it can be said that the fiducial
points output was the more accurate of the two. Figure 10 is a copy of the polynomial mosaic
map and Figure 11 is a copy of the fiducial point mosaic map. The fiducial point mosaic and map
is the more accurate of the two maps.
With the addition of fiducial points around the air photos the total and average error of
each of the photos used in the creation of the mosaic decreased by a sizable degree. With the
lower error the mosaic processing produced an output with fewer discrepancies. The orthoimages
lines up alone the seams with a smaller amount of offset than the polynomial mosaic. In addition,
the road lines layer and building polygons layer align with the fiducial mosaic more than the
polynomial mosaic.
Pixel size variation can be attributed to the errors from the ground control points. The
greater error for each photo, the greater that warping that will have to occur when a resample is
taken and the more the pixels will be distorted. Due to the fiducial point images having a smaller
error than the polynomial images the pixels in those resamples would be more consistent in size.
The accuracy of the control point tool increases with additional control points. The first
control point that the tool predicts is often off by a large margin, but the accuracy of further
predicted points increases with each point added. Points near the edge of the image are often less
accurate than points that are located towards the center of the image. This is because the center
of the image is often encapsulated by several surrounding points, allowing for better predictions.
Furthermore, the addition of fiducial points increased the rate at which the accuracy if the tool
improved.

96 Cole Crescent, Niagara-on-The-Lake, ON, L0S 1J0


Tel: (519) 841 4151 Email haydenpolski@gmail.com
Figure 10: Polynomial Mosaic Map

96 Cole Crescent, Niagara-on-The-Lake, ON, L0S 1J0


Tel: (519) 841 4151 Email haydenpolski@gmail.com
Figure 11: Fiducial Point Mosaic Map

96 Cole Crescent, Niagara-on-The-Lake, ON, L0S 1J0


Tel: (519) 841 4151 Email haydenpolski@gmail.com
Conclusion
During the mosaicking process, the use of fiducial points resulted in a significantly more
accurate output than using the polynomial method. There is less disconnect between the photos,
the RMS error is less and the ground features (buildings and roads) align more accurately.
Using the polynomial is acceptable when a mosaic will not be examined closely or when
precision is not overly important. However if precision is important or it will be examined
closely then it is worth sparing the extra time to do a polynomial model.

96 Cole Crescent, Niagara-on-The-Lake, ON, L0S 1J0


Tel: (519) 841 4151 Email haydenpolski@gmail.com
References
Google Earth. (2017, 4 13). Aberfoyle. 43°26'43.28" N 80°08'58.64"W.

96 Cole Crescent, Niagara-on-The-Lake, ON, L0S 1J0


Tel: (519) 841 4151 Email haydenpolski@gmail.com

You might also like