Introduction
This lab will be fairly similar to the Processing Data Using Pix4D lab. The only difference will be the use of Ground Control Points (GCPs). An overview of GCPs will be given as well as an explanation about why they are important and how they are obtained. This will be followed by a methods section which will include a detailed summary of processing the data in Pix4D. The orthomosaic and DSM created in this processing will then be used to create maps and to generate discussion. The orthomosaic and DSM generated with the use of GCPs will then be compared to the orthomosaic and DSM generated without the use of GCPs from the Processing Data Using Pix4D lab. The differences between these will be discussed and displayed using gifs and maps. Lastly, the importance of GCPs and how they relate to data quality will be addressed.
Ground Control Points (GCPs)
GCPs are locations on the earth's surface which are used to help tie down imagery taken using UAS platforms, satellites, and airplanes. They are used to gereference the images taken from these devices. Without the use of GCPs the imagery can be up to 50 meters off of its correct location. If the images taken are already geolocated, GCPs are not mandatory, but will help to make the image location more accurate. If the images used in processing are not geolocated, then GCPs become a necessity and are required to process the images.
Methods
Create a New Project in Pix4D
A new project was created in Pix4D with the name 20160616_Litchfield1&2_Phantom60_GCP. The project/folder name is based off of some metadata. This includes the date the images were taken, the location of the images, the UAS platform used, the flight at which the platform was flown at, and an indication that GCPs were used.
Add Images and Change Camera Shutter Model Properties
Images from both Litchfield flight 1 and Litchfield flight 2 were added. Adding both sets of images will make the processing take a bit longer, but the processing is done together and no merge will be needed. In total 155 images were added.
Pix4D assumes that the camera shutter model is Global Shutter or Fast Read Out but it is actually Linear Rolling Shutter this is changed and is shown below in figure 5.0 in the black ellipse.
 |
| Fig 5.0: Changing Camera Shutter Model Properties |
Set the Coordinate System and Processing Template
The coordinate system was left to the default
WGS1984(egm96) and the processing template 3D model was chosen
Import GCPs
Next, GCPs were imported from a .txt file which was provided with the assignment. The GCP format was in UTM meters so the coordinate order was changed to Y, X, Z. The Y value was the false easting, the X value was the false northing, and the Z value was the elevation. This can be seen below in figure 5.1.
 |
| Fig 5.1: Import GCPs Window |
Use the Basic Editor to Tie Down GCPs Before Initial Processing
The basic editor was then used to tie down GCPs using one image per GCP. Doing this before initial processing would make for tying down the GCPs using the ray cloud later easier. Figure 5.2 shows the Basic Editor window. The GCP in the image is labeled #1 and is tied down by finding the GCP on the image and clicking in the middle of the GCP. Doing this will put a yellow crossbar on the GCP. After tying down some GCPs using this method, the data is ready for initial processing.
 |
| Fig 5.2: Tie Down GCPs Using the Basic Editor |
 |
| Fig: 5.3: Quality Report Summary and Quality Check |
Start Initial Processing, and then View the Quality Report After
Initial processing was then started. After the initial processing, the Quality Report was reviewed to make sure that the initial processing went well. Figure 5.3 on the right shows the summary and quality check of the initial quality report. The summary shows that the ground sampling distance was 2.56 cm and that the area covered was 0.1022 square km. The quality check shows that all 155 images were calibrated, that there was a 6.3% difference between the initial and optimized camera parameters, and that some GCPs were tied down, but not all of them. In the Quality Report there was also a 2D keypoint matches section. This shows the amount of coverage there is with the images and were the Phantom was flown. This is shown below at left in figure 5.4. The areas with the best coverage are where the two flights overlap. This is shown by the thicker black lines and the relatively small ellipses around the yellow circles in these areas.
Another way the quality report shows the coverage of the data is through the overlap section. Figure 5.5 displayed below at right shows what the overlap between images is. The green areas represent good overlap, and the yellow to orange areas represent poor overlap.
 |
| Fig 5.4: Keypoint Matches |
 |
| Fig 5.5: Overlap of Images |
Overall, the output in this Quality Report looked good, so further processing can be done. If the Quality Report showed that the data didn't process well, then it is a good idea to redo the processing and make sure all the settings were correct before starting.
Tie Down GCPs Using the Ray Cloud
Next, the ray cloud was used to tie down GCPs. GCPs in the ray cloud are shown as a blue circle with a blue pin intersecting it. To tie these points to the surface, first the GCP was clicked on. This then brought up a window which would allow one to tie down the GCP using multiple images. This is shown at left, below in figure 5.6. The process of finding the GCP in the image provided was similar to the basic editor option which was used before initial processing. However, by using the ray cloud, multiple images were used to tie down the GCPs making the location of these points much more accurate. This process was used to tie down all of the GCPs which could be tied down. Three GCPs could not be tied down because they were nowhere to be found, or they could only be located in a single image. Figure 5.7, shown at right below, shows what the ray cloud looks like after tying down the GCPs.
 |
| Fig 5.7: Ray Cloud After Tying Down GCPs |
 |
| Fig 5.6: Tying Down GCPs using the Ray Cloud |
Reoptimize the Images
Next, a reoptimize was performed by clicking on the reoptimize button under the processing ribbon. Reoptimizing the image will bring down the GCPs to the surface. If this isn't done, the tying down of the GCPs with the ray cloud will have no effect in the final processing. This takes about 20 minutes to perform. Once finished the ray could will look like figure 5.8. The green circles will be coincident with the blue circles. If these are not coincident, then the GCPs can be retied down with the ray cloud and reoptimizing will have to be run again.
 |
| Fig: 5.8: Ray Cloud After Reoptimizing |
Also, the GCPs accuracy can be checked by clicking on the GCPs in the ray cloud. This will bring up an image such as the one shown in figure 5.9 below. As seen here, this GCP looks to be properly tied down with all of the images and in the ray cloud.
 |
| Fig 5.10: Double Checking the GCP Accuracy in the Ray Cloud |
Run Steps 2 & 3 of Processing, and Examine its Quality Report.
 |
| Fig 5.11: Final Quality Report |
Now, the data is ready for steps 2 and 3 of the processing. This took about 45 minutes to process. The summary section and the quality check section of the Quality Report can be seen on the right in figure 5.11. Some of the important things include the camera optimization which is 2.48% and the georeferencing which shows that the GCPs were processed correctly with a margin of error of 2.1 cm. Notice that all of the indicators are green checks which show that the data was processed well. Another way to see if the GCPs were processed correctly is to look at the computed tie points section. This is shown below in figure 5.12. The small ellipses around the green circles is a good sign that the GCPs processed correctly.
Overall, this quality report looks good, and the orthomosaic and DSM generated by it are ready to be used to create maps.
 |
| Fig 5.12: Computed Image/GCPs/Manual Tie Points |
Results/Discussion
Maps
Maps were created using the GCP processed data. The map shown below in figure 5.13 shows the orthomosaic overlaid with a hillshade generated in ArcScene from the DSM. The high level of accuracy and precision can be seen in this map. It is very clear where the sand piles are placed. In the southwest region, there are very few sand piles, and in the central region there are quite a few. Vegetation is pretty sparse though out the mine with the exception in the southern part of the map. This is where there are some deciduous trees. The map has high enough resolution that the location of where the sand trucks drive on the sand piles can be seen. This is most clear in the two sand piles just north of the deciduous trees in the southern part of the map. In general, this map does a great job in displaying the shape, location, and detail of the objects in the mine such as the sand piles, trucks, and other equipment.
 |
| Fig 5.13: Orthomosaic Overlaid with Hillshade |
This next map, shown below in figure 5.14 displays the DSM overlaid with the hillshade. The maximum elevation was 247.553 meters, the minimum elevation was 228.141 meters, and the mean elevation was 234.751 meters. In general, this map depicts a relatively flat, low laying area with exceptions where the sand piles are present. The sand piles can be easily identified with the yellow and red coloring. The area where the deciduous trees are in the south didn't processes very well and this is because the image overlap in this area was not very good compared to the rest of the mine site, and there are many trees which cause interference with elevation values. The lowest area on the map can be found along the water's edge, and the highest point (besides the misprocessed trees) is located in the center of the map on the top of the cone shaped sand pile. Almost all of the sand piles are located in the center third of the map. Overall, this map does a good job of showing the differences in the height of the sand piles as well as their shape and size.
 |
| Fig 5.14: DSM Overlaid with Hillshade |
GCPs are important to help tie down accurate points on the image to the surface. A series of GIFs were created to show the difference between the orthomosaics and DSMs with and without the use of GCPs. The orthomosaic and the DSM used in the images without the use of GCPs were created in the
Processing Data Using Pix4D lab.
This first GIF shown below in figure 5.15. alternates the orthomosaic generated using the GCPs with the orthomosaic and DSM generated without the use of GCPs. Both images are labeled. Overall, the image with the text "Use of GCPs" is more accurate. This can especially be seen along the edges. In the southwest portion of the GIF the "No GCPs" image appears to move farther southwest when compared to the "Use of GCPs" image. The "No GCPs" image tends to move farther from the "Use of GCPs" image where the coverage and overlap of the processed images is of poorer quality.
 |
| Fig 5.15: A GIF Showing the Difference Between Using GCPs and Not Using GCPs |
These next two GIFS are a zoomed in portions of the GIF above. The first one (Figure 5.16) shows the portion of the image located in the northwest next to the water's edge. This zoomed in GIF clearly shows that the waters edge is more aligned in the "Use of GCPs" image. The second GIF (Figure 5.17) shows the road located in the southeast corner of the GIF above. Figure 5.17 shows that the "Use of GCPs" image is more accurate as the road lines line up better with the base map in the background.
 |
| Fig 5.16: Water's Edge GIF |
 |
| Fig 5.17: Road Comparison GIF |
This final GIF, shown below in figure 5.18 shows the difference in the DSMs with and without the use of GCPs. There is a stark contrast between the two images. The main difference is the overall elevation profile of the mine site. The "No GCPs" image suggests that the mine's overall elevation goes from high in the northeast and southwest to low in the middle. The "Use of GCPs" image, which is more accurate, suggests that the mine's overall elevation is fairly flat with the exception of the sand piles. The "Use of GCPs" DSM is more refined. This is seen by looking at the location of the sand piles. In the "No GCPs" DSM the sand piles look more like blobs than sand piles.
 |
| Fig 5.18: DSM GIF With and Without GCPs |
Conclusion
In conclusion, the GCPs help to make the data more accurate. Although there are only small differences seen between the orthomosaics, the difference between the two DSMs is quite large. This proves that the use of GCPs when processing data is important even though it's not mandatory. Also, processing the data using GCPs wasn't much more complicated than without it.
The difference between using GCPs can be seen in the Quality Reports as well. In the quality check sections the Quality Report generated using the GCPs had a camera optimization of 2.48%, and the Quality Report generated not using the GCPs had a camera optimization of 5.22%. Overall, GCPs help improve data quality and accuracy.
Sources
GIF Maker, Make a GIF
UAS images provided by Joe Hupy
No comments:
Post a Comment