DRONE IMAGERY MAPPING
GEOG 5170 - Geospatial Field Methods
The drone mapping exercise was designed to take the structure from motion concepts we were learning in the classroom and apply them to a real site of our choosing at the Rio Mesa Field Station. After finding our area, we were to set out some ground control points (GCPs) and then use a drone armed with a camera to take a couple hundred photos. We would input the exact coordinates of the ground control points into a handheld GPS device, and then take everything home for post-processing in Agisoft and ArcGIS.
​
Our group picked an area to the east of the basecamp tucked back in a canyon-like bowl. It was a large and smoothed rock landscape with a middle area filled with more vegetation dipping down lower. We picked this location because we felt that it would be an interesting location to model where we could see clear elevation differences in our final products, without having too many obstacles for the drone or its camera.
Methods/Tools
The first step that our group had to take was to scout out the landscape and find an effective place to conduct our project. Once we had the done, we returned the next day to conduct the drone study. We spread eleven canvas GCPs around our area as evenly as possible and then used the GPS device to store their exact location. In case data was accidentally lost, we wrote down the coordinates as well.
​
A Phantom 4 drone was used to conduct our study. We surveyed our landscape using two lawnmower patterns, but the conditions of each pattern were different. The camera on the drone was put at a 90-degree angle facing straight down for one of the runs, and then turned it to a 45-degree angle for the other. By using these two different angles, we wanted to see how out final results differed from one other and if using one camera angle versus the other would prove beneficial.
​
Before getting started, we noted the exact start and end times of each of our flights, the flight conditions that might affect the drone, the weather, the observers, and the serial number on the drone. It is important to notate anything that might affect the data gathering of an experiment, and these surveys were no exception. The drone captured images of the landscape every 3 seconds as it flew, giving us a full view of the area but also potentially creating small gaps. All five of us in the group took turns flying the drone and we all tried to fly around 3 miles per hour. However, particularly when some of us were first handed the controller, the speed of the drone was more sensitive than we realized. Because of this, the drone did at times fluctuate between about 2 and 10 miles per hour, which means we might have gotten more photos of one part of the area than others.
​
When we were all finished with both lawnmower patterns at the two different camera angles, we headed home to process the data. Using the Agisoft Metashape program, two separate files for each angle were created. In the 45-degree angle file, there were 229 images that we uploaded for use, while the 90-degree angle file only had 174. From there, the images were aligned and used to create dense clouds. We could then use these dense clouds to create the DEMs for each angle as well as the orthomosaics. The GCP coordinate points were used to georeference the model as well as the DEMs separately in the ArcGIS. Finally, a difference DEM was created to show the strongest points of variance between the DEMs of the two camera angles.
Orthomosaics & Hillshades (Scroll over the Images)
The orthomosaics are images produced straight from Agisoft. The models that you see without the background imagery (figures on the left) were pieced together using all of the photos that we captured in the drone flight. The hillshades (figures on the right) are a different way to symbolize digital elevation models (DEMs) that allow for the details of the terrain to be more easily visible than that of a standard DEM. The orthomosaics and the hillshades show that the 45-degree angle has captured a substantially larger amount of area than the 90-degree angle, but actually did not lose any quality in the process.
Figures 1-4: Orthomosaic and Hillshade Maps
90-Degree minus 45-Degree Difference DEM
Producing the difference DEM in Figure 7 involves a raster math calculation in ArcGIS. Using both DEMs produced in Agisoft, we ran the raster calculator subtracting the 90-degree DEM by the 45-degree DEM. The result shows that both DEMs have very similar elevation values, as the biggest difference between them is just over 10 feet.
​
This DEM displays the landscape taken into account by both camera angles, shown without the hillshade because it can more effectively communicate the true differences in elevation. It only displays the areas that are present in both DEMs and because the 90-degree angle DEM is the smaller of the two, it is the one that the general shape of the difference DEM resembles.
Figure 5: Difference DEM
Issues Encountered
We did run into some issues during our experiment, only one of them not software related. On the day that we flew the drone, there were occasional gusts of wind that blew a few of our canvas GCPs away during the flights. We had to retrieve them and try to figure out exactly where they belonged so we could record their location in the GPS device. A software issue that a few of us had were that we didn’t own powerful enough computers to run Agisoft very quickly, and some dense clouds and orthomosaics took an hour plus to build. Another problem we encountered was that we originally intended to do checkerboard flight patterns for both angles, and in fact did. Once we were back home for post processing however, we found that we only had half of our photos from our 90-degree angle flight. This meant we essentially only had a lawnmower pattern for that angle, while we had all the photos for the 45-degree angle. Not sure how that happened, other than there may have been a user error with the drone in the field. To compensate and create an equal comparison between the two, we only used half of the 45-degree angle photos that we had, but it was difficult to pick out photos that belonged to one of our lawnmower pattern flights versus the other. Therefore, our 45-degree angle model does not completely display what a single lawnmower pattern flight looks like, which is important to note in a comparison experiment such as this.
Conclusion
Despite a few technical difficulties, I feel that our drone experiment did a good job at displaying the effects that two different camera angles might have on a final structure-from-motion model built in Agisoft. I think the 90-degree camera angle would be best to use when possible, as it can build the most accurate model and DEM to the landscape. But if one could not use a flight plan or the location being studied prevents using the checkerboard flight pattern, the 45-degree angle is not a bad option to go with. It appears that angle is in fact more forgiving for manual drone flying, and is only slightly less accurate than the 90-degree angle. Drone data collection requires a good amount of attention to detail and I think that, given another chance at a project like this, I could go through the entire process of image capturing and post-processing far more error free.
Skills Learned / Used
Communication Skills
Project Design
Project Management
Cartography and Graphic Design