Balloon Mapping II

Introduction:

To wrap up our balloon mapping activity, we took out our balloon and rig for a final run and later mosaicked our new images together. Having learned from our test run with the balloon last week, our mapping activity went much more smoothly this time. The balloon was able to reach a much higher elevation and the rig was adjusted so as to better the camera angles. We were able to take the balloon to an extended area for more coverage as well. For the mosaicking portion of the activity, we divided the class into groups (the same groups we were working with for our past activities) and were given a specific area of campus to work with (Figure 1).  I used a different program for this process and found it to be slightly frustrating. We will later stitch together each area to create an entire aerial map of the University of Wisconsin-Eau Claire campus.

Figure 1: This is an image of the area of the UWEC campus that my group and I were assigned for the mosaicking task.

 

Methods:

Flight of the Balloon:

The second and final balloon outing was fairly simple to set up because we had learned from the mistakes of the past week. We had to fill up a new balloon (since the last balloon flew off into space), but the process was exactly the same. Once the balloon was filled, we took it out onto campus to be released. This week the weather conditions were much more seemly as there was very little wind, which allowed the balloon to reach the intended elevation of 400 feet more or less. The balloon was walked around almost the entire campus including the other side of the river and “upper campus” which resides on top of a large hill that we did not get to in the test session (Figure 2). The balloon was taken down between major areas of the campus are re-released once it was in the intended area. This helped to keep the line from getting caught up in any trees or on any buildings; the aerial images were much better as a result, but they were still imperfect and difficult to work with.

Figure 2: This image depicts the two areas of the UWEC campus that we were able to take the balloon during the final mapping session that we were unable to get to during the test run.

My task for this activity was to collect ground control points to use for georeferencing. Georeferencing is the process of matching an aerial photo or raster to a base image to geometrically correct the image. In other words, the image will be correctly positioned to match real world features. Georeferencing is essential for mosaicking because the photos cannot match up to make a seamless image if they are not corrected. The ground control points mark features to which the photos can be matched. We decided to use mostly light posts as our points because they don’t move over time and they can be spotted on the aerial photos relatively easily.

To collect the points, a small group and I used several different types of GPS units—a Juno and Nomad by Trimble, and a handheld Topcon—and mapped the light posts around campus. We went to each light post and marked them as a waypoint in the GPS. These points were later converted into point shapefiles. Unfortunately, the GPS units were fairly inaccurate. The most truthful results came from the Juno unit which is supposedly the least accurate of the three units. The light posts were found to be several meters off in some cases which is not appropriate for georeferencing.

Figure 3: This is a photo of the Juno GPS unit we used to collect ground control points for georeferencing. This was the most accurate GPS unit of the three we used and also the cheapest.

 

Georeferencing/Mosaicking:

Once all the data was collected and uploaded, I were able to begin the mosaicking process .This began with choosing appropriate images for our area. We had thousands of photos to choose from. This was advantageous because it allowed for a wide variety of images, but it also made it difficult to find the images that fit into my study area. I chose to use about eight images in the end. After I chose my images I uploaded them into ArcMap. I decided to use ArcMap to mosaic and georeference as opposed to Mapknitter (which I used last week to mosaic my images) because I thought it would provide more accurate results.

Next, I had to georeference these images. I used a .tif of an aerial image of the UWEC campus that was previously collected for my base and set about matching my balloon photos to the base image. Instead of using the ground control points I collected during the balloon mapping, I chose to use distinguished features on the photos instead. Some of these features included trees, sidewalk ends/edges, small divots or juts of land on the shoreline of the river, and garbage cans. The tops of buildings couldn’t be used unless necessary because they are offset depending on the angle of the photo. Each photo needed a minimum of nine control points so as to align it as accurately as possible. I found myself using up to 25 points for some images to try to fit them more seamlessly. Due to the nature of the photos, however, it was still extremely difficult to georeference the photos so that they matched the base image well despite the number of control points I used.

Figure 4: This is an image of all my georeferenced aerial photos. They appear to line up somewhat well from a distance, but upon closer inspection, they have quite a few faults.

 

I manipulated the photos as best I could, then I ran the Mosaic to New Raster tool. This Data Management tool allows the user to mosaic—combining several overlapping rasters to create one final raster. It is different from the Mosaic tool because it allows the user to create an entirely new output layer instead of overwriting one of the input rasters as the Mosaic tool does. The most important factor of the Mosaic to New Raster tool is making certain that the images are stacked correctly. To create the best mosaic possible, one must choose which raster should be the top image, or the most visible image. The top image should be the highest quality image, and each successive image beneath it should be of higher quality than the one it overlays. The tool allows the user to specify the order of the layer stacking, so it is vital to make sure that this is done properly. The output of the Mosaic to New Raster tool is a final mosaic of the input images.

Results/Discussion:

Although the mapping portion of this activity went very smoothly, the georeferencing and mosaicking of the images was very difficult. The images, though much better than those collected last week, were still hard to work with.  A large portion of the photos were blurry because we did not have the camera set to “scenery” and the camera was unable to focus. Some of the photos also include the string that was used to hold the balloon down. Though I tried to eliminate the string by overlaying photos, I could not find enough suitable images to do this over the river (Figure 5).

 Figure 5: This is an image of one of the anchoring strings that appears in my mosaic. It is partially covered by another photo, but could not be entirely covered.

Another major problem was lining up the images. Because there were so many different angles, none of the images exactly matched the base image. This also meant that none of them lined up well with each other when they were overlaid either. My final mosaic, unfortunately, reflects this issue because there are many fissures where two photos did not match up (Figure 6).

Figure 6: This is an image of the overlay of several of the aerial photos collected by the balloon that I used while georeferencing. This is an example of the disconnect between photos because they could not be completely geometrically corrected while georeferencing.

 

Figure 7: This is an area, along the road, where the base image and the aerial image collected by our balloon line up properly. However, there is still some misalignment near the building in the lower right hand corner of the image.

 Another challenge was the area my group and I were given. A large portion of our area included the river and a forest, which could not be covered well by our balloon since it was inaccessible (Figure 5). For the photos that did cover the river, the georeferencing was more problematic than other photos because there were few features to use as reference points The portion of land that was included in these photos was covered by a parking lot that has been recently renovated. This proved to be difficult because the base image I was using is older and did not include any of the renovation; therefore, there were even fewer features to use as reference points because many of the trees, the sidewalks, and the parking lot had been changed (Figure 8).

Figure 8: This image displays the areas that were most difficult to find aerial photos of. The top section is the river and the bottom is a forest that did not get a lot of coverage from the balloon.

 

Figure 9: This is a photo of a parking lot along the river before its renovation. This is how it appears on the basemap.

Figure 10: This is an image of one of my aerial images overlaying the basemap. As can be seen, there were a lot of changes made with the renovation that make it difficult to find a reference point.

If I had been able to, I would have liked to have worked in ERDAS instead of ArcMap to improve my mosaic as well. I felt more comfortable working in ArcMap because I have worked with it more extensively than ERDAS which led to my decision to use ArcMap. However, there were some issues trying to run the Mosaic to New Raster tool in ArcMap that may have been a bug. I also know that ERDAS allows for more color matching techniques and can produce a more unified final raster.

Conclusion:

Overall, the mapping portion of this task was fairly successful. We were able to improve our photos from last week and chose a much better day to fly the balloon. Collecting the ground control points did not work quite as well—many of the points were several meters from their intended features and couldn’t be used in the georeferencing process. This can be attributed to error within the GPS units we were working with. The mosaicking and georeferencing also did not go as intended either. The area my group and I had to mosaic was not well covered because a large portion of it was covered by the river or an area of woods that wasn’t covered. The photos also couldn’t be aligned properly to the basemap or to one another because they were taken from many varying angles and elevations. This resulted in a faulted mosaic.