Lab 9: Mosaicking Aerial Images Collected by a Helium Balloon

Introduction:

There are a variety of mediums available to gain access to aerial images. Using an airplane or a satellite are common ways, with each equipped with a high resolution camera to capture the desired images. Each of these techniques tends to produce great quality pictures that can then be used in a variety of applications. But they also are extremely expensive ways to collect the data. A cheaper alternative to getting aerial images is through the process of balloon mapping. An earlier post in this blog detailed the construction of the various rigs that would be needed in order to house the camera securely to the balloon. This post covers the testing, flight, collection of pictures, and the mosaicking of those pictures.

Methodology:

Figure 1: The mapping rig, airborne and taking pictures

With our rigs constructed in Lab 3, and the weather cooperating, we were ready to obtain some aerial pictures with the helium balloon. The total airborne object would consist of the camera with a 32GB memory card, Styrofoam case, a Garmin E-Trex, a tracking device, and the balloon, all of which can be seen in Figure 1. Once in the air, a walker on the ground would pull the string attached to the balloon with them as they walked around the UWEC campus. The continuous shot mode on the camera enabled many pictures to be taken in each outing. The first balloon trip resulted in 321 images, and the second trip managed to give us a staggering 4,846 images. The first round of pictures were taken around the main academic center of campus, while the second balloon trip covered a significantly greater area. In addition to the main academic area, the Davies Student Center, Haas Fine Arts building, and some sections of upper campus, where the residential halls are located, were mapped.
After collecting the images, they were uploaded for the next step of mosaicking. A number of software options were available to us for complete this task. MapKnitter, a website that provides a freeware mosaicking program, ArcGIS 10.1, and ERDAS Imagine, provided the needed functions that we needed. In the first activity, MapKnitter was used to produce a mosaicked image, and ArcGIS was used in the second. The first task that needed to be done in any of the programs was to georeference the pictures. Georeferencing is a process involving relating the picture to a physical area in space. To do this, an aerial image of campus was used as a base map, and the pictures from the balloon were laid over the image, paying attention to scale of the picture. In MapKnitter, this was done by resizing, rotating, and stretching the images to produce a fluid grouping of pictures that looked like one image, instead of over ten. ArcGIS required setting ground control points, or GCPs for each picture on the base map. Placing good GCPs is a challenge itself, especially considering the current aerial images for the UWEC campus does not show the addition of the new student center, and all of the construction that is currently taking place. So good GCPs were obtained by using things like corners of existing buildings, light poles, lines painted on the streets, and, when nothing else was available, gaps in sidewalks. Images were selected based on the amount of overlap, which allows the operator to produce a better looking image that is more consistent, rather than simply selecting single images that show the area of interest, and then moving to another area that did not have any of the features from the first image in it. This also ensures that buildings, lightpoles, and streets are in the same perspective, generally speaking. In order for a visually appealing image, the parallax of each building should look like one is directly over a feature, instead of on the side of it, which would result in a warped view.
Unfortunately for this assignment, the lack of communication was an issue. The assignment was to use the images collected through the mapping rig to create a seamless image of the entire UWEC campus that had coverage. As this was an extremely time consuming task, campus was split into six sections. This seemed like a good idea at the time, but there was minimal communication between the groups, and this resulted in hardly any of the mosaicked TIFF files being uploaded into the geodatabase so that a total mosaic could be built.    

Discussion:

Figure 2: Mosaicked image of the campus mall, produced with MapKnitter software

When mosaicking the pictures and experimenting with the software that was available for us, each had its benefits and drawbacks. MapKnitter, for example, was extremely user friendly, using simple functions like rotate, re-size, distort, and different transparency levels to place each picture in the correct spot at the correct resolution. This made it simple to place the sidewalks in the correct location, as it was only a matter of making one image transparent, and then lining it up with the image that was already laid behind it. Figure 2 shows the finished mosiac of the images from the first mapping exercise, using MapKnitter. The ease of use for this software was also its biggest drawback, however. Toned down for simplicity's sake, MapKnitter lacked the more advanced functions of mosaicking programs, notably the inability to place GCPs. Additionally, the limited number of pictures taken in this first round, and the relatively small area covered, did not allow for a complete scan, which can be clearly demonstrated by the distortion of Schneider Hall, the building on the far right in Figure 2. If more passes with the balloon were taken near that particular area, more overlapping pictures could have been taken, which would have resulted in a better quality image.

Figure 3: Mosaicked image of our group's area using ArcGIS Desktop 10.1 to georeference and mosaic

Mosaicking in ArcGIS was a different process altogether. The placement of GCPs was critical to having a correct georeferenced image. To ensure this, at least 20 GCPs were used per image, to eliminate the possibility that the image would shift due to the placement of a point in an isolated section of the image. Having these GCPs, in addition to a significantly larger area covered and pictures taken, did make for a better finished image, which you can see in Figure 3. Despite the number of pictures taken, the mosaicked images appears more or less seamless, the goal of the project. The only real problem with this mosaic is the lack of coverage on the south side of Haas Fine Arts center, the large building in the middle of the picture. This is due to us pulling in the balloon, to check if the camera was still taking pictures. After checking that everything was still functional, we did not deploy the balloon again until we crossed the footbridge back to the main area of campus. Had we had the balloon up in the air during that walk, a better group of images would have been taken on the south side of Haas, and would have resulted in a more complete picture.

Conclusion:

This activity was an incredibly valuable experience. To the best of my knowledge, not too many people have done balloon mapping, especially in the scale that we had. The second balloon outing especially confirmed this to me, with close to 5,000 images being taken. This large amount of pictures ensured that issues like inadequate coverage, which was present in the first balloon activity, would never happen. The large number of pictures did create quite a daunting task, however, as going through them to find the image with the right amount of overlap, the desired features being clearly displayed, and ones with the rope from the balloon in the correct area to be covered up by another picture, was a time consuming process. After using both MapKnitter and ArcGIS to mosaic, it is clear that, despite its user friendly nature, MapKnitter was the worst of the two programs. The act of georeferencing and the placement of GCPs is critical to a good image, and ArcGIS is the software to use to do this effectively.