Introduction:
In early February, I posted about the early stages
in perparing for a method of collecting aerial imagery using helium balloons
known as Balloon
Mapping. Now, for the past two weeks the class has been making good on the
first half of that project: the low-altitude ballon launch has come to
fruition. On the 1st and 8th of April, the launched our noble apparatus and
towed it around the campus here at the University of Wisconsin -- Eau Claire as
it collected thousands of aerial images from which to develop a single image of
campus tied to defined geographic reference points. The following is an account
of the methodology of our launch and image processing, as well as what was
learned in the class's two launches.
Methodology:
Launch 1:
The first launch was conducted on Monday, April 1st. April 1st in Eau Claire, WI was windy, to say the least. Naturally, the wind was highest during the period from 3 to 6 p.m. which was exactly when it was decided to launch the aerial rig. Noticing wind speeds around 15 mph and gusts in excess of 25mph, it was decided to use the more substantial platforms to house the digital camera today; so the HABL rig was attached to the helium balloon instead of the Low-Altitude rigs developed for the project.
 |
| figure 1: wunderground historical weather data for Eau Claire, WI on Apr 1. Note the wind gusts around 4 PM. |
The HABL provided certain advantages over the other aerial mapping platforms: it was more insulated, painted with hydrophobic material for waterproofing
and heavier which it was reasoned would provide more stability in the wind. Its design was based around a Styrofoam bait
container, the type used in fishing which would prove fatefully ironic. Perhaps most importantly, it was heavier and therefore (in theory) less susceptible to the prevailing winds, however the end results of this project indicate that the added weight was inadequate in this case.
Bessie, as the big red weather balloon came to be called, was attached to 400 meters of rope in addition to the camera platform and released into the air while attached to a ground crew whose duty was to guide the balloon around campus (and around obstacles on campus, such as light poles, buildings etc.). In the wind, however, Bessie's actual distance from the ground was significantly reduced from 400m because she was pushed some distance away from the ground control crew in the breeze. It was attempted to find her precise height using the radar distance finder (see distance azimuth exercise) but this attempt met poor results. The camera rig was tossed around rather harshly in the turbulence, which had the side effect of providing some nice oblique images of campus caused some concern for those on the ground. Regardless, the exercise continued as scheduled and the ground crew guided Bessie through the area until she broke free and made for the horizon somewhere over the bridge connecting campus across the Chippewa River. Fortunately, she was kind enough to drop her payload in the river before lifting off, and being waterproof the images retrieved from the day were quite usable after the rig was fished from the River by Professor Hupy.
Launch 2:
Monday the 8th provided far clearer weather than the week prior, and as such it was decided that a less weight-intensive payload could be attached to the second balloon (named Bertha). The old Low-Altitude, bottle-based design for the rig was improved by Stacey’s addition of an old arrow whose fletching was augmented with cardstock to stabilize the rig in the air and keep it from spinning much in the wind this time.
The rig was launched to a height of 550 meters and guided through campus much like it had been one week prior, but today the balloon was recovered as well as the imagery. Because of the good weather conditions, the balloon was taken over significantly more territory on this occasion than earlier.
Processing:
The class was allowed to use any combination of three programs to mosaic the images from each rig into a pair of georeferenced orthophotographs of campus: ArcMap, Imagine, and MapKnitter. Some students used the online services of MapKnitter, but I decided to use ArcMap, having mosaicked orthophotos in Imagine before and feeling some trepidation about the quality of Mapknitter’s abilities to provide a seamless image. Before Images can be mosaicked however, they must be georeferenced to locations on the earth.
The first task is to select the images to be used for processing from the camera. When looking for a good image, one looks for an image that is taken as close to directly overhead as possible: oblique images are not very useful for mosaicking. In the first set of images, this restriction made selecting appropriate images difficult as many of them were very oblique indeed. The second set of images was difficult to sift through simply because there were so many usable images, which is a good problem to have I suppose.
After the enough images to cover the area of interest have been selected, they need to be georeferenced. One does not simply piece each individual image together and call it done: the pixels in the image need to be tied to coordinates in reality using georefrencing. To do this, several “control points” in the image are selected which are readily identifiable and relatively stationary such as lamp posts, building corners, or road edges. These points in the image are then “tied” to the geographic coordinates of that point in reality; this can be done in several ways.
One way is to use another image, which is already
georeferenced, on which to overlay the new image being referenced. The points in the new image are selected, and
then related to the points in the original image in order to warp the new image
into an appropriate approximation of reality.
Another possibility is to use GPS or Surveying in order to physically
collect each point in the field, and use these points to connect to the image
being georeferenced. For both sets of
images, I used a georeferenced basemap to tie my GCP’s to, but for the second
set of images a group of students also set out to collect GCP’s using GPS
equipment.
In Arc, this requires the Georeferencing
toolbar. First, a dataset with spatial
reference is opened, and the new image is imported on top of it. Not having any spatial information, the new image
will not be displayed with the currently open map and so must be added to the
display by using the “fit to display” tool on the georeferencing toolbar. Then, the rotate, shift and scale tools are
used to roughly position the image where it belongs in relation to the control
points of the map. Then, the control
points on the map are selected individually and connected to the control points
of the reference image. Finally, a
transformation is applied to the raster which warps the image to best fit the
changes in each point. Most of the
images I mosaicked used 2nd-order polynomial transformation.
Finally, once all of the images are selected and georeferenced, they need to be mosaicked together into a single orthophoto. In my case, this is accomplished by using the mosaic to new raster tool in ArcMap, which creates a fresh new image from my multiple component images. For the first set of images, I used eight photographs to create a rough mosaic of campus. For the second set of images, there were so many photographs covering so much of campus that the class divided the area into six sections and worked on each section in groups of three. Then the class uploaded each section into a class geodatabase to be finally pulled together in and a single, high-quality image of campus from the class will have been made.
No comments:
Post a Comment