 |
| Search all USDA |
 |
|---|
|
Aerial Vegetation Survey
Ryan Becker, Project Leader
Mosaicking UAV Images Data from two flights in Hawaii were selected for mosaicking trials. The trials included various automated mosaicking software as well as several image processing services. The final products showed widely varying levels of success, as well as total cost. The trials revealed many shortfalls of UAV-collected data for this application, but these shortfalls can generally be minimized by upgrading the UAV with existing technology, as well as monitoring and controlling flight plans and conditions more closely. Ultimately, however, while mosaicked UAV-collected images do retain better resolution than commercially available satellite data, a considerable amount of resolution will be lost compared to the original images, given the current limitations of commercial mosaicking software. It should also be mentioned that images may be mosaicked into two basic types of products, called first- and second-generation orthophotos. A first-generation orthophoto is created by merging images using only the positional information about the images, or metadata. The positional information must be extremely precise, and the image must contain virtually no distortion, to successfully create a first-generation orthophoto. A second-generation orthophoto relies on an existing orthophoto or other map as a reference for placing the new images. Common objects can be aligned between the original and new images, and the new image can be moved, stretched, and skewed to align with the old image. The inherent imprecision of UAV positional information, combined with the inevitable lens distortion, eliminate the possibility of creating satisfactory first-generation orthophotos with existing technology. All subsequent discussion on this page assumes that a second-generation orthophoto creation process is followed. UAV Metadata Image metadata is information about the spatial location and orientation of the camera when an image was taken. This information is essential for easily overlaying an image onto a map, or for making pictures into a map. A combination of GPS data, aircraft attitude, and occasionally time are required for this process. This metadata is also required for UAV flight control software, so basically any UAV that can carry a camera and record its flight data can be used for this application. Several other factors determine whether the images and metadata collected can produce useful mosaics. The Bat carries good equipment for this application, while MLB plans to incorporate significant upgrades in many areas. The combined effect of the upgrades will be a significant increase in the speed, accuracy, and quality of mosaicking possible with Bat-collected data. GPS data GPS accuracy and update frequency are major factors in determining mosaicking quality and success. One reason for this is the relative inflexibility of existing mosaicking software packages. None located during this project could shift images more than ten pixels from their GPS-reported positions when seeking overlap. For a resolution of two inches, this means an image must be located within twenty inches of its true position for automated mosaicking to be successful. Of course, this resolution pushes the limits of GPS technology at any expense and in any condition. One way of circumventing the pixel limitation is by down-sampling the image to be placed until automated overlapping is successful, then replacing the down-sampled image with a higher resolution version and repeating the process. This technique is not currently automated in any package, so it can be even more time-consuming than purely manual overlapping. GPS update rate plays an important role in collecting timely positional information. Images may be created as fast as once per second by the camera, depending on the UAV's ground speed. Many commercial GPS units update just once per second, leaving open the possibility of a difference in records of as much as half a second, or eighteen feet, given a UAV cruise speed of 25 miles per hour. While software interpolates between GPS points given the measured lag, this cannot account for influences such as gust-caused course correction. A new GPS unit will provide an update rate of four measurements per second in the upcoming Bat controller hardware. Aircraft Attitude Aircraft roll, pitch, and yaw (collectively called aircraft attitude) determine the deviation from vertical of the camera's viewing angle. This affects both deviation of image position from aircraft position and image perspective. Software calculates image position based on roll, pitch and yaw data, but perspective cannot be reliably removed. This reduces image similarity between adjacent photos, complicating the location of common features. A more elegant, but heavy and expensive, solution used by larger aircraft is to mount the camera in a gyroscopically stabilized housing that maintains orientation and verticality regardless of aircraft attitude. An entirely different, larger aircraft would be required to carry a housing of sufficient size to stabilize even a commercial digital camera. The accelerometers used by the Bat are currently accurate enough to re-orient the photo properly for manual placement. The greatest potential improvement for aircraft attitude inaccuracy lies in flying in higher, more stable air masses. This solution awaits resolution of the FAA issue. Lens and Perspective The resolution page shows examples of the physical origins of lens- and perspective-based distortion in low-altitude aerial imagery. The steps required to mathematically remove such distortion are complex and consume large amounts of processor time, and consequently none of the mosaicking software packages were able to perform such corrections to the Bat data. Most of the software packages do perform similar corrections to satellite data, but several factors allow software makers to successfully limit the complexity of the problem in this special case. Satellites peer down on Earth from a height where a 35-millimeter lens could show the entire globe in a single image. The amount of magnification employed by satellites to produce images allows the virtual elimination of lens and perspective distortion. Satellites use sophisticated location and stabilization systems that ensure high-quality metadata. The end result is typically a set of precisely placed, overlapping images that are relatively easy to tie together with very little stretching or skewing. A moderately priced UAV flying one thousand feet above ground level cannot match the spatial precision of a satellite, and a high magnification at such an altitude yields a tiny field of view. It is therefore likely that UAVs will use relatively wide-angle lenses, with all their associated benefits and drawbacks, for the foreseeable future. While these lens-related issues necessarily deteriorate the uniformity of the collected images, the processes to compensate for them and enable automated mosaicking are attainable. Image processing companies are not likely to develop these tools until demand increases. |
 
|
