From Thousands of Images, a Clear Picture
Lockheed Martin Engineers Develop an Algorithm to Mosaic Images from Low-Flying UAVs
Thousand-piece puzzles are difficult, but imagine assembling thousands of images of a 250-acre field without the use of curved pieces that conveniently fit together. Using an algorithm that links images from low-flying unmanned aerial vehicles (UAVs), Lockheed Martin engineers have found a way to assemble these pieces to create a single high resolution image in a matter of hours instead of days.
Gathering imagery from UAVs is up to ten times cheaper than imagery from manned aircraft, and it’s rapidly becoming an affordable imaging alternative for disaster response, infrastructure maintenance and precision agriculture. Current Federal Aviation Administration restrictions prohibit small UAVs from flying higher than 400 feet, which enables a narrow, but high-resolution aerial view for imagery. In order to capture imagery of a wide geographic area, those images must be mosaicked together to create one large high-resolution photograph.
The conventional algorithm which currently performs these “orthorectified mosaics” is only equipped to manage tens or hundreds of images at a time, and even then, can require several hours – even days to complete. Engineers at Lockheed Martin created a new algorithm that projects tie points to the ground using estimates of the image cameras and adjusts these projections to reduce tie point errors. So rather than simultaneously varying all the camera estimates to reduce the projection error, this new algorithm reduces the projection error and then fits the estimates independently.
When coupled with Lockheed Martin’s automated georegistration software called GeoMI, thousands of images can be mosaicked much faster to enhance the precision of large-scale mapping.
“We didn’t set out to create this new algorithm, because at the beginning of our research, we didn’t know we’d need it,” said Dr. Mark Pritt, Lockheed Martin systems engineer. “We were working with a small agricultural company that needed to mosaic thousands of images of a farm, but their commercial software couldn’t handle the amount of data we were throwing at it.”
While UAVs fly their predetermined pattern over a chosen area of interest, they take snapshots every couple of seconds. Due to variables like wind and aircraft stability, the images don’t perfectly align when overlapped. Using the algorithm repeatedly to reduce projection error on the images causes the tie point overlap errors to fade, leaving a single blended image as seen below.
The Indago unmanned aircraft system gives warfighters, first responders and others in small, cluttered, urban environments an eye-in-the-sky in just minutes. More on Indago.
Using Photographs to Guide UAVs
Small UAVs are an effective and economical option for capturing imagery, but Pritt’s research also explores the use of imagery to guide the aircraft themselves.
Most aircraft, whether manned or unmanned, depend on the Global Positioning System (GPS) for navigation, but GPS signals are susceptible to jamming or spoofing. Using only a digital camera, the system works by taking a photograph of the ground and georegistering it to a reference image or digital elevation model (DEM) with the GeoMI™ software.
By transferring geospatial information from the DEM to the photo, the system calculates the position and orientation of the aircraft. The process is fast enough to perform in real time and suitable for use as a primary or backup navigation system for aircraft, UAVs, and cruise missiles. For nighttime navigation, the system could be extended to infrared or radar imagery.
From mapping flood damage to monitoring power lines and pipelines, small UAVs are already an effective and economical option for remote sensing. For example, Benjamin Dittbrenner and Chris DiTomaso, post graduate students at the University of Washington’s School of Environmental and Forestry Sciences use UAVs to routinely monitor hydrological impacts made by relocated beaver populations, and they required the use of image mosaicking during their research.
“We had to find a way to prove that these changes were being made to the landscape and measure those changes over time – remote sensing was the best way to do that,” said Dittbrenner. “As researchers, we’re not in the business of learning how to process data – we’re in the business proving science. Creating more robust software that can do mosaicking faster and more accurately will be incredibly helpful to the scientific community.”
Published April 20, 2015