Real-Time Implementation of Vision-Aided Navigation for Small Fixed-Wing Unmanned Aerial Systems

Small Fixed-wing Aerial Positioning Using Inter-vehicle Ranging Combined with Visual Odometry

“…In [33] David G. Lowe developed SIFT which uses a process of Gaussian blurring to find features within an image. In [34] SIFT was used as the main image feature detection and the results of SIFT in a SUAS application can be explored further in that study. SURF is very similar to SIFT, except it is computationally more efficient making it a more viable solution for real-time use on large image sets.…”

“…More information on SIFT and SURF as well as a comparison between the two, can be found in [35]. In [34] a UAV position estimate was calculated by comparing aerial images of the ground with georefrenced imagery, called Perspective-n-Point (PnP), using SIFT as a feature detector based on an algorithum described in [41]. Using a 3 degree of freedom solution, where the altitude of the aircraft was determined by the autopilot, the lowest distance root mean squared (DRMS) error was 8.22 meters.…”

“…This means the location of the aircraft would be known to 8.22 meters 67% of the time. [34] focused on comparing georefrenced imagery with real-time aerial monocular imagery.…”

“…From [34] it was noted that using the autopilot attitude in the 3 degree of freedom solution allowed for a better position estimate. Allowing the software to estimate the attitude (6 degrees of freedom) produced a worse position estimation then the 3 degrees of freedom solution.…”

“…Allowing the software to estimate the attitude (6 degrees of freedom) produced a worse position estimation then the 3 degrees of freedom solution. In this research the same autopilot was chosen and the autopilot attitude was used in the VO to track the camera rotation between images as it appeared to be sufficient in [34].…”