Drone navigation possible in GPS-free environment using Google Maps
Wed 5 Apr 2017
A team of researchers from the National University in Singapore created a framework to navigate drones in a GPS-denied environment using Google Maps.
The team was able to use Google Maps to create geo-referenced navigation for unmanned aerial vehicles (UAVs) in a flight where GPS is unavailable to the drone pilot.
The issue with tracking an operational UAV without GPS involves simultaneous localization and mapping, or SLAM. This refers to the problem of constructing a map of a new environment while tracking the drone’s activity and location within the environment. There are SLAM algorithms that address this issue, however, estimations of position based on inertial measurements and optical flow tend to suffer from drift.
Geo-referenced navigation combining onboard camera images and information from Google Maps was shown to be both reliable and drift-free in the testing environment.
The process used by the researchers involved searching Google Maps for the drone’s position by using the onboard camera, and comparing the gradient pattern of the captured photo frame to Google Maps, matching patterns to provide a precise location.
Matching patterns efficiently poses a challenge that the team met by using Histograms of Oriented Gradients to register the image, and particle filters to expedite the gradient match.
This gives the operator a confidence map of the UAV’s position in near real time. Optical filtering then provides translation between consecutive frames, so that the predicted position of the drone is constantly updated for the navigator.
The combined use of Histograms of Oriented Gradients (HOG), with Optical Filtering (OF) and Particle Filtering (PF) was shortened to the acronym HOP, which describes the new framework for efficient GPS-free navigation.
In the experiment, the team found that the HOP method was an accurate and reliable method for drone navigation in an environment where GPS is denied.
Images transmitted from the drone and subjected to the HOP framework failed to register only 7% of the time, which the researchers determined was due either to a homogenous area with few gradient patterns, or a significant illumination change from one frame to the next.
Not only did the combined effect of HOP improve upon the results obtained from optical filtering alone, it had a localization accuracy that was comparable to GPS.