Feasibility Study for Pose Estimation of
Small UAS in Known 3D Environment
Using Geometric Hashing
Julien Li-Chee-Ming and Costas Armenakis
Abstract
A novel self-localization method for small Unmanned Aerial
Systems (
UAS
) is presented. The algorithm automatically
establishes correspondence between the First-Person View
(
FPV
) video streamed from a
UAS
flying in an urban envi-
ronment and its 3
D
building model. The resulting camera
pose provides a precise navigation solution in the densely
structured environment. Initially, Vertical Line Features are
extracted from the
FPV
video frames, as the forward and
downward-looking camera is kept stabilized through a gimbal
camera mount. Geometric hashing is then used to match the
extracted image features with a database of Vertical Line
Features extracted from synthetic images of the 3
D
build-
ing models. The exterior orientation parameters of the
FPV
video frames for localizing the
UAS
frames are determined
by photogrammetric bundle adjustment. The results demon-
strate that sub-meter accuracies in the
UAS
's X, Y, Z positional
coordinates are achievable from flying 40 m above ground.
Introduction
First-Person View (
FPV
) unmanned aerial systems (
UAS
) are
equipped with a small forward-looking video camera and
a transmitter to downlink the video signal wirelessly in
real-time to a ground station monitor or to virtual reality
goggles.
FPV
gives the operator of a radio-controlled
UAS
a per-
spective view from its “cockpit.” This allows the aircraft to be
flown more intuitively than by visual line-of-sight and beyond
the pilot’s visual range, i.e., where the aircraft’s separation
from the pilot is limited only by the range of the remote con-
trol and video transmitter.
FPV
systems are commonly used
solely as a visual aid in remotely piloting the aircraft. The
obtained quantitative information on position and angular
orientation of the aerial platform supports the
UAS
operator in
navigation and path planning. The need for precise naviga-
tion is increased in urban missions, where the possibility of
crashing is high, as
UASs
fly at low altitudes among buildings,
avoid obstacles, and perform sharp maneuvers. This is espe-
cially useful in
GPS
-denied environments, or in dense-signal
multipath environments such as in urban canyons. As such,
this work does not require an
a priori
position and attitude
estimate of the
UAS
, perhaps provided by an autopilot.
This work contributes to the localization process of
UAS
video frames because an
FPV
video sequence typically consists
of thousands of image frames. We propose an approach to
obtain the
FPV
video camera’s position and orientation (pose)
as it travels through a known 3
D
environment by matching
video image features with features from the 3D model of the
environment.
Quickly estimating the
UAS
’s position and orientation is
crucial, as the
UAS
travels rapidly. This ultimately requires an
efficient search of the 3
D
model database. Geometric hash-
ing (Wolfson and Rigoutsos, 1997) has been widely used to
determine the feature correspondence between video frames
and the 3D model of the environment in on-line applications.
Geometric hashing is well-suited for this application as the
extracted image features are not compared with every fea-
ture in the database; instead, the search space is strategically
reduced such that only relevant information is accessed.
Geometric hashing is a simple, efficient, and robust feature
pattern matching method between two datasets.
The developed self-localization process requires a metric
3
D
map of the environment. The main steps of the proposed
approach are:
1. Generate a database of model Vertical Line Features
from synthetic images of the 3
D
map.
2. Extract Vertical Line Features from the video frames.
3. Use geometric hashing to match the extracted image
features with their corresponding model features in the
database.
4. Determine the camera position and orientation param-
eters, as a function of the matched feature locations in
the map using a photogrammetric bundle adjustment
solution.
Notably, the proposed approach estimates the required initial
approximations for the exterior orientation parameters, thus
the onboard sensors (i.e., the
GPS
,
IMU
, and magnetometer)
were not used in the proposed solution to determine the
UAS
’s
pose. If an initial approximation is available, Li-Chee-Ming
and Armenakis (2013) propose a viable solution to refine the
navigation solution.
Related Work
Different solutions have been proposed to this model-to-image
registration problem, Treiber (2010), Chin and Dyer (1986),
and Besl and Jain (1985) provide a comprehensive survey.
Among them is the geometric hashing technique (Gavrilla
and Groen, 1992), an algorithm used in computer vision to
match features against a database of such features. It is a
highly efficient technique as matching is possible even when
Geomatics Engineering, Department of Earth and Space
Science and Engineering, Lassonde School of Engineering,
York University, 4700 Keele Street, Toronto, Ontario, M3J 1P3
Canada (
).
Photogrammetric Engineering & Remote Sensing
Vol. 80, No. 12, December 2014, pp. 1117–1128.
0099-1112/14/8012–1117
© 2014 American Society for Photogrammetry
and Remote Sensing
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
December 2014
1117
