The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008
964
Integrated multi-sensor systems are increasingly used to provide
cost-effective and robust solution to navigation. Recently, some
efforts have been made to improve GPS/INS navigation by
visual aiding. The horizon line can be detected by an onboard
camera (Winkler et al., 2004) to provide pitch and roll angles of
a Micro Air Vehicle (MAV). A sequence of stereo imagery is
processed to determine the platform trajectory, which can
bridge the poorly determined sections of the platform trajectory
by GPS/INS (Tao et al., 2001).
An integrated GPS/INS/Vision navigation system for UAVs is
investigated in this paper. A CCD video camera and LRF based
vision system are used to sense the environment and observe
relative vertical and horizontal movements over the ground. The
system modelling strategies and the data fusion procedure for
this sensor integration scenario are investigated. Numerical
analysis is included to show the potential performance of the
proposed triple integration.
where v bxy are the horizontal translation velocities; i\ y is the
optical flow measurement of angular rate; cp xy at two horizontal
axis; rotation rates; r gz is the LRF measurement of the relative
height from the ground. The integration flow chart is shown in
Figure 1.
2. VISION AIDED MOVEMENT ESTIMATION
A wide rang of vision sensors are available to meet the
requirement of this particular application, which provides a
flexible enhancement to the integrated system. The study of
visual motion analysis consists of two basic issues. One is to
determine optical flow and/or feature correspondences from
image sequences, and the other is to estimate motion parameters
using them. Huang and Netravali (1994) have made a review
on the algorithms for estimation of motion/structure parameters
from image sequences in the computer vision context. In order
to optimally integrate vision component into a GPS/INS
navigation system, the vision navigation performance should be
investigated first.
Figure 1. Vision based navigation flow chart
There are several error sources in this model. The height from
the LRF may contain a small fixed offset (up to 10cm) and a
small scale error (<1%). Optic flow has scale errors. Gyro rates
also have bias and drift. Other errors include initial attitude
error and the ground slope etc. The major error sources can be
estimated using the GPS measurements as discussed below.
3. INTEGRATED GPS/INS/VISION NAVIGATION
The integrated GPS/INS/vision navigation system flow chart is
shown in Figure 2. Two Kalman filters (KF) are employed in
the system.
The image sequences taken from the UAV can be used as a
separate set of self-contained spatial measurements. Given that
close objects exhibit a higher angular motion in the visual field
than distant objects, optic flow can be used to calculate the
range to stationary objects in the field of view, or the true
velocity of objects with known ranges. In this project, optic
flow is calculated on the UAV helicopter in real-time at 50Hz
using an image interpolation algorithm (Srinivasan, 1994),
which is robust in natural outdoor environments and in the form
of angular rates of visual motion.
Two steps are needed to determine translation velocities from
the optic flow derived angular rates. Firstly, the effects of
rotation are separated from those translations by subtracting the
known rotation rates, measured by the onboard rate gyroscopes,
from the optic flow rates. Secondly, the image motion rate is
multiplied by the range above the ground estimated by a LRF to
estimate the mean-centred measurement of both lateral and
longitudinal velocities (Garratt and Chahl, 2003). The vertical
velocity relative to the ground can be calculated through the
measurement of LRF. As all the sensors have measurement
errors, the key issue here is to model and estimate the errors and
extract the navigation information from the vision and INS data
streams.
Therefore, the UAV horizontal velocity in the body frame can
be calculated from the optical flow, LRF and gyro angular rate
with the following formula:
V*,, = (n„ -<O x r ,. (1)