Aerospace and Electronic Systems - November 2018 - 26

Optical Flow Navigation System for a Flying Robot
based on ground truth data created with
the help of Google Earth by choosing four
waypoints around a specific location in
Kooye Daneshgah, as shown in Figure 8a.
For better understanding and verification
of the results, the estimation outcomes are
illustrated in a 3D plot, together with an
augmented graphical view with the aid of
Google Earth (Figure 8b).
In order to demonstrate the effects of
lighting on the OF predictions, we prepared
an additional examination in the same
place. Unlike the former test, under an
adverse illumination condition, the image
quality turned to be undesirable (Figure 9).
As previously explained, the first test
was done under desired illumination conditions to ensure that captured images by
Px4flow sensor be suitable for OF calculations in almost all the moments of the flight
test. Afterwards, the pixel velocity of the
vehicle, the metric velocity, and position
are calculated as well. In the new test, as
can be seen in Figure 9, from seconds 10
to 65, there is almost no any suitable image for OF calculations. In other words,
through applying HCD, there is not any
suitable pixel location for pixel velocity
computations. Accordingly, pixel velocity
is estimated nearly zero through this period
of time and at the succeeding moments of
the test.
In the following, Figure 10 exhibits that
the velocity estimations suffer from negative spikes in the y-direction and symmetrical ones in x components. As uncertainty
increases under adverse lighting conditions,
statistical characteristics of the predictions,
such as standard deviation of pixel velocFigure 8.
(a) The 3D traveled path. (b) Augmented graphical observation on Google Earth.
ity, confront negative impacts, prominently
from unpredicted degradations of the images. Consequently, low image quality due to improper illumination
11c) and position estimations (Figure 11d) is inevitable in indoor
conditions causes an increase in gross errors in the velocity predicas well as outdoor conditions. Similar to the outdoor flight test,
tions (Figure 10). Accordingly, this discrepancy appears in position
scientific evidence of flight path is provided based on ground
results, in comparison with the ground-truth data provided with the
truth data created with the help of Google Earth by choosing four
help of Google Earth that are acquired by choosing four waypoints
waypoints around a specific location in Guidance and Navigation
around a specific location in Kooye Daneshgah, and in the flight
Laboratory of University of Tehran, as shown in Figure 11d.
path under desired illumination conditions, as well (Figure 11a).
Finally, to evaluate the robustness of the presented method, an
CONCLUSIONS
indoor flight test was performed to evaluate the ONS under different illumination conditions (Figure 11b). Analogous to the outdoor
In this article, we first implemented three off-line OF algorithms
tests in adverse illumination conditions, shown in Figure 11b, an
(LK, HS, and BM), in order to compare them in terms of precision
appropriate image for OF reckoning can hardly be found during
and stability of velocity and position estimations. For the predictions,
all of the flight time. Hence, pixel velocity is estimated practically
during the OF process, the image velocity is assumed as the average
zero in this period of time as well. Due to the adverse illumination
of all pixel velocities of the considered frame. According to data disconditions, increasing of the gross errors on the velocity (Figure
persion criteria in every frame, the BM-based OF is more accurate
26

IEEE A&E SYSTEMS MAGAZINE

NOVEMBER 2018

Aerospace and Electronic Systems - November 2018

Table of Contents for the Digital Edition of Aerospace and Electronic Systems - November 2018