IEEE - Aerospace and Electronic Systems - July 2022 - 33

Image licensed by Ingram Publsihing
application is often autonomous racing, where the gates and
windows have synthetic, predetermined shapes [18]-[21].
Although these methods are more robust against disturbing
factors in the task of racing gate detection, they do not provide
enough navigation data required for safe passage
through a gap; the output ofsuch networks is a bounding box
containing target window, while the exact boundaries ofwindow
are required for a safe aligned crossing.
Despite the impressive accuracy, detection algorithms
are computationally extensive; hence, it is tricky to deploy
them framewise in an online and real-time setup. The
common workaround is to perform the detection only periodically
and deploy a relatively fast out-of-the-box tracking
algorithm to track the detected targets framewise.
Among several trackers reported in the literature, certain
optical flow variations are more suited for online and realtime
implementation (see Table 1 for a list of standard and
most popular built-in trackers in the open-source computer
vision library OpenCV). It is likely for any tracking algorithm
to lose or drift from the target as time grows. In our
preliminary experiments, it was revealed that such a drift
could possibly lead to a complete failure, especially when
the window portal is quite inclined with respect to the
camera optical axis in the mission initialization.
Taking these observations into account, we urge that the
sole implementation ofenhanced optical flow algorithms for
the entrance mission could only offer a limited success, due
to the safety and reliability issues. To establish a collisionfree
guidance strategy for the intended mission, a novel yet
simple technique is developed to prevent the tracker from losing
sight ofthe portal or even drifting, which is referred to as
the dark area extraction (DAE). The main objective is to provide
the tracker with an accurate real-time estimation of the
window portal, i.e., boundaries, which effectively aids the
guidance algorithm in the tracking and entrance phases.
The Dark Area Extraction (DAE) method relies on the
fact that the portal section of the window is the least lightreflective
part (in daytime), and therefore, extracting this
section will lead to a good estimation of the window portal
boundaries. To cope with several disturbing factors affecting
the success of the DAE, we introduce a dynamic thresholding
technique, whose tuning procedure is addressed from
two different but complementary viewpoints.
Using the DAE in conjunction with the tracker offers the
following advantages. First, the boundary estimation can be
used to compensate for the tracking drifts, since it provides
information that is less sensitive to noise compared to the
appearance-based features typically used in the tracking algorithm.
Second, a 3D perception ofthe MAV orientation relative
to the window can be established using the information
associated with the estimated window boundaries [22]. This
navigation data aids the guidance strategy in the final
entrance phase such that the MAV can pass through the center
ofthe window in a perpendicular path, hence, minimizing
the risk ofany collision with the window boundaries.
The rest of this article is structured as follows. In the secFigure
1.
Schematic of the window entrance mission by an MAV equipped
with a single monocular camera. The three coordinate systems
shown and all the symbols are defined in the " Guidance " section.
JULY 2022
tion " Entering Through Window-Overview, " we give an
overview of the guidance strategy developed for the entrance
mission. In the section " Dark Area Extraction, " we provide a
detailed presentation of the DAE method and its tuning.
IEEE A&E SYSTEMS MAGAZINE
33
IEEE - Aerospace and Electronic Systems - July 2022

Table of Contents for the Digital Edition of IEEE - Aerospace and Electronic Systems - July 2022
