Aerospace and Electronic Systems Magazine November 2016 Tutorial X - 100

Sense and Avoid for Unmanned Aircraft Systems
radar, since this process can be exploited on a small rectangular
window of pixels inside the image. This window is chosen as
a function of the LOS estimated by the radar and the expected
level of uncertainties in the angular measurements of the radar
[30].
2. Template matching. In this case, the processor can access a
proper database reporting an adequate coverage of all the expected images of object of interest. The object can be detected
by performing a correlation between the stored image and the
one acquired by the sensor. This technique works well for objects that have a small variability in terms of shape, but it tends
to be degraded if the number of objects in the database becomes
large.
3. Morphological filtering. In this case, specific morphological
operators are applied in a proper sequence in order to highlight and separate candidate objects to be detected. Subsequently, the image is binarized by selecting a proper thresholding strategy. At the end of the application, all connected
clusters that have expected size and pixel distributions are
classified as objects.
In general, a detected object is extended over several pixels.
As a consequence, the position if its image center can be computed with subpixel accuracy by averaging the position of pixels
weighted with the measured intensities as happens when the center of mass of a point mass distribution. In this case, the resulting
value of q can be represented to a level of accuracy in the order of
1/10 of a pixel.
In order to assess the detection range for EO cameras, two
terms must be considered.
1. Radiometric detection, i.e., the standard term that is regulated
by (27). In the case of the EO cameras, this condition must be
checked for a single pixel. Indeed, if the measured level is in
the order of the level of noise, no object can be detected.
2. Geometric detection, i.e., an object can be detected if it is distributed over an area of pixels that is larger than a minimum
value.
Standard detectors are realized so that the noise is contained in
the first few levels of the digital output for each pixel. Moreover,
the level of sunlight reflected by most aircraft or the heat emitted
by an aircraft engine is always much greater than the sensor noise
in practical application. As a consequence, radiometric detection is
not the one that limits the detection range.
Regarding geometric detection, a basic support is provided
by the Johnson rule [31], which establishes that an object cannot
be detected by an imaging sensor unless it is contained in at least
two pixels. It is worth noting that the Johnson rule is a purely
theoretical limit. A practical approach requires that the object is
extended over 4 or 5 pixels. As a matter of fact, if an object has
a characteristic size equal to l, e.g., the wingspan of an aircraft, q
is the minimum number of pixels needed for the detection, m is
the pixel size, then the detection distance Rdet for an EO camera
is given by (40)
100

(40)
An aircraft with a wingspan of l = 9 m can be detected by a 1
Mpixel EO camera with focal length f=2 mm and pixel size of 5
µm at a distance Rdet= 720 m, if a 5-pixel criterion is applied.
It is worth underlining that the boundary between obstacle detection and tracking may be blurred, as many vision-based SAA
approaches use multiframe techniques such as track-before-detect
logics.
In summary, EO cameras can be an option as a primary source
of SAA information for small UAS flying into class G airspace.
Alternately, they can be used as secondary sensors of a radar in
order to increase the overall angular resolution and the data rate.

D. GROUND-BASED SENSOR CONSIDERATIONS
The sensor descriptions so far have assumed that the sensors are
onboard the UA. As mentioned in Section II, ground-based sensors
can also be used to inform SAA for the UAS. The math to describe
the encountered geometry is the same as outlined in Section IIIA-the major change is the perspective of the sensor in that geometry. The ground-based sensor senses the intruding aircraft (location and velocity) in absolute terms, which can be mathematically
compared with that of the UA (assuming that the UAS knows the
location and velocity of its "ownship"). In contrast, the airborne
sensor senses the intruding aircraft in relative terms-the location
and velocity are calculated as a difference from the location and
velocity of the UA.
Sensors that are located on the ground offer two advantages-they are not as constrained in SWaP as onboard sensors, and
they are not dependent on RF links to transmit the information to
the GCS and the pilot. These sensors can also provide their data
directly to onboard systems for detection and avoidance processing, especially for the collision avoidance situations that require
very fast response times. As outlined in Section II, ground-based
sensors are environment-centric, rather than UA-centric, so they
are able to detect intruding aircraft anywhere around the UA.
This is particularly useful when the mission of the UAS is limited to a specific fixed geographic area (i.e., a farm or a power
station) as the ground-based sensor can be set up to maximize
the desired coverage of the fixed area.
The main sensors that can be used in ground-based sensing are
noncooperative sensors. Cooperative sensors only apply to the airborne part of the systems, though processing the cooperative environment can be accomplished in the air or on the ground. Many of
the noncooperative sensor types described in Section III-C can also
be applied as ground-based sensors.
Monostatic or traditional radar is a very mature technology
with proven systems that are available commercially. Some are
based on military transportable systems, such as the LSTAR [32]
or Sentinel, which have been adapted and tested for civil use.
These systems tend to be more expensive, though there are a wide
variety of vendors that offer many lower cost portable systems. Active radar provides the advantage of providing accurate range and
bearing information. Some radar systems have been produced for

IEEE A&E SYSTEMS MAGAZINE

NOVEMBER 2016, Part II of II

Table of Contents for the Digital Edition of Aerospace and Electronic Systems Magazine November 2016 Tutorial X