Aerospace and Electronic Systems Magazine September 2016 - 12

Prototype Sense and Avoid System Onboard a ScanEagle Unmanned Aircraft
During post flight-test analysis it was found that using a smaller
(3×3) structuring element in the morphological filter improved the
detection performance of the SAA system (as shown by the red
SOC curve of Fig. 7). Not only did the new structuring element
increase the average detection range from 1.6 km to 2.1 km, for a
false alarm rate of zero, but it also allowed the SAA system to detect
the threat aircraft successfully in the two nondetection encounters.

DISCUSSION
DETECTION RANGE
The FT&E results presented in the previous section represent a
world first: real-time SAA demonstrated on a small unmanned
aircraft in nonsegregated civilian airspace. The achieved detection
ranges of the ScanEagle-based SAA system are consistent with results that were published for the Cessna-based SAA system [10].
This consistency provides a high-level validation of the process of
migrating the SAA system to the ScanEagle unmanned aircraft.
Fig. 8 shows the ScanEagle's SAA detection-range performance, for the 5 mm lens, in the context of other published results.
The high take-off weight SAA results are from manned aircraft,
acting as surrogate unmanned aircraft, and two optionally-piloted
aircraft. The low take-off weight SAA results are from obstacleavoidance detection systems that are designed for unmanned aircraft. Our world first claim is based on the lack of other real-time
SAA results from small UAS as shown by Fig 8.
Previous research has found that low-time pilots have an average aircraft detection range of 1670 m [20]. Fig. 8 and Table 2
show that the FT&E results from the ScanEagle's SAA system provided similar detection ranges. Thus, the ScanEagle's SAA system
has demonstrated a capability that is similar to a pilot's ability to
see and avoid other aircraft.

RECOGNITION AND REACTION TIME
A Federal Aviation Administration (FAA) advisory circular that
discusses the pilots' role in collision avoidance suggests that, for a
manned aircraft, the average person has a recognition and reaction
time of 12.5 s [33]. The advisory circular also provides a breakdown of the 12.5 s into its constituent components. The 12.5 s has
also been used as a general guide for unmanned aircraft SAA systems. The recognition and reaction time for the prototype ScanEagle SAA system removes or reduces some of the human-related
components of the 12.5 s, which reduces the overall recognition
and reaction time. The average detection ranges and times shown
in Table 2 were achieved with all of the image stabilisation, detection, and communication processing times included.
One additional factor to be considered for a UAS is the data
link latency. The data link latency for the ScanEagle SAA system
is less than 50 ms, which is a minor contributor to the overall recognition and reaction time.
During the FT&E campaign one human-in-the-loop collision
avoidance manoeuvre was performed successfully. The ASL was
detected at a range of more than 3 km (see Fig 6). The ScanEagle
was subsequently commanded to change its trajectory, to avoid the
12

Figure 6.

Flight paths of the ScanEagle (black line) and the ASL (red line) during one encounter (8 mm lens configuration). The blue stars mark the
position of each aircraft when the ASL was detected by the SAA system
(t = 78 s). The remote pilot subsequently commanded the ScanEagle to
execute an avoidance manoeuvre. The red circles mark the position of
each aircraft at the minimum miss-distance point (t = 127 s). The '+'
symbols on the flight paths indicate the positions of each aircraft, in 30
s increments, from the start of the experiment (t = 0 s).

collision threat, 2.7 s after the detection. Automating the execution
of avoidance manoeuvres will put the remote pilot "on-the-loop",
in a supervisory role, rather than "in-the-loop" and consuming
valuable time making a decision. Thus, automation will reduce the
overall recognition and reaction time, which provides additional
time to execute a collision avoidance manoeuvre.

SOC CURVES
Figure 7 presented the SOC curves for a system that used the original structuring element in the morphological filter and for a system with configuration adjustments made during post flight-test
analysis. The detection range and false alarm rate are inextricably
linked by the detection threshold [15]. The SOC curves show that
longer detection ranges can be achieved at the expense of higher
false alarm rates.
An important function of an SOC curve is that it allows the
selection of a detection threshold that will provide the desired false
alarm performance. Typically, the aim is to operate the SAA system
using a detection threshold that provides zero false alarms. Figure
7 shows that this point occurs where the SOC curve intersects the
y-axis (i.e., selecting a threshold of 0.38 for the red SOC curve). As
the figure shows, increasing the detection threshold from this point
does not reduce the false alarm rate, but it does decrease the detection range. Similarly, decreasing the detection threshold from this
point increases the detection range at the expense of also increasing

IEEE A&E SYSTEMS MAGAZINE

SEPTEMBER 2016

Table of Contents for the Digital Edition of Aerospace and Electronic Systems Magazine September 2016