Aerospace and Electronic Systems Magazine September 2016 - 26

Onboard Visual Sense and Avoid System for Small Aircraft

Figure 9.

Flight test results showing aircraft paths from the GPS track of an avoidance maneuver. The intruder approaches on a straight path from the left
side. The solid blue line represents the own craft, and the red dashed
line represents the intruder craft. The additional lines connect the time
synchronized points on the trajectories.

Figure 10.

Flight test results (same as in Figure 9), overlaid on Google Earth map.
The solid blue line represents the own craft, and the solid red line represents the intruder craft.

Figure 11.

Flight test results showing the aircraft paths from the GPS track of an
UAS-UAS encounter when no avoidance is required. The solid blue line
represents the own craft, and the red dashed line represents the intruder
craft. The additional lines connect the time synchronized points on the
trajectories.

Figure 12.

Flight test results (same as in Figure 11 but overlaid on Google Earth
map. The solid blue line represents the own craft, and the solid red line
represents the intruder craft.)

gorithms are applied onboard in real-time using the Jetson TK1
hardware achieving 7-8 FPS performance. The final results of image processing and intruder tracking are the velocities of intruder
image centroid and size (dx/dt and dS/dt). Upon violation of noncollision or collision thresholds the collision decision is achieved.
An avoidance maneuver is initiated if required by applying bank
turns. Sideslip turns will be tested in the future.
Both own craft and intruder positions are logged onboard.
Straight paths of 20 m and 50 m from each other are generated and
tracked applying an altitude separation between aircrafts to avoid
accidental collision in case of system malfunction. This way, only
camera x coordinates (horizontal parameters) are used in collision
decision. First results are shown in Figures 9-12. Figures 9-10
show that if the two straight paths are close (20 m) an avoidance
maneuver is started and the distance between the aircrafts is increased. The logged minimum distance is 61 m, which is much
larger than 20 m, so the avoidance is successful. When avoidance
starts, the intruder is on the left side of own craft so the decision
about direction is good. However, if the distance between two aircraft is large (50 m (logged 51.5 m minimum)) no avoidance is
done as Figures 11-12 show.
These results mean that the proof of concept is achieved;
however, there are missed detections (MD rate 60% (6/10)), false
alarms (FA rate 43% (3/7)) (caused mainly by the cloudy weather), and the aircraft trajectories are sometimes far from straight.
This means that the system needs further development to achieve
reliable avoidance.

CONCLUSION
In this article the background of collision avoidance, in particular a visual sense and avoid system, are described. The system
considers only one intruder aircraft at a time; multiple intruders
are not considered. Also, a small survey on current implementations for UAS SAA with details of the authors' current solution is
shown. Visual SAA can be a feasible solution for smaller RPAS
and also for manned airplanes as a backup or complementary system in collision avoidance. Based on the background complexity,
two kinds of detection strategies are used. The detection system
uses the multi-fovea approach in order to minimize the computational costs of sophisticated image processing algorithms. A
GPU and an FPGA implementation of the processing part is also
performed.
Regarding collision risk estimation and possible avoidance
strategies, two specific solutions are briefly presented, building on
the relevant literature of the field. The closest point of approach
and related formulae are used for collision risk estimation. Characterization of possible intruder threat size and velocity are also part
of threshold selection. Regarding the avoidance strategy, a three
step procedure with successive turns is proposed with which own
craft flies around the intruder. Preliminary flight test results with
a real intruder aircraft are presented. In case of intruders within
minimum separation distance, successful avoidance is accomplished. In case of intruders outside of the required separation, no
avoidance is applied. However, the proposed system needs further
development in several aspects, which will be the scope of our fu-

IEEE A&E SYSTEMS MAGAZINE

SEPTEMBER 2016

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