Aerospace and Electronic Systems Magazine September 2016 - 24

Onboard Visual Sense and Avoid System for Small Aircraft
The detailed strategy proposed in [24] is to turn in the direction
of the intruder for a given time (vision system will possibly lose
intruder during this maneuver). Following this maneuver, own will
start to turn back to original path. If the intruder is again observed
and CPA threshold violation occurs, another turn is started in the
direction of the intruder. Turns and back turns are repeated until the
intruder is out of FOV or no CPA threshold is violated.
This can be a good strategy, but requires intensive maneuvering of own craft which increases power consumption. There is also
a chance to lose the intruder by the vision system during the first
maneuver and to be incapable to recover (re-detect) because the
detection system has usually a transient time.
Considering this and the danger to turn in front of the intruder
and collide (see Figure 7) the authors propose a modified strategy
which is also based on pure image data.

PROPOSED INTRUDER AVOIDANCE STRATEGY
The first problem to be solved is the decision about direction of
avoidance. Small UAVs usually turn better then ascend or descend,
so turning maneuvers are considered. Another issue is camera FOV
in which we should hold the intruder at least at the first part of an
avoidance maneuver. By an ascending/descending maneuver it is
easier to lose the intruder from the FOV because camera FOV is
usually a horizontal rectangle (vertical angle is much smaller than
horizontal). Of course, it is also possible to lose the intruder from
the camera FOV in the case of horizontal (turning) maneuvers as
pointed out in [25]. That's why sideslip turns should be considered.
The direction of avoidance is decided based on the direction
(bearing) of the intruder in [19]. However, turning into the direction of the intruder can lead to collision as visualized in Figure 7.
As a result, it is better to turn away from the intruder except for the
crossing scenario, which is also shown in Figure 7. The best possibility is to consider not only intruder bearing (position on image),
but also its velocity in the image. This way a decision table can be
constructed as shown in Table 1.
The table shows two types of maneuvers. If the intruder is on one
side and moves into that direction that means that it will pass (or hit)
own craft on that side, so avoidance should be done into the other
direction. If the intruder is on one side and it moves into the other
direction (towards the other side) that means the crossing scenario
and so avoidance should be done to the side of the intruder.

Table 1.

Decision Table for Avoidance Direction
Intruder
Position
Velocity
Direction

Avoidance
Maneuver
Direction

Right

Left

Right

Left

Right

Left

Right

Left

Right

Left

Intruder
Position on
Image

Figure 8.

Fly around the intruder with 90° angle or below.

The second problem to be solved is the tracking and release
of intruder during the avoidance maneuver. A simple strategy can
be to turn until the intruder moves to the side of the image, then
hold the intruder on the side of the image (hold it at a given bearing angle) until own craft faces its original straight path again and
can continue path tracking. This strategy can guide the own craft
to fly around the intruder; however, it could lead to a collision if
the bearing angle reference is not 90° because in that case the own
velocity vector has a component pointing towards the intruder
(see Figure 8).
Examining the possibility to hold the intruder at a given bearing angle will give a course angle rate reference which should
be followed. In the GPU test system, the horizontal FOV is 116°
which means that a 50° bearing can be safe to track the intruder.
However, this could lead to a collision. That is why the theoretical
possibility to hold the intruder at 90° is also examined. Avoidance
scenarios were calculated obtaining the required course angle rate
to hold the intruder at the given bearing angle. Results are summarized in Tables 2 and 3.
In the tables, Vi and Vo are the intruder and own craft velocities, respectively, beta is the angle of the intruder path relative to
own path (positive if the intruder comes from the right), om0 is
the maximum course angle rate of own craft in a sideslip turn,
and Tbeta is the target bearing angle where the intruder should be
held. Distance is the distance between the intruder and own path
when the intruder is beside own craft (Z = 0, negative on left side).
max(om) is the required maximum course angle rate which should
be realized by own craft to hold intruder at the given angle.
The tables show that max(om) is well above the maximum possible rate in all of the cases either for a 50° or 90° intruder angle.
This means that it is impossible to track the intruder with own craft.
In case of 50°, the intruder angle tracking means that the own craft
will approach and possibly hit the intruder. Incapability to track the
intruder can be advantageous. This observation is applied in designing the avoidance strategy, which consists of the following steps:
1. Turn with maximum possible course angle rate into the decided
direction until the intruder is lost from camera FOV.
2. Start decreasing turning rate towards the reverse maximum.
This finally means a turn back to the original own path.
3. If own direction is close to own path direction again, continue
path tracking.

IEEE A&E SYSTEMS MAGAZINE

SEPTEMBER 2016

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