Graphics Reference
In-Depth Information
Fig. 4.20
Third indoor experiment: The MAV is commanded to take off autonomously and rise
up to 1.5 m. Then we continuously set waypoints to explore the environment (the
solid line
follows
the flight trajectory). Once a valid landing spot is detected, the two approach waypoints are generated
(
stars
) which are used by the MAV for the landing maneuver. Note that each waypoint includes a
tolerance radius (the trajectory does not hit the
red star
at 1.5 m altitude exactly). We reduced this
tolerance for the actual landing spot on the surface to enable high-precision landing
over the landing zone by defining manual waypoints, which were approached by the
vehicle autonomously while executing the landing site detection algorithm to analyze
the area beneath the MAV. As soon as an appropriate landing spot was detected,
the two approach waypoints were submitted and executed by the vehicle (Fig.
4.18
a).
Figure
4.20
depicts the 3D point cloud of the reconstructed box and ground surface
together with the flight trajectory and the issued approach waypoints. The vehicle took
off to the left of the illustrated scene and landed correctly on top of the box. Example
scene views, together with a resulting landing map are illustrated in Fig.
4.17
.All
pixels in the middle of the box have been labeled correctly as safe to land (blue),
whereas pixels close to the edges of the box are detected as either unsafe (red) or
provide not enough space to land on (orange).
4.4.4.2 Outdoor Evaluation
For the outdoor experiments, we conducted overflights over a one story building
(Fig.
4.21
) and recorded image sequences from the downward-looking camera
together with pose data for offline analysis. A quantitative evaluation for two differ-
ent overflights is given in Table
4.2
row 3-4. The average altitude of the first flight
was 6.5 m which lead to an average required baseline of 1.16 m. The second over-
flight was at a higher altitude of approximately 10.15 m requiring a slightly higher
minimum baseline of 2.28 m on average. From all frames, were at least a part of the
safe landing zone on top of the building was visible in the disparity images, we could
successfully identify a valid landing target in over 90 % for both flights.
