Computer Vision for Micro Air Vehicles - Advances in Embedded Computer Vision - page 69

Chapter 4
Computer Vision for Micro Air Vehicles

Roland Brockers, Martin Humenberger, Yoshi Kuwata, Larry Matthies
and Stephan Weiss

Abstract Autonomous operation of small UAVs in cluttered environments requires
three important foundations: fast and accurate knowledge about position in the world
for control; obstacle detection and avoidance for safe flight; and all of this has to
be executed in real-time onboard the vehicle. This is a challenge for micro air vehi-
cles, since their limited payload demands small, lightweight, and low-power sensors
and processing units, favoring vision-based solutions that run on small embedded
computers equipped with smart phone-based processors. In the following chapter,
we present the JPL autonomous navigation framework for micro air vehicles to
address these challenges. Our approach enables power-up-and-go deployment in
highly cluttered environments without GPS, using information from an IMU and a
single downward-looking camera for pose estimation, and a forward-looking stereo
camera system for disparity-based obstacle detection and avoidance. As an exam-
ple of a high-level navigation task that builds on these autonomous capabilities, we
introduce our approach for autonomous landing on elevated flat surfaces, such as
rooftops, using only monocular vision inputs from the downward-looking camera.

4.1 Introduction
Miniature rotorcrafts are an ideal platform for exploration and reconnaissance
missions, since they can operate in highly cluttered environmentsforest, close to
the ground) or confined spaces (indoors, collapsed buildings, caves) and allow with

B

· · ·

R. Brockers ( S. Weiss

Jet Propulsion Laboratory, California Institute of Technology, Pasadena, USA
e-mail: [e-mail]
Y. K u w a t a
e-mail: [e-mail]
L. Matthies
e-mail: [e-mail]
S. Weiss
e-mail: [e-mail]
M. Humenberger
AIT Austrian Institute of Technology, Vienna, Austria
e-mail: [e-mail]

) Y. K u w a t a L. Matthies