Page 60 - Autonomous Mobile Robots

Millimeter Wave RADAR Power-Range Spectra Interpretation 43

techniques [24]. The results show the merit of the proposed algorithm which
can detect features in typically cluttered outdoor environments with a higher
success rate compared to other feature detection techniques. This work is a step
toward robust outdoor robot navigation with MMW-RADAR-based continuous
power spectra.
Millimeter wave RADAR can penetrate certain nonmetallic objects, mean-
ing that multiple line-of-sight objects can sometimes be detected, a property
which can be exploited in mobile robot navigation in outdoor unstructured
environments. This chapter describes a new approach in predicting RADAR
range bins which is essential for simultaneous localization and map building
(SLAM) with MMW RADAR.
The third contribution of this chapter is a SLAM formulation using an
augmented state vector which includes the normalized RADAR cross sections
(RCS) and absorption cross sections of features as well as the usual fea-
ture Cartesian coordinates. The term “normalized” is used as the actual RCS
is incorporated into a reflectivity parameter. Normalization results, as it is
assumed that the sum of this reflectivity parameter and the absorption and
transmittance parameters is unity. This is carried out to provide feature-rich rep-
resentations of the environment to significantly aid the data association process
in SLAM.
The final contribution is a predictive model of the range bins, from differ-
ing vehicle locations, for multiple line-of-sight targets. This forms a predicted
power–range observation, based on estimates of the augmented SLAM state.
The formulation of power returns from multiple objects down-range is derived
and predicted RADAR range spectra are compared with real spectra, recor-
ded outdoors. This prediction of power–range spectra is a step toward a full,
RADAR-based SLAM framework.
Section 2.2 summarizes related work, while Section 2.3 describes FMCW
RADAR operation and the noise affecting the range spectra, in order to under-
stand the noise distributions in both range and power. Section 2.4 describes how
power–range spectra can be predicted (predicted observations). This utilizes the
RADAR range equation and an experimental noise analysis. Section 2.5 ana-
lyzes a feature detector based on the CFAR detection method. The study also
shows ways to compensate for the inaccuracies of the power–range compens-
ating high-pass filter, contained in FMCW RADARs, and thereby improve the
feature detection process. A method for estimating the true range to objects from
power–range spectra is given in Section 2.6 in the form of a new robust feature
detection technique based on target presence probability. Section 2.6.1 shows
the merits of the target presence probability-based algorithm which can detect
ground level features with greater reliability than other feature detection tech-
niques such as constant threshold on raw RADAR data and CFAR techniques.
An augmented state vector is introduced in Section 2.8 where, along with the
vehicle and feature positions, normalized RCS and absorption cross sections of

© 2006 by Taylor & Francis Group, LLC

FRANKL: “dk6033_c002” — 2006/3/31 — 17:29 — page 43 — #3