Page 89 - Autonomous Mobile Robots

72 Autonomous Mobile Robots

where c is an over-subtraction factor (c ≥ 1) and d is spectral floor parameter
(0 < d < 1). The values of c and d are empirically determined for
obtaining an optimal noise subtraction level at all ranges and set to be 4
and 0.001.
Althoughareducednoiserangebincanbeusefulinotherdetectionmethods,
the target presence probability estimate (Equation [2.20]), will be demon-
strated further in the results. This method shows improved performance over
CFAR methods as the threshold can be applied on the target presence probab-
ility instead of SNP. Setting an arbitrary threshold value on the probability of
target presence (≥0.8) is sufficient for target detection. Based on the results,
this is a robust method and requires no adjustments when used in different
environments.

2.6.1 Target Presence Probability Results

The results of the proposed target detection algorithm are shown in Figure 2.17
where a noisy RADAR range bin (Figure 2.17a), the corresponding estimated
target presence probability (Figure 2.17b) from Equation (2.20) and the reduced
noise range spectra (Figure 2.17c) have been plotted. In Figure 2.17a, the range
bin contains three distinct peaks of differing power values, whereas the target
presence probability plot shows the three peaks with a more uniform range
width and similar probabilistic values. This result shows that although the return
power values varies from different objects, the corresponding target presence
probability values will be similar.
The target presence probability-based feature detector is easier to interpret
as shown in Figure 2.18 and Figure 2.19 where the target presence probability
plot is shown along with the corresponding raw RADAR data. Figure 2.18a and
Figure 2.19a show the raw RADAR data obtained in an indoor sports hall and
outdoor sports field, respectively. The corresponding target presence probab-
ilities are shown in Figure 2.18b and Figure 2.19b, respectively. Figure 2.18b
shows the target presence probability plot of an indoor stadium. The four walls
of the stadium are clearly obtained by the proposed algorithm. The other prob-
ability values at higher ranges arise due to the multipath effects in the RADAR
range spectrum. Figure 2.19b is obtained from an outdoor field. The detec-
ted features are marked in the figure. The clutter shown in Figure 2.19b is
obtained when the RADAR beam hits the ground due the unevenness of the
field surface.
The merit of the proposed algorithm is shown in Figure 2.20 where plots
obtained using different power thresholds applied to raw RADAR range spectra
are shown and compared with the threshold (0.8) applied to the probability plot.
Figure 2.20a shows the comparison of 2D plots obtained by choosing a constant
threshold of 25 dB applied to the raw RADAR data and the target presence prob-
ability plot. Figure 2.20b shows the comparison of plots obtained by constant

© 2006 by Taylor & Francis Group, LLC

FRANKL: “dk6033_c002” — 2006/3/31 — 17:29 — page 72 — #32