Phantom: Prediction of Human Motion with Distributed Body Sensors - Prediction and Classification of Respiratory Motion - page 51

the normalized root mean squared error (NRMSE) and the prediction overshoots,
in Sects. 3.5.4 and 3.5.5 , respectively. We also show CPU time used for the
computational time in Sect. 3.5.6 .

3.5.1 Motion Data

We have used three kinds of motion data, i.e., chest motion, head motion, and
upper body motion. Motion data was collected using a Polhemus Liberty AC
magnetic tracker in Fig. 3.7 , operating at 240 Hz for approximately 20 s (4,800
sample dataset) [ 54 ]. Eight sensors were attached on the target motion surface with
the magnetic source rigidly mounted approximately 25.4 cm from the sensors.
Each motion data was randomly selected based on the motion speed for Monte
Carlo analysis with three sets of motion data—the first datasets for slow motion,
the second datasets for moderate, and the rest for the violent motion. For the target
estimation, the experimental tests have been conducted based on repeated random
sampling to compute their results for Monte Carlo analysis. Each of the datasets
was taken with great care to limit target movement to the type based on Table 3.2 .

3.5.2 Collaborative Grouping Initialization

Before the efficient target tracking, the proposed collaborative method needs to
make the grouping for distributed sensory data. The objective of this Chapter is to
find out the optimal group number with an adaptive hyper-parameter. First, we
need to find out the initial hyper-parameter (b y ) in (1) below and then calculate the
group number (G) based on the adaptive (ADT) posterior probability p ADT (y | z j ).
In (2) and (3) is compared the difference (D) of the consecutive log-likelihood
functions between non-collaborative grouping method described in Sect. 3.2.2 and
collaborative grouping method described in Sect. 3.3.1 .

(1) Calculation of Hyper-parameter (b y )

The objective of this Chapter is to calculate the initial hyper-parameter (b y )with
potential group numbers. To find out the hyper-parameter, we iterate expectation
and maximization steps with sample training data (approximately 2,400 sample

Fig. 3.7 Polhemus liberty
AC magnetic tracker

System Electronics Units
with 8 and 16 sensor
channels

Long Range Source

Sources

Body Sensors