Phantom: Prediction of Human Motion with Distributed Body Sensors - Prediction and Classification of Respiratory Motion - page 39

dynamic models operate in a parallel way with Markov switch probability [ 16 ].
The proposed solution is to employ a tracking relationship among distributed body
sensors by adding switching probability for multiple models and grouping method
to figure out the interactive relation within the sensors. IMME algorithm can be
used to combine different filter models with improved control of filter divergence.
As a suboptimal hybrid filter [ 16 - 18 ], IMME makes the overall filter recursive by
modifying the initial state vector and covariance of each filter through a proba-
bility weighted mixing of all the model states and probabilities [ 1 , 19 - 26 ].
The overall contribution of this research is to minimize the prediction overshoot
originating from the initialization process by newly proposed Multi-channel
IMME (MC-IMME) algorithm with the interactive tracking estimation. MC-
IMME can estimate the object location as well as the velocity from measured
datasets using multiple sensory channels. For this MC-IMME, we have extended
the IMME to improve overall performance by adding switching probability to
represent the conditional transition probability and a collaborative grouping
method to select a proper group number based on the given dataset. The technical
contributions of this study are twofold: First, we propose a cluster number
selection method for distributed body sensors based on Dirichlet hyper-prior on the
cluster assignment probabilities. Second, we present a new prediction method to
reduce the initial estimate error by employing a tracking relationship among dis-
tributed sensory data. For the performance improvement, we added switching
probability to represent the conditional transition probability from a previous
channel state to a current channel state and a collaborative transition probability to
select a proper group number based on the given datasets.
This Chapter is organized as follows. In Sect. 3.2 , the theoretical background
for the proposed algorithm is briefly discussed. In Sects. 3.3 and 3.4 , the proposed
grouping criteria with distributed sensors placement based on EM algorithm and
the proposed estimate system design for distributed body sensors are presented in
detail, respectively. Section 3.5 presents and discusses experimental results of
proposed methods—grouping methods and adaptive filter design. A summary of
the performance of the proposed method is presented in Sect. 3.6 .
3.2 Related Work
3.2.1 Kalman Filter
The Kalman filter (KF) provides a general solution to the recursive minimized
mean square estimation problem within the class of linear estimators [ 43 , 44 ]. Use
of the Kalman filter will minimize the mean squared error as long as the target
dynamics and the measurement noise are accurately modeled. Consider a discrete-
time linear dynamic system with additive white Gaussian noise that models
unpredictable disturbances. The problem formulation of dynamic and the mea-
surement equation are as follows,