Parameterization of non-linear/non-gaussian data distributions for efficient information sharing in distributed sensor networks

• US 20080215298A1
• Filed: 10/10/2007
• Published: 09/04/2008
• Est. Priority Date: 10/10/2006
• Status: Active Grant

• Current Assignee: Frantorf Investments GmbH LLC (Intellectual Ventures LLC)
• Original Assignee: Frantorf Investments GmbH LLC (Intellectual Ventures LLC)
• Inventors: Haney, Philip J., Fiore, Paul D., Edelson, Geoffrey S.
• Granted Patent: US 7,698,108 B2
• Time in Patent Office: Days
• US Class Current: 702/189
• CPC Class Codes: G06F 18/251 of input or preprocessed data
• CPC Class Codes: G06T 7/277 involving stochastic approaches, e.g. using Kalman filters

Abstract

A method for parameterization of data distributions for efficient information sharing in distributed sensor networks including a plurality of sensors, comprising the steps of performing Bayesian multi-source data fusion and sharing probalistic data information.

Claims

1. A method for parameterization of data distributions for efficient information sharing in distributed sensor networks including a plurality of sensors, comprising the steps of:

performing Bayesian multi-source data fusion; and

sharing probalistic data information.

2. The method of claim 1 wherein target tracking in a decentralized sensor network is accomplished.

3. The method of claim 2 wherein target tracking is accomplished for application to a distributed sensor network sensor network that stores probalistic information.

4. The method of claim 3 wherein there are a plurality of particle filters and said particle filters are used to estimated posterior densities for each sensor.

5. The method of claim 4 wherein the particle filters provide consistent local posterior estimation.

6. The method of claim 5 wherein local posterior densities are parameterized.

7. The method of claim 6 wherein local posteriors are reconstructed.

8. The method of claim 7 wherein global posteriors are estimated.

9. The method of claim 1 wherein the data distribution is non-linear/non-Gaussian.

10. The method of claim 1 wherein the method is used for target tracking.

11. The method of claim 10 wherein there is a single decentralized data fusion (DDF) node and the method includes the steps of:

(a) collecting observations corresponding to a local image of a fusion environment;

(b) updating or initiating local target tracks using collected observations;

(c) communicating local target tracks throughout the sensor network;

(d) assimilating external target tracks received from other nodes; and

(e) propagating assimilated target to time of next expected measurement.

12. The method of clam 1, wherein the method makes us of a lambda distribution defined by the quantile function

$\begin{array}{c}Q\end{array}$

( u ) = { u λ

- ( 1 - u ) λ

, λ

≠

0 log 

( u ) ( 1 - u ) , λ

= 0 ( 3.3 

.1 )

for 0=u=1, the generalized lambda distribution parameterizes (3.3.1) as follows

$\begin{array}{cc}Q<br><br>\left(u\right)={\mathrm{\lambda <br><br>}}_1+\frac{u^{{\mathrm{\lambda <br><br>}}_3}-{\left(1-u\right)}^{{\mathrm{\lambda <br><br>}}_4}}{{\mathrm{\lambda <br><br>}}_2}& \left(3.3<br><br>\mathrm{.2}\right)\end{array}$

where λ

₁ acts as a location parameter, ζ

₂ acts as a scale parameter, and the combination of λ

₃ and λ

₄ jointly capture the shape of the empirical distribution.