Multi-Modal Sensor Fusion Utilizing Statistical Dependence and Compressive Sampling

ABSTRACT: Our goal is to develop theoretical frameworks for efficient multisensor fusion of high dimensional data for target detection localization and tracking. We plan to develop novel algorithms based upon our previous research on copula theory to perform inference with multi-modal correlated sensor data. Copulas describes the dependence between random variables and allow one to optimally exploit the inherent low-dimensional characteristics of high dimensional data. They are popular in high-dimensional statistical applications as they allow one to easily model and estimate the distribution of random vectors by estimating marginals and copulae separately. There are many parametric copula families available that capture more information than traditional approaches while still maintaining low data communication rates to the central fusion center. We will evaluate achievable performance limits of the developed algorithms and compare the efficiency to several benchmark algorithms. Testing will be performed on physics based simulated data sets. The benchmark techniques we will consider will include conventional fusion of Kalman state spaces, principal component analysis (PCA) based methods and parametric likelihood-based approaches. The research output is expected to have significant implications in coping with the data deluge problem at individual sensors employed for detection, estimation and tracking in sensor and radar networks. ; BENEFIT: The proposed approach will permit improved sensor fusion of heterogeneous data sets with only a small increase in communication requirements over traditional fusion approaches. The technology benefits Air Force ISR and missile defense systems. Commercial applications include aviation radar systems as well as emerging multisensor systems.