Grant Number: N000140710625
Funding Agency: ONR
PI: Mauro Maggioni
Additional Researchers: Yoon Mo
Effective Dates: 2007/04-2010/06
Amount: $254,943
Description: The goal of multimodal sensor fusion is to integrate the data acquired from the different sensors in order to obtain more precise information about a target. The different modalities can be leveraged in order to obtain more accurate information, and allow us to achieve greater specificity than any single sensor. Potential modalities include: visual band, near and far IR, microwave, radar and sonar. Our research will focus on three related aspects, each of which is important and of independent interest: Geometric properties of data concentrated along low-dimensional sets in high-dimensional space. Key issues: definition of local similarities, dimensionality reduction, parametrizations of the data, stability with respect to perturbation of the data (e.g. measurement noise, instrument normalization etc...). Properties of functions on the data. Functions of interest, or in any case functions that can be learnt, can be modeled as having certain smoothness properties with respect to the geometry of the data (but not necessarily of the geometry of the high-dimensional embedding space!). Key issues: modeling such functions and designing efficient approximation schemes, stability with respect to noise in the underlying data as well as in the function itself. Manifold matching and sensor fusion. The problem of matching two manifolds is related to the problem of parametrizing the manifolds and then viewing one manifold parametrization as a set of functions on a second manifold..