Inference for Identity Management
Abstract
A computational framework has been developed to carry out identity management, that is, the automatic inference of the identities of targets tracked by surveillance systems that cover wide areas such as a shopping mall or a large harbor. People or vehicles may remain invisible to the system for long periods of time as they move between sensors. Identity management attempts to infer from uncertain measurements who or what is where at all times. The following work was performed in this short-term project: Fleshed out and streamlined the mathematical framework for identity management. This required significant changes at the core of the framework, and several of the ideas built on top of this had to be adapted or reinvented as well, prompting a systematic reformulation of the mathematics. Studied and tested algorithms from the literature to be used, either directly or in modified form, in the core inference engine of an identity management system. Developed a computationally efficient method for finding high-likelihood identity assignments given a graph of association probabilities between sensor observations. This method efficiently solves the batch version of the main estimation problem underlying identity management.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 16, 2010
- Accession Number
- ADA535031
Entities
People
- Carlo Tomasi
Organizations
- Duke University