Authors: Kalyan Veeramachaneni, Lisa Osadciw
Addresses: Department of Electrical Engineering and Computer Science, Syracuse University, Syracuse, NY 13244, USA. ' Department of Electrical Engineering and Computer Science, Syracuse University, Syracuse, NY 13244, USA
Abstract: In this paper an adaptive sensor management algorithm is presented for a biometric sensor network. A distributed detection framework is adapted for varying security requirements in the network, by considering the trade-offs between accuracy and time. Accuracy and time are tied into a single weighted objective function and a particle swarm optimisation algorithm is designed to achieve best possible configurations for a given set of weights. Results are presented for different weights applied to the bi-objective problem. A Bayesian framework is proposed for estimating the a priori of the imposter in real time. This determines the security requirements of the network. The estimation uses the observations collected from the sensors for different individuals accessing the network via the distributed detection framework. The distributed detection framework is redesigned for the new updated a priori, resulting in a closed loop control of a biometric sensor network. Results show that the new adaptive sensor management algorithm leads to lower false acceptance and false rejection rates when compared to a network without the adaptive algorithm.
Keywords: biometrics; biometric sensor networks; network management; distributed detection; PSO; particle swarm optimisation; Bayesian learning; adaptive sensor management; distributed detection; network security; false acceptance; false rejection.
International Journal of Biometrics, 2009 Vol.1 No.4, pp.495 - 524
Published online: 19 Jul 2009 *Full-text access for editors Access for subscribers Purchase this article Comment on this article