Software Framework for Multimodal Fusion in Ubiquitous Computing Applications

As users tend to bypass inconvenient security measures, access control systems can be made more user-friendly and secure by varying verification strength depending on current security risks; by using continuous unobtrusive verification with whatever modalities are available and requesting explicit verification only if unobtrusive one fails; by allowing users to choose among available biometric and non-biometric modalities. Recognition accuracy and security can be also improved by using person-dependent and environment-dependent fusion models and classifier ensembles. This work presents software framework for multimodal fusion that provides the above-listed capabilities, and the experiments on the data of 150 users and four biometric modalities, confirming the feasibility of the proposed approach. The experimental results have shown that the proposed framework is capable of providing verification of different strengths while reducing significantly the need in explicit verification, allowing users to choose among available modalities and ensuring fairly low False Rejection Rates despite fairly high error rates of individual biometric modalities. For example, for a security requirement of False Acceptance Rate not greater than 0.2% it is possible to verify users unobtrusively in 72% of cases, while the overall False Rejection Rate could be as low as 0.1% if data of certain modalities is available within short time interval.