Evaluation of Sensor Calibration in a Biometric Person Recognition Framework based on Sensor Fusion

Bernhard Fröba, Constanze Rothe, Christian Küblbeck

Abstract

Biometric person authentication is a secure and user-friendly way to identify persons in a variety of everyday applications. In order to achieve high recognition rates, we propose an audio-visual person recognition system based on voice, lip motion and still image. The combination of these three data sources (called sensor fusion) may be performed in several ways. In this paper we compare different approaches to this problem. We present a method for a sensor normalization based on statistical properties which we call sensor calibration.The final fusion simplifies to a multiplication or addition of the outputs of each sensor.This approach is then compared to a traditional fusion method which handles fusion as a classification problem. These two methods are compared with the voting based fusion system presented in an earlier version of our system.