Probabilistic fusion of gait features for biometric verification

This paper examines the fusion of various gait metrics in a probabilistic framework. Using three gait modalities we describe a process for determining probabilistic match scores using intra and inter-class variance models together with Bayes rule. We then propose to fuse these modalities based on established fusion rules with weights determined in a principled manner. Using a large publicly available database we show improvements through fusion, both in terms of verification accuracy and class separation; we also consider how the accuracy of each modality and the correlation between the modalities affects overall performance.