In this paper, we describe a multi-modal Bayesian network for person recognition in a HRI context, combining information about a person's face, gender, age, and height estimates, with the time of interaction. We conduct an initial study with 14 participants over a four-week period to validate the system and learn the optimal weights for each of the metrics. Several normalisation methods are compared for different settings, such as learning from data, face recognition threshold and quality of the estimation. The results show that the proposed network improves the overall recognition rate by at least 1.4% comparing to person recognition based on face only in an open-set identification problem, and at least 4.4% in a closed-set.

Publication Date


Organisational Unit

School of Engineering, Computing and Mathematics


Person recognition, Bayesian network, multi-modal data fusion, soft biometrics, personalisation