A Multimodal People Recognition System for anIntelligent Environment

In this paper, a multimodal system for recognizing peoplein intelligent environments is presented. Users are identified and trackedby detecting and recognizing voices and faces through cameras and microphonesspread around the environment. This multimodal approachhas been chosen to develop a flexible and cheap though reliable system,implemented through consumer electronics. Voice features are extractedthrough a short time spectrum analysis, while face features are extractedusing the eigenfaces technique. The recognition task is achieved throughthe use of some Support Vector Machines, one per modality, that learnand classify the features of each person, while bindings between modalitiesare also learnt through a cross-anchoring learning rule based onthe mutual exclusivity selection principle. The system has been developedusing NMM, a middleware software capable of splitting the sensorsprocessing in several software nodes, making the system scalable in thenumber of cameras and microphones