People identification for domestic non-overlapping RGB-D camera networks
Boris Takač, Andreu Catalá, Matthias Rauterberg, Wei Ren Chen · 2014
The ability to identify the specific person in a home camera network is very relevant for healthcare applications where humans need to be observed daily in their living environment. The appearance based people identification in a domestic environment has many similarities with the problem of re-identification in public surveillance systems, but there are also some additional beneficial and constraining factors (e.g., less people, non-pedestrian behaviour, unusual camera viewpoints). In this paper, we are considering the problem of people identification in a small home RGB-D camera network, for which we have developed a method based on appearance learning and classification using a combination of SVM and the Naive Bayes classifier. The method is evaluated using the prototype of a real-time multiple camera system on a 16 people dataset.