Multi-agent based framework for person re-identification in video surveillance

Multi-agent based systems have been used in a number of practical application domains. However their use in computer vision based systems that often provide solutions to automated video surveillance, remains in its infancy. Addressing this gap in research this paper proposes a novel design, based on multi-agents, to address one of the most important open research problems in video surveillance, i.e. person re-identification. The re-design of a typical computer vision based solution for the problem is based on our analysis of the problem considering how a human observer would successfully carry out person re-identification. Hence the proposed approach mimics human behavior and promises many advantages over the existing approaches that do not consider such a human behavior based approach. We provide preliminary experimental results to justify the contribution of the proposed novel approach to person re-identification and conclude the paper with insights to future research that has potential for a new paradigm of research in person re-identification in video surveillance.