Multi-camera invariant appearance modeling for non-rigid object identification in a real-time environment

摘要

Surveillance of wide areas requires a system of multiple cameras to keep observing people, the non-rigid objects. In such a multiple view system, the appearance of people obtained in one camera is usually different from the appearance obtained in other cameras. In order to correctly identify people, the unique appearance model of each specific object should be invariant to such changes. Unlike previous methods building an appearance model by using only single camera, our appearance modeling, in this paper, is based on the multi-camera environment to fit real cases. Our appearance model is represented by two hierarchical tree structures that are responsible for color and texture information, respectively, where each layer of a tree is maintained by a Gaussian mixture model (GMM). The identification process is performed with a delicate voting scheme without complicated computations to meet the requirements of real-time applications. Experimental results show that our unique appearance model is robust to translation, rotation, scaling, and shape variations. Furthermore, it is equipped with automatic model updating, and it achieves a high precision rate and high processing performance.