A new general 3D object model is required in the literature of smart camera networks to facilitate future research. This paper presents a novel hierarchical and structural 3D model description which is well suited for both events detection and real-time free viewpoint surveillance. With this 3D model, sparse points are used to reconstruct objects. In addition, the state of the model is easy to track and estimate, which can be used to reduce time and computation when reconstructing the model. Further, data flow in the network is reduced to a level that smart cameras can afford. A concrete data structure of the model is described in this paper and its reconstruction method, the fusion method, is provided. Finally, experiment results show its feasibility, efficiency and effectiveness.