A Bayesian Hierarchical Framework for Multitarget Labeling and Correspondence With Ghost Suppression Over Multicamera Surveillance System

Abstract

In this paper, the main purpose is to locate, label, and correspond multiple targets with the capability of ghost suppression over a multicamera surveillance system. In practice, the challenges come from the unknown target number, the interocclusion among targets, and the ghost effect caused by geometric ambiguity. Instead of directly corresponding objects among different camera views, the proposed framework adopts a fusion-inference strategy. In the fusion stage, we formulate a posterior distribution to indicate the likelihood of having some moving targets at certain ground locations. Based on this distribution, a systematic approach is proposed to construct a rough scene model of the moving targets. In the inference stage, the scene model is inputted into a proposed Bayesian hierarchical detection framework, where the target labeling, target correspondence, and ghost removal are regarded as a unified optimization problem subject to 3-D scene priors, target priors, and foreground detection results. Moreover, some target priors, such as target height, target width, and the labeling results are iteratively refined based on an expectation-maximization (EM) mechanism to further boost system performance. Experiments over real videos verify that the proposed system can systematically determine the target number, efficiently label moving targets, precisely locate their 3-D locations, and effectively tackle the ghost problem. Note to Practitioners-As cooperative multicamera surveillance networks become more and more popular, the demand of multicamera information fusion and user-friendly representations becomes crucial. Motivated by the demand, this paper demonstrates a new system to efficiently integrate, summarize, and infer video messages from multiple client cameras. The ultimate goal is to provide a global view of the surveillance zone so that the managers in the control room may monitor the scene in an easier way. The main functions of the proposed system include the fusion of detection results from many client cameras, the summarization of consistent messages, and the inference of the target movement in the 3-D scene. In the near future, we will also take into consideration more information, like temporal clues, photometric clues, and object-level clues, in order to perform advanced scene analyses like abnormal behavior detection.