Multi-module switching and fusion for robust video surveillance

In this paper, we address two of the common faults of indoor background modeling, namely the light switch and the bootstrapping problems. Light switch concerns sudden changes in lighting conditions that cause the failure of a background model of the scene. Bootstrapping problems occur when a training sequence free of moving objects is not available for model building. Our study investigates how rearrangements in the structure of multi-modular vision systems can improve the system performance in a changing environment. In other words, we want to introduce in the system the capability to select the most reliable method for extracting useful information among those available, and to exclude inadequate modules from the flow of signal analysis.