Background Subtraction Based on Perception-Contained Piecewise Memorizing Framework

keywords: Long-term background memory, piecewise stationary test, Gaussian mixture model, background subtraction, foreground detection
A key issue for full-time video surveillance is to search or establish a reference image of background which corresponds to current video frame. However, background that was ever in presence long time ago is enclosed or discarded due to background forgetting assumption. How to rapidly pick up or even rebuild long-term background needs to be discussed. This paper aims to present a framework for background maintenance in order to solve the problem. A piecewise memorizing framework is proposed for matching, updating and even rebuilding long-term background. Based on the metaphors of psychological selective attention theory, the framework is composed of a prior piecewise perception processor for intensity stationary test. Besides, a hierarchical memorizing mechanism constitutes the other part of the framework for overcoming the exponential forgetting of long period background appearances. Applied to Gaussian mixture model (GMM), this framework is capable of maintaining short-term background states, identifying long period background appearances, and rapidly adjusting to new background states according to different expressions derived from the prior perception of scene intensity changes. Its effectiveness can be demonstrated by experimental results for solving various typical problems.
mathematics subject classification 2000: 68-T45
reference: Vol. 37, 2018, No. 4, pp. 865–893