Improved background subtraction based on word consensus models

Abstract

The motion detection approach plays a crucial role in the intelligent video surveillance technology. A universal background subtraction algorithm called PAWCS (Pixel-based Adaptive Word Consensus Segmenter), based on word consensus models, is proven that it performs better in video motion detection recently. In this paper, we present an algorithm to improve the robustness of PAWCS. Specifically, the background models' update can be inhibited when the pixels locate in the edge of foreground objects. Then, the bi-updating approach is used in the models updating strategy, and the persistence of the word will be updated according to their matching accuracy. Finally, the experiments' results demonstrate the effectiveness of our method.