Background Subtraction via Regional Multi-Feature-Frequency Model in Complex Scenes

Abstract Background subtraction is one of the most popular techniques for the detection of moving objects. This paper presents a regional multi-feature-frequency (RMFF) method that utilizes the frequency contributions of regional features to detect changes. This approach allows the spatial relationship between pixels in a neighborhood and the frequencies of features over time to be considered, so that both spatial and temporal information is taken into account while constructing a model of an observed scene. Instead of using a global segmentation threshold, an adaptive strategy is utilized to dynamically adjust the foreground/background segmentation threshold for each region without user intervention. This adaptive threshold is defined for each region separately, and can adjust dynamically based on continuous monitoring of the background changes. The use of multi-scale superpixels for exploiting the structural information existing in real scenes also enhances robustness to noise and environmental variations. Experiments on the 2014 version of the ChangeDetection.net dataset demonstrated that the proposed method outperforms the twelve state-of-the-art algorithms in terms of overall F-Measure and worked effectively in many complex scenes.