REGION-BASED MIXTURE OF GAUSSIANS MODELLING FOR FOREGROUND DETECTION IN DYNAMIC SCENES

 

Abstract

One of the most widely used techniques in computer vision for foreground detection is to model each background pixel as a Mixture of Gaussians (MoG). While this is effective for a static camera with a fixed or a slowly varying background, it fails to handle any fast, dynamic movement in the background. In this paper, we propose a generalised framework, called region-based MoG (RMoG), that takes into consideration neighbouring pixels while generating the model of the observed scene. The model equations are derived from expectation maximisation theory for batch mode, and stochastic approximation is used for online mode updates. We evaluate our region-based approach against ten sequences containing dynamic backgrounds, and show that the region-based approach provides a performance improvement over the traditional single pixel MoG. Comparison with four state-of-the art approaches shows that RMoG outperforms the others in reducing false positives whilst still maintaining reasonable foreground definition. Lastly, using the ChangeDetection (CDNet 2014) benchmark, we evaluated RMoG against numerous surveillance scenes and found it to be amongst the leading performers for dynamic background scenes, whilst providing comparable performance for other commonly occurring surveillance scenes.


Consider a sequence of n images of size 8×8 pixels. Each image can viewed as made up of two areas of different pixel intensities, dark (~70) and light (~180) . At time instant n+1, there is a small movement in the scene that causes the edge to shift down by one row, Fig. (a). Now, let us look at the central pixel in the 3×3 region given by the red box. In the case of standard mixture modelling, this pixel will be considered a foreground pixel as the model at that location is learnt solely from data due to the light area, see Fig. (b) Left. However, in the case of region modelling, this pixel will still be considered a background pixel as the model in the 3×3 region are learnt from data in both the light and dark areas, see Fig. (b) Right. Hence, the pixel will be able to be classified as belonging to one of the mixtures from the top row in the 3×3 region.

 

Results

 Quantitative Metrics for RMoG algorithm for each category in CDNet 2014. More results from the benchmark can be found here

 

Citation

Region-based Mixture of Gaussians modelling for foreground detection in dynamic scenes

Sriram Varadarajan, Paul Miller, Huiyu Zhou

Pattern Recognition, Volume 48, Issue 11, November 2015, Pages 3488-3503

 

bibtex:

@article{Varadarajan20153488,
title = "Region-based Mixture of Gaussians modelling for foreground detection in dynamic scenes",
journal = "Pattern Recognition ",
volume = "48",
number = "11",
pages = "3488 - 3503",
year = "2015",
doi = "http://dx.doi.org/10.1016/j.patcog.2015.04.016",
url = "http://www.sciencedirect.com/science/article/pii/S0031320315001557",
author = "Sriram Varadarajan and Paul Miller and Huiyu Zhou",
}

 

Videos

 

The RMoG algorithm is able to handle extrinsic disturbances in the scene caused by the camera jitter

 The RMoG algorithm is able to handle dynamic variations in the background such as ripples in the water and the waving tree.