Memory-based Gaussian Mixture Modeling for moving object detection in indoor scene with sudden partial changes

Abstract

In this paper, a memory-based Gaussian Mixture Model (MGMM) is proposed inspired by the way human perceives the environment. The human memory mechanism is introduced to model the background, which can make the model remember what the scene has ever been and help the model adapt to the variation of the scene more quickly. Experimental results show the effect of the memory mechanism in segmenting moving objects with sudden partial changes in the background scene.

Locally Statistical Dual-Mode Background Subtraction Approach

“…With other specific background scenarios, LSD can successfully detects moving objects, however, the pixel-wise accuracy is unremarkable. In the last experiment, we compare our proposed LSD background subtraction method with several state-of-theart approaches on the Wallflower dataset (i.e., Gaussian Mixture Model (GMM) [12], texture-contained Gaussian Mixture Model (TGMM) [15], Gaussian mixture shadow model (GMSM) [17], memorizing GMM (MGMM) [18], piecewise memorizing GMM (P-MGMM) [20], Lightness-Red-Green-Blue (BF-LRGB) [44]) and on the CDnet2014 dataset (i.e., Gaussian Mixture Model (GMM) [12], improved Gaussian Mixture Model (EGMM) [13], Region-based Mixture of Gaussian (RMoG) [21], Kernel Density Estimation (KDE) [27], Visual Background Subtractor (ViBE) [32], Spatially Coherent Self-Organizing Background Subtraction (SC_SOBS) [36], and Graph Cut algorithm (GraphCut) [39]).…”

“…• Based on (19), two decision thresholds of mean τ µ and standard deviation τ σ are figured out over (18).…”

Long-term background memory based on Gaussian mixture model

“…In order to demonstrate the effectiveness of P-MGMM to foreground detection under complex environments, we tested and compared P-MGMM with seven statistical or structural background subtraction methods (including ALPCA [6], LBP [3], GMM [1], TGMM [7], HGMM [8], GMSM [9] and MGMM [10]) using nine benchmark datasets from two databases (i.e., PETS2001 1 and Wallflower 2 ). Firstly, we select two benchmark datasets (i.e., PETS01-D3-T-C2 and PETS01-D1-T-C2) containing the problems of light switch (i.e., a kind of fast intensity change of background appearances) and ghosts (corresponding to a forgetting of long-term background appearances) for qualitative analysis 3 .…”

“…Prevailing methods are competent for rapid adaptation to the changes of background; yet, they are unequal to long-term background memory because of the exponential decay of background states. A memorizing GMM (MGMM) [10] is applied to tackling this problem. However, the memorizing states in MGMM are restricted during the learning period, for backgrounds are labeled excessively without any designed identification of the states representing long period background under complex environments.…”