Compressive Background Modeling for Foreground Extraction

Abstract

Robust and efficient foreground extraction is a crucial topic in many computer vision applications. In this paper, we propose an accurate and computationally efficient background subtraction method. The key idea is to reduce the data dimensionality of image frame based on compressive sensing and in the meanwhile apply sparse representation to build the current background by a set of preceding background images. According to greedy iterative optimization, the background image and background subtracted image can be recovered by using a few compressive measurements. The proposed method is validated through multiple challenging video sequences. Experimental results demonstrate the fact that the performance of our approach is comparable to those of existing classical background subtraction techniques.

A block-based background model for moving object detection

“…7, the qualitative and quantitative results for one test frame of each video sequence obtained by each method are represented in Table 1. We can observe that, for almost every test sequence, the generated binary image of our proposed approach is more precise than those obtained by the SDE [17], MDPS [18], ISBS [19] and CBMFE [27] methods. The proposed method attains the highest quantitative and qualitative values such as the accuracy rate is over 72% for Similarity measures and over than 82% F-measure values in the almost sequences.…”

“…Those sequences contain different types of objects such as vehicles, human bodies and others, moving or waiting in the scene. We compare our results with four other background subtraction methods: (1) SDE [17]; (2) MDPS [18]; (3) ISBS [19]; and (4) CBMFE [27]. We implement the four methods, using the default values of all the parameters given in each paper.…”

“…It represent binary image of the moving objects thus the original frames and the ground truth frames. These results were generated by SDE [17], MDPS [18]; ISBS [19], (4) CBMFE [27] and PRO methods. For almost every video sequence, we can observe that the generated binary objects mask of the our approach is more precise than those obtained the other methods.…”

“…Different methods are explored in this context. Most of them are classified as background subtraction methods [8][9][10][11][12], [14][15][16][17][18][19][20][21][22][23][24][25][26][27][28][29]. Where, the Background Modeling is the principal step in these different methods.…”

“…The second approach is the calculation of the optical flow [3], [4] which provides all information about the movement, but the real-time implementation is difficult and calculation of flow being generally slow. The background subtraction approaches start by modeling the background(model of static scene), it can detect the most precise foreground [5][6] [ [24][25][26][27][28][29][30][31], , but it has many limitations like sensor noise (noise of acquisition and digitization) and management of homogeneous areas when the luminance difference between two moments is less than a threshold.…”