臺灣博碩士論文加值系統

本論文的重點是移動物體偵測的問題。它不僅偵測移動物體，而且也能偵測出移動物體的陰影區域。一般來說，最常見和最基本用來偵測移動物體的方法是背景相減法。傳統的背景相減方法是在背景為靜止的假設下進行的。然而，它並不適用於動態背景，其背景圖像隨著時間變化。在這篇論文中，我們提出了一個具有適應性和參考附近區域的高斯混合模型來為每一個像素建立背景模型。我們修改了原始的高斯混合模型(GMM)變成附近區塊高斯混合模型(LPGMM)。因此，LPGMM是用來解決在動態背景下移動物體的偵測問題。大部分在動態背景下的背景相減方法不會考慮陰影的問題。由於物體的陰影是隨著移動物體來移動，所以它是很難在動態背景下去區分移動物體和移動陰影的區域。我們使用支持向量機(SVM)來偵測在動態背景環境中的陰影區域。

This thesis is focused on the problem of moving object detection. It is not only to detect the moving object but also to detect the object shadow areas. Generally speaking, the most common and basic approach to detect the moving object is background subtraction. Traditional background subtraction methods work under the assumption that the background is stationary. However, it is not applicable to dynamic background, whose background image changes over time. In this thesis, we propose an adaptive and local mixture-of-Gaussians model for each pixel to build the background model. We modify the original Gaussian Mixture Model (GMM) to the Local-Patch Gaussian Mixture Model (LPGMM). Thus, the LPGMM is utilized to solve the problem of detecting the moving object under dynamic background. Most traditional background subtraction methods in dynamic background do not consider the problem of cast shadow. Since the object shadow moves with the moving object, it is difficult to differentiate moving shadow from moving objects under dynamic background. We use the support vector machine (SVM) to detect cast shadow areas under the dynamic background environment.

List of Figure…………………………………………………………………………ii
List of Tables…………………………………………………………………………iii

1 Introduction……………………………………………………………………..1
2 Previous Work…………………………………………………………………..4
3 Proposed Method………………………………………………………………..7
3.1 MVGMM Background Modeling……………………………...……...9
3.1.1 Original [1] GMM Background Modeling…………………….....9
3.1.2 Modifying GMM to Multi-Variable GMM (MVGMM)………..11
3.2 Cast Shadow Removal by Using SVM……………………...……….15
3.2.1 Introduction of Support Vector Machine (SVM)……………….15
3.2.2 Feature Extraction of SVM Training……………………………18
3.2.2.1 Features of Color information……………………………..19
3.2.2.2 Features of Texture information…………………………...22
3.2.2.3 Features of other information……………………………...23
4 Experimental Results………………………………………………………….27
4.1 The First Stage-MVGMM segmentation…………………………….27
4.1.1 The Qualitative Results…………………………………………28
4.1.2 The Quantitative Results………………………………………..31
4.2 The Second Stage-SVM Shadow Detection………………………….33
4.2.1 The Qualitative Results…………………………………………35
4.2.2 The Quantitative Results………………………………………..41
4.3 Other Results…………………………………………………………43
4.3.1 Results under different parameters……………………………...43
4.3.2 Results from stationary background videos…………………….47
5 Conclusion……………………………………………………………………...52
6 References……………………………………………………………………...53

[1] C. Stauffer and C. Grimson, “Learning patterns of activity using real-time tracking,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 22, pp. 747–757, August 2000.
[2] Huang, S.-S., “Region-Level Motion-Based Background Modeling and Subtraction Using MRFs,” IEEE TRAN`SACTIONS ON IMAGE PROCESSING, vol. 16, NO. 5, MAY 2007.
[3] N. Martel-Brisson and A. Zaccarin, “Kernel-based learning of cast shadows from a physical model of light sources and surfaces for low-level segmentation,” Computer Video and Pattern Recognition (CVPR) conference, in June 2008.
[4] Liu, Z. and Huang, K., “Cast Shadow Removal Combining Local and Global Features,” Computer Video and Pattern Recognition (CVPR) conference, in June 2007.
[5] Zeng, H.-C. and Lai, S.-H., “Adaptive foreground object extraction for real-time video surveillance with lighting variations,” IEEE Conference on Acoustics, Speech, Signal Processing, 2007.
[6] Li, L., Huang, W., Gu, I.-Y.H. and Tian, Q., “Foreground object detection from videos containing complex background,” ACM international conference on Multimedia, 2003.
[7] Y. Sheikh and M. Shah, “Bayesian modeling of dynamic scenes for object detection,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 27, no. 11, November 2005.
[8] Zhang, S., Yao, H. and Liu, S., “Spatial-temporal nonparametric background subtraction in dynamic scenes,” IEEE international conference on Multimedia and Expo, 2009.
[9] Cheng, L. and Gong, M., “Realtime background subtraction from dynamic scenes,” IEEE international Conference on Computer Vision, 2009.
[10] J. Sun, W. Zhang, X. Tang, and H. Shum, “Background Cut,” Proc. Ninth European Conf. Computer Vision, 2006.
[11] L. Li and M. Leung, “Integrating Intensity and Texture Differences for Robust Change Detection,” IEEE Trans. Image Processing, vol. 11, no. 2, pp. 105-112, Feb. 2002.
[12] O. Javed, K. Shafique, and M. Shan, “A Hierarchical Approach to Robust Background Subtraction Using Color and Gradient Information,” Proc. Workshop Motion and Video Computing, 2002.
[13] S. Jabri, Z. Duric, H. Wechsler, and A. Rosenfeld, “Detection and Location of People in Video Images Using Adaptive Fusion of Color and Edge Information,” Proc. 15th Int’l Conf. Pattern Recognition, 2000.
[14] Z. Zivkovic, and F. van der Heijden, “Recursive unsupervised learning of finite mixture models,” IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol.26, No.5, pp 651–656, May 2004.
[15] C.-C. Chang and C.-J. Lin. LIBSVM: a library for support vector machines, 2001. Software available at http://www.csie.ntu.edu.tw/~cjlin/libsvm.
[16] A. Bhattacharyya, “On a measure of divergence between two statistical populations defined by their probability distributions,” Bulletin of Calcutta Mathematical Society, vol. 35, pp. 99–110, 1943.
[17] Joshi, A.J., Papanikolopoulos, N.P.: Learning to detect moving shadows in dynamic environments. IEEE Trans. Pattern Analysis and Machine Intelligence 30(11), 2055–2063 (2008)