視訊監控是一門需要應用於全天候的課題，因此探討不同天氣時的監控情形則有其必要性。而背景模型之建立則是智慧型監控系統的主要關鍵技術，其中又以高斯混合模型具有最為廣泛之應用。本研究將雨天視訊監控分為兩部分進行探討，其一是應用非負矩陣分解法(NMF)將畫面中的雨痕去除，其二則是改良高斯混合模型(GMM)對雨中緩慢移動物體的偵測。由於傳統的GMM在偵測雨中的移動物體時容易將雨誤判為移動物，使得訓練出的背景模型不穩定，因此在偵測之前必須先去除雨的影響。本研究提出一種利用NMF去除影像中的雨的方法。由於雨通常出現在中頻的部分，因此先將含有雨的影像經由高斯濾波器分為低頻和高頻，從低頻的部分使用NMF的方法重建不含雨資訊的低頻影像；在高頻部分則利用Canny 邊緣偵測的概念以及區塊複製的技巧將雨排除。解決雨的問題以後，下雨時的移動物如行人，移動的速度可能會由於雨天而加快或變慢，移動加速的部分對於傳統的GMM來說並非難題；但緩慢移動的物體則有可能因為移動太慢而被更新為背景。因此本研究提出一種利用誤差函數(ERF)的機制來調整GMM的權重與標準差，可有效的解決緩慢的移動物體被誤判為背景的問題，加強對雨天環境視訊監控的可靠性。

Video surveillance is a study that needs to consider every environmental condition. Therefore, the surveillance condition under different weathers is important topic within the field. The background modeling plays a key role of event detection in intelligent surveillance systems. Gaussian Mixture Model (GMM) is the most popular background modeling method in latest surveillance systems. This thesis study surveillance in raining condition in two parts. The first is adapting the non-negative matrix factorization (NMF) to remove the raining effect in the image, and the second is to improve the Gaussian Mixture Model to increase the accuracy of slow moving object under raining condition. Due to the fact that utilizing original GMM to detect the moving objects in the rain result in misjudging the rain as moving objects, Rain streak in any given image should be removed before detection. Therefore, the rain component, which is usually in the middle frequency, could be discarded in high and low frequency domains. In this thesis, the rain image is decomposed into a low-frequency part and a high-frequency part by a Gaussian filter. Then the NMF method is applied to deal with the rain streak in the low frequency; while in the high frequency part, the concept of Canny edge detection and block copy strategy are utilized separately to remove the rain hidden in high frequency and improve the image quality. After the rain streak is removed, the moving pattern in the rain is studied. The moving objects like pedestrians in the rain could move faster or slower. The faster moving object easy to filter by the traditional GMM; However, the model has some disadvantageous when the object moves slowly. In this thesis, we propose a mechanism which takes the advantage of Gaussian error function (ERF) to adjust the growths of each Gaussian's weights and variances, to solve the problem that traditional GMM misjudged with the slow moving object as background. The mechanism improves the GMM model on detecting the slow moving object accurately and enhance the robustness of surveillance systems in raining weathers.

中文審定書 i
英文審定書 ii
誌 謝 iii
摘 要 v
Abstract vi
List of Figures ix
Chapter 1 Introduction 1
1.1 Rain Streaks Removal 1
1.2 Background Modeling 3
1.3 Thesis Organization 4
Chapter 2 Previous Works 5
2.1 Sparse Representation 5
2.2 Non-negative Matrix Factorization (NMF) 7
2.3 Sparse Coding Image De-raining and Kang et al.’s Method [3] 10
2.4 Gaussian Mixture Model (GMM) 12
2.5 Gaussian Error Function 14
Chapter 3 Proposed methods 15
3.1 Rain removal based on NMF 15
3.2 Frequency Separation and de-raining with NMF 18
3.3 Edge Connectivity 22
3.4 Reconstruction of the non-rain image 25
3.5 Dark Channel Replacement 27
3.6 Sparse Representation Rain Removal 28
3.7 Conventional Weight Updating 32
3.8 Observation of Variance Updating 34
3.9 The Proposed Weight Updating 36
Chapter 4 Experimental Results 38
Chapter 5 Conclusions 48
REFERENCES 49

[1] K. Garg, and S. K. Nayar, “Vision and rain,” International Journal of Computer Vision, vol. 75, no. 1, pp. 3–27, 2007.
[2] P. C. Barnum, S. Narasimhan, and T. Kanade, “Analysis of rain and snow in frequency space,” International Journal of Computer Vision, vol. 86, no. 2–3, pp. 256–274, 2010.
[3] D.-A. Huang, L.-W. Kang, Y.-C. Wang, and C.-W. Lin, “Self-Learning Based Image Decomposition with Applications to Single Image Denoising,” IEEE Transactions on Multimedia, vol. 16, no. 1, pp. 83–93, 2014.
[4] C.-H. Yeh, L.-W. Kang, and C.-Y. Lin, “Image haze removal via haze density analysis based on pixel-based dark channel prior,” Proceedings of Computer Vision & Graphic Image Processing, 2012.
[5] J. Bossu, N. Hautière, and J.-P. Tarel, “Rain or snow detection in image sequences through use of a histogram of orientation of streaks,” International Journal of Computer Vision, vol. 93, no. 3, pp. 348–367, 2011.
[6] N. Brewer, and N. Liu, “Using the shape characteristics of rain to identify and remove rain from video,” Lecture Notes in Computer Science, vol. 5342, pp. 451–458, 2008.
[7] K. Garg, and S. K. Nayar, “Detection and removal of rain from videos,” Proceedings of IEEE Computer Society Conference on Computer Vision and Pattern Recognition, vol. 1, pp. 528–535, 2004.
[8] K. Garg, and S. K. Nayar, “When does a camera see rain?,” Proceedings of IEEE International Conference on Computer Vision, vol. 2, pp. 1067–1074, 2005.
[9] K. Garg, and S. K. Nayar, “Photorealistic rendering of rain streaks,” ACM Transactions on Graphics, vol. 25, no. 3, pp. 996–1002, 2006.
[10] X. Zhang, H. Li, Y. Qi, W. K. Leow, and T. K. Ng, “Rain removal in video by combining temporal and chromatic properties,” Proceedings of IEEE International Conference on Multimedia and Expo, pp. 461–464, 2006.
[11] C.-H. Yeh, L.-W. Kang, Y.-W. Chiou, C.-W. Lin, and S.-J. F. Jiang, “Self-learning-based post-processing for image/video deblocking via sparse representation,” Journal of Visual Communication and Image Representation, vol. 25, pp. 891-903, 2014.
[12] L.-W. Kang, B. Zhuang, C.-C. Hsu, C.-W. Lin, and C.-H. Yeh, “Self-learning-based low-quality single image super resolution,” Proceedings of IEEE International Workshop on Multimedia Signal Processing, 2013.
[13] L. Wixson, “Detecting Salient Motion by Accumulating Directionally-Consistent Flow,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 22, no. 8 pp. 774-780, 2000.
[14] C. R. Wren, A. Azarbayejani, T. Darrell, and A. P. Pentland, “Pfinder: Real-Time Tracking of the Human Body,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 19, no. 7, pp. 780-785, 1997.
[15] C. Stauffer, and W. E. L. Grimson, “Adaptive Background Mixture Models for Real-Time Tracking,” Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, pp. 246-252, 1999.
[16] C. Stauffer, and W. E. L. Grimson, “Learning Patterns of Activity Using Real-Time Tracking,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 22, no. 8, pp. 747-757, 2000.
[17] Z. Zivkovic, “Improved Adaptive Gaussian Mixture Model for Background Subtraction,” Proceedings of International Conference on Pattern Recognition, pp. 28-31, 2004.
[18] Y. C. Pati, R. Rezaiifar, and P. Krishnaprasad, “Orthogonal matching pursuit: Recursive function approximation with applications to wavelet decomposition,” Asilomar Conference on Signals, Systems and Computers, pp. 40-44, 1993.
[19] D. D. Lee, and H. S. Seung, “Learning the parts of objects by non-negative matrix factorization,” Nature, vol. 401, pp. 788-791, 1999.
[20] D. D. Lee, and H. S. Seung, “Algorithms for Non-negative Matrix Factorization,” Advances in neural information processing systems, 2001.
[21] J. Mairal, F. Bach, J. Ponce, and G. Sapiro, “Online dictionary learning for sparse coding,” Proceedings of International Conference on Machine Learning, pp. 689-696, 2009.
[22] J. Mairal, F. Bach, J. Ponce, and G. Sapiro, “Online learning for matrix factorization and sparse coding,” The Journal of Machine Learning Research, vol. 11, pp. 19-60, 2010.
[23] K. He, J. Sun, and X. Tang, “Single image haze removal using dark channel prior,” Pattern Analysis and Machine Intelligence, IEEE Transactions on, vol. 33, no. 12, pp. 2341-2353, 2011.
[24] G. Grindlay, “NMFlib v0.1.3,” 2010.