多干擾環境下之自適應性視訊監控系統

A Self-adapted Video Surveillance System under Multi-interference Environment

陳松鴻(Sung-Hung Chen)

指導教授 ： 方志鵬

國立臺北科技大學/電資學院/電機工程系研究所/碩士(2013年)

摘要

在視覺化監控系統中，多數的演算法係先建立參考背景影像，再藉由背景相減法來取得前景影像，亦即移動目標物偵測。然而背景相減法的準確率卻極易因外界環境變化而受到嚴重的干擾；例如光源強度的改變、樹的搖晃、陰影的變化以及雨滴造成的紋路等，這些干擾都會影響系統分析與判斷的準確率。因此，有部分研究提出以統計的方式來建立背景影像，並將像素值出現機率較高者歸類為背景。然而，以此方法所建立的適應性背景仍無法解決因光源遭遮蔽而造成的陰影現像，使陰影仍會被當作前景般被擷取出來，造成判斷錯誤。此外在下雨時，除了會使視線能見度降低外，雨滴的紋路也會讓視訊畫面變的模糊，導致系統判斷錯誤。為了改善上述問題，本論文提出多干擾環境下之自適應性背景演算法。在所提出的演算法中，主要包含了三部分。首先為去雨演算法；我們觀察雨滴在一段影像中其色彩模型的數值變化，發現雨滴會使像素的亮度值增加。根據這個特性我們提出一套能將雨滴所造成紋路移除的演算法。第二部分為背景模型，在影像的同一位置可能會出現兩種以上的背景像素值，我們採用統計的方式對每一個像素計算出數個不同的背景，以建立該像素的背景模型。第三部分則是陰影偵測，我們根據物體遮蔽光源所產生的陰影不影響其色彩值的特性，提出能夠剔除陰影之演算法，擷取出正確無陰影之前景影像。考量視覺化監控系統具有即時性的應用需求，本論文所提出之演算法具有極低之運算複雜度，且經實驗證明能有效對抗各種不同的干擾，具有良好的移動目標物判別率。

關鍵字

多干擾環境 ； 適應性背景模型 ； 陰影偵測 ； 去雨演算法 ； 背景相減

並列摘要

Among all the visual surveillance systems, most of the algorithms apply the so-called background subtraction approach for the detection of a foreground, i.e., for the detection of a moving target. However, the background subtraction approach is susceptible to making an incorrect judgment for a time-varying environment, such as the change of light source, the interference of waving trees, the change of shadow, and the texture caused by raindrops. To conquer this problem, some of the researches propose the use of a statistical approach for the construction of an adaptive background model, and pixel values with higher probability would be classified as a background. Nevertheless, the problem of a shadow caused by light masking still exists, and a shadow can still be deemed as the foreground, resulting in an incorrect judgment. In addition, when the system is operating in a rainy day, the effect of the raindrops not only reduce visibility but also make the video screen blurred, causing the system hard to function well. To solve these problems, we propose in this dissertation a self-adapted video surveillance algorithm for multi-interference environment. The proposed algorithms can be divided into three parts. The First part is for rainfall interference removal. We find that only the intensity component of a pixel will be affected when a raindrop exists. That is, the illumination tends to be increased but with chrominance components kept unaltered when a raindrop exists. Based on this observation, we propose in this dissertation an algorithm for rainfall interference removal. The second part is in proposing an efficient approach by using statistical methods for the construction of an adaptive background model for time-varying environments. The third part is for shadow detection. Based on the observation that a shadow doesn’t affect the chrominance components of a pixel, we propose in this dissertation an algorithm for shadow detection so that a shadow will be excluded from being regarded as the foreground. Considering the real-time processing requirement in most of the visual surveillance systems, the proposed algorithm has extremely low computational complexity. Experimental results show that the proposed algorithm works very well under various test environments with a variety of interferences, which justifies the superiority of the proposed approach.

並列關鍵字

Multi-interference environment ； Adaptive background model ； Shadow detection ； Rainfall interference removal ； Background subtraction

參考文獻

[1] K. Garg and S. K. Nayar, “Detection and removal of rain from videos,” in Proc. IEEE Conference Comput. Vis. Pattern Recognit., vol. 1, June 2004, pp. 528–535.

[3] X. Zhang, H. Li, Y. Qi, W. K. Leow, and T. K. Ng, “Rain removal in video by combining temporal and chromatic properties,” in Proc. IEEE Int. Conference Multimedia Expo, Toronto, Ont. Canada, July 2006, pp. 461–464.

[5] 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, July 1997, pp. 780-785.

[6] C. Stauffer, W. Grimson, “Adaptive background mixture models for real-time tracking,” in Proc. IEEE Computer Society Conference on Computer Vision and Pattern Recognition, vol. 2, 1999, pp. 246-252.

[7] S. Jabri, Z. Duric, H. Wexhsler and A. Rosenfeld, “Detection and location of people in video images using adaptive fusion of color and edge information,” in Proc. International Conference on Pattern Recognition, vol. 4, 2000, pp. 627-630.