臺灣博碩士論文加值系統

本篇論文提出一種利用單一頂照式魚眼攝影機的人物偵測和追蹤的新方法，這此之前有關人物偵測的研究有很大一部分都是使用投射型攝影機(projective camera)且已經十分完善；而在魚眼攝影機這個領域的研究相對較少，且都是應用在較為簡單的場景，效率也尚有提升的空間，除此之外，魚眼鏡頭的優勢在於只用單一攝影鏡頭就能涵蓋非常廣的範圍，而且在人物較為靠近、容易重疊的場合，使用頂照式的魚眼攝影機相較其他鏡頭有更好的視野。本論文的目標是提出一個新的方法能在比較複雜的場景中偵測、追蹤並計算人數，且期望能應用在實時(real-time)的場合。我們的偵測方法是利用橢圓形的遮罩搭配HOG特徵，接著套用數種SVM(support vector machine)分類器來判斷每一個遮罩是不是涵蓋人物在內。追蹤的方法則是透過前後時間的位移，加上一些顏色的特徵來推斷人物的落點。

This paper proposes a new algorithm for human detection using single downward-viewing fish-eye camera. In recent years, methods for human detection using projective camera have been studied extensively, but researches on human detection from fish-eye camera images are very limited and time-consuming, and most of them were used in simple and uncluttered environments. In addition, the advantage of fisheye lenses is that they can cover a very wide range with only one camera. When people are close, or they may be blocked, using fish-eye camera has a better view than other cameras. The main purpose of this paper is to propose a new method which can detect, track and estimate the number of people in more complicated scenes and expect to be applied in real-time situations. Our detecting algorithm makes use of elliptic templates and HOG features, and then apply a set of support vector machines (SVMs) to find out whether there are people in the template or not. Meanwhile, we track people’s position by applying color features and analyzing their movement in a specific time.

摘要 I
Abstract II
致謝 III
目錄 IV
圖目錄 VI
表目錄 VII

第一章 緒論 1
1.1 背景介紹 1
1.2 研究動機與方法 1
1.3 論文架構 3
第二章 文獻探討 4
第三章 研究方法 7
3.1 實驗用資料庫 8
3.1.1 影片來源與前處理 8
3.1.2 Ground Truth的標記 10
3.2 背景模型 13
3.3 橢圓遮罩與場景的建模 14
3.4 人物偵測 18
3.4.1 以徑向距離將畫面分段 18
3.4.2 HOG + SVM分類器 20
3.4.3 人物偵測、追蹤 21
第四章 研究結果 23
4.1 偵測/追蹤之準確度評分方法 24
4.2 統計結果 26
第五章 結論 30
參考文獻 31

[1] N. Dalal and B. Triggs, "Histograms of oriented gradients for human detection," Proc. CVPR, pp. 886–893, 2005.
[2] Y. Kubo, T. Kitaguchi, and J. Yamaguchi, "Human tracking using fisheye images," Proc. SICE Annual Conference, 2007.
[3] L. Meinel, M. Findeisen, M. Heß, A. Apitzsch, and G. Hirtz, "Automated real-time surveillance for ambient assisted living using an omnidirectional camera," Proc. ICCE, pp. 396-399, 2014.
[4] M. Saito, K. Kitaguchi, G. Kimura, and M. Hashimoto, "Human detection from fish-eye image by Bayesian combination of probabilistic appearance models," Proc. SICE Annual Conference, 2011.
[5] F. Vandewiele, F. Bousetouane, and C. Motamed, "Occlusion management strategies for pedestrians tracking across fisheye camera networks," Proc. ICDSC, 2013.
[6] A.T. Chiang and Y, Wang, "Human detection in fish-eye images using HOG-based detectors over rotated windows," Proc. ICME Workshops, 2014.
[7] E. Schwalbe, "Geometric modelling and calibration of fisheye lens camera systems." Proc. 2nd Panoramic Photogrammetry Workshop, Int. Archives of Photogrammetry and Remote Sensing, vol. 36, no. Part 5. 2005.
[8] K. Bernardin and R. Stiefelhagen. "Evaluating multiple object tracking performance: the CLEAR MOT metrics," EURASIP Journal on Image and Video Processing, 246309, 2008.