臺灣博碩士論文加值系統

本論文提出一套即時行人偵測系統，本系統利用即時輸入影像擷取移動物體區塊後，抽取該移動物體的輪廓特徵，最後結合本論文提出的SVM混合模型類神經網路完成移動物體辨識，此SVM混合模型類神經網路包含有六層的架構，第一層至第三層為一個全相連順向類神經網路，第四層為正規化層，第五層為SVM混合模型層，最後一層為輸出層。由於單一SVM不適用於訓練較大樣本數量的分類問題，解決的方法是將訓練樣本空間分割成許多小的訓練樣本集合，每一個小的訓練樣本集合交予一個SVM訓練，最後將每一個SVM的輸出與類神經網路輸出權重值加總整合輸出，因此稱為SVM混合模型類神經網路，並且針對此網路提出一個新的學習演算法，除學習類神經網路相關參數外，亦學習各SVM中RBF的參數使得有最佳效能，經實驗驗證，即使有光影與非行人的移動物體影響，也不會影響辨識結果，能有效的提高行人偵測系統的正確率，針對數量龐大的訓練樣本集合，本論文提出的方法其訓練時間也可比利用單一SVM快速，測試的正確率也比使用單一的SVM效果好。

This paper proposes a real-time pedestrian detection system, which gets the object area image through the real-time input image and extraction it signature feature, it classifies the feature of the real-time pedestrians through the proposed SVMs-mixture model neural networks, it classifies the feature of the real-time pedestrians through the proposed SVMs-mixture model neural networks, which is composed of six layers. From the first to the third layer, it is the full-connected forward neural networks. The fourth layer is the normalization layer. The fifth layer is the SVMs-mixture model, while the last one is the output layer. the single SVM is unsuitable to the classification of large scale training samples, it will divide the training samples into many smaller sub-sets. It trains a SVM for each set. Finally, it uses the neural network to put out the weight sum of these SVMs results. Therefore, it is called SVMs-mixture model neural network. Moreover, a new learning algorithm is proposed for it, which not only adjusts the related parameters on the artificial neural network, but also adjusts the parameters of RBF function in SVM to get the optimum parameters for kernel function. Through the experimental verification, even if influenced by the shadow and the mobile object other than pedestrians, the image identification results won’t be influence at all. It can effectively improve the accuracy of pedestrian detection system. For large scale of training sample sets, the calculation time of the proposed method is shorter than that of using the single SVM, and the test accuracy also gets the better results than that of the single SVM.

摘要......................................................I
Abstract.................................................II
致謝....................................................III
目錄.....................................................IV
表目錄...................................................VI
圖目錄..................................................VII
第一章緒論.................................................1
1.1 研究動機與目的.....................................1
1.2 相關文獻探討.......................................3
1.3 論文架構...........................................7
第二章、即時行人偵測系統架構...............................9
第三章、候選區域偵測......................................11
3.1 前景偵測..........................................11
3.2 背景影像建立與更新................................13
3.3 影像去雜訊........................................14
3.4 候選區域切割......................................15
第四章、行人特徵抽取......................................21
4.1 邊緣偵測..........................................21
4.2 輪廓點簽名........................................24
4.3 方向梯度直方圖 (Histogram of Oriented Gradient)...27
第五章、行人辨識..........................................31
5.1 支持向量機器 (Support Vector Machines) ..........31
5.2 SVM混合模型類神經網路 (SVMs-Mixture Model
Neural Networks) .....................................36
5.3 學習演算法........................................40
5.4 IRIS資料分類範例..................................42
第六章、實驗..............................................48
第七章、結論..............................................57
參考文獻..................................................59
表5.4.1、Iris資料庫.......................................47
表6.1、行人資料庫.........................................50
表6.2、不同特徵抽取方法的辨識結果與所需時間...............51
表6.3、利用不同數量的SVM結果..............................52
表6.4、利用不同類神經隱藏層數量結果.......................53
表6.5、比較不同分類器結果.................................54
圖 2.1、系統流程圖........................................10
圖3.1.1、背景相減法範例 (a)背景影像 (b)輸入影像 (c)移動物體影像......................................................12
圖3.2.1、統計式的背景影像更新方法.........................13
圖3.3.1、孤立點去除法示意圖...............................14
圖3.3.2、過濾前景影像雜訊 (a)有孤立點的前景影像 (b)去除孤立點的前景影像..............................................15
圖3.4.1、4個相連掃瞄示意圖................................16
圖3.4.2、Labeling Algorithm操作之二值影像範例.............17
圖3.4.3、Labeling Algorithm操作過程示意圖.................18
圖3.4.4、利用Labeling Algorithm框取移動物件：場景1........19
圖3.4.5、利用Labeling Algorithm框取移動物件：場景2........20
圖4.1.1、3 3區域的影像像素點..............................23
圖4.1.2、3 3區域的影像像素點 (a)x方向 (b)y方向............23
圖4.1.3、針對整張影像做邊緣偵測 (a)二值化影像 (b)邊緣偵測影像........................................................24
圖4.2.1、輪廓像素點的排序示意圖...........................25
圖4.2.2、輪廓曲線平滑 (a)平滑前 (b)平滑後.................25
圖4.3.1、直方圖統計結果圖(4方向示意圖) ...................28
圖4.3.2、二維資料轉成一維資料.............................29
圖5.1.1、SVM分類示意圖....................................32
圖5.2.1、SVM混合模型類神經網路架構圖......................37
圖5.3.1、類神經網路結構圖.................................40
圖5.4.1、單一SVM結果圖(線性核函數) .......................44
圖5.4.2、SVM混合模型類神經網路結果圖(線性核函數)..........44
圖5.4.3、單一SVM結果圖(RBF核函數) ........................45
圖5.4.4、SVM混合模型類神經網路結果圖(RBF核函數) ..........46
圖6.1、行人與非行人的部份樣本.............................50
圖6.2、行人偵測部分結果：場景1............................55
圖6.3、行人偵測部分結果：場景2............................56
圖6.4、行人偵測失敗的例子.................................56

[1].Lucire. (February 9, 2010). “Volvo Announces New S60, with Pedestrian Detection and Automatic Braking,” Retrieved May 8, 2010, from http://lucire.com/insider/20100209/volvo-announces-new-s60-with-pedestrian-detection-and-automatic-braking/.

[2].Carzi. (June 22, 2008). “Toyota Night Vision Pedestrian Finder Technology,” Retrieved May 8, 2010, from http://www.carzi.com/2008/06/22/toyota-night-vision-pedestrian-finder-technology/.

[3].C. Papageorgiou and T. Poggio, “A Trainable System for Object Detection,” International Journal of Computer Vision, vol. 38, no. 1, pp. 15-33, Sept. 2000.

[4].A. Mohan, C. Papageorgiou, and T. Poggio, “Example-Based Object Detection in Images by Components,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 23, no. 4, pp. 349-361, Apr. 2001.

[5].M.-B. Luis and F. Olac, “Object Detection Using Image Reconstruction with PCA,” Image and Vision Computing. vol. 27, no. 1-2, pp. 2-9, 2009.

[6].P. A. Viola and M. J. Jones, “Robust Real-time Face Detection,” International Journal of Computer Vision, vol. 57, no. 2, pp. 137-154, 2004.
[7].N. Dalal and B. Triggs, “Histograms of Oriented Gradients for Human Detection,” IEEE Conference on Computer Vision and Pattern Recognition, pp. 886-893, June 2005.

[8].Gonzalez, “Digital Image Processing 2/e,” Addison-Wesley Publishing Company, 2002.

[9].J. Heikkila and O. Silven, “A Real-time System for Monitoring of Cyclists and Pedestrians,” Second IEEE Workshop on Visual Surveillance, pp.74-81, Jul. 1999.

[10].C.-F. Juang and L.-T. Chen, “Moving Object Recognition by a Shape-based Neural Fuzzy Network,” Neurocomputing, vol. 71, no 13-15, pp. 2937-2949, Aug. 2008.

[11].R. Cutler and L. S. Davis, “Robust Real-time Periodic Motion Detection, Analysis, and Applications”, IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 22, no 8, 2000.

[12].R. Fablet and M. Black, “Automatic Detection and Tracking of Human Motion with a View-based Representation,” in Proc. ECCV, pp. 476-491, 2002.

[13].L. Andrade, S. Blunsden, and B. Fisher, “Characterisation of Optical Flow Anomalies in Pedestrian Traffic,” Symposium on Imaging for Crime Prevention and Detection, pp. 73-78, 2005.



[14].T. Zhao and R. Nevatia, “Tracking Multiple Humans in Complex Situations,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 26, no. 9, pp. 1208-1221, Sep. 2004.

[15].J. Deutscher, A. Blake, and I. D. Reid, “Articulated Body Motion Capture by Annealed Particle Filtering,” IEEE Conference on Computer Vision and Pattern Recognition, vol. 2, pp. 126-133, 2000.

[16].M. Spengler and B. Schiele, “Towards Robust Multi-cue Integration for Visual Tracking,” Machine Vision and Applications, vol. 14, no. 1, pp. 50-58, 2003.

[17].S. Munder, C. Schnörr, and D. M. Gavrila, “Pedestrian Detection and Tracking Using a Mixture of View-Based Shape-Texture Models,” IEEE Transactions on Intelligent Transportation Systems, vol. 9, no. 2, June 2008.

[18].D. M. Gavrila, J. Giebel, and S. Munder, “Vision-based Pedestrian Detection: The PROTECTOR System,” IEEE Intelligent Vehicles Symposium, pp. 13-18, Jun. 2004.

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



[20].M. Oren, C. Papageorgiou, P. Sinha, E. Osuna and T. Poggio, “Pedestrian Detection Using Wavelet Templates”, IEEE Computer Society Conference on Computer Vision and Pattern Recognition, pp. 193-99, 1997

[21].S. Kang, H. Byun and S. W. Lee, “Real-time pedestrian detection using support vector machines,” International Journal of Pattern Recognition and Artificial Intelligence, vol. 17, no. 3, pp. 405-416 , 2003.

[22].V. Vapnik, “The Nature of Statistical Learning Theory,” Springer-Verlag, New York, 1995.

[23].C. J. C. Burges, “A Tutorial on Support Vector Machines for Pattern Recognition,” Data Mining and Knowledge Discovery, vol. 2, no. 2, pp. 1-47, 1998.

[24].C. Nello and S.-T. John, “An Introduction to Support Vector Machines and Other Kernel-based Learning Methods,” Cambridge University Press, 2000.

[25].N. Gilardi and S. Bengio. “Local Machine Learning Models for Spatial Data Analysis” Geographical Information and Decision Analysis, vol. 4, no. 1, pp. 11-28, 2002.

[26].R. Collobert, S. Bengio, and Y. Bengio, “Parallel Mixture of SVMs for Very Large Scale Problems,” Neural Computation, vol. 14, pp. 1105-1114, 2002.