臺灣博碩士論文加值系統

近年來監視攝影機在日常生活中的應用已越來越普及，在高速公路上目前也部署數以百計的IP攝影機來提供路況的即時影像，以往車流量的計算大部份是利用S線圈，但由於S線圈在部署與維護上相當不便，因此本文提出二種以高速公路IP攝影機之即時影像為基礎的車輛計數方法，第一種方法是採用類似S線圈的概念，在影像上設定一條判別線(Trap wire, TW)與一條輔助線(Aux. wire, AW)，分別觀察兩條線上像素值的變化藉此得知車輛移動的方向與數量，此方法的運算只有針對這兩條線段，因此運算量極低。而第二種方法是以移動物體偵測方法為主，並設定一條判別線與一個橢圓偵測區域，為了克服網路延遲可能造成車輛在影像上出現瞬間位移的現象，在移動物體偵測時加入陰影移除方法，接著使用最短距離與樣本比對方法來追蹤橢圓偵測區域內的車輛，並計數通過判別線的車輛。針對六台不同地點之攝影機的實驗結果中，第一種方法的平均計數成功率為85.8%，第二種方法的平均計數成功率為92.4%，雖然第一種方法降低了整體的運算量，但是成功率也略低於第二種方法。整體而言，本論文提出的方法在高速公路車輛計數上具有相當的實用性。

In recent years, the installation of cameras is getting more and more ubiquitously. Hundreds of IP (Internet Protocol) cameras are also installed along the freeway for providing real-time traffic images. S-coil is the traditional way for counting vehicles. But, the installation and maintenance of S-coil is not easy. In this paper, two vehicle counting methods are proposed and compared for IP cameras on the freeway. The first method is similar to the principle of S-coil, a trap wire (TW) and auxiliary wire (AW) are set on the image of an IP camera. Then, the change of pixel values on the wires is used to detect the movement of the vehicles. The computation cost of this method is very low since only those pixels on two wires are needed to be processed. The second method is based on the vehicle detection. A TW and an ellipse as detection area are set on the camera image. For an IP camera, the traffic condition causes the video frames are not stable. That is, the movement of vehicles is not smooth. Sometimes, a vehicle moves a large distance among two successive frames. Therefore, shadow removal technique is used to separate two or more vehicles connected by their shadows on camera image. Shortest distance and template matching techniques are also used to tracking vehicles in the ellipse detection area. A vehicle is counted when it passed the TW. According to the experimental results on six IP cameras, the counting success rate of the first and the second method is 85.8% and 92.4%, respectively. Although the computation cost of the first method is low, its success rate is also slightly lower than that of the second method based on vehicle detection. Taken as a whole, the proposed methods are feasible for vehicle counting for IP cameras on the freeway.

主目錄
中文摘要 I
Abstract II
表目錄 VIII
圖目錄 IX
第一章 簡介 1
第二章 文獻探討 4
2.1硬體式自動計數方法 4
2.2軟體式自動計數方法 7
第三章 Trap wire系統設計與實作 12
3.1系統流程 12
3.2初始設置 14
3.3背景建構 15
3.4線段物件化 16
3.5車輛跨越狀態 18
第四章 Trap wire by motion系統設計與實作 24
4.1系統流程 24
4.2初始設置 25
4.3移動物體偵測 27
4.4車輛追蹤 30
4.5車輛跨越狀態 32
第五章 實驗分析 38
5.1實驗設計與步驟 38
5.2Trap wire system實驗結果 40
5.3Trap wire by motion實驗結果 43
第六章 結論及未來工作 49
參考文獻 51
附錄一 Trap wire系統使用者介面 55
附錄二 Trap wire by motion系統使用者介面 56

表目錄
表1：線段物件定義表 17
表2：短軸一半長度對準確率的影響 27
表3：各樣本錄製frame數、偵測方向與車道數 39
表4：Trap wire系統之實驗結果統計表 41
表5：Trap wire by motion系統之實驗結果統計表 45
表6：兩系統運算時間之實驗結果統計表 48

圖目錄
圖1：影像延遲範例 3
圖2：紅外線偵測團塊 5
圖3：光學感測器計數大型車輛之模擬環境 6
圖4：機車與汽車分類計數方法 7
圖5：計數人群人數之範例 8
圖6：基於臉部偵測計數人數結果 9
圖7：虛擬閘道分別計數行人與車輛 10
圖8：Trap wire系統流程圖 13
圖9：判別線設置範例圖 14
圖10：判別線設定範例 15
圖11：線段產生範例 16
圖12：車行方向的定義 18
圖13：Trap wire系統之狀態轉換圖 19
圖14：車輛跨越狀態圖 19
圖15：車輛進入判別線後延遲範例 20
圖16：延遲後車輛進入狀態2範例 21
圖17：延遲後車輛進入狀態3範例 22
圖18：延遲後錯過判別線範例 23
圖19：Trap wire by motion系統流程圖 25
圖20：判別線與橢圓設置範例 26
圖21：橢圓錯誤設置範例 27
圖22：影像前處理 27
圖23：受到陰影影響兩車合併為一個blob 28
圖24：陰影偵測結果 29
圖25：陰影濾除範例 30
圖26：Trap wire by motion系統之狀態轉換圖 33
圖27：車輛跨越狀態圖 34
圖28：進入橢圓後發生延遲範例 35
圖29：進入橢圓後發生延遲以及延遲後進入橢圓範例 36
圖30：因延遲使車輛跳過橢圓區域 37
圖31：6種不同拍攝角度的測試樣本 39
圖32：幾種造成計數錯誤的範例 42
圖33：陰影濾除造成團塊碎裂 45
圖34：本論文兩系統之直方圖 46
圖35：Trap wire系統使用者介面 55
圖36：Trap wire by motion系統使用者介面 56

[1] M. Tanaka, “Development of traffic counters by laser range scanners,” Proceedings of 2010 SICE Annual Conference, pp. 3128-3134.
[2] G. Lee, H. Kim, J. Yoon, J. Kim, and W. Kim, “Pedestrian counting using an IR line laser,” Proceedings of 2008 Convergence and Hybrid Information Technology (ICHIT’2008), pp. 482-485.
[3] L. Kovavisaruch, T. Sanpechuda, J. Chinrungrueng, U. Sununtachaikul, S. Kittipiyakul, and S. Samphanyuth, “Accuracy improvement method for vehicle detection using optical sensors,” Proceedings of 2009 Intelligent Transport Systems Telecommunications (ITST’2009), pp. 218-222.
[4] A. Treptow, G. Cielniak, and T. Duckett, “Real-time people tracking for mobile robots using thermal vision,” Robotics and Autonomous Systems, Vol. 54, Issue. 9, Sept. 2006, pp. 729-739.
[5] C. Dai, Y. Zheng, and X. Li, “Pedestrian detection and tracking in infrared imagery using shape and appearance,” Computer Vision and Image Understanding, Vol. 106, Issue. 2-3, May-June 2007, pp. 288-299.
[6] A. Fernández-Caballero, J. C. Castillo, J. Martínez-Cantos, and R. Martínez-Tomás, “Optical flow or image subtraction in human detection from infrared camera on mobile robot,” Robotics and Autonomous Systems, Vol. 58, Issue. 12, Dece. 2010, pp. 1273-1281.
[7] M. Bertozzi, A. Broggi, C. Caraffi, M. Del Rose, M. Felisa, and G. Vezzoni, “Pedestrian detection by means of far-infrared stereo vision,” Computer Vision and Image Understanding, Vol. 106, Issue. 2-3, May-June 2007, pp. 194-204.
[8] R. O’Malley, E. Jones, and M. Glavin, “Detection of pedestrians in far-infrared automotive night vision using region-growing and clothing distortion compensation,” Infrared Physics & Technology, Vol. 53, Issue. 6, Nove. 2010, pp. 439-449.
[9] I. K. E. Purnama, A. Zaini., B.N. Putra, and M. Hariadi, “Real time vehicle counter system for intelligent transportation system,” Proceedings of 2009 Instrumentation, Communications, Information Technology, and Biomedical Engineering (ICICI-BME’2009), pp. 1-4.
[10] P. Kilambi, E. Ribnick, A. J. Joshi, O. Masoud, and N. Papanikolopoulos “Estimating pedestrian counts in groups,” Computer Vision and Image Understanding, Vol. 110, Issue. 1, April 2008, pp. 43-59.
[11] T. Chen, C. Chen, D. Wang, and Y. Kuo, “A people counting system based on face-detection,” Proceedings of 2010 Genetic and Evolutionary Computing (ICGEC’2010), pp. 699-702.
[12] K. Zu, F. Liu, and Z. Li, “Counting pedestrian in crowded subway scene,” Proceedings of 2009 Image and Signal Processing (CISP’2009), pp. 1-4.
[13] J. Lu, Y. Xu, and X. Yang, “Counting pedestrians and cars with an improved virtual gate method,” Proceedings of 2010 Computer Application and System Modeling (ICCASM’2010), pp. 448-452.
[14] K. Wang, Z. Li, Q. Yao, W. Huang, and F. Wang, “An automated vehicle counting system for traffic surveillance,” Proceedings of 2007 Vehicular Electronics and Safety (ICVES’2007), pp. 1-6.
[15] Y. Cong, H. Gong, S. Zhu, and Y. Tang, “Flow mosaicking: Real-time pedestrian counting without scene-specific learning,” Proceedings of 2009 Computer Vision and Pattern Recognition (CVPR’2009), pp. 1093-1100.
[16] G. Garcia-Bunster, and M. Torres-Torriti, “Effective pedestrian detection and counting at bus stops,” Proceedings of 2008 Robotic Symposium (LARS’2008), pp. 158-163.
[17] C. C. C. Pang, W. W. L. Lam, and N. H. C. Yung, “A method for vehicle count in the presence of multiple-vehicle occlusions in traffic images,” Intelligent Transportation Systems, Vol. 8, Issue. 3, Sept. 2007, pp. 441-459.
[18] N. U. Rashid, N. C. Mithun, B. R. Joy, and S. M. M. Rahman, “Detection and classification of vehicles from a video using time-spatial image,” Proceedings of 2010 Electrical and Computer Engineering (ICECE’2010), pp. 502-505.
[19] W. Zhang, Q. M. J. Wu, and H. Yin, “Moving vehicles detection based on adaptive motion histogram,” Digital Signal Processing, Vol. 20, Issue. 3, May 2010, pp. 793-805.
[20] J. Choi, Y. J. Yoo, and J. Y. Choi, “Adaptive shadow estimator for removing shadow of moving object,” Computer Vision and Image Understanding, Vol. 114, Issue. 9, Sept. 2010, pp. 1017-1029.