本論文提出基於立體視覺於人流計數的演算法。以往的方式通常以人臉辨識或者背景相減法來達成人流計數。但是，人臉辨識只能在人臉朝向攝影機時才能偵測出來；而背景相減法雖然可以解決前者的問題，但對於交疊的人群，則無法被各別區分開來。立體視覺是以雙鏡頭的架構獲得視差資訊，不但可以不用透過偵測人臉完成計數，也能利用視差資訊來將交疊的行人各別偵測出來。
然而，對於立體視覺計數人流而言，大部分的方法都是以獲得整張影像的視差資訊，即視差圖來完成。而本論文則是只透過垂直邊緣的視差來達成，不僅節省在區塊比對上花費的時間，也能夠偵測出正確的人數。本論文也提出更新區塊來改善區塊比對在重複紋理區域容易比對錯誤的問題，利用搜尋範圍內找出所有相似於最小代價的代價作為第二次比對的區塊大小。接著以連通物件濾除雜訊後，統計影像垂直邊緣的視差為直方圖，透過逐一尋找波峰的方式，利用最小平方法將高斯函數做曲線擬合，自動選取波峰的範圍，投影出不同的視差平面，最後將視差平面作垂直投影，更進一步改善定位行人的方框。實驗結果顯示出本論文提出的人流計數演算法能夠有效計數出正確的人數。

關鍵字：立體視覺、立體匹配、區塊比對、人流計數、垂直邊緣偵測。

Based on stereo vision, this thesis proposes an algorithm for human flow counting. In the past, the face recognition or background subtraction method is usually used to achieve the flow counting. However, face recognition can only be detected when the face is facing the camera; while the background subtraction method can solve the former problem, the overlapping people still cannot be distinguished. Stereoscopic vision, based on a two-lens architecture, not only does not need to detect the face to complete counting, but also can use parallax information to detect overlapping pedestrians.
However, for stereoscopic human flow counting, most of the methods are to obtain the parallax information of the entire image, that is, the parallax map. In this paper, it is achieved only by the parallax of the vertical edge, which not only saves time spent on block comparison, but also detects the correct amount of people. This paper also proposes the method of updating the block to improve the block alignment with comparing errors easily caused in the repeated texture regions. We use the search range to find all the costs similar to the minimum cost as the block size of the second comparison. After filtering the noise by the connected object, the parallax of the vertical edge of the image is calculated as a histogram. By searching for the peak one by one, the Gaussian function is curve-fitted by the least square method, and the range of the peak is automatically selected to project different parallaxes. In the plane, the parallax plane is finally projected vertically, which further improves the square of positioning pedestrians. Experimental results show that the proposed flow counting algorithm in this thesis can effectively count the correct amount of people.

Keywords: Stereo vision, stereo matching, block matching, human flow counting, vertical edge detection.

[1] T.-Y. Chen, C.-H. Chen, D.-J. Wang, and Y.-L. Kuo, “A people counting system based on face-detection,” in Genetic and Evolutionary Computing (ICGEC), 2010 Fourth International Conference on, 2010, pp. 699–702.
[2] T. Ahonen, A. Hadid, and M. Pietikäinen, “Face Recognition with Local Binary Patterns,” Proc. Eighth European Conf. Computer Vision, 2004, pp. 469-481.
[3] T. Ma, Q. Ji and N Li, “Scene Invariant Crowd Counting Using Multi-Scales Head Detection in Video Surveillance,” IET Image Processing, vol. 12, pp. 2258-2263, Aug. 2018
[4] C.-C. Chang and C.-J. Lin, “LIBSVM: A Library for Support Vector Machines,” ACM Transactions on Intelligent Systems and Technology (TIST), vol.2, no.3, pp. 1-27, Apr. 2011.
[5] Q. Ye, “A Robust Method for Counting People in Complex Indoor Spaces,” in Education Technology and Computer (ICETC), 2010 2nd International Conference on, 2010, pp. 450-454.
[6] N. Bernini, L. Bombini, M. Buzzoni, P. Cerri and P. Grisleri, “An Embedded System for Counting Passengers in Public Transportation Vehicles, ” in Proc. IEEE/ASME 10th Int. Conf. Mechatronic Embedded Syst. Appl. (MESA), pp. 1-6, Sep. 2014.
[7] T. Darrell, G. Gordon, M. Harville, and J. Woodfill, “Integrated Person Tracking Using Stereo, Color, and Pattern Detection,” Proc. Conf. Computer Vision and Pattern Recognition, pp. 601-609, June 1998
[8] V. Kolmogorov and R. Zabih, “Computing visual correspondence with occlusions using graph cuts,” in International Conference for Computer Vision, vol. 2, pp. 508–515, 2001.
[9] S. Birchfield and C. Tomasi, “Depth discontinuities by pixel-to-pixel stereo,” in Proceedings of the Sixth IEEE International Conference on Computer Vision, Mumbai, India, January 1998, pp. 1073–1080.
[10] R. Zabih and J. Woodfill, “Non-Parametric Local Transforms for Computing Visual Correspondence,” Proc. European Conf. Computer Vision, 1994, vol. 2, pp. 151-158.
[11] R. K. Gupta and S.-Y. Cho, “Window-based approach for fast stereo correspondence,” IET Comput. Vis., vol. 7, no. 2, pp. 123–134, May. 2013.
[12] K. Zhang, J. Lu, and G. Lafruit, ‘‘Cross-based local stereo matching using orthogonal integral images,’’ IEEE Trans. Circuits Syst. Video Technol., vol. 19, no. 7, pp. 1073–1079, Jul. 2009.
[13] H. Hirschmüller, “Stereo processing by semiglobal matching and mutual information,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 30, no. 2, pp. 328–341, Feb. 2008.
[14] “OpenCV Stereo BinarySGBM Class Reference,” https://docs.opencv.org/3.4/d1/d9f/classcv_1_1stereo_1_1StereoBinarySGBM.html
[15] C. C. Pham and J. W. Jeon, “Domain transformation-based efficient cost aggregation for local stereo matching,” IEEE Trans. Circuits Syst. Video Technol., vol. 23, no. 7, pp. 1119–1130, Jul. 2013.
[16] “ADLINK 影像擷取卡PCIe-2602,” http://www.adlinktech.com/Products/Machine_Vision/FrameGrabbers_VideoCaptureCards/PCIe-2602?lang=zh-Hant
[17] “15-CAH22 HD-SDI高清攝影機產品規格,” http://www.best-cctv.com.tw/showroom.php?mode=detail&item_id=I24228339
[18] “Stereo Vision: Algorithm and Applications,” www.vision.deis.unibo.it/smatt
[19] R. A. Hamzah and H. Ibrahim, “Literature survey on stereo vision disparity map algorithms,” Journal of Sensors, 2016.
[20] “Richard’s blog: 攝像頭校正,” http://wycwang.blogspot.com/2012/10/camera-calibration-part-2-calibration.html
[21] “Pinhole camera: 相機校正(Camera Calibration),” http://silverwind1982.pixnet.net/blog/post/153218861-pinhole-camera%3A-%E7%9B%B8%E6%A9%9F%E6%A0%A1%E6%AD%A3-%28camera-calibration%29
[22] Z. Zhang, ‘‘A flexible new technique for camera calibration,’’ in Proc. IEEE Trans. Pattern Anal. Mach. Intell., vol. 22, no. 11, pp. 1330–1334, Nov. 2000.
[23] D.-W. Yang, L-C. Chu, C.-W. Chen, J. Wang, M.-D. Shieh, “Depth reliability-based Stereo Matching Algorithm and its VLSI Architecture Design,” IEEE Transactions on Circuits and Systems for Video Technology, Vol.25, no.6, pp.1038–1050, Jun. 2015
[24] S. Birchfield and C. Tomasi, “Depth Discontinuities by Pixel-to Pixel Stereo,” Proc. Sixth IEEE Intl. Conf. Computer Vision, pp. 1073- 1080, Jan. 1998.
[25] “Middlebury立體視覺影像資料庫,” http://vision.middlebury.edu/stereo/
[26] L. He, Y. Chao, K. Suzuki, and K. Wu, “Fast connected-component labeling,” Pattern Recog., vol. 42, no. 9, pp. 1977–1987, Sep. 2008.
[27] H. Guo, “A simple algorithm for fitting a Gaussian function [DSP Tips and Tricks]”, IEEE Signal Processing Magazine, vol. 28, no. 5, pp. 134-137, 2011.