本論文探討基於尺度不變光流法之物件追蹤，並應用於一個穩健的行人監控系統。這個系統由兩部分組成，第一部分是行人偵測系統，第二部分是以動態偵測系統來輔助行人偵測系統，以此補足行人偵測系統無法辨識的部分，使行人偵測系統更加穩健。
在行人偵測的部分：文獻回顧顯示，在行人偵測系統的研究上，大多藉由HOG截取特徵向量再由SVM分類器去做分類行人和背景，但是對於基本型態和一般行人不太一樣的例子，像是撐傘的人、坐輪椅的人或是牽車的行人就有辨識上的困難，以上述方法則會無法辨識或無法完整辨識。
在物體追蹤的部分：Lucas-Kanade光流法是一種稀疏光流法，先取出欲追蹤的前景後，再尋找角點當作特徵點，接著利用光流法追蹤特徵點的座標位置，即可計算特徵點移動的像素差。另外，在SIFT演算法中為了降低程式運算量，其於抓取關鍵點時會依據某些規則省略很多關鍵點，以此加快關鍵點比對的進行，但省略關鍵點會造成圖形比對精確度的降低。
本論文提出結合LOG-SIFT (優化後的尺度不變特徵)擷取關鍵點的光流追蹤法，此法可改善原始光流法在物體做大動作、旋轉、翻轉或是移動過快時，追蹤的特徵點容易遺失或產生位移的狀況，可以解決在追蹤過程中遺失過多特徵點或導致特徵點無法繼續追蹤的情況。而此法(LOG-SIFT加上光流法)在程式執行時間上，也優於原始單純以LOG-SIFT逐幀進行匹配，可增加追蹤系統的即時性。在應用方面，可將此研發系統整合於交通號誌或線上即時路徑規劃系統，以保持交通順暢。

This thesis discusses the object tracking based on optical flow with SIFT (Scale Invariant Feature Transform) feature extraction, and then applies it to a robust pedestrian detection system. This system is constructed by two parts: 1) a pedestrian detection subsystem; 2) a dynamical detection subsystem, capable of making up the unrecognized part in the former subsystem to assist the former subsystem.
Literature survey indicates that in most of the pedestrian detection research HOG (Histogram of Oriented Gradients) features are extracted and then the pedestrian and background are distinguished by using the SVM (Support Vector Machines) classifier. However, it is difficult to recognize by using HOG & SVM for the cases with basic patterns different from general pedestrian, e.g., pedestrian holding umbrellas, in wheel chairs, and holding bicycles.
Concerning the Lucas-Kanade (LK) optical flow method in object tracking, one should extracts the foreground, detects corner point as feature points, tracks feature points, and then calculates the pixel moving distance. For SIFT (Scale Invariant Feature Transform), in order to reduce the computation in programming, some rules will be used to omit many feature points. However, it will result in the lower accuracy in the graphical matching.
This thesis first proposes the optical flow method integrated with LOG-SIFT (Laplacian of Gaussian – Scale Invariant Feature Transform) to extract features. This method can improve the cases with displacements or feature points easily to be lost, caused by the large action, rotation, overturning, and moving too fast. Therefore, it can solve the problem that the feature points cannot be tracked due to missing too many feature points. Concerning the execution time in programs, this method (LOG-SIFT integrated with the optical flow method) also behaves better than the original LOG-SIFT method, so it can improve the real-time performance. This developed system can be applied to traffic light control or online real-time route planning to reduce traffic congestion and to maintain smooth flow of traffic.

[1] N. Dalal and B. Triggs, “Histograms of oriented gradients for human detection,” IEEE Conference on Computer Vision and Pattern Recognition, vol. 1, pp. 886-893, Jun. 25, 2005, CA, USA.
[2] W. L. Lu and J. Little, “Simultaneous tracking and action recognition using the pca-hog descriptor,” Computer and Robot Vision, pp. 6-13, Jun. 07-09, 2006, Quebec, Canada.
[3] T. Kobayashi, A. Hidaka, and T. Kurita, “Selection of histograms of oriented gradients features for pedestrian detection,” International Conference on Neural Information, pp. 598-607, Nov. 25-28, 2008, Heidelberg, Berlin.
[4] H. Xiong, “Fast pedestrian detection based on HOG-PCA and gentle AdaBoost,” International Conference on Computer Science and Service, pp. 1819-1822, Aug. 11-13, 2012, Nanjing, China.
[5] H. Salmane, Y. Ruichek, and L. Khoudour, “Object tracking using Harris comer points based optical flow propagation and Kalman filter,” International IEEE Conference on Intelligent Transportation Systems, pp. 67-73, Oct. 5-7, 2011, Washington DC, USA.
[6] C. Li, L. Guo, and Y. Hu, “A new method combining HOG and Kalman filter for video-based human detection and tracking,” IEEE Conference on International Congress on Image, vol. 1, pp. 290-293, Jun. 16-18, 2010, Yantai.
[7] D. Wei, Y. Zhao, R. Cheng, and G. Li, “An enhanced histogram of oriented gradient for pedestrian detection”, International Conference on Intelligent Control and Information Processing, pp. 459-463, Jun. 9-11, 2013, Beijing, China.
[8] Z. Chen, J. Cao, Y. Tang, and L. Tang, “Tracking of moving object based on optical flow detection,” IEEE International Conference on Computer Science and Network Technology, vol. 2, pp. 1096-1099, Dec. 24-26, 2011, Harbin.
[9] M. B. Ka'aniche, and F. Br'emond, “Tracking HoG descriptors for gesture recognition,” IEEE Conference on Advanced Video and Signal Based Surveillance, pp. 140-145, Sep. 2-4, 2009, Genova.
[10] A. J. Lipton, H. Fujiyoshi, and R. S. Patil, “Moving target classification and tracking from real-time video.” IEEE Workshop on Applications of Computer Vision, pp. 8-14, Oct. 1, 1998, Princeton, NJ.
[11] S. P. Zhu, J. Gao, and Z.-C. Guo “Video object segmentation based on change detection and frame difference accumulation,” Journal of Optoelectronics Laser, pp. 1592-1599, Oct. 24, 2013, Beijing, China.
[12] I. Huerta, A. Amato, X. Roca, and J. Gonzalez, “Exploiting multiple cues in motion segmentation based on background subtraction,” Neurocomputing, vol. 100, pp. 183-196, Jan. 16, 2013.
[13] W. Eric and L. Grimson, “Learning patterns of activity using real-time tracking,” Pattern Analysis and Machine Intelligence, vol. 22, pp. 747-757, Oct. 22, 2000, Washington, DC, USA.
[14] B. D. Lucas and T. Kanade, “An iterative image registration technique with an application to stereo vision,” Proceedings of Imaging Understanding Workshop, vol. 2, pp. 674-679, 1981, San Francisco, CA, USA.
[15] P. Premaratne, S. Ajaz, and M. Premaratne, “Gesture tracking and recognition system using Lucas-Kanade algorithms,” Neurocomputing, vol. 116, pp. 242-249, Sep. 20, 2013.
[16] W. Zhao and C. Ngo, “Flip-invariant SIFT for copy and object detection,” IEEE Transactions on Image Processing, vol. 22, no. 3, pp. 980-991, Jan. 25, 2013.
[17] S. Wang, Y. Liu, L. Daxiang and H. Yan, “An Improved SIFT feature extraction method for Tyre tread patterns retrieval,” Computational Intelligence and Design, vol. 1, pp. 539-543, Dec. 13-14, 2014, Hangzhou.
[18] C. Geng and X. Jiang, “SIFT features for face recognition,” Computer Science and Information Technology, pp. 598-602, Aug. 8-11, 2009, Beijing, China.
[19] S. P. Nasholm, “Post-processing enhancement of reverberation-noise suppression in dual-frequency SURF imaging ,” IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, vol. 58, no. 2, pp. 338- 348, Feb. 22, 2011.
[20] S. P. Nasholm, R. Hansen, S. E. Masoy, and T. F. Johansen, “Transmit beams adapted to reverberation noise suppression using dual-frequency SURF imaging,” IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, vol. 56, no. 10, pp. 2124 -2133, Oct. 30, 2009.
[21] S. Nashol, R. Hansen, and A. J. Angelsen, “Post-processing enhancement of reverberation-noise suppression in dual-frequency SURF imaging,” IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, vol. 58, no.2, pp. 338-348, Feb. 22, 2011.
[22] M. B. Kaaniche and F. Bremond, “Tracking HoG descriptors for gesture recognition,” Advanced Video and Signal Based Surveillance, pp. 140-145, Sep. 2, 2009, Genova.
[23] C. Yao, F. Wu, H. Chen, and X. Hao, “Traffic sign recognition using HOG-SVM and grid search,” International Conference on Signal Processing, pp. 962-965, Oct. 19-23, 2014, Hangzhou.
[24] D. Sun and J. Watada, “Detecting pedestrians and vehicles in traffic scene based on boosted HOG features and SVM,” Intelligent Signal Processing, pp. 1-4, May. 15-17, 2015, Siena.
[25] Y. Tan and J. Wang, “A support vector machine with a hybrid kernel and minimal Vapnik-Chervonenkis dimension,” IEEE Transactions on Knowledge and Data Engineering, vol. 16, no. 4, pp. 385-395, Apr. 3, 2004.
[26] S. Vassiliadis, E.A. Hakkennes, J. S. S. M. Wong, and G. G. Pechanek, “The sum-absolute-difference motion estimation accelerator,” Euromicro Conference, vol. 2, pp. 559-566, Aug. 25-27, 1998, Vasteras.
[27] J. Vanne, E. Aho, T.D. Hamalainen, and K. Kuusilinna,“A high-performance sum of absolute difference implementation for motion estimation,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 16, no. 7, pp. 876-883, Jul. 24, 2006.
[28] X. Cai, F. H. Ali, and E. Stipidis, “Robust online video background reconstruction using optical flow and pixel intensity distribution,” International Conference on Communications, pp. 526-530, May 19-23, 2008, Beijing, China.
[29] J. Tang, X. Jing, and D. He, “Research on background reconstruction based on prediction and filtering,” Computer Science and Software Engineering, vol. 1, pp. 969-972, Dec. 12-14, 2008, Wuhan, Hubei.
[30] G. Farneback, “Two-frame motion estimation based on polynomial expansion,” Scandinavian Conference, pp. 363-370, Jul. 29, 2003, Heidelberg, Berlin.
[31] G. Farneback, “Very high accuracy velocity estimation using orientation ensors parametric motion and simultaneous segmentation of the motion field,” IEEE International Conference on Computer Vision, vol. 1, pp. 171-177, Oct. 21, 2001, Vancouver, Canada.
[32] G. Farneback, “Polynomial expansion for orientation and motion estimation.PhD thesis,” Linkoping University, Sweden, no. 790, 2002, Linkoping.
[33] D. Lowe, “Distinctive image features from scale invariant keypoints,” International Journal of Computer Vision, pp. 91-110, 2006.
[34] Y. Ke and R. Sukthankar, “ PCA-SIFT: a more distinctive representation for local image descriptors,” Computer Vision and Pattern Recognition, IEEE Computer Society Conference, vol.2, pp. 506-513, Jul. 27, 2004.
[35] 朱庭葳，「以邊緣強化之SIFT偵測移動物件」，國立臺灣科技大學電機工程系碩士學位論文，1月，26日，2016。
[36] H. Bay, A. Ess, T. Tuytelaars, and L. V. Gool, “ Speeded-up robust features (SURF),” Similarity Matching in Computer Vision and Multimedia, vol.110, no 3, pp. 346-359, Jun. 23, 2008.
[37] L. Juan and O. Gwun, “A comparison of SIFT, PCA-SIFT and SURF,” International Journal of Image Processing, vol. 3, no. 4, pp. 561-756, Aug. 21, 2009, South Korea.
[38] K. P. Horn and G. Rhunck, “Determining optical flow,” Massachusetts Institute of Technology, 1980, Cambridge, MA, USA.
[39] D. Lucas and T. Kanade, “An iterative image registration technique with an application to stereo vision,” Imaging Understanding Workshop, vol. 2, pp. 121-130, Aug. 24-28, 1981, CA, USA.