本篇論文的目的為利用單一相機實現辨識與追蹤行人，運算上使用梯度方向直方圖(Histogram of Oriented Gradient, HOG)擷取行人特徵，並使用行人特徵訓練支持向量機(Support Vector Machine, SVM)，作為行人偵測器，利用此偵測器找出圖像中是否存在行人。為了提高偵測器的偵測率，針對容易偵測錯誤之影像，將其列為Hard Example，再次進行訓練提升行人之偵測率以及降低誤判率。

偵測行人後取得行人所在的位置與範圍，利用TLD演算法(Tracking Learning Detection)即時追蹤行人目標，並採用監督自助法(Supervised Bootstrapping)使用帶標籤樣本訓練檢測器，在追蹤過程中所取得的新樣本將再次更新分類器，使得檢測器得以適應目標與環境的變化。 在追蹤過程透過錯誤評估方法，得以有效的選擇效果較佳的追蹤點，追蹤的結果將與檢測結果彼此監督，綜合彼此的結果作為行人追蹤結果。 在應用上，利用取得的行人二維座標估算出目標實際三維位置，使得本系統在應用上更加具有彈性。

The purpose of this thesis is using a RGB camera to implement recognition and tracking of pedestrian. In this thesis, Histogram of Oriented Gradient(HOG) is used to extract the features of pedestrian, these features to are used to train a SVM as our pedestrian detector to detect pedestrians existing in the image acquired by camera.
To achieve higher detection accuracy rate, we categorize easily misclassified images as Hard Examples, later retrain the detector to increase detection rate and reduce false-positive rate.

After the pedestrian has been detected, Tracking Learning Detection algorithm is applied to track target pedestrian, also Supervised Bootstrapping method is used to train a classifier with labeled samples and update the classifier with new samples acquired during tracking procedure, such that the classifier adapts better to the variation of target position and surrounding. Through forward-backward error method feature points with better performance and be selected during tracking procedure, with the tracking results and the classifier under each other’s supervision, combing both results as pedestrian tracking result. Estimating target’s actually 3-D position through acquired pedestrian 2-D coordinate gives the proposed system more flexibility with other applications.

[1] Dalal,N. and Triggs,B. “Histograms of Oriented Gradients for Human Detection”, in Computer Vision and Pattern Recognition, 2005.

[2]Kalal,Z., Mikolajczyk,K. and Matas,J. “Tracking-Learning-Detection”,in Pattern Analysis and Machine Intelligence, IEEE Transactions ,2011

[3]B. D. Lucas and T. Kanade . “An iterative image registration technique with an application to stereo vision,” in International Joint Conference on Artificial Intelligence, 1981

[4] Kalal,Z., Mikolajczyk,K. and Matas,J. “Forward-Backward Error: Automatic Detection of Tracking Failures”, in Pattern Recognition (ICPR), 20th International Conference onYear 2010,2010.

[5] Kalal,Z., Mikolajczyk,K. and Matas,J. “P-N learning: Bootstrapping binary classifiers by structural constraints”, in Computer Vision and Pattern Recognition (CVPR), IEEE Conference on Year: 2010,2010

[6]Geronimo, D.; Lopez, A.M. , Sappa. A.D. and Graf, T. “Survey of Pedestrian Detection for Advanced Driver Assistance Systems”, in Pattern Analysis and Machine Intelligence, IEEE Transactions ,2009.

[7]Enzweiler, M. , Gavrila, D.M. “Monocular Pedestrian Detection: Survey and Experiments”, in Pattern Analysis and Machine Intelligence, IEEE Transactions on
Year: 2009, Volume: 31, Issue: 12,2008.

[8]林哲宇 “基於視訊之公路車輛偵測與追蹤”,中原大學資訊工程學系碩士班,2014

[9]林可薇“以HOG為基礎的Adaboost方法做行人的頭部與肩部偵測”,國立清華大學電機工程學系碩士班,2011.

[10]齊楠楠,揭斐然,謝熙,吳巍“基於TLD 的艦船目標追蹤方法研究” ,中航工業洛陽電光設備研究所光電控制技術重點實驗室,河南 洛陽2013

[11]毛銓毅“語意驅動式HOG行人偵測”,國立高雄大學電機工程學系碩士班,2011.

[12]顧詩涵 “結合銓忠樣本與HOG之行人偵測”,國立高雄第一科技大學電腦與通訊工程系,2013

[13]台灣大學林智仁教授“LIBSVM”,http://www.csie.ntu.edu.tw/~cjlin/libsvm/

[14]OpenCV2.4.6,http://opencv.org/downloads.html

[15]Qt5.4.1, http://www.qt.io/developers/

[16]TLD Zdenek Kalal, http://personal.ee.surrey.ac.uk/Personal/Z.Kalal/tld.html