臺灣博碩士論文加值系統

本文提出一個利用影像梯度直方圖與支援向量機器學習於智慧車流之應用(Smart Traffic Flow Statistics Based on Histogram of Gradient and Support Vector Machine)，以下稱為本系統，本系統是智慧車流的應用範例。智慧城市中車流是非常重要的領域，有了車流數據才能得知道路路況，目前常見的方法有路面裝設感應地磁、人工計數、AI影像辨識等。前二者所消耗的成本及人力過大，而後者又因深度學習的運算複雜度高，一般電腦沒辦法達到趨近即時的辨識，導致在場域中難以實現。
因此本文提出影像梯度直方圖與支援向量機器學習於智慧車流之應用系統，可以用一般等級的電腦搭配攝影機即可達成路口車流計數，本系統分為三個部分，分別為影像前處理、車種辨識引擎、物件追蹤引擎。前處理負責接收及處理輸入影像，提取感興趣區域(region of interest, Roi)。車種辨識引擎從感興趣區域中將影像利用影像梯度直方圖進行編碼後，交由支援向量機進行辨識，當有多輛車重疊的時候，則會進行在多輛車中偵測內含車輛，辨識出的結果在送回非極大值抑制進行合併。物件追蹤引擎將輸入的物件位置進行追蹤，有出現新的物件就加入新的追蹤列表，其餘則持續追蹤直到確定消失，短暫消失時可以利用短期物件預測進行猜測。
本文的研究貢獻為多物件偵測模組及物件追蹤模組，前者可在多物件貼合的情況下用較少辨識次數偵測出重疊的個體，後者利用物理特性進行物件追蹤的預測。

In this thesis, we proposed a Smart Traffic Flow Statistics Based on Histogram of Gradient and Support Vector Machine. The system is an example in vehicle counting. In smart city, vehicle counting is an important domain as we obtain traffic information by it. Currently, commonly use wireless magnetometer, count by human or video-based Artificial Intelligence.Former two option cost highly on time-consuming and and later one get hinder by deep-learning’s high Computational complexity, Make them unrealistic to achieve.
So we proposed this system, which can be achieve vehicle counting by PC and camera. The system has three parts, namely image pre-processing, vehicle recognition engine and object tracking engine. Image pre-processing receive input image and obtain region of interest(Roi). Vehicle recognition engine extract Histogram of Gradient form input Roi as feature, then identified the input Roi as which kind of vehicle. If Roi be identified as multiple vehicle combined, we scan in Roi to separate combined vehicle. Object tracking engine takes the result of vehicle recognition engine. When an object appear no-where near current tracking object, it add into tracking list as new tracking object. Otherwise it count as new location as former tracking object and update it’s location.
The main contributions of this paper are as follows: multi-object-scan and object tracking. Former can separate multiple object combined together and later can predict short-term disappear object by it’s trajectory.

摘要 i
ABSTRACT ii
誌謝 iii
目錄 iv
圖目錄 vi
表目錄 viii
壹、 緒論 1
一、 研究動機及目的 1
二、 本文貢獻 1
三、 文獻探討與本文特色 2
(一) 文獻探討 2
(二) 本文特色 3
貳、 系統框架 4
一、 影像前處理 5
(一) 背景模型建立、前景判斷 6
(二) 非極大值抑制 7
二、 車種辨識引擎 9
(一) 車種辨識模組 10
1. 影像梯度直方圖 10
2. 支援向量機 12
(二) 多物件偵測模組 14
三、 物件追蹤引擎 17
(一) 物件追蹤模組 18
(二) 車流方向計算模組 20
參、 實驗 21
一、 實驗環境 21
二、 實驗設計 21
(一) 車輛辨識模組 21
(二) 多尺度SVM 27
(三) 非極大值抑制 28
(四) 軌跡追蹤 29
(五) 車流計數 30
(六) 文獻比較 33
(七) 實驗結果分析 35
肆、 結論與未來展望 36
一、 結論 36
二、 未來展望 36
參考文獻 37
附錄 38
一、 距離轉換 38
二、 局部二值模式 38
三、 舊版計數結果 39
四、 GPU執行VGG16及YOLO 40
(一) VGG16 40
(二) YOLO 40

[1] .Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556. 2014 Sep 4.
[2] .Redmon J, Farhadi A. Yolov3: An incremental improvement. arXiv preprint arXiv:1804.02767. 2018 Apr 8
[3] .Zhe Dai, HuanSheng Song, Xuan Wang, Yong Fang, Xu Yun, ZhaoYang Zhang, and HuaiYu Li, “Video-based vehicle counting framework,” IEEE Access, vol. 7, pp. 64460 – 64470, May. 2019.
[4] .Ruimin Ke, Zhibin Li, Sung Kim, John Ash, Zhiyong Cui, Yinhai Wang, "Real-time bidirectional traffic flow parameter estimation from aerial videos,” IEEE Transactions on Intelligent Transportation Systems, vol. 18, pp. 890 - 901 Aug. 2016.
[5] .Tsuyoshi Idé, Takayuki Katsuki, Tetsuro Morimura, Robert Morris, “City-wide traffic flow estimation from a limited number of low-quality cameras,” IEEE Transactions on Intelligent Transportation Systems, vol. 18, pp. 950 - 959, Apr. 2017
[6] .Honghong Yang, Shiru Qu, “Real-time vehicle detection and counting in complex traffic scenes using background subtraction model with low-rank decomposition,” IET Intelligent Transport Systems, vol.12, pp. 75 - 85, Jan. 2018.
[7] 張家峻，2018，基於圖像檢索之觀賞魚識別系統，國立高雄科技大學，碩士論文.
[8] .N. Dalal, B. Triggs, “Histograms of oriented gradients for human detection,” IEEE Computer Society Conference on Computer Vision and Pattern Recognition, vol. 1, 2005.
[9] .Cortes, C. and V. Vapnik, Support-Vector Networks. Machine Learning, 1995. 20(3): p. 273-297.
[10].Tom Fawcett, “An introduction to ROC analysis,” Pattern Recognition Letters, vol. 27, pp. 861 – 874, 2006