論文中，我們將基於Kinect之靜／動態機車車牌辨識系統分為：深度資訊提取、車牌定位、文字辨識三個階段處理。
首先主要提取Kinect所偵測前方機車車輛的深度資訊，即可解決複雜背景等問題，並且再利用演算法將多機車之影像依序分割並提取為單一機車之影像進行後續處理。接著利用Marker Detection搜尋方框物件的特性定位車牌，而此演算法優於可即時偵測並定位以及擁有良好的抗旋轉、變型、歪斜與遮蔽等強健性。
最後將定位車牌傾斜校正與字元切割後，並給予樣板比對進行字元辨識。待測之機車車牌分為靜態與動態兩種狀況，動態為行進間拍攝前方行駛中機車車輛進行偵測，而靜態則是以行進間偵測路邊停放之車輛。而本系統以靜態偵測為主要研究目的。

In this paper, we divide the system into three parts: the extraction of the depth information, the orientation of the license plate and the identification of the words.
First, we exact the depth information of the license plate on the motorcycle in front of the Kinect to solve the problem of the complex background. Then we use the algorithm to divide the image of the motorcycles sequentially and abstract the image of one motorcycle to do the following process. After that, we apply the Marker Detection Algorithm with the characteristic of searing the square object to orientate the license plate. The algorithm can not only detect in the real-time but also have the robust resistance to rotation, deform, slope and shelter, etc.
	Finally, we correct the slope of the retaining license plate and divide the words. We base on the sample to identify the words. The training motorcycle license plates divide into the static and dynamic state. The dynamic state is to detect the driving motorcycle license plate under moving. But, the static state is to detect the parking motorcycle in the roadside under moving. However, our system’s main purpose is the static detection.

致謝 I
中文摘要 II
英文摘要 III
目錄 IV
圖目錄 VII
表目錄 IX
第1章緒論 1
1.1研究背景 1
1.2研究動機 3
1.3研究目的 4
1.4論文架構 5
第2章相關技術與研究 6
2.1 相關技術 6
2.1.1 KINECT 6
2.1.2深度影像建立方法 8
2.1.3彩色模型的轉換 12
2.1.4中值濾波器 14
2.1.5形態學(Morphology) 14
2.2車牌定位相關研究 17
2.3車牌文字辨識研究 28
2.3.1 分類器分類法 28
2.3.2 統計分類法 30
2.3.3 樣板比對法 31
第3章多車牌定位技術 33
3.1 系統架構 33
3.2 Kinect架設與輸出之影像 34
3.3影像前處理 35
3.3.1 車輛偵測 35
3.3.2 影像二值化 37
3.3.3侵蝕影像 38
3.3.4區域分割 39
3.4車牌定位 41
3.5校正傾斜車牌 45
3.6字元切割 46
3.6.1 水平投影 46
3.6.2 垂直投影 47
3.7字元辨識 48
3.7.2 文字辨識步驟 52
第4章實驗結果 53
4.1 實驗環境 53
4.2 車牌定位實驗 54
4.2.1 複雜背景之多車牌定位 55
4.2.2 傾斜車牌之定位 56
4.2.3 髒污與遮蔽車牌之定位 57
4.2.4 不同時間點之車牌定位 58
4.2.5 車牌定位效果分析 59
4.3 車牌文字分割實驗 60
4.4車牌辨識實驗結果 62
第5章結論與未來展望 71
 
圖目錄
圖1.1含有排氣檢驗合格標籤與保險標章等的車牌 5
圖2.1 Kinect................................................................................................ 6
圖2.2 Kinect之架構圖引用於PrimeSensor[7] ........................................ 7
圖2.3 Kinect鏡頭輸出之彩色圖像與深度圖.............................................. 8
圖2.4雙眼三角測距[8]................................................................................. 9
圖2.5 結構光掃描[9].................................................................................. 10
圖2.6 Light Coding技術示意圖[11] ...........................................................11
圖2.7 3x3的結構元件................................................................................ 16
圖2.8 H. F. Zhang等人[14]的系統流程..................................................... 18
圖2.9 S. W. PARK等人[16]的系統流程.................................................... 19
圖2.10 分割區塊........................................................................................ 21
圖2.11 邊緣檢測......................................................................................... 21
圖2.12 邊緣搜尋........................................................................................ 23
圖2.13合併區域線條................................................................................. 24
圖2.14 合併後線條.................................................................................... 25
圖2.15 定位角點........................................................................................ 26
圖2.16 定位方框區域................................................................................ 26
圖2.17類神經網路架構[20]....................................................................... 29
圖2.18結合統計是與結構是兩階段混合辨識系統[21] .......................... 30
圖3.1 系統流程圖...................................................................................... 33
圖3.2 Kinect架設示意圖............................................................................ 34
圖3.3 行車中的複雜背景.......................................................................... 35
VIII
圖3.4 Kinect輸出之深度影像.................................................................... 36
圖3.5深度影像二值化............................................................................... 37
圖3.6侵蝕後深度影像............................................................................... 38
圖3.7 地面雜訊.......................................................................................... 39
圖3.8 緊密且多車輛資訊.......................................................................... 39
圖3.9垂直投影圖....................................................................................... 40
圖3.10垂直投影取閥值............................................................................. 40
圖3.11機車區域分割.................................................................................. 41
圖3.12車牌定位流程圖............................................................................. 42
圖3.13 Marker Detecetion搜尋方框物件結果.......................................... 44
圖3.14校正後車牌..................................................................................... 46
圖3.15影像邊界點累積量的水平投影..................................................... 47
圖3.16車牌文字切割結果。..................................................................... 48
圖3.17字元結構的區域示意圖................................................................. 50
圖3.18字元尺寸為45x95 的原始字元樣板............................................ 51
圖4.1 微軟發售的Kinect........................................................................... 53
圖4.2為複雜背景環境之校正與定位....................................................... 55
圖4.3 傾斜車牌校正之定位...................................................................... 56
圖4.4髒污與遮蔽車牌之定位................................................................... 57
圖4.5 不同時間點之車牌定位.................................................................. 58
圖4.6 車牌定位失敗之影像...................................................................... 59
圖4.7 字元分割結果.................................................................................. 60
圖4.8 車牌字元切割失敗.......................................................................... 61 

表目錄
表3.1 Kinect深度影像不同距離的灰階值變化........................................ 37
表4.1文字辨識結果................................................................................... 63
表4.2時速5~10公里之車牌辨識率......................................................... 65
表4.3時速10~15公里之車牌辨識率....................................................... 66
表4.4時速15~20公里之車牌辨識率....................................................... 67
表4.5各時速之車牌整體辨識率............................................................... 68
表4.6時速5~10公里之車牌辨識之成功率............................................. 68
表4.7時速10~15公里之車牌辨識之成功率........................................... 69
表4.8時速15~20公里之車牌辨識之成功率........................................... 69

[1]梁智凱，多車牌辨識系統之研究，淡江大學電機工程學系碩士論文，民國九十五年。
[2]莊祥琳，動態車牌辨識系統之研究，淡江大學電機工程學系碩士論文，民國九十六年。
[3]陳祥傑，自動化車牌辨識系統設計，國立中央大學電機工程研究所，民國九十四年。
[4]Yan-qing Zheng, Dai-ping Li, Shu-wen Zhang， “License Plate Location Based on Combinatorial Feature”, IEEE, Guangdong University of Technology Guangzhou, 2011.
[5] Shaohong Wu, Yunkuan Wang, Tao Sun, “A Novel Accurately Automatic License Plate Localization Method,” IEEE, Institute of Automation Chinese Academy of Sciences, 2011.
[6] http://www.xbox.com/zh-TW/kinect
[7] http://www.primesense.tw/?p=741
[8]張量鈞，基於立體視覺之手掌位移辨識系統，淡江大學電機工程學系碩士論文，民國九十九年。
[9] http://www.stockeryale.com/i/lasers/structured_light.htm
[10]Pasi Palojarvi “Integrated electronic and optoelectronic circuits and devices for pulsed time-of-flig6ht laser range finding,” Department of Electrical and Information Engineering and Info tech Oulu, University of Oulu, 2003.
[11]http://www.bb.ustc.edu.cn/jpkc/guojia/dxwlsy/kj/part2/grade3/LaserSpeckle.html

[12] Albitar, I.C., Graebling, P., Doignon, C., “Robust Structured Light Coding for 3D Reconstruction” IEEE 11th International Conference on Computer Vision, p.1-6.2007.
[13] 王振興， 多標的汽機車車牌辨識系統之研究，私立元智大學，資訊管理所碩士論文，民國九十四年。
[14] H. Zhang, W. Jia, X. He, and Q. Wu, “A Fast Algorithm for License Plate Detection in Various Conditions,” in IEEE International Conference on Systems, Man, and Cybernetics, manuscript received 2006.
[15]Y. Freund, “An Adaptive Version of the Boost by Majority Algorithm,” Machine Learning, 2001, pp. 293-318.
[16] S. W. Park, S. C. Hwang, and J. W. Park, “The Extraction of Vehicle License Plate Region Using Edge Directional Properties of Wavelet Subband,” in IEICE Trans. on Information & System, vol. E86-D, no. 3, 2003, pp. 664-669.
[17] Z. B. Huang, and Y. F. Guo, “Classifier Fusion-Based Vehicle License Plate Detection Algorithm,“ in IEEE International Conference on machine Learning and Cybernetics, vol. 5, 2003, pp. 2984-2989.
[18] Martin Hirzer, Marker Detection for Augmented Reality Applications, 2008.
[19] Donald F.-S., Journal Neural Networks, Elsevier Science Ltd. Oxford, UK, vol.3, Issue 1, 1990.
[20] Gao, Q., X. Wang, and G. Xie, “License plate recognition based on prior knowledge,” in Proc. Int. Conf. on Automation and Logistics, Jinan, China, 2007, pp.2964-2968.

[21] Pan, Xiang, Ye Xiuzi, and Zhang Sanyuan, “A hybrid method for robust car plate character recognition,” in Proc Int. Conf. on Systems Man and Cybernetics, Hague, Netherlands, 2004, pp.4733-4737.
[22] Xu, L., A. Krzyzak, and C. Y. Suen, “Methods of combining multiple classifiers and their application to handwriting recognition,” IEEE Trans. Systems Man and Cybernetics, vol.22, no.3, 1992, pp.418-435.
[23] Trier, O. D., A. K. Jain, and T. Taxt, “Feature extraction methods for character recognition - a survey,” Pattern Recognition, vol.29, no.4, 1996 pp.641-662.
[24] Han, P., H. Wei, D.-F. Wang, and Y.-J. Zhai, "Car license plate feature extraction and recognition based on multi-stage classifier," in Proc. Int. Conf. on Machine Learning and Cybernetics, Hebei, China, 2003, pp.128-132.
[25] Lu, Y., "On the segmentation of touching characters," in Proc. 2nd Int. Conf. on Document Analysis and Recognition, Tsukuba Science City, Japan, 1993, pp.440-443.
[26] Hegt, H. A., R. J. Haye, and N. A. Khan, “A high performance license plate recognition system,” in Proc. Int. Conf. on Systems, Man, and Cybernetics, San Diego, CA, 1998, pp.4357-4362.
[27] Duan, T.-D., T.-L. Du, T.-V. Phuoc, and N.-V. Hoang, "Building an automatic vehicle license-plate recognition system," in Proc. Int. Conf. Computer Science, Can Tho, Vietnam, Feb.21-24, 2005, pp.59-63.
[28] Comelli, P., P. Ferragina, M. N. Granieri, and F. Stabile, "Optical recognition of motor vehicle license plates," IEEE Trans. Vehicular Technology. vol.44, no.4, pp.790-799, Nov. 1995.
[29] 陳客智，照像手機的車牌偵測與辨識，國立中央大學資訊工程研究所碩士論文，民國一百年。