監控以及攝影系統在現行社會的使用極為廣泛。舉凡路口監視系統、閉路電視防盜攝影系統，抑或是當今車輛所配備的行車輔助系統等，以及大多數車輛駕駛人都會裝設的行車記錄器。電腦影像所提供人類的輔助應用多元，而對於此方面的研究、論文也不在少數。
本論文著重於提供機器視覺影像以協助車輛駕駛人在雨天駕駛之用。以車內行車記錄器為基礎，面對雨滴落在車輛擋風玻璃上造成影像的模糊及混亂，利用影像處理工具消除上述的干擾。並還原為較清晰的影像，提供駕駛人後續記錄、採證、甚至應用於發展即時行車輔助影像等的運用。
本論文參考文獻所提供偵測雨滴之方法進行實驗。利用門檻值濾波器、梯度檢、形態學之處理擷取雨滴之資訊。實驗過程中發現部分雜訊殘留及特殊形態之影像邊界會造成誤判。因此本論文修正文獻設定之參數，並加入包含形態前處理之流程與原流程相比較。由此得知各有其優點，遂將兩種流程所得之結果結合達到互補之作用。最後將偵測到之雨滴移除後再利用近似背景之方式還原影像，以提供駕駛者較清晰之影像。

The monitoring video camera systems have been widely applied in our daily life. Typical examples include the surveillance system at intersections, the anti-theft CCTV security system, the drivers’ assistance system, and the event data recorder, which already became the most popular device installed on current vehicles. Machine vision is apparently capable of providing supplementary image information for versatile applications in real world. As a result, researches on this specific field of machine vision have been growing rapidly.
This thesis focuses on the implementation of machine vision for drivers in rainy days. Based on the event data recorder inside the vehicle, develop image processing techniques to deal with the blurred image caused by raindrops on windshield and to restore the image to a clearer status. This achievement provides a more efficient way for event recording and evidence collection, and can also be extended to real-time applications for assisting driving vehicles.
Experiments were carried out by adopting methods for raindrop detection presented in literatures. The techniques including the threshold filter, the gradient detection, and morphological processing were applied to acquire information of raindrops. Experimental results indicate that residual noises and some forms of image boundary can bring about false positives. In this thesis, therefore, parameters modification and morphological preprocessing are added. Comparing with the original process, each approach has its own advantages. Complementary performance can be obtained by combining those two processes. Finally, removal of detected raindrops is followed by image recovery according to background pixels so that a better image can be provided for the drivers.

目錄
論文審定書 i
摘要 ii
Abstract iii
目錄 v
圖次 vi
第一章　緒論 1
1.1 研究動機與目的 1
1.2 文獻回顧 3
1.3 論文架構 4
第二章　影像處理工具[2] 5
2.1 梯度檢測 5
2.2 形態學處理 7
2.3 閉合運算 9
2.4 標籤化與型態鑑別 10
第三章　影像處理遭遇問題與實驗環境 13
3.1 透鏡效應與光源影響 13
3.2 實驗環境與設備 15
第四章　實驗過程與結論分析 16
第五章　論文貢獻與未來展望 30
5.1 論文貢獻 30
5.2 未來展望 30
參考文獻 32
 
圖次
圖1.1 擋風玻璃上的雨滴 2
圖1.2 移動中雨刷的干擾 2
圖2.1 座標關係圖 5
圖2.2 邊緣檢測之示意圖 6
圖2.3 影像形態學之膨脹處理 8
圖2.4 影像形態學之侵蝕處理 8
圖2.5 閉合運算結構種類 9
圖2.7 標籤化等值表範例 11
圖2.8 物件尺寸大小示意圖 12
圖3.1 光線經過透鏡之折射效應 13
圖3.2 雨滴折射光線後在擋風玻璃上呈現不同的形象 14
圖3.3 車燈與紅綠燈和雨滴極為相像 14
圖3.4 DVD Video Soft Free Studio & Mathworks.Matlab.R2010b 15
圖4.1 影像處理流程圖 16
圖4.2 平均連續六張影像所得之結果 17
圖4.3 經過梯度運算後所得垂直與水平方向之邊界影像 18
圖4.4 邊緣檢測後之完整邊界影像 18
圖4.5 梯度濾波過後所得之強化梯度影像 19
圖4.6 經強化梯度濾波所得之結果 20
圖4.7 閉合運算填補空洞後之影像 21
圖4.8 閉合運算R=3之處理結果 22
圖4.9 閉合運算R=3與R=2之比較 22
圖4.10 閉合運算R=2與R=1之比較 23
圖4.11經標籤化後各獨立區塊皆有相同之編號 23
圖4.12 經尺寸濾波器處理之影像 24
圖4.13 經形狀濾波器處理之影像 25
圖4.14 偏心率之判別 25
圖4.15 消除雨滴之影像 26
圖4.16 雨滴還原計算示意圖 26
圖4.17 影像還原之結果 27
圖4.18 影像處理結果範例一 27
圖4.19 影像處理結果範例二 28
圖4.20 影像處理結果範例三 28
圖5.1 地面積水反射對向車道之車燈 31

[1] 國際能源組織IEA ，http://www.iea.org/。
[2] 交通部公路總局，http://www.thb.gov.tw/TM/Default.aspx。
[3] 交大智慧視覺晶片系統設計實驗室，http://ivs.ee.nctu.edu.tw/~jiguo/JIGUO_main.htm。
[4] 陳冠名，具深度資訊之道路辨識影像處理技術，國立中山大學機械與機電工程學系碩士論文，中華民國100年7月。
[5] R. Jain, R. Kasturi, and G. Schunck, Machine Vision, Mc Graw-Hill, Inc., USA, 1995.
[6] A. Cord and D. Aubert, ”Towards rain detection through use of in-vehicle multipurpose cameras,” 2011 IEEE Intelligent Vehicles Symposium (IV) Baden-Baden, Germany, June 5 - 9, 2011, pp. 833 - 838.
[7] H. Kurihata, T. Takahashi, Y. Mekada, I. Ide, and H. Murase, ”Raindrop detection from in-vehicle video camera images for rainfall judgment,” Proceedings of the First International Conference on Innovative Computing, Information and Control (ICICIC), Beijing, China, Augusut 30 - September 1, 2006, pp. 544 - 547.
[8] M. Roser and A. Geiger, ”Video-based raindrop detection for improved image registration,” 2009 IEEE 12th International Conference on Computer Vision Workshop, Kyoto, Japan, September 27 - October 4, 2009, pp. 570 - 577.
[9] J. C. Halimeh and M. Roser, “Raindrop detection on car windshields using geometric-photometric environment construction and intensity-based correlation,” 2009 IEEE Intelligent Vehicles Symposium(IV), Xi'an, China, June 3 - 5, 2009, pp. 610 - 615.
[10] H. Bay, T. Tuytelaars, and L. Van Gool, ”SURF: Speeded up robust features,” Lecture Notes in Computer Science, vol. 3951, 2006, Springer Berlin / Heidelberg, pp. 404 - 417.
[11] M. Zhou, Z. Zhu, R. Deng, and S. Fang, “Rain detection and removal of sequential images,” 2011 Chinese Control and Decision Conference(CCDC), Mianyang, China, May 23 – 25, 2011, pp. 615 - 618.
[12] M. Shen and P. Xue, “A fast algorithm for rain detection and removal from videos,” 2011 IEEE International Conference Multimedia and Expo (ICME), Barcelona, Spain, July 11 - 15, 2011, pp. 1 – 6.
[13] Y. Miao, H. Hong, and H. Kim, “Size and angle filter based rain removal in video for outdoor surveillance systems,” 2011 8th Asian Control Conference (ASCC), Kaohsiung, Taiwan, May 15 - 18, 2011, pp. 1300 – 1304.