色彩學習的交通標誌偵測與辨識;Traffic Sign Detection and Recognition with Color Learning

NCU Institutional Repository > 資訊電機學院 > 資訊工程研究所 > 博碩士論文 > Item 987654321/65672

jsp.display-item.identifier=請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/65672

题名:	色彩學習的交通標誌偵測與辨識;Traffic Sign Detection and Recognition with Color Learning
作者:	廖政權;Liao,Jheng-cyuan
贡献者:	資訊工程學系
关键词:	先進駕駛輔助系統;交通標誌偵測;交通標誌辨識;三維凸包;八分樹;樣板比對;支援向量機;Advanced driver assistance systems;Traffic sign detection;Traffic sign recognition;3D convex hull;Octree;Template matching;Support vector machine
日期:	2014-07-28
上传时间:	2014-10-15 17:07:59 (UTC+8)
出版者:	國立中央大學
摘要:	本研究目的是以電腦視覺技術完成一個完整的交通標誌偵測與辨識系統，以協助駕駛人正確辨識出交通標誌。可以有效的減少駕駛人注意力的負擔，以達到降低交通事故的目的。顏色是偵測交通標誌的主要依據，不同一般直接定義顏色範圍的方式，我們以實際拍攝交通標誌樣本學習標誌色彩，以三維凸包演算法 (3D convex hull algorithm) 填補色彩空間中樣本擷取不足的色彩空缺，並以八分樹 (Octree) 資料結構紀錄標誌的色彩範圍。整個標誌偵測與辨識分成以下五個步驟。首先，透過學習到的色彩範圍擷取標誌色彩之像素，並將該像素連結成獨立區塊。接著，標誌區塊根據最小矩形框的長寬比、面積大小等幾何條件作初步篩選。第三，通過篩選的標誌候選區塊，將正規到固定大小，再與事先定義好的圓形、半圓形、三角形、及矩形做樣板比對，判斷屬於那一類標誌。第四，根據不同顏色與形狀的標誌，我們對其符號所在區域使用群間最大差異法 (Otsu′s method) 進行二值化，再把標誌符號正規到固定的大小。最後，再將符號二值化影像使用支援向量機 (support vector machine, SVM) 進行標誌的辨識。我們在不同日間環境下拍攝諸多實驗影片；並針對含有紅色及藍色標誌的3591張影像，利用我們所提的方法擷取色彩，其擷取的準確度為90.25%。在擷取到的1574個標誌中，交通標誌分類平均正確率為94.26%。經過前兩階段處理後，本系統在1488個標誌中辨識出18種標誌，辨識正確率平均達96.63%。 ;Traffic signs provide drivers with important information and help them to drive more safely and more easily by guiding and warning them and thus regulating their actions. Generally, drivers pay much visual attention to gaining the traffic information. The drivers would be more eased if there is a traffic sign detection and recognition system. In this thesis, we propose a traffic sign detection and recognition system to increase road traffic safety by helping drivers notice the traffic situation on the roads. There are three stages in the proposed system: i. colored sign detection, ii. shape of sign classification, and iii. traffic sign recognition. The detection task is the most difficult due to the variation of colors in different weather conditions. Here we propose a color learning method to extract the proper pixels to detect traffic signs. The color distributions of traffic sign is analyzed in the YCbCr color space. The color distribution is built by a 3D convex hull method and described by an Octree data structure. According to the learned color distribution, candidates of sign are extracted from the image. Then regions of candidates of sign are verified by geometric conditions: size, aspect ratio, and ratio of number of extracted red or blue pixels. We classify the red signs into three classes: circle, semicircle, and triangle by using template matching. Correspondingly, the blue signs are classified into two classes: circle and rectangle. In the sign recognition, the symbols of detected sign are extracted by Otsu’s method and some image processing. Support vector machine (SVM) is employed to recognize the extracted symbols of sign. The proposed systems are evaluated in variant environments. The accuracy of traffic sign with border being red or blue color detection is 90.25%. The average classification rate of traffic sign shapes is 94.26%. The average recognition rate of symbols in speed limit, prohibition, warning, and obligation signs is 96.63%.
显示于类别:	[資訊工程研究所] 博碩士論文

文件中的档案:

档案	描述	大小	格式	浏览次数
index.html		0Kb	HTML	418	检视/开启

在NCUIR中所有的数据项都受到原著作权保护.

社群 sharing

数据加载中.....