基于多特征协同的交通标志检测  

Traffic Sign Detection Based on Multiple Features Cooperation

作　　者：汤凯[1] 李实英[1] 刘娟[1] 李仁发[1] 

机构地区：[1]湖南大学信息科学与工程学院嵌入式系统与网络省重点实验室,长沙410082

出　　处：《计算机工程》2015年第3期211-217,共7页Computer Engineering

基　　金：国家自然科学基金资助项目(61272396)

摘　　要：针对交通场景图像中交通标志因颜色失真、形状失真和尺度变化导致漏检严重的问题,提出一种颜色特征、形状特征和尺度特征的多特征协同方法。从交通场景图像计算颜色增强图,利用多阈值分割方法和闭合轮廓曲率直方图链码表达,并对提取的形状轮廓的曲率直方图进行尺度归一化处理,融合颜色特征和归一化后的形状特征构成区域的特征向量,采用支持向量机分类获得检测结果。实验结果表明,该算法在较低时间复杂度下,能有效提高交通标志检测精度。A new method is presented to improve traffic sign detection with cooperation of color,shape and scale features,especially under conditions of color distortion,shape deformation and scale variance. Color enhancement maps are generated from traffic scene images. Regions of interest are then extracted from the color enhancement maps using multiple thresholds of color,chain codes of the curvature histograms of closed contours are calculated and scale normalized for the contours. The Support Vector Machine（SVM）classifier is applied to classify the chain codes of the extracted traffic signs and the template signs. Experimental results demonstrate that this method is capable of improving traffic sign detection,with low time complexity.

关 键 词：交通标志 多特征协同 多阈值 曲率链码 尺度归一化 支持向量机分类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]