摘要
提出一种检测城市和高速公路环境中字符型交通标志的新方法。首先,在输入图像中分割出蓝色和墨绿色区域,用形态滤波和形状标记图判断交通标志的候选区域;其次,将候选区域的彩色图像灰度化,用Otsu算法计算候选区域灰度分布直方图的阈值,并对其进行分割,得到包含字符的二值图像;然后,将候选区域的二值图像向垂直方向上投影,用3次样条拟合算法对其进行拟合,利用曲线的性质,找到拟合曲线中的局部极小值点,分割出包含字符条形区域;最后,将条形区域向水平方向上进行投影和曲线拟合,查找局部极小值点并分割出单个字符区域,再进行形态过滤,分割并定位交通标志中的字符。实验结果表明:该算法的字符查全率高于84%,准确率超过92%。
A novel detection algorithm for traffic signs with character in the urban and highway scenes was proposed. Firstly, blue and green regions were segmented from the input image, and candidates of traffic sign were obtained by morphology filter and signature of shape. Then, color image of candidate was coverted to gray image, and binary images contain characters were obtained by segmentation with threshold of gray distribution histogram computed by Otsu algorithm. Thirdly, the binary image of candidate was projected to vertical orientation and fitted a curve with cubic spline interpolation. The local minimum points were found using the nature of the curve, and the bar region with character rows were segmented from binary image. Finally, bar region was projected to horizontal orientation and fitted a curve, the local minimum points were found to segment the single character regions. The characters in traffic sign were segmented and localized by morphology filter. The results indicate that recall rate of character exceeds 84%, and precision rate is more than 92%.
出处
《中南大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2013年第5期1861-1868,共8页
Journal of Central South University:Science and Technology
基金
国家自然科学基金资助项目(90820302
60805027)
国家博士点基金资助项目(195470)
湖南省院士基金资助项目(20010FJ4030)
湖南省自然科学基金资助项目(12JJ6058)
关键词
交通标志检测
字符定位
曲线拟合
标记图
traffic sign detection
character localization
curve fitting
signature
