摘要
研究了常用的28类机动车和非机动车的车符识别问题.针对手绘车符不规则形变和样本间相似度高的特点,提出一种基于弹性网格方向梯度直方图(histograms of oriented gradients,HOG)特征提取方法,该方法利用了水平与垂直投影直方图较好地表征图符的全局特征、局部特征及HOG特征,结合弹性网格吸收形变的能力,使得在现有的训练样本库中总识别率达94.8%,在手写数字等3个不同领域的手绘图符样本库中,平均识别率达92.5%.实验结果表明,该方法对手绘车符识别具有较高的识别率,对其他手绘图符也有较好的识别效果.
This paper focus on the problem of recognizing 28 kinds of motored and non-motorized hand-drawn vehicle symbols which are commonly used in traffic accident description. To solve the problems of the irregular transformation of sketch vehicles and similar- ity among different vehicles during recognition, an elastic meshes HOG feature is proposed. In this feature, horizontal and vertical pro- jection histogram which is a good choice for global features is used to calculate the elastic meshes,and then elastic meshes combine with HOG feature instead of HOG's cells for absorbing deformation. The total recognition rate reaches 94.8% in our set of samples, and 92.5 % for different areas of the pen digits, common diagram shapes, and electrical circuit symbols. The experimental results show that:our proposed algorithm has a high recognition rate for the hand-drawn vehicle symbols and it has a good promotion prospects.
出处
《厦门大学学报(自然科学版)》
CAS
CSCD
北大核心
2013年第4期466-472,共7页
Journal of Xiamen University:Natural Science
关键词
手绘车符识别
弹性网格
方向梯度直方图
投影直方图
特征提取
hand-drawn vehicles recognition
elastic meshes
histogram of oriented gradient
projection histogram
feature extract