Abstract:

In order to solve the problem that the appearance size is similar, the actual vehicle type is different, this paper proposes the classification method of vehicles by identifying the number characters of bus body load. First of all, the two-value graph combined with the character’s format tower features to achieve the fine positioning of the text area, and the split-out number characters are identified by using a neural network, the result eventually translates the identified result into the corresponding bus category for model classification. The algorithm is tested on the samples of three-class or four-class buses collected at the dot, with a comprehensive recognition rate of 88.5%, which is nearly 10 percentage points than the model recognition based on the geometric features of the appearance, and the recognition rate based on the Psauvola algorithm is more than double compared with other binarization algorithms for character recognition.

Key words: binarization, area detection, character segmenting, character recognition, Psauvola

摘要：

针对外观尺寸相近，处于分类区间边缘的客车车型识别率不高的问题，提出通过识别车身限载数字字符进行车型分类的方法。由二值图结合字符的格式塔特征实现文字区域精定位，将分割出的数字字符使用神经网络进行识别，最终将识别的结果对应转换成相应客车类别以实现车型分类。将该算法在道口采集的三类、四类客车样本上进行实验，综合识别率为88.5%，相比基于外观几何特征的车型识别，识别率提高了将近10个百分点，且基于改进二值化算法（Psauvola）的字符识别相比使用其他二值化算法，识别率提升了一倍多。

关键词: 二值化, 区域检测, 字符分割, 字符识别, Psauvola