一种字符组合词包模型的少约束车牌定位算法

A Less Constrained License Plate Location Algorithm Based on the Bag of Words Model Using Combination of Characters

针对如何在复杂背景、拍摄角度和车牌尺度发生变化等少约束条件下实现基于字符组合词包模型的车牌定位算法。本文首先构造包含车牌数字字符和英文字符的数据库;然后利用本算法识别提取车牌字符的SIFT特征,并精准计算识别特征点在字符识别区域的相对位置、物理方向等信息组成视觉识别词汇;最后把本车牌字符的视觉词汇聚合后搭建车牌字符视觉词包数据库。在识别阶段,提取待识别图像SIFT特征与视觉词包中的视觉词汇进行匹配,并聚合所有有效投票位置来实现车牌区域的准确识别定位。仿真结果表明,本文算法对于背景复杂下的车牌区域定位具有较好的效果。

Aiming at how to realize the less constrained license plate location algorithm based on character combination word packet model under the conditions of complex background, camera angle and license plate scale changing.Firstly, a database containing license plate number characters and English characters is constructed;Then, SIFT features of license plate characters are recognized and extracted by this algorithm,The relative position of the recognition feature points in the character recognition area is calculated accurately,Physical direction and other information constitute the visual recognition vocabulary, and then the visual vocabulary of the license plate characters is aggregated together to build the license plate character visual word pack database. In the recognition stage, SIFT features of the image to be recognized are extracted to match the visual words in the visual word pack, and all the valid voting positions are aggregated to realize the accurate recognition and location of the license plate area. The simulation results show that this algorithm has a good effect on the license plate region location under complex background.