摘要
自然场景下复杂多变的影响因素给车牌检测带来困难,为检测并定位自然场景下移动车辆的车牌区域,通过分析信息融合和多类特征提取的特点,提出基于多类别特征信息融合的车牌检测方法。该算法在两种不同场景数据集上的测试效果验证了信息融合和多类特征提取能显著提高车牌检测的检测率和场景鲁棒性。
Due to the variable influence factors, detecting license plates in natural scenes is difficult. To detect and locate the mo- bile vehicle license plate area in natural scenes, based on the characteristics of information fusion and multi-class feature extrac- tion, a license plate detection algorithm based on the multi-class information fusion was proposed. Performances were significant- ly improved in two different scene datasets using the proposed algorithm. The experimental results show that the information fu- sion and the multi-class feature extraction significantly improve the license plate detection performance and robustness.
出处
《计算机工程与设计》
北大核心
2015年第1期250-253,共4页
Computer Engineering and Design
基金
国家自然科学基金项目(60901078)
郑州市科技领军人才基金项目(10LJRC189)
关键词
车牌检测
信息融合
卷积神经网络
颜色检测
文本验证
license plate detectiom information fusion
CNN
color checking
text verification
