基于L1-BRD的视频图像车牌识别  

Multi-frame license plate recognition method by using L1-BRD

作　　者：薛云松[1] 冯国瑞[1] 刘立庄[2] 

机构地区：[1]上海大学通信与信息工程学院,上海200444 [2]中国科学院上海高等研究院,上海201210

出　　处：《电视技术》2017年第1期73-78,共6页Video Engineering

基　　金：国家自然科学基金项目(61373151;U1536109);上海市自然科学基金项目(13ZR1415000);上海市教委创新基金项目(14YZ019)

摘　　要：针对实时视频的车牌识别系统应用,利用车牌的边缘特征和形态学操作对车牌进行粗定位,通过支持向量机(SVM)结合方向梯度直方图特征(HOG)方法对粗定位车牌进行鉴别真伪。在对字符进行分割后,取连续多帧车牌,对字符特征利用L1-BRD(L1-norm Bin Ratio-Based Histogram Distance)自适应计算融合权重,使L1-BRD能够从图像间特征相似度评估扩展到实时视频中。把L1-BRD用于车牌字符识别,可以降低单帧图像中噪声及字符分割误差产生的影响,有助于后期基于多帧加权直方图特征进行字符识别,提高车牌字符识别的准确性和稳定性,并保证了实时性。For the application of license plate recognition system based on real-time video, this paper firstly locates the license plates in coarser-grained way by using the edge features and morphological operations. Then it locates the license plates exactly by combining SVM（ Support Vector Machine） and HOG（ Histogram of Oriented Gradient） features, after segmenting characters, it se- lects a series of plates and calculates weights adaptively by using L1-BRDs, making L1-BRD be used in the real-time video for li- cense plate character recognition from measuring feature＇ s similarity to a single image, it can reduce the effects because of noise and character segmentation in single-frame image, contribute to characters recognition by multi-frames weighted histogram fea- tures, this method can improve the accuracy, stability and instantaneity.

关 键 词：车牌字符识别 L1-BRD 车牌定位 多帧加权直方图特征 SVM 

分 类 号：TN911.73[电子电信—通信与信息系统]