摘要
车牌检测是车牌识别系统中关键的第一步,传统的车牌检测算法存在大角度下车牌检测精度不高的问题。East算法将车牌检测看成场景文本检测问题来处理,直接预测图像中任意角度的车牌位置和倾斜信息。采用改进的East网络对车牌进行检测,特征提取采用PVANet来加速运算,整个网络端到端训练,有效避免不必要的中间步骤,直接得到车牌区域。实验结果表明,构建的车牌检测算法具有较好的检测效果,可以准确检测大角度倾斜车牌的位置坐标信息,提高检测精度。
License plate detection is the key first step in the license plate recognition system, the traditional license plate detection algorithms have the problem that the detection accuracy of the license plate is not high under large angle. The East algorithm treats the license plate detec-tion as a scene text detection problem, and the license plate position and tilt information at any angle in the image is directly predicted. The improved East network is used to detect the license plate, the feature extraction uses PVANet to accelerate the calculation, the whole net-work performs end-to-end training, which effectively avoids unnecessary intermediate steps and directly obtains the license plate area. The experimental results show that the constructed license plate detection algorithm has better detection effect, it can accurately detect the position coordinate information of the large-angle tilted license plate, and improve the detection accuracy.
作者
杨飚
周文婷
YANG Biao;ZHOU Wen-ting(Beijing Key Laboratory of Urban Road Traffic Intelligent Control Technology, North China University of Technology, Beijing 100144)
出处
《现代计算机》
2019年第12期53-56,共4页
Modern Computer
基金
国家自然科学基金资助项目(No.61374191)
北京市教育委员会科技计划资助项目(No.KM201710009001)
关键词
车牌检测
神经网络
East网络
License Plate Detection
Neural Network
East Network
