基于深度学习的车牌识别系统设计

Design of License Plate Recognition System Based on Deep Learning

随着国内机动车辆数目的不断增加,如何对众多的机动车进行有效管理已成为当前交通管理机构面临的主要问题。利用深度学习技术,通过对车牌定位、车牌字符分割和车牌字符识别技术进行研究,提出了一种车牌识别原型系统方案。在车牌预处理模块,通过图像灰度化处理等一系列操作,抑制了非车牌区域的噪声;在车牌定位模块,提出使用基于深度学习的目标检测方法对车牌进行定位,进行二值化、倾斜校正后使用垂直投影法分割出车牌字符,最后通过改进的Hausdorff距离计算待识别图像与模板之间的相似程度,利用模板匹配的方法识别出车牌字符。实验结果显示,该系统车牌识别准确率高。

With increasing of vehicles in domestic,howto effectively manage a large number of motor vehicles has become the main problem in the current traffic management institutions.In this paper,we propose a prototype system of license plate recognition through the research of license plate location,license plate character segmentation and recognition by using deep learning technology.In the plate preprocessing module,a series of operations,such as image grayscale processing,are used for denoising in the non-license plate area.In the plate location module,a target detection method based on deep learning is adopted.After the binarization and tilt correction,license plate characters are segmented by vertical projection.Finally,the similarity between the image and the template is calculated by the improved Hausdorff distance,and the license plate characters are recognized by the template matching method.The experiment shows that the accuracy of license plate recognition system is higher than others.