雾霾环境下车牌图像预处理及识别算法研究

发布时间：2018-05-11 11:14

本文选题：雾霾环境 + 车牌识别　；参考：《郑州大学》2017年硕士论文

【摘要】：随着智能交通系统的发展,车牌自动识别技术越来越广泛地应用于生活中的各种场景。但是由于现今雾霾天气的增多,传统的车牌识别算法在雾霾天气下的准确率会大幅度下降,很难满足人们的需求。这就急需在车牌自动识别过程中加入去雾算法,提高雾霾条件下的车牌自动识别准确率。本文在车牌自动识别算法中引入暗原色先验去雾算法,同时利用指导滤波对暗原色先验去雾算法中透射率优化的方法进行改进,在保证去雾效果的同时缩短了去雾过程的时间,提高了去雾算法的实时性。对去雾后得到的图像,先进行灰度化处理,然后进行区域增强,最后利用边缘检测的方法来确定车牌的上下边界。接着,利用基于先验知识的方法确定车牌的左右边界,完成车牌的定位。对得到的定位后的车牌图像先进行二值化处理,再利用垂直投影法,通过垂直方向的像素累计图进行字符分割,然后对分割后的车牌图像进行归一化处理,最后将归一化的图像转化为粗特征矩阵,以便进行车牌识别。由于BP神经网络抗干扰性差,其在识别去雾后的字符图像时准确率下降,本文选用鲁棒性强的径向基函数(Radial Basis Function,RBF)神经网络对去雾字符图像进行识别。但RBF神经网络参数确定较为复杂,偶然性大,故选用粒子群优化算法(Particle Swarm Optimization,PSO)对其参数进行优化。实验证明,利用基于粒子群算法优化的径向基函数(Radial Basis Function Optimized by Particle Swarm Optimization,PSO-RBF)神经网络对字符进行识别可以有效提高车牌识别的准确率。大量实验结果证实,本文算法可以有效提高雾霾条件下的车牌识别准确率,同时保证车牌识别的实时性。
[Abstract]:With the development of intelligent transportation system, license plate recognition technology is more and more widely used in all kinds of scenes in life. However, because of the increasing fog and haze weather, the accuracy of the traditional license plate recognition algorithm will decrease greatly in haze weather. It is difficult to meet the needs of people. In this paper, the dark original color priori fog removal algorithm is introduced in the automatic recognition algorithm of the license plate. At the same time, the method of improving the transmittance optimization of the dark original color prior fog algorithm is improved by using the guiding filtering, and the time of the fog removal process is shortened and the time of the fog removal is shortened, and the time of the fog removal process is shortened. The image of the fog removal is real-time. The image obtained after the fog is gray, then the region is enhanced. Finally, the edge detection method is used to determine the upper and lower boundary of the license plate. Then, the left and right boundary of the license plate is determined by using the prior knowledge, and the location of the license plate is completed. First, the license plate image after the location is first obtained. Two value processing is carried out, and then the vertical projection method is used to divide the characters through the vertical pixel accumulative graph, and then the segmented license plate image is normalized. Finally, the normalized image is converted into a rough feature matrix to carry out the license plate recognition. Because of the poor anti-interference ability of the BP God channel network, the character is identified after the fogging character. The accuracy rate of the symbol is decreased. In this paper, the robust radial basis function (Radial Basis Function, RBF) neural network is used to identify the fog character images. But the parameters of the RBF neural network are more complex and the chance is larger, so the particle swarm optimization (Particle Swarm Optimization, PSO) is selected to optimize the parameters. The experiment proves that the parameters are optimized. Using the radial basis function (Radial Basis Function Optimized by Particle Swarm Optimization, PSO-RBF) neural network for character recognition can effectively improve the accuracy of the license plate recognition. A large number of experimental results confirm that the algorithm can effectively improve the accuracy rate of license plate recognition under the haze condition. The real-time performance of the license plate recognition is guaranteed.

【学位授予单位】：郑州大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】