一种基于神经网络的车牌定位方法

Neural network based vehicle license plate location

针对过去车牌定位难的问题,提出了一种基于神经网络的车牌定位方法,算法通过神经网络训练、图像预处理以及用训练好的网络进行车牌的定位,依照上述算法对编制的软件检验,从测试的 600 幅 320×240(像素×像素)汽车图像,正确率达到了 95.1%,每幅图像的运行时间小于 2s,基本上达到了实时处理的要求。

Presents an approach based on neural network to locate the license number plate aimed at solving the difficulties of accurate location, which is caused by factors such as complex picture background, variable categories of license plate, versatile colors of license plate and different illumination as a result of weather variation. This approach is to locate license plate through three steps including training the neural network, preprocessing the image and locating the number plate. We have developed a software to test 600 pieces of car pictures whose resolution is 320×240 (pixel × pixel) according to abovementioned approach. Relevant experiment shows that its accuracy rate is 95.1% and that its runtime is within 2 seconds per picture, which has met the requirement for real time processing.