交通肇事逃逸车辆的车牌自动识别算法

摘要

针对交通肇事中智能交通监控系统由于事故车辆快速逃逸导致其车牌难以辨识的问题,研究了智能交通监控系统中如何克服监控图像模糊,实现逃逸车辆的车牌自动识别.根据逃逸车辆和监控摄像设备之间的空间几何关系以及车牌图像的灰度特点,提出了一种融合监控录像视频复原和车牌号码局部HOG特征的交通肇事逃逸车辆的车牌自动识别算法.算法主要包含两个关键步骤:监控录像的复原和车牌号码的自动识别.算法首先将监控录像中的关键帧进行频域分析,得到车辆的逃逸方向和逃逸速度,进而判断造成监控图像模糊的关键参数,并据此对监控图像进行复原；然后通过车牌号码局部HOG特征提取和构建特征字典,并求解字符图像HOG特征的稀疏系数,进而完成车牌的自动识别.实验结果表明,算法能很好地复原交通监控录像中的关键帧,在车牌识别功能上,改进方法与BP神经网络法相比有很高的正确识别率,具有很大的应用价值.%The license plate recognition for abscond vehicle in traffic trouble suffered from the car's rapid escape which can lead to the fuzzy of monitor's recording. In order to complete the automatic recognition for license plate with overcoming the fuzzy of recording in intelligent traffic monitor system, an algorithm was proposed, which include fuzzy record recovering and license plate recognition. According to the geometrical relation between the escape car and the camera and also the gray degree character of the license plate, this method fused the monitor records rebuid-ing and the local HOG feature extracting. Firstly, the key frame in records was transformed to frequency where the vehicle's escape orientation and velocity which were the key index for image fuzzy can be determined. And then making use of the features of license plate's character and based on local HOG and dictionary construction, this algorithm can get the sparse coefficient of image frame on the feature - dictionary which can complete the automatic recognition. The experimental results demonstrate that the algorithm can recover the key frame in monitor records finely and that this method has the superiority of reliability and stability compared with the traditional BP Neural Network.