车辆识别技术本身存在着识别难度大、识别结果精度低等问题,本文提出一种基于统计模式识别理论的车辆识别方法,利用非线性支持向量机(SVM)对目标车辆进行识别.首先,该算法通过车载CCD摄像头采集自车前后方车辆的图像信息,对所采集到的图像进行小波去噪以及图像二值化处理,剔除噪声干扰.通过坐标变换使图像中车辆跟实车建立一一对应关系,进而对目标车辆进行准确定位;其次,对处理后的图像进行8×8网格划分,将各网格内满足要求的像素点数跟网格内总像素点数的比值作为每个网格输出值(0,1)的唯一判定条件,将每一行网格输出值的总和作为特征向量的元素.以遗传算法为搜索模式,采用交叉验证技术确定SVM的最佳参数组合,最后将自车前后方10—20m内车辆作为训练样本对模型进行训练和测试;并用ROC曲线(受试者工作特征曲线)对模型进行评价.

Vehicle identification is a very difficult problem and the accuracy of identification results is very low, a method to identify vehicle is proposed based on statistical pattern recognition, adopting nonlinear Support Vector Machine(SVM) to identify the target vehicle. First, the image information of vehicles in the front and the back of the vehicle is collected by using the vehicle-mounted CCD camera. The collected images are filtered by wavelet denoising and processed with image binaryzation in order to eliminate the noise interference. Through the coordinate transformation, one-to-one correspondence relationship between the vehicles in image and the real ones is established. Then, the target vehicle is correctly positioned. Secondly, the processed images are partitioned into 8×8 grids. The ratio of the number of pixels meeting the requirements to the total pixels in each grid is served as the only decision condition for the output (0 or 1) of each grid. The total output for each row could be taken as the characteristic vector's element. The best parameter combination is determined by using the cross validation and genetic algorithm. The vehicles located at10-20m before and after the subject vehicle are taken as the training sample to train the model. The model is verified. Test results show that the algorithm is able to accurately distinguish the types of the vehicles.