基于相关向量机和方向导数的车辆识别方法

Vehicle Recognition Method Based on Relative Vector Machine and Direction Derivative Function

为解决车辆识别过程中图像匹配对数据库大小和灰度梯度值依赖严重的问题,提出一种基于相关向量机和灰度方向导数的车辆识别方法.对图像进行编码,确定图像中车道线的搜索域.根据灰度均值确定网格的属性标签,并采用原图和特征向量的2范数建立相关向量机模型,实现对车辆的分类和定位.对车辆灰度区域进行函数拟合,确定车辆的区域中心,并将满足双阈值的像素点作为车辆轮廓边缘,从而实现对车辆的识别.利用真实道路图像对模型的测试,结果表明,所建模型所需的图像库容量小,识别耗费时间短,准确率高.模型对图像灰度梯度的依赖性小,且通过双阈值能有效剔除伪轮廓点,细化了车辆边缘.

In order to solve the problem that image matching depends heavily on the size of database and gray gradient value in vehicle identification process, a vehicle identification method based on correlation vector machine and gray directional derivative was proposed. The search domain of lane mark was constructed by encoding the images. According to the gray mean value, the attribute label of the grid was determined, and the correlation vector machine model was established by using the 2-norm of the original image and the feature vector to realize the classification and positioning of vehicles. Finally, the function fitting was carried out on the gray scale region of the vehicle to determine the region center of the vehicle, and the pixel points meeting the dual threshold value were taken as the contour edges of the vehicle to realize the recognition of the vehicle. The test results of the model using real road images show that the image database required by the model has the advantages of small capacity, short recognition time and high accuracy. Moreover, the model has little dependence on the gray gradient of the image, and can effectively eliminate false contour points and refine the edge of the vehicle through dual threshold.