基于图像不变矩特征和BP神经网络的车型分类方法

文中提出了一种基于图像分析的车型分类方法。首先，根据视频序列图像建立路面背景，利用背景差分将图象中的车辆区域分割出来，计算车辆区域的不变矩特征量。为了加快特征的提取，利用Canny算子检测车辆区域的边缘，提取出车辆轮廓，直接计算车辆轮廓的矩不变量作为车型分类的特征量。然后建立具有3层结构的BP神经网络，将不变矩特征量作为神经网络的输入，根据神经网络的输出实现车型的分类。试验证实了该方法的有效性。

Vehicle Classification with Image Moment Invariant Feature and Neural Networks

The paper put forward the vehicle classification using image analysed. First of all, road background is set up according to the serial images, and the vehicle region are segmented using background divide, then its moment invariant features are calculated. In order to reduce the exhaust, Canny operator is used to draw the vehicle outline, which is substitute for vehicle region. The moment invariant features of vehicle outline will be the input of BP neural networks with three layers, and the vehicle type is classified according to the output of the BP neural networks. The test has verified the validity of this method.