华南理工大学学报(自然科学版) ›› 2009, Vol. 37 ›› Issue (2): 87-91.

• 交通运输工程 • 上一篇    下一篇

基于图像不变矩特征和BP神经网络的车型分类

秦钟   

  1. 华南理工大学 土木与交通学院, 广东 广州 510640
  • 收稿日期:2007-10-26 修回日期:2008-01-15 出版日期:2009-02-25 发布日期:2009-02-25
  • 通信作者: 秦钟(1972-),男,高级工程师,博士,主要从事智能交通、图像处理研究. E-mail:qzscut@yahoo.com.cn
  • 作者简介:秦钟(1972-),男,高级工程师,博士,主要从事智能交通、图像处理研究.
  • 基金资助:

    建设部软科学研究项目(2008-K5-6)

Vehicle Classification Based on Image Moment Invariant Feature and BP Neural Network

Qin Zhong   

  1. School of Civil Engineering and Transportation, South China University of Technology, Guangzhou 510640, Guangdong, China
  • Received:2007-10-26 Revised:2008-01-15 Online:2009-02-25 Published:2009-02-25
  • Contact: 秦钟(1972-),男,高级工程师,博士,主要从事智能交通、图像处理研究. E-mail:qzscut@yahoo.com.cn
  • About author:秦钟(1972-),男,高级工程师,博士,主要从事智能交通、图像处理研究.
  • Supported by:

    建设部软科学研究项目(2008-K5-6)

摘要: 为实现车型的有效分类,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了路面背景,利用背景差分将图像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,提取车辆轮廓,直接计算车辆轮廓的矩不变量,将其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.

关键词: 车型分类, 不变矩, 神经网络, 图像分割, 边缘检测

Abstract:

In order to effectively classify vehicles, a new classification method based on image analysis is proposed. In this method, the road background is set up according to video sequence images, and the moment invariant fea- tures of the vehicle region segmented via the background division are calculated. Then, in order to speed up the ex- traction, a Canny operator is used to detect the edge of the vehicle region and to extract the vehicle outline, and the moment invariant is directly obtained as the characteristic quantity of vehicle classification. Afterwards, a three-layer BP neural network, with the moment invariant as the input, is established, and the vehicle is then classified according to the network output. Some experiments are finally carried out to verify the effectiveness of the proposed method.

Key words: vehicle classification, moment invariant, neural network, image segmentation, edge detection