复杂背景下的车型自动分类研究  

Research of automatic vehicle classification under complex backgrounds

作　　者：霍炜[1] 刘大维[1] 王江涛[2] 

机构地区：[1]青岛大学车辆工程系,青岛266071 [2]南京理工大学计算机系,南京210094

出　　处：《青岛理工大学学报》2008年第1期107-110,115,共5页Journal of Qingdao University of Technology

基　　金：山东省自然科学基金项目(Q99F11)

摘　　要：针对复杂背景下的车辆自动分类问题,提出了一种新的车型识别方法.该方法是首先采用自适应高斯混合模型进行背景抽取,然后以背景抽取后的车辆侧面图像作为特征图像,设计了顶长比、顶高比和前后比3种参数对车型进行表达,最后通过训练—BP神经网络分类器对车型进行自动分类.对9个样本进行分类测试时,取得了均方误差小于0.0023、识别率达到100%的测试结果,表明笔者设计的车型分类系统可有效地实现车型的自动分类.A novel approach is presented to solve the vehicle classification problem under complex backgrounds. We first adopt a Gaussian mixture model to model the backgrounds, and then vehicle side image is gotten through the background abstraction. Take foreground image as feature images, three coefficients are designed to represent the feature image. Finally, a BP neural network is constructed to class the car types automatically. In the experiment, 12 samples are tested, the error is below 0. 0023, and the identification rate is achieved 100%, this indicates that the vehicle classification system is effective.

关 键 词：高斯混合模型 BP神经网络 车型分类 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]