摘要
车型识别技术作为智能交通系统中关键技术,特征识别法具有较高的识别精度、鲁棒性、实时性,是车型分类技术的主要方法。但是该方法存在两个主要问题:车型分类网络需要优化,目标特征提取算法性能与工程应用要求尚有差距。针对上述问题展开研究,引入遗传算法、动量项等对BP算法优化车型分类网络、采用耗时低的Surendra背景提取算法、迭代阈值分割算法改善目标特征提取工作的实时性,仿真结果表明,基于遗传BP算法构建的车型识别系统的精确性、鲁棒性等关键性能达到系统设计要求。
As the key technology of Intelligent Transportation System (ITS),Character Distilling are the main methods of Vehicle Classifier,because of higher identification accuracy and the latter is superior to the former about robustness and real-time. But it has two main issues:Vehicle Classifier network need to be optimized,the performance of Character Distilling algorithm still has gaps with the demand of engineering applications.This thesis studies on these problems and presents improvements.GA and momentum algorithm of LM are introduced to optimize BP Neural network.The real-time quality of object character distilling is improved by background extraction algorithm.
出处
《工业控制计算机》
2007年第9期77-79,共3页
Industrial Control Computer
基金
科技部国际合作项目(050296)