摘要
针对换道预警系统中换道行为识别识别率较低的问题,提出了以实际道路试验所得到的换道样本数据为基础,采用支持向量机(SVM)对换道行为进行识别的方法。为进一步提高模型的识别率和缩短识别耗时,对样本数据进行卡尔曼滤波、归一化处理、主成分分析,利用遗传算法对核参数进行调优。对优化后SVM模型进行训练与测试,测试结果表明:在1.2秒时窗下,优化后模型的识别准确率达到了97%以上,能够满足车载换道预警系统对识别有效率和实时性的要求。
For the recognition rate of lane changing behavior in lane changing warning system is low, lane changing behavior identification method based on support vector machine （SVM） is proposed and the data from practical road test are as the founda tion of this method. To further enhance identification accuracy and shorten the time of identification consuming, the collected da ta are filtered by kalman filtering technology, data normalization, principal component analysis, and the SVM model is optimized with genetic algorithm. The optimized SVM model is trained and tested, and the result shows that the recognition accuracy of the optimized model got to 97% when the window is the length of 1. 2 second. It can meet requirements of vehiclemounted lane changing warning system for realtime and recognition effectiveness.
出处
《计算机工程与设计》
CSCD
北大核心
2013年第2期643-648,共6页
Computer Engineering and Design
基金
国家自然科学基金项目(51178053)
国家科技支撑计划基金项目(2009BAG13A05)
关键词
换道行为
支持向量机
卡尔曼滤波
主成分分析
遗传算法
lane-change support vector machine Kalman filter principal component analysis genetic algorithm
