作　　者： (胡杰）; (刘志成）; (杨慧萍）; (李泽西）;

机构地区： 武汉理工大学汽车工程学院,湖北武汉430070

出　　处： 《物流技术》 2017年第8期117-121,共5页

摘　　要： 为了提高车辆检测的准确率和效率,为智能交通系统提供可靠的参考信息,针对传统Harr特征车辆检测算法存在特征向量维数过大、训练时间过长的问题,提出采用非负矩阵分解Harr特征得到低维的Harr-NMF特征,对Harr-NMF特征分类并由Adaboost算法训练得到基于Harr-NMF特征的车辆分类器,对车身进行有效识别。针对实车测试时出现的重复检测、错误检测等问题,进一步优化了算法。测试结果表明,改进后的算法提高了车辆检测率并有效降低了误检率。 In this paper, in order to improve the accuracy and efficiency in vehicle inspection to provide reliable reference information for the intelligent traffic system, and in light of the problems of the traditional vehicle inspection algorithm based on the Hart＂ feature, we proposed to decompose the Harr feature by a non-negative matrix to yield a lower-dimension Harr-NMF feature. Next we categorized the Harr- NMF feature and obtained through Adaboost training the Harr- NMF based vehicle classifier for the effective identification of the vehicle body. At the end, in view of the problems encountered in an empirical nractice of the method, we fnrther ontimized the algorithm

关 键 词： 智能交通 车辆检测 特征 算法