摘要
基于随机蕨算法,在快速"车脸"定位的前提下对车辆实时匹配.结合车辆区域定位和车牌快速定位提出了快速车脸区域定位方法.结合车辆的特征,提出了一种快速多尺度特征点检测算法.先建立离线随机蕨分类器,在线阶段用训练好的分类器进行分类,形成初始匹配.提出了一种改进的顺序抽样一致性(PROSAC)算法,对初始匹配进行快速精确的匹配.实验结果表明,基于随机蕨的车辆匹配算法能够快速实时地进行匹配.
We propose a real-time vehicle matching algorithm based on random ferns. In order to reduce the influence of complex traffic background,the paper combine vehicle location and quick license plate location to locate the "car face" area which contains most of the significant features of vehicles.According to random fern characteristics and texture characteristics of cars,we propose a fast multi-scale feature point detection method.To generate initial match, we produce large amounts of virtual diagram for training the classifier in the offline training stage.We detect feature points and put the area where contains feature points into the classifier for rapid classification in online operation stage.An improved algorithm of PROSAC is introduced to do fast and exact match on the initial match.
出处
《厦门大学学报(自然科学版)》
CAS
CSCD
北大核心
2014年第2期206-211,共6页
Journal of Xiamen University:Natural Science
基金
福建省高校产学合作重大项目(2012H6024)
关键词
实时车辆匹配
车脸定位
随机蕨
real-time vehicle matching
cars face location ~ random ferns