交通事故现场摄影测量标定点识别  

Recognizing reference points for photogrammetry of traffic accident scene

作　　者：鲁光泉[1] 许洪国[2] 李一兵[1] 

机构地区：[1]清华大学汽车安全与节能国家重点实验室,北京100084 [2]吉林大学交通学院,吉林长春130025

出　　处：《计算机应用》2006年第3期711-713,716,共4页journal of Computer Applications

摘　　要：对交通事故现场测量的现场标定点进行自动识别是提高交通事故摄影测量速度的重要技术。在自然光照条件下,标定点区域不可避免地存在阴影等干扰,采用无限脉冲响应(IIR)递归滤波器对脊线边缘进行检测,可以有效排除阴影干扰。在进行直线提取和聚类的基础上,对标定直线进行识别和排序,并以此为基础对标定点进行识别和排序。直线分类采用动态聚类方法。为了保证聚类的鲁棒性,对传统动态聚类方法的聚类中心生成方法进行了改进,用中位数代替传统方法中的均值,并采用最大、最小距离算法确定初始聚类中心。The recognization of reference points is important to survey traffic accident scene fleetly using photographs. In the nature light, there are shadows on the area of reference points. Infinite impulse response （IIR） filter was used to detect ridge edge and avoid the effect of shadows in nature light. After line was detected, reference line could be recognized and ordered based on clustering. Dynamic Clustering Analysis （DCA） was used to cluster lines. In order to robust result of clustering, DCA was improved： median value was used to replace mean value as the clustering center. Maximal and minimal distance analysis arithmetic was used to initialize clustering centers.

关 键 词：事故现场测量 标定点识别 动态聚类 模式识别 

分 类 号：U491[交通运输工程—交通运输规划与管理] TP391.41[交通运输工程—道路与铁道工程]