基于混沌动力学模型的群体目标检测与分类  

Crowd object detection and classification based on a chaotic dynamic model

作　　者：乔伟[1] 王汇源[1,2] 吴晓娟[1] 刘鹏威[1] 

机构地区：[1]山东大学信息科学与工程学院,山东济南250100 [2]阿维罗大学电子与电信研究所

出　　处：《山东大学学报（工学版）》2010年第2期19-23,共5页Journal of Shandong University（Engineering Science）

基　　金：国家自然科学基金资助项目(60675024)

摘　　要：传统目标跟踪检测方法对多目标情况处理已较为困难,对群体目标监测和跟踪更是难以解决的问题。因此考虑在拉格朗日混沌动力学基础上实现对视频群体流的分割以及相关检测。把群体运动系统作为混沌动力学系统处理,而群体目标则作为流动的粒子群处理,粒子的流动以流图(flow map,FM)进行跟踪。在此基础上可以获得有限时间李亚普诺夫指数(finite time Lyapunov exponent,FTLE)场,形态学处理后,对群体流动区域进行提取并对结果分类。实验结果验证了该方法的有效性。Multi-object detection and tracking is an arduous task for traditional methods,and objects in crowds are more difficult to deal with.In this paper,segmentation and detection to video crowd flow was realized by a chaotic dynamics based method.The crowd moving system was treated as a chaotic dynamic system,and object in crowds were treated as particles,whose flow is tracked by flow map.As a result,the finite time Lyapunov exponent（FTLE）field was obtained.After morphological processing,region extraction and classification can be achieved.Experimental results indicate the effectiveness of the proposed approach.

关 键 词：群体目标检测 有限时间李亚普诺夫指数 轮廓提取 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP242.62[自动化与计算机技术—计算机科学与技术]