基于轨迹分析的异常行为识别与管控研究  

Abnormal Behavior Detection and Control based on Trajectory Analysis

作　　者：王晓龙[1] 郭捷[1] 邱卫东[1] 徐鹏[2] 郭曼 张菡[4] 

机构地区：[1]上海交通大学电子信息与电气工程学院,上海200240 [2]山东省警官培训学院信息技术研究所,山东济南250000 [3]科学技术部火炬高技术产业开发中心,山东济南250000 [4]山东省警官培训学院,山东济南250000

出　　处：《信息安全与通信保密》2015年第3期76-80,共5页Information Security and Communications Privacy

基　　金：国家科技支撑计划课题(No.2014BAK06B00);上海市教育委员会科研创新项目(No.12ZZ019)

摘　　要：人类行为识别是模式识别领域的一个重要研究方向,是综合了计算机视觉与机器学习的多领域学科。而异常行为的识别与管控,因为其行为主体与行为模式的复杂性,是人类行为识别领域中的重点与难点。本文采用了一种新的运动目标轨迹提取算法,通过计算轨迹的特征向量,采用词袋模型进行建模,再把提取到的特征投入到支持向量机SVM(Support Vector Machine)中进行学习。我们在国际标准暴力行为视频数据库Hocky Fights上实现了实验仿真,获得了很好的识别分类效果。监狱内罪犯的暴力行为的识别将是我们下一步研究的重点。Human behavior recognition,as an important branch of pattern recognition,is a multi- field subject integrating computer vision and machine learning. Abnormal- behavior detection and control,due to the complexity of behavers and behavior pattern,is the focal point and difficult point as well in human- behavior recognition. A novel algorithm to extract motion trajectories is proposed in this paper. By calculating feature vectors,a bag- of- words model is constructed,and the extracted vectors are put into muti- kernel SVM( Support Vector Machine) for study. Experienmental simulation on the violent behavoir video dataset Hocky Fights achieves fairly good result of recognition and classification. In addition,the subsequent research would focus on violent- behavior recognition of prisoners in jail.

关 键 词：行为识别 轨迹分析 机器学习 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]