基于多源时空分析的复杂活动识别方法  

Approach of high-level activity recognition based on spatio-temporal analysis of multisource data

作　　者：王方 崔海英 方亚东 郎瑞祥 叶剑[2] WANG Fang;CUI Hai-ying;FANG Ya-dong;LANG Rui-xiang;YE Jian(Inspur Cloud Information Technology Limited Company,Inspur Group,Jinan 250010,China;Institute of Computing Technology,Chinese Academy of Sciences,Beijing 100190,China)

机构地区：[1]浪潮集团有限公司浪潮云信息技术有限公司,山东济南250010 [2]中国科学院计算技术研究所,北京100190

出　　处：《计算机工程与设计》2020年第7期2076-2081,共6页Computer Engineering and Design

基　　金：国家重点研发计划基金项目(2016YFB1001100);国家自然科学基金项目(61401040)。

摘　　要：为在时空领域内能够更好地进行人类活动识别,解决活动识别中活动种类多、构成复杂、个体差异性强等问题,利用复杂活动与时空轨迹关系,通过定义活动区域并对活动区域进行标注,建立区域-活动关联模型,设计并实现区域特征耦合多源时空分析的复杂活动识别方法,实现分区域复杂活动识别,降低活动识别误差。实验结果表明,该方法相对于复杂活动整体识别方法,具有较高的识别准确率。To better recognize human activities in the field of space-time,solving the problems of various kinds of activities,complex composition and strong individual differences in activity recognition,based on the relationship between complex activities and space-time trajectory,a region-activity correlation model was established,embedded with the features of region of activity(ROA)and with definition and marking of ROA.A method of complex activities recognition based on multi-source space-time analysis coupled with regional features was designed and implemented to realize complex activities recognition in the sub-region,which effectively reduced the error of activity recognition.Experimental results show that the proposed approach has advantages over existing approaches of overall recognition in accuracy.

关 键 词：活动识别 时空分析 轨迹挖掘 密度聚类 滤波 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]