基于时空周期模式挖掘的活动语义识别方法

PDF下载HTML

Title:

Active semantic recognition method based on spatial-temporal period pattern mining

作者:

郭茂祖^1,2, 邵首飞^1,2, 赵玲玲³, 李阳^1,2

1. 北京建筑大学 电气与信息工程学院，北京 100044;
2. 北京建筑大学 建筑大数据智能处理方法研究北京市重点实验室，北京 100044;
3. 哈尔滨工业大学 计算机科学与技术学院，黑龙江 哈尔滨 150001

Author(s):

GUO Maozu^1,2, SHAO Shoufei^1,2, ZHAO Lingling³, LI Yang^1,2

1. School of Electrical and Information Engineering, Beijing University of Civil Engineering and Architecture, Beijing 100044, China;
2. Beijing Key Laboratory of Intelligent Processing for Building Big Data, Beijing University of Civil Engineering and

关键词:

时空轨迹; 时空紧密相连性; 密度聚类; 停留时间; 活动语义识别; 周期模式挖掘; 随机森林

Keywords:

spatial-temporal trajectory; spatial-temporal close connection;  density clustering;  stay time;  active semantic recognition; period pattern mining; random forest

分类号:

TP181

DOI:

10.11992/tis.202012035

摘要:

传统的活动语义识别研究侧重从时空轨迹的空间信息中提取人类的活动语义，对时空轨迹数据的时间特性挖掘不足。本文兼顾时间和空间特征，提出了一种基于周期模式挖掘的活动语义识别方法。首先将分离出的活动轨迹数据通过空间距离进行密度聚类分成不同轨迹簇；然后，根据轨迹簇的时序特征挖掘个体对特定位置的访问周期，基于该访问周期，并结合在该位置的停留时间，及其附近兴趣点分布等特征构建分类模型，识别人类个体的活动语义。基于签到数据和仿真数据的实验结果表明，结合周期特征的活动语义识别方法相比没有加入周期特征的实验结果有效提升识别精度20%以上，在2个相同的签到数据集下，对比其他的识别方法提升精度10%以上。

Abstract:

Active semantic recognition aims to mine people’s activities from spatial-temporal data recording through the smart equipment they carry. Traditional studies paid more attention to studying the spatial features of spatial-temporal data but failed to mine temporal features adequately. Considering both features, this work proposes an active semantic recognition method based on period pattern mining. First, trajectories that have already been separated from raw trajectories are clustered based on the spatial distance. The periods of reference spots that are frequently visited by the people are then mined according to the sequence of clustering. Based on the visit period and combined with the residence time at the location and the distribution of interest points nearby, a classification model is constructed to identify the activity semantics of human individuals. The experimental results on the check-in dataset and simulation data show that the valid recognition accuracy of active semantic recognition combined with periodic characteristics increases by 20% more than that without periodic characteristics. Under the same two check-in datasets and compared with other recognition methods, the accuracy is improved by more than 10%.