开放性公共场所的人群聚集密度高且人员活动模式多样，客观上存在着全局或者局部的人群聚集风险。因此，有必要对场所内的人群活动进行监测，在对到访人员进行分类的基础上解析不同类型人群活动的时空特征，以减少拥挤踩踏事件的发生。Wi-Fi探针作为一种新兴的位置感应装置，可以方便快捷地采集个体活动的时空信息，适用于开放性公共场所的人群监测。 针对开放性公共场所人群分类的两个关键问题，即活动轨迹的提取和聚类方法的选择，本文构建了基于Wi-Fi数据的人群分类模型。首先，针对Wi-Fi数据的离散性问题，通过时空行为研究解析人群活动的总体特征，确定时间维度的降维规则；通过对POI（Point of Interest）进行语义分析解析活动地点的功能差异，确定空间维度的降维规则。基于上述规则构建时空矩阵，对人群活动轨迹进行降维处理，量化个体活动的时空特征，为人群分类提供依据。其次，借助K-均值法对时空矩阵进行分类。基于聚类中心的时空矩阵解析类别含义，为分类结果赋予人群类别属性。本文以北京市什刹海景区为研究区域，通过在研究区内布设一定数量的Wi-Fi探针采集人群活动数据，将人群分类模型系统地应用于研究区的人群活动分析中，并在此基础上解析各类型人群活动的时空特征，得到以下主要结论。 （1）研究区内的到访人群可分为居民、工作人员和游客三种类型，其中游客类型可进一步分为工作日常访游客、节假日常访游客、工作日偶访游客、节假日日间偶访游客和节假日早晚偶访游客等五类。通过实地调研，验证该方法的分类精度为72.97%。 （2）不同类型人群的平均到访人流量在一天内有较平稳的变化趋势，但节假日日间偶访游客和节假日早晚偶访游客的到访人流量存在快速、显著增加和减少的时间段。 （3）针对节假日日间偶访游客和节假日早晚偶访游客到访人流量快速、显著变化的时间段进行空间分析，发现在荷花市场、银锭桥及烟袋斜街及其周围区域发生了显著的人群聚集，且存在“快聚集、慢消散”的特征，表明该区域在节假日的正午至凌晨为游客活动的高密度区域，存在一定的人群聚集风险。

Open public places are densely populated with people of diverse activity patterns across space and time, and thus characterized by a high crowd gathering risk. Therefore, it is of great significance to monitor the crowd activities in open public places, classifying the population and exploring the spatio-temporal activity characteristics of distinct classifications to reduce the occurrence of stampedes. As an emerging position sensing device, Wi-Fi probe can conveniently collect the spatio-temporal information of individual activity, which is suitable for crowd monitoring in open public places. This paper constructs a population classification model based on Wi-Fi data aiming to solve the two key problems of crowd classification in open public places, which are the extraction of activity trajectories and the selection of clustering method. Firstly, in order to reduce the sparsity of Wi-Fi data, the general characteristics of crowd activities are analysed through spatio-temporal behavior research and the rules of dimensionality reduction in time dimension is determined. Also, the functional differences of activity sites are analysed through the semantic analysis of POI (Point of Interest) and the rules of dimensionality reduction in space dimension is determined. Based on the above rules, the space-time matrix is constructed to reduce the dimension of activity trajectories, quantifying the spatio-temporal characteristics of individual activities and provide a basis for population classification. Secondly, the space-time matrices are classified by K-means method. Based on the space-time matrices of clustering center, the category attributes are given to the classification results. In this paper, Shichahai scenic area in Beijing city is taken as study area and 29 Wi-Fi probes are deployed to collect individual activity trajectories. The proposed method is applied systematically for population classification in study area. The main conclusions are as follow: (1) The population in study area can be classified into residents, staff and tourists, where the tourists can be further divided into five types, including frequent-visiting tourists on weekday, frequent-visiting tourists on holiday, occasional-visiting tourists on weekday, occasional-visiting tourists on holiday daytime, occasional-visiting tourists on holiday nighttime. By comparing with the results of field survey, the classification accuracy of the proposed method can reach 72.97%. (2) The average crowd flow of all population categories has a stable trend of change in one day. However, two categories of occasional-visiting tourists, the occasional-visiting tourists on holiday daytime and nighttime, have significantly different trends of crowd flow in one day comparing with the average one, with rapid increases and decreases of crowd flow on specific time intervals. (3) Spatial characteristics of two categories of occasional-visiting tourists on holiday are analyzed, finding that the main gathering areas of these tourists include Lotus Market, Yinding Bridge, Yandai By-way and their peripheral regions. The crowd activity has the pattern of fast gathering and slow dissipation, indicating that these regions are densely populated with tourists from midday to night on the holidays, foaming a certain degree of crowd gathering risk.