The recent controversy about the size of crowds at candlelight protests in Korea raises an interesting question regarding the methods used to estimate crowd size. Protest organizers tend to count all participants in the event from its start to finish, while the police usually report the crowd size at its peak. While several counting methods are available to estimate the size of a crowd at a given time, counting the total number of the participants at a protest is not straightforward. In this paper, we propose a new estimator to count the total number of participants that we call the size of a dynamic crowd. We assume that the arrival and departure times of the crowd are randomly observed and that the number of the attendees in the crowd at a specific time is estimable. We estimate the number of total attendees during the entire gathering based on the capture-recapture model. We also propose a bootstrap procedure to construct a confidence interval for the crowd size. We demonstrate the performance of the proposed method with simulation studies and the data from Korea's March for Science, a global event across the world on Earth Day, April 22, 2017.

最近在韩国举行的关于烛光抗议的人群规模争议引起了一个有趣的问题，即用于估算人群规模的方法。抗议组织者往往会从事件开始到结束都计算事件的所有参与者，而警察通常会在事件高峰时报告人群的数量。虽然有几种计数方法可用来估计给定时间的人群规模，但对抗议活动参与者的总数进行计数并不是一件容易的事。在本文中，我们提出了一种新的估算器来计算我们称为动态人群规模的参与者总数。我们假设随机观察人群的到达和离开时间，并且可以估计特定时间人群中的参与者人数。我们根据捕获-捕获模型估算整个聚会期间的总出席人数。我们还提出了一种引导程序，以构建人群数量的置信区间。我们通过仿真研究和韩国科学三月号（March for Science）的数据证明了该方法的性能，该数据是2017年4月22日地球日在世界范围内举行的全球性活动。