ATD: New Approaches for Analyzing Spatiotemporal Data for Anomalies

近年来，人口密度高的城市地区越来越多地成为恐怖主义的目标，因为在拥挤的环境中可能造成大量伤亡，并造成高影响，特别是在室外环境中。该项目开发了新的方法，以补充和验证基于视频监控和通过手机使用估算人口密度的现有方法所产生的结果，以监测城市地区的异常情况和威胁级别。它将有利于需要规划和分配资源的机构和地方政府，以便及时保护受到更高威胁的地点。使用共享交通（例如，共享单车、出租车）的数据来监测异常和威胁提供了对城市地区的额外覆盖，并可以捕获其他数据源未涵盖的“盲点”。为此，恐怖分子在一个城市发动袭击的成功率可以大大降低。这个为期三年的项目的目标是：(i)开发一种新的数据驱动的混合微分方程（DE）建模方法，用于物体（如人、自行车等）的密度和流量估计，以模拟人类动力学；（ii）开发一种实时异常检测算法，利用DE模型和实时观测数据来识别城市环境中的异常人群密度和交通。本项目使用的主要时空数据是来自美国三个城市的公共共享单车数据。在人体动力学建模时，由于位置的相互依赖性，将时空数据转换为随时间变化的图作为数据表示。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

High population density urban areas have increasingly been the targets of terrorism in recent years due to the possibility of inflicting a large number of casualties in a crowded environment and causing high impact disruption especially in the outdoor environment. The project develops new approaches to complement and validate results produced by existing approaches based on video surveillance and human population density estimation via cellphone usage to monitor anomalies and threat level in urban area. It will benefit agencies and local governments that require the planning and allocation of resources to secure locations with higher levels of threat in a timely manner. The use of the data on shared transportation (e.g., bikesharing, taxi) to monitor anomalies and threats provides additional coverage of an urban area and could capture "blindspots" not covered by the other data sources. Towards this end, the success of a terrorist attack in a city can be significantly reduced. The goals of the three-year project are to (i) develop a new data-driven hybrid differential equations (DE) modeling approach for object (e.g., human, bike, etc.) density and flow estimation to model human dynamics, and (ii) develop a real-time anomaly detection algorithm utilizing the DE model and real-time observed data to identify anomalous crowd density and traffic in an urban environment. The main spatiotemporal data used in this project is the publicly available bikesharing data from three US cities. Due to the interdependency of locations when modeling human dynamics, spatiotemporal data are transformed into time-evolving graphs as the data representation.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.