Collaborative:RAPID:Leveraging New Data Sources to Analyze the Risk of COVID-19 in Crowded Locations.

该项目的目标是创建一个软件基础设施，帮助科学家调查COVID-19传播的风险，并使用实时公共网络摄像头视频和基于位置的服务（LBS）数据分析拥挤地区未来的流行病。其动机是观察到COVID-19聚集性病例经常出现在人口密度高的地点。目前的战略建议采取粗略的干预措施来防止这种情况，例如取消活动，这会造成巨大的经济和社会成本。对拥挤地区人群的移动和互动模式进行更详细的精细分析，可以提出干预措施，例如改变人群管理程序和建筑环境的设计，从而产生社会距离，而不会对人类活动和经济造成破坏性影响。行人动力学领域提供了能够产生如此详细见解的数学模型。然而，这些模型需要有关人类行为的数据，而这些数据因环境和文化的不同而有很大差异。该项目将利用新的数据流，如公共网络摄像头和基于位置的服务，为行人动态模型提供信息。相关的数据、模型和软件将提供给在这个领域工作的其他研究人员，但要遵守隐私限制。项目小组还将与决策者进行外联，以便科学见解产生有助于公共卫生的可行政策。最终结果将是至关重要的科学见解，可以对应对COVID-19大流行（包括可能的第二波疫情）产生变革性影响，从而保护公众健康，同时最大限度地减少干预措施的不利影响。我们将通过以下方式和创新来完成上述工作：LBS数据可以在几十米的范围内识别出拥挤的地点，并通过分析个人在那里的远距离移动，帮助筛选潜在的风险。世界范围的视频流可以产生更精细的社会亲密性细节和其他行为模式，这些都是精确建模所需要的。另一方面，这些视频可能无法用于潜在的高风险地点，也无法直接回答“如果”的问题。与正在建模的背景相似的视频将用于校准行人动力学模型参数，例如步行速度。然后在目标位置模拟单个行人的运动轨迹来估计社会亲密度。感染传播模型将应用于这些轨迹，以产生感染传播的估计。这将产生一种新的方法，将各种实时数据纳入行人动力学模型，以便他们能够在新的和不断变化的情况下快速准确地捕捉人类的运动模式。网络基础设施将自动发现互联网上的实时视频流，并对其进行分析，以确定行人密度、运动和社交距离。行人动力学模型将从目前基于力的定义重新制定为使用行人密度和个人速度的定义，这两者都可以通过视频分析有效地测量。修订后的模型将用于提供科学见解，为政策提供信息，例如减轻局部COVID-19疫情的步骤，以及系统地重新开放，可能重新关闭以及经济和社会活动的永久性变化。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。

The goal of this project is to create a software infrastructure that will help scientists investigate the risk of the spread of COVID-19 and analyze future epidemics in crowded locations using real-time public webcam videos and location based services (LBS) data. It is motivated by the observation that COVID-19 clusters often arise at sites involving high densities of people. Current strategies suggest coarse scale interventions to prevent this, such as cancellation of activities, which incur substantial economic and social costs. More detailed fine scaled analysis of the movement and interaction patterns of people at crowded locations can suggest interventions, such as changes to crowd management procedures and the design of built environments, that yield social distance without being as disruptive to human activities and the economy. The field of pedestrian dynamics provides mathematical models that can generate such detailed insight. However, these models need data on human behavior, which varies significantly with context and culture. This project will leverage novel data streams, such as public webcams and location based services, to inform the pedestrian dynamics model. Relevant data, models, and software will be made available to benefit other researchers working in this domain, subject to privacy restrictions. The project team will also perform outreach to decision makers so that the scientific insights yield actionable policies contributing to public health. The net result will be critical scientific insight that can generate a transformative impact on the response to the COVID-19 pandemic, including a possible second wave, so that it protects public health while minimizing adverse effects from the interventions.We will accomplish the above work through the following methods and innovations. LBS data can identify crowded locations at a scale of tens of meters and help screen for potential risk by analyzing the long range movement of individuals there. Worldwide video streams can yield finer-grained details of social closeness and other behavioral patterns desirable for accurate modeling. On the other hand, the videos may not be available for potentially high risk locations, nor can they directly answer “what-if” questions. Videos from contexts similar to the one being modeled will be used to calibrate pedestrian dynamics model parameters, such as walking speeds. Then the trajectories of individual pedestrians will be simulated in the target locations to estimate social closeness. An infection transmission model will be applied to these trajectories to yield estimates of infection spread. This will result in a novel methodology to include diverse real time data into pedestrian dynamics models so that they can quickly and accurately capture human movement patterns in new and evolving situations. The cyberinfrastructure will automatically discover real-time video streams on the Internet and analyze them to determine the pedestrian density, movements, and social distances. The pedestrian dynamics model will be reformulated from the current force-based definition to one that uses pedestrian density and individual speed, both of which can be measured effectively through video analysis. The revised model will be used to produce scientific insight to inform policies, such as steps to mitigate localized outbreaks of COVID-19 and for the systematic reopening, potential re-closing, and permanent changes to economic and social activities.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.