Graph-based real-time monitoring of physical distancing and estimation of secondary infection probability
Connected material: https://doi.org/10.1371/journal.pone.0240963
Caspar A. S. Pouw, Andrea Di Benedetto, Federico Toschi, Alessandro Corbetta
How likely is a Covid-19 infection in a trafficked facility such as a train station? As of today monitoring and ensuring physical distancing, together with face masks, are the conventional mechanisms to prevent secondary infections in public facilities. In this contribution we investigate the physical-distance patterns in a train station and, on this basis, estimate the secondary infection probabilities using basic contagion models. This work relies on an improved version of the graph-based analytic framework we proposed in Pouw et al., PLoS ONE 15(10): e0240963, 2020. The method uses a sparse network that represents, in real-time, pedestrian-pedestrian interactions via vector-weighted graph connections and is capable of autonomously recognizing family-group structures. First, using privacy-respectful pedestrian data collected at a trafficked Dutch train platform, we report on the distance-time contact patterns considering different Covid-19 regulations, e.g. with or without face masks. Then, on the basis of these contact patterns, we tackle our initial question: “If a randomly picked person in our network is assumed infectious, how likely is this person infecting others?”. We estimate the probability of such secondary infection by combining distance-time contact patterns with basic epidemic and physics-based models taken from the literature, encompassing both droplet and airborne transmissions. Despite the actual contagion probabilities are probably susceptible to many other factors, we believe that this study helps understand dependencies on variables related to crowd dynamics.
Covid-19 automated physical distancing analysis – Contagion probability in public spaces – High-statistics pedestrian dynamics – Statistical mechanics of human crowds – Privacy-respectful tracking – Real-time trajectory and group analysis
Benchmarking high-fidelity pedestrian tracking systems for research, real-time monitoring, and crowd control
Connected material: https://arxiv.org/abs/2108.11719
Caspar A.S. Pouw, Joris Willems, Frank van Schadewijk, Jasmin Thurau,
Federico Toschi, Alessandro Corbetta
High-fidelity pedestrian tracking has been an important tool in fundamental pedestrian dynamics research allowing us to quantify the statistics of relevant observables such as walking velocities, mutual distances, and body orientations. As this technology advances, it is becoming increasingly useful also in societal applications connected to the public domain. Continued urbanization is overwhelming existing pedestrian infrastructures like transportation hubs and stations, generating an urgent need for real-time highly-accurate individual tracking, aiming both at flow monitoring and dynamics understanding. To successfully employ pedestrian tracking techniques in research and technology, it is crucial to validate and benchmark them for accuracy. This is not only necessary to guarantee data quality, but also to identify systematic errors. Currently, there is no established policy in this context. In this contribution, we present a benchmark suite for privacy-respectful pedestrian tracking techniques. The suite is universal as it is technology independent and it is applicable for scientific, open-source and commercial pedestrian tracking systems, which operate both in lab environments and real-life conditions. The benchmark suite consists of 5 tests addressing specific aspects of pedestrian tracking quality and includes accurate line-based crowd-flux estimation, density estimation, position detection, and path registration. The tests follow from an effort to systematically identify the most accurate large-scale pedestrian tracking system. We provide the benchmark results for two tracking systems, both operating in real-life, one commercial, and the other based on overhead depth maps developed in academia. We discuss the results on the basis of the synthetic indicators and report on the typical sensor behavior. This enables us to highlight the current state of the art, its limitations and provide installation recommendations (with specific attention to multi-sensor setups and data stitching). We conclude with an outlook for further refinements of the benchmark suit.
High-fidelity pedestrian tracking – Sensor benchmarking – Crowd monitoring – Real-life pedestrian measurements – Industrial and societal applications
Multi-scale analysis of pedestrian inter-arrival time in large-scale public facilities
Caspar A. S. Pouw, Alessandro Gabbana, Alessandro Corbetta, Federico Toschi
Pedestrians are commuting continuously around the globe traveling to and from different destinations e.g. to work, school, supermarket, sports club. The exact moments and the frequency in which pedestrians enter and leave certain places or facilities defines a complex random arrival process inherently connected to the crowd flows and their many timescales. The arrival process is often modulated by characteristics such as office hours, train frequencies, and queue formations, besides, pedestrian groups generate very frequent and correlated arrivals. However, all this complexity is usually disregarded: in the practice, pedestrian arrival processes are often simplified by Poisson models, i.e. arrivals are completely defined by an average inter-arrival time. This is most likely due to the absence of high-resolution data sets that span sufficiently large time scales. Thanks to the recent emergence of accurate privacy-respectful pedestrian tracking we can build sophisticated high-statistics data sets which enable us to analyze the arrival process far more accurately. For this purpose, we present an in-depth analysis of the pedestrian arrival process from a privacy-respectful data set measured at a real-life Dutch train platform. From the high-statistics trajectory data, we determine the frame-interpolated time of entering or leaving a train platform. The signal contains approximately 13 million pedestrian arrivals between 2017 and 2019. The timescales in this data set range from sub-second intervals to seasonal changes. Guided by a renormalization flow principle we analyze the signal at different timescales seeking patterns and self-similarities. The practical implications of this work are arrival models that go beyond Poisson processes and are thus capable of describing the statistics of real-life arrival processes with far higher accuracy.
Real-life pedestrian arrival processes – Pedestrian models – Coarse-grained inter-arrival times