Predicting the variability in pedestrian travel rates and times using crowdsourced GPS data
Predicting the variability in pedestrian travel rates and times using crowdsourced GPS data
复制标题
DOI：
10.1016/j.compenvurbsys.2022.101866
复制
发表时间：
2022-08-20
影响因子：
6.8
通讯作者：
Thompson, Matthew P.
中科院分区：
文献类型：
--
作者：
Campbell, Michael J.;Dennison, Philip E.;Thompson, Matthew P.
准确预测行人出行时间在应急响应、野外消防、灾害管理、执法和城市规划中至关重要。然而，行人运动和景观条件之间的关系因人而异，因此很难估计大量人群步行从一个地方到另一个地方需要多长时间。虽然存在预测出行速度的功能，但它们通常通过假设景观对运动的影响是普遍的，从而过度简化了行人出行的内在可变性。在本研究中，我们提出了一种利用大型众包GPS轨迹数据库预测行人出行率和时间变化的方法。这些足迹来自户外娱乐网站AllTrails，代表了美国犹他州和加利福尼亚州各种各样的小径上近2000次徒步旅行。我们通过生成一系列非线性百分位数模型，将旅行速率建模为地形坡度的函数，这些模型从2.5百分位数到97.5百分位数。与独立测试数据集相比，第50百分位模型（代表典型个体的徒步速度）比现有的斜坡旅行速率函数有显著改善。我们的研究结果展示了一种估算旅行时间变异性的新方法，模型百分位数能够以小于10%的误差预测实际百分位数。旅行速率函数也可以应用于最低成本路径分析，以提供旅行时间的可变性。
Accurately predicting pedestrian travel times is critically valuable in emergency response, wildland firefighting, disaster management, law enforcement, and urban planning. However, the relationship between pedestrian movement and landscape conditions is highly variable between individuals, making it difficult to estimate how long it will take broad populations to get from one location to another on foot. Although functions exist for predicting travel rates, they typically oversimplify the inherent variability of pedestrian travel by assuming the effects of landscapes on movement are universal. In this study, we present an approach for predicting the variability in pedestrian travel rates and times using a large, crowdsourced database of GPS tracks. Acquired from the outdoor recreation website AllTrails, these tracks represent nearly 2000 hikes on a diverse range of trails in Utah and California, USA. We model travel rates as a function of the slope of the terrain by generating a series of non-linear percentile models from the 2.5 th to the 97.5 th by 2.5 percentiles. The 50 th percentile model, representing the hiking speed of the typical individual, demonstrates marked improvement over existing slope-travel rate functions when compared to an independent test dataset. Our results demonstrate novel ca-pacity to estimate travel time variability, with modeled percentiles being able to predict actual percentiles with less than 10% error. Travel rate functions can also be applied to least cost path analysis to provide variability in travel times.