Machine-learning prediction models for pedestrian traffic flow levels: Towards optimizing walking routes for blind pedestrians

Achituv Cohen, Sagi Dalyot

Research output: Contribution to journalArticlepeer-review

7 Scopus citations

Abstract

Navigation and orientation while walking in urban spaces pose serious challenges for blind pedestrians, sometimes even on a daily basis. Research shows the practicability of computerized weighted network route planning algorithms based on OpenStreetMap mapping data for calculating customized routes for blind pedestrians. While data about pedestrians and vehicle traffic flow at different times throughout the day influence the route choices of blind pedestrians, such data do not exist in OpenStreetMap. Quantifying the correlation between spatial structure and traffic flow could be used to fill this gap. As such, we investigated machine-learning methods to develop a computerized model for predicting pedestrian traffic flow levels, with the objective of enriching the OpenStreetMap database. This article presents prediction results by implementing six machine-learning algorithms based on parameters relating to the geometrical and topological configuration of streets in OpenStreetMap, as well as points-of-interest such as public transportation and shops. The Random Forest algorithm produced the best results, whereby 95% of the testing data were successfully predicted. These results indicate that machine-learning algorithms can accurately generate necessary temporal data, which when combined with the available crowdsourced open mapping data could augment the reliability of route planning algorithms for blind pedestrians.

Original languageEnglish
Pages (from-to)1264-1279
Number of pages16
JournalTransactions in GIS
Volume24
Issue number5
DOIs
StatePublished - 1 Oct 2020
Externally publishedYes

Fingerprint

Dive into the research topics of 'Machine-learning prediction models for pedestrian traffic flow levels: Towards optimizing walking routes for blind pedestrians'. Together they form a unique fingerprint.

Cite this