Crowdsourced pedestrian map construction for short-term city-scale events

This paper targets the construction of pedestrian maps for city-scale events from GPS trajectories of visitors. Incomplete data with a short lifetime, varying localisation accuracy, and a high variation of walking behaviour render the extraction of a pedestrian map from crowd-sourced data a difficult task. Traditional network or map construction methods lean on accurate GPS trajectories typically obtained over longer time periods from vehicles at high speeds with less variation in locomotion. Not designed to operate under mobility conditions of pedestrians at large scale events they cannot be directly applied. We present an algorithm based on a crowd-sensing scheme to construct the pedestrian network during city scale events. In a thorough evaluation, we investigate the effect of trajectory quality and quantity on the map construction. To this end, we use a real world dataset with 25M GPS points obtained from 28.000 users during a three-day public festival event. Results indicate that with a short observation window of 30min the estimated pedestrian network can represent previously unseen trajectories with a median map-matching deviation in matching of only 5m and a map accuracy of more than 85%.