Rule-based Mapping of Parked Vehicles using Aerial Image Sequences

Mapping of parking spaces in cities is a prerequisite for future applications in parking space management like community-based parking. Although terrestrial or vehicle based sensors will be the favorite data source for parking space mapping, airborne monitoring can play a role in building up city wide basis maps which include also parking spaces on ancillary and suburban roads. We present a novel framework for automatic city wide classification of vehicles in moving, stopped and parked using aerial image sequences and information from a road database. The time span of observation of a specific vehicle during an image sequence is usually not long enough to decide unambiguously, whether a vehicle stopped e.g. before a traffic light or is parking along the road. Thus, the workflow includes a vehicle detection and tracking method as well as a rule-based fuzzy-logic workflow for the classification of vehicles. The workflow classifies stopped and parked vehicles by including the neighbourhood of each vehicle via a Delaunay-Graph. The presented method reaches correctness values of around 86.3%, which is demonstrated using three different aerial image sequences. The results depend on several factors like detection quality and road database accuracy