Development of a Data-Driven On-Street Parking Information System Using Enhanced Parking Features

Abstract

On-street parking information (OSPI) systems help reduce congestion in the city by lessening parking search time. However, current systems use features mainly relying on costly manual observations to maintain a high quality. In this paper, on top of traditional location-based features based on spatial, temporal and capacity attributes, vehicle parked-in and parked-out events are employed to fill the quality assurance gap. The parking events (PEs) are used to develop dynamic features to make the system adaptive to changes that impact on-street parking availability. Additionally, a parking behavior change detection (PBCD) model is developed as an OSPI supplementary component to trigger potential parking map updates. The evaluation shows that the developed OSPI availability prediction model is on par with state-of-the-art models, despite having simpler but more enhanced and adaptive features. The foundational temporal and aggregated spatial parking capacity features help the most, while the PE-based features capture variances better and enable adaptivity to disruptions. The PE-based features are advantageous as data are automatically gathered daily. For the PBCD model, impacts by construction events can be detected as validation. The methodology proves that it is possible to create a reliable OSPI system with predominantly PE-based features and aggregated parking capacity features.

Keywords

• Change detection
• connected vehicles
• geospatial analysis
• intelligent transportation systems
• machine learning
• parking