Road safety research has raised concerns about the quality and reliability of police reports, which often lack pivotal features such as crash coordinates, even in developed countries. To explore possible solutions, this study aims to link geospatial and telematics data to the intersections of a specific study area in Athens and identify and locate intersection crashes registered in the police crashes database. Using the geocodes from the Hellenic Statistical Authority (ELSTAT), which features police datasets, the street names forming the intersection in the police report were identified. Subsequently, the intersections’ coordinates have been retrieved from OpenStreetMap (OSM) for the study area by using the street names. Telematics data for the study area were also obtained. To resolve discrepancies between the street names obtained through geocoding and those from OSM, a Natural Language Processing (NLP) technique was applied. Specifically, a string similarity task which involves encoding strings into contextual embeddings and calculating their similarity. Nonetheless, some data losses occurred while attempting to retrieve the names, primarily due to the limited coverage of the telematics dataset which defined the study area excluding some of the crashes reported by the police. By creating a buffer around each node within the study area, each node has been characterized by the road centroids’ attributes falling within the buffer, from a multi-source dataset containing geographic, transport network and telematics attributes per road. Therefore, the intersections recorded by the police were successfully flagged. The created datasets are analyzed with spatial models and relevant insights are obtained for intersection crash occurrences. Speeding and road angle were identified as significant factors highly correlated with the intersection crash occurring as well as a crash index.
| ID | pc577 |
| Presentation | |
| Full Text | |
| Tags |
