Title: Unveiling the Reality: A Segment-Level Examination of Aggressive Driving in Milan Through Telematics, Street Networks, and Google Street View Data

Abstract:

While police-reported crash data serves as the conventional basis for evaluating urban road safety, its utility is often compromised by incomplete records and delays in reporting, hindering the development of timely, precise interventions. Although harsh acceleration and braking incidents are frequently utilized as proxies for safety risks, previous research has largely been confined to limited urban samples. This study expands the scope by examining these harsh events across Milan’s entire urban road network. The analysis integrates high-resolution telematics data from over 4.2 million vehicles equipped with On-Board Units, segment-level traffic indicators from TomTom, infrastructure and street-network details from OpenStreetMap, and visual streetscape characteristics derived from Google Street View. To extract these visual features, we employed semantic segmentation via a OneFormer model.

Our analytical approach merges supervised machine-learning regressors with non-parametric Mann–Whitney U tests, comparing the distribution of segment features between groups exhibiting high and low harshness levels. The findings reveal that, after adjusting for exposure, specific environmental factors correlate with increased harsh-event intensity. These include wider carriageways, the presence of crossings and transit stops, and more open visual fields, indicated by higher proportions of sky and road pixels. Conversely, denser built frontage is linked to reduced intensity.

Additionally, a case study focused on cycling infrastructure highlights a distinct gradient in harsh-event intensity across different facility types. Compared to physically separated cycle paths, markings-only cycle lanes are associated with a 19.5% increase in harshness scores, while mixed-traffic configurations show an 11.5% increase, both conditional on the included control variables. These outcomes advocate for targeted, context-specific urban safety measures rather than one-size-fits-all approaches. Furthermore, the study demonstrates how merging large-scale telematics with open geospatial and visual datasets can effectively guide Vision Zero initiatives at a metropolitan level.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC