Hard-braking events as indicators of road segment crash risk (opens in new tab)

Google Research has established a statistically significant correlation between hard-braking events (HBEs) collected via Android Auto and actual road crash rates. By utilizing HBEs as a "leading" indicator rather than relying on sparse, lagging historical crash data, researchers can proactively identify high-risk road segments with much greater speed and spatial granularity. This validation suggests that connected vehicle data can serve as a scalable proxy for traditional safety assessments.

Data Density and Scalability

• HBEs—defined as forward deceleration exceeding -3m/s²—provide a signal that is 18 times denser than reported crash data.
• While crashes are statistically rare and can take years to provide a valid safety profile for a specific road segment, HBEs offer a continuous stream of information.
• This high density allows for the creation of a comprehensive "safety map" that includes local and arterial roads where crash reporting is often inconsistent or sparse.

Statistical Validation of HBEs

• Researchers employed negative binomial regression models to analyze 10 years of public crash data from California and Virginia alongside anonymized HBE data.
• The models controlled for confounding factors such as traffic volume, segment length, road type (local, arterial, highway), and infrastructure dynamics like slope and lane changes.
• The results confirmed a consistent positive association between HBE frequency and crash rates across all road types, proving HBEs are a reliable surrogate for risk regardless of geography.

High-Risk Identification Case Study

• An analysis of a freeway merge connecting Highway 101 and Highway 880 in California served as a practical validation of the metric.
• This specific segment was found to have an HBE rate 70 times higher than the state average, correlating with a historical record of one crash every six weeks.
• The HBE signal successfully flagged this location as being in the top 1% of high-risk segments without needing years of collision reports to confirm the danger, demonstrating its utility in identifying "black spots" early.

Real-World Application and Road Management

• Validating HBEs transforms raw sensor data into a trusted tool for urban planners and road authorities to perform network-wide safety assessments.
• This approach allows for proactive infrastructure interventions, such as adjusting signage or merge patterns, before fatalities or injuries occur.
• The findings support the integration of connected vehicle insights into platforms like Google Maps to help authorities manage road safety more dynamically.