Assessing Surrogate Safety Measures using a Safety Pilot Model Deployment Dataset

Abstract

Emerging data sources such as Safety Pilot Model Deployment (SPMD) provide a great opportunity to gain a better understanding of collision mechanisms and to develop novel safety metrics. The SPMD program was a comprehensive data collection effort under real-world conditions in Ann Arbor, Michigan, covering over 73 lane-miles and including approximately 3,000 pieces of onboard vehicle equipment and 30 pieces of roadside equipment. In-vehicle data (e.g., speed, location) collected by the SPMD program can potentially be an important supplement to traditional crash data-oriented safety analysis. The goal of this study was to assess roadway link-level surrogate safety measures using the vehicle trajectory data from SPMD. The study’s objectives included: 1) developing a framework to process the SPMD dataset using big-data analytics; 2) converting raw vehicle motion data from SPMD to surrogate safety measures; and 3) analyzing the statistical relationship between crash records and the calculated safety index. The statistical models showed that modified time to collision (MTTC) outperforms time to collision (TTC) and deceleration rate to avoid collision (DRAC) with respect to its goodness of fit. The findings are promising in that augmenting safety analysis with surrogate measures and vehicle performance (e.g., speed and brake duration from connected vehicles) improves the overall model performance. Such information is vital for safety analysis, especially in the absence of detailed roadway and traffic data.

Review of Usage of Real-World Connected Vehicle Data

“…The newly available CV data collected from CV devices and RSEs give researchers new opportunities to conduct SSM research. Some researchers have attempted to develop SSMs from the vehicle trajectory data of the SPMD program ( 20, 21 ).…”

“…He et al ( 20 ) proposed a framework to process CV data, calculate SSMs and their safety indices, and analyze the correlation between crash records and the calculated safety indices. They calculated three SSMs: time-to-collision, modified time-to-collision, and deceleration rate to avoid collision, at the vehicle level and their safety indices at both the trip level and the link level.…”

“…Both studies He et al and Xie et al ( 20, 21 ) presented detailed procedures for cleaning and processing the CV data from the SPMD program. They determined that CV data could help predict high-risk locations in a very short period (one month and two months) before a substantial number of crashes occur.…”

“…The purpose of this type of study is to conduct proactive safety measures on specific intersections before the occurrence of crashes, and to seek solutions for improving their safety. Two studies ( 20, 21 ) attempt to identify high-risk locations which are not limited to intersections, whereas others ( 25–27 ) focused on evaluating location-based intersection safety because of the high crash frequency detected at intersections.…”

“…Jahangiri et al(16) Random forest classification of risky driving behaviors. Surrogate safety measures He et al(20) Negative binomial regression model: Statistical relationship between the link developed safety surrogate measures and crash frequency. Xie et al(21)…”