A unifying view on traffic conflicts and their connection with crashes

Area of Science:

• Traffic safety research
• Transportation engineering
• Human factors in driving

Background:

• Decades of research on traffic conflicts and crash surrogates highlight the need for consistent definitions and crash connections.
• External observations of traffic events do not fully capture safety-critical situations, necessitating consideration of unobservable factors.
• Existing methods for crash prediction and traffic conflict analysis face challenges in accurately modeling rare, severe crash events.

Purpose of the Study:

• To establish a consensus on defining traffic conflicts and connecting them to actual crashes through a unified theoretical construct.
• To introduce the concept of 'crash nearness' using both externally observed and driver-preferred values to better understand safety-critical events.
• To identify conditions for accurately detecting traffic conflicts and estimating crash probability, particularly for safety analysis with limited data.

Main Methods:

• Theoretical discussion of traffic encounters and conflicts, integrating empirical evidence and established theories.
• Analysis of two distributions of crash nearness: observed (external) and preferred (driver-internal).
• Identification of conditions for traffic conflict detection based on the SHRP2 study, including speed thresholds and encounter types.

Main Results:

• The difference between preferred and observed crash nearness is defined as the delay in response to a violation of preference.
• Traffic conflicts stemming from driver errors violating minimum crash nearness are recommended for safety analysis when only external data is available.
• Sufficiently high speeds and the elimination of self-clearing encounters are identified as critical conditions for proper traffic conflict identification.

Conclusions:

• High speeds in traffic encounters correlate with serious, reportable crashes, validating their use in safety analysis.
• Autonomous vehicles offer potential for improved safety analysis by providing known preferred crash nearness values.
• Accurate modeling of crash risk requires aligning observed event distributions with crash event distributions, a challenge autonomous vehicles may help overcome.