Task Order 5208
Transportation Safety Research


Investigation of Driver Behavior at Rail Crossings

David R. Ragland
School of Public Health, Traffic Safety Center
University of California, Berkeley

Introduction

Rail crossing crashes have declined in the past 30 years, both nationally and in California. This is largely because of deployment of a wide range of countermeasures at rail crossings, including signal systems, gating and grade separation programs. However, the number of crashes and subsequent injuries and deaths is still unacceptably high. In the three years 2000-2002 there were 572 highway-rail incidents at California at-grade crossings, resulting in 111 deaths. The problem could increase because of numerous recently constructed or planned light rail or commuter rail systems that cross busy urban streets in many cities (California cities include Los Angeles, San Francisco, Sacramento, San Diego, and San Jose) and also the increased automobile travel demand statewide and subsequent increased exposure.

Rail crossings provide different levels of warnings, from descending barriers or four-quadrant gates down to mere stop signs at private crossings. Subsequently, crashes are either caused by people violating the signs, signals, and gates or people not perceiving an approaching train. Gating systems that cannot be violated are difficult and expensive (as there are 12,784 crossings in California, 7,847 public and 4,777 private) and it is imperative that we conduct research to determine the reason for violations and misjudgment. A considerable amount of research has already been conducted, identifying some of the following factors that may lead to violation or misjudgment.

  • Perception of speed/distance of the train (i.e., time of arrival of the train to the crossing)
  • Perception of waiting time (i.e., waiting time until the train arrives plus waiting time for the train to pass [related to speed and length of the train])
  • Perceived probably of injury given a crash
  • Value placed on different outcomes (e.g., avoiding a crash, having to wait for a train to pass, etc.).
  • Perceived frequency of trains at the crossing
  • Environmental factors (e.g., weather, lighting, traffic conditions, etc.)

Methodology

The goal of this study is to develop a comprehensive model of rail crossing violations, based on Signal Detection Theory (SDT) concepts and incorporating previous research, that can (i) predict violations under different conditions (listed above) and (ii) predict driver response to different countermeasure configurations (e.g., variation in design, timing, etc.). The development of the model will begin with a thorough review of the available literature, and will be an iterative process, providing hypotheses for, and being modified based on, the results of the following two tasks:

  • Develop and demonstrate a video/radar observation model for driver behavior at rail crossings. Driver behavior (e.g., approach behavior, violation, crossing speed) will be studied as a function of objectively measured variables (e.g., speed/distance of train, frequency of trains, length of train, etc.)
  • Develop and demonstrate a model using an instrumented vehicle to study the behavior of individual drivers at rail crossings. Individual differences in behavior as a function of driver characteristics (e.g., age, gender) and driver perception (e.g., perception of speed, distance, frequency, length) will be observed.
This comprehensive model of rail crossing violations will be tailored for use in California, and can be used by Caltrans to (i) conduct a comprehensive evaluation of rail crossings throughout the state and (ii) prepare a cost-effective plan for continued reduction of rail crossing incidents. The project will benefit greatly by leveraging already existing assets at PATH, the Institute of Transportation Studies, the School of Public Health, and the School of Optometry, including extensive prior experience in rail crossing projects, expertise in SDT concepts, expertise in studies of perception, expertise in video monitoring, availability of, and expertise in use of, an instrumented vehicle, and access to bibliographic sources and relevant data bases.

The project will consist of five tasks: (i) Conduct literature/data review, (ii) Develop a comprehensive SDT model; (iii) Develop video/radar observation model; (iv) Develop naturalistic data collection model; (v) Prepare final report.

Task 1. Literature/Data Review

  1. Causal factors/behavioral theories and models/empirical studies In this step we will thoroughly review rail crossing studies, including studies based on analyses of crashes, studies of the judgment and perception of drivers, and models of driver behavior at rail crossings.
  2. COTS signal equipment now available or being developed A wide range of signs, signals, and gates are being used for rail crossings. We will review available control devices with respect to conditions warranting use, cost, reliability, and effectiveness. The review will cover devices that have been in use for some time as well as devices that are being developed (red traffic signals with centered embedded white strobe (Florida), embedded pavement lights studied by Prof. Ted Cohn (UCB/PATH) under a contract with the Office of Traffic Safety.
  3. Collision records We will use a list of California crashes and a database of crossings on the San Joaquin Corridor prepared in a previous Caltrans-funded project (Innovative Grade Crossing Safety Measures for the San Joaquin Rail Corridor) and other databases on line with descriptions of crossing characteristics (e.g. public or private, type of warning, number of tracks, daily traffic). An important outcome will be a list of locations at which to perform tests. We will consider two crossings in close proximity along the same line: one "bad (crash-prone) and one "good" (not crash-prone). The aforementioned "bad" crossing will potentially host one of the new signal developments uncovered in the literature review, or an elaborated wig-wag borne from an OTS-sponsored project, making the two symbiotic.

Task 2. Develop a Comprehensive Signal Detection Theory (SDT)-based Model

This task is initially based on the literature review, and will result in a comprehensive model for predicting violations and misjudgments under different conditions and for predicting driver response to different signal configurations (e.g., variation in design, timing, etc.) or other countermeasures. The development of the model will be an iterative process, providing hypotheses for, and being modified based on the results, of the following two activities.

The first iteration of the model will generate hypotheses, requirements, and specifications for video observation of rail crossings to observe driver behavior under different conditions (Task 3) and direct observation of drivers as they perceive speed and arrival time of trains, and as they make decisions in an instrumented vehicle (Task 4).

Task 3. Develop and Test a Video/Radar Observation Model

In this task we will develop and demonstrate an observation model for driver behavior at rail crossings. Driver behavior (e.g., approach behavior, violation, crossing speed) will be viewed as a function of objectively measured variables (e.g., speed/distance of train, frequency of trains, length of train, etc.). The system will be composed of cameras and radar sensors. The cameras will be mounted to view the grade crossing in each direction of travel of the highway and railroad as well as crossing signal status. The radar sensors will be mounted to record highway traffic in each direction of the highway traffic and one for the railroad traffic. The radar will be used to record the velocity profile, direction, and distance of the vehicles and trains

Task 4. Develop and Pilot a Naturalistic Data Collection Model

This task will consist of two subtasks. The goal of this data collection is to describe nominal driver behavior at a crossing in terms of vehicle control (speed, acceleration) and scanning patterns (data collected via an eye tracker). In the first subtask we will collect data on perception of speed (of a train) by human observers. We will ask our observers to carefully study an approaching train and then to make a judgment as to whether or not a closely following (~ 2 minutes delay) rail-equipped pick-up truck traveling the same direction on the same tracks is going faster or slower (it will have been going the identical speed). We will test the important hypothesis that people underestimate the speed of large objects (e.g., trains). Speed perception can then be factored into our model.

In the second subtask we will study the perception and understanding of rail crossings by developing and demonstrating a model using an instrumented vehicle to study the behavior of individual drivers at rail crossings. Individual differences in behavior as a function of driver characteristics (e.g., age, gender) and driver perception (e.g., perception of speed, distance, frequency, length) will be observed.

Task 5. Final Report

The report will contain (i) summary of literature review, analysis of existing data, and description of existing and newly developed signs, signals, and gates, (ii) description of the model for predicting driver violations and response to countermeasures, (iii) results from the video observation of drivers, (iv) results of observation using the instrumented vehicle and of speed perception survey, and (v) recommendations for a comprehensive plan for reducing rail crossing incidents in California.