Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
1This report describes work done during Phase 2 of SHRP 2 Safety Project S01A, Development of Analysis Methods Using Recent Data. In the report submitted at the end of Phase 1, three research problems were identified on which progress was needed. The first of these was identi- fication of an appropriate class of structural models describing how crash and near-crash events developed, together with analytic tools for fitting these models to data expected from the vehicle- based and site-based field studies. The second problem involved counterfactual screening of supposed near-crash events to determine their similarity to crashes. The third problem involved developing plausible models of how drivers select evasive actions as functions of the situations in which they find themselves. Solutions to the second and third research problems are contin- gent on solution of the first, so the bulk of the project teamâs effort during Phase 2 has been devoted to structural modeling of crash and near-crash events, using data from the 100-car vehicle-based field study, from site-based video data collected by the Minnesota Traffic Observa- tory (MTO), and from site-based Doppler shift data obtained from the Cooperative Intersection Collision Avoidance Systems (CICAS) project. The teamâs focus has been on crashes and near crashes involving more than one vehicle, of the sort that occurs at intersections. Background Chapter 1 of this report outlines the context within which this research has taken place. In the United States, there are two substantial national efforts related to road traffic safety. The first is the development of the first edition of a Highway Safety Manual (HSM). The second is the design and execution of the SHRP 2 Safety field studies. The project team points out that it may be possible for the data collected in the SHRP 2 field studies to support the development of microscopic (i.e., individual event) crash models, which can be incorporated into traffic simu- lation models to supplement or replace the macroscopic regression methods used in the first edition of the HSM. For this to occur, though, it will be necessary to identify and fit plausible microscopic models of crash-related events using the SHRP 2 field data. The team illustrates how this might be accomplished using a simple braking-to-stop model applied to trajectory data extracted from site-based video and then illustrates how once a fitted model is at hand, it is pos- sible to quantify the expected number of crashes in a set of noncrash events. Chapter 2 takes up the problem of extending these ideas to more complicated situations, and the project team proposes a modeling strategy where driver behavior is treated as a piecewise constant sequence of acceleration changes. Given such an acceleration history and initial values for a vehicleâs location and speed, it is logical to move toward a system of ordinary differential equations to get predicted histories of the vehicleâs speed and position. Fitting such a model then involves identifying the appropriate break points in the acceleration profile, the corresponding acceleration levels, and the initial conditions that best fit observed trajectory data. The team Executive Summary
2illustrates model identification and estimation using speedometer, radar range, and radar range- rate data for a near-crash event from the 100-car vehicle-based study. The team also illustrates a what-if counterfactual analysis where the final deceleration of the following vehicle is varied over a range of values, for each of which, other things being equal, the probability that a collision would have resulted is computed. Findings Chapter 3 describes the project teamâs work with data from the 100-car study. Data were obtained for 33 crash and near-crash events, each consisting of approximately 30 s of forward video and 30 s of instrumentation measures at 10 Hz. The instrumentation measures included speeds from the instrumented vehicleâs speedometer, range and range rate for objects ahead of the instru- mented vehicle from its forward radar, accelerometer measures, GPS positions, heading, yaw measures, and indicators of brake, accelerator, and turn signal use. After reviewing the data, seven of the 33 events were determined to have data sufficient to attempt modeling, and model identification, estimation, and goodness-of-fit evaluation using speedometer, range, and range- rate data are described. Six of the events involved a leading vehicle and a following vehicle decel- erating in the same lane of traffic, and it was possible to identify plausible acceleration profiles for the leading and the following drivers. From these measures, it was then possible to estimate the following driverâs reaction time, together with measures describing the situation at the time his or her reaction phase began. The team also reconstructed one event involving a swerving maneuver by the following driver, where the forward radar data were limited. Chapter 4 describes work using site-based video data from the University of Minnesotaâs Beholder system. This is a set of video cameras, computers, and wireless communication equip- ment positioned to overlook a section of Interstate 94 in downtown Minneapolis. Vehicle posi- tions, extracted from video recordings and rectified for camera position effects, provided the raw data for the analyses. Four rear-end crash events in JulyâOctober 2008 were analyzed, with two of these events also including trajectories for vehicles not involved in the crash. In each case, it was possible to identify plausible acceleration profiles for each of the involved drivers, which in turn provided information on reaction times and conditions at the start of the reaction phase. Chapter 5 describes pilot work using site-based Doppler shift data collected by the University of Minnesotaâs Intelligent Vehicles Laboratory as part of the CICAS project. This configuration consists of a coordinated set of radar units collecting information on the positions and speeds of major approach vehicles approaching a two-way-stop controlled intersection, and LIDAR units collecting information on the positions of minor approach vehicles. Since the instrumentation configuration was designed to support a prototype driver-information system and not to collect data on vehicle trajectories, data acquisition and preparation were not as straightforward as with the other data sets; but with some data mining and postprocessing, it was possible to identify one event suggesting braking on the part of a major approach vehicle in response to minor approach crossing. It was then possible to identify and fit plausible acceleration profiles for both drivers. Chapter 6 presents the conclusions and recommendations for further research. Conclusions 1. At least for situations where direction of travel is roughly constant, trajectory-based recon- struction of crash-related events, where trajectory data are used to fit parsimonious models of driver behavior, is feasible using both vehicle-based and site-based data. 2. It is possible to extend the methods of counterfactual analysis to more complicated structural models involving differential equations. 3. At least for rear-ending events, there is some limited evidence that the distributions of evasive actions for crashes and near crashes share some overlap, so that it should be possible to find near-crash events that are similar in other respects to crashes.
34. Although the CICAS system as currently configured was not designed to collect and process crash and near-crash trajectory data, with technical modifications it could support site-based field research, at least at lower-volume intersections. 5. The usefulness of the data produced by the SHRP 2 vehicle-based field study will be strongly dependent on the ability to calibrate and maintain the data-collection systems. Recommendations 1. The modeling methods presented in this report should be extended to handle two-directional trajectories. 2. When attempting to include crash events in a microscopic traffic simulation, plausible models that close the feedback loop between existing conditions and driver actions will be necessary. This issue should be pursued using the data from the SHRP 2 field studies. 3. The trajectory modeling methods described in the report should be enhanced to allow for possible serial correlation in trajectory data. 4. Compiling data on gap-selection and other intersection-related events will require a data setup different from the SHRP 2 vehicle-based field study, so the SHRP 2 vehicle-based field study should be complemented with site-based research. 5. Clear descriptions of data collection and processing and associated metadata should be required in future major data-collection efforts.