U.S. Department of Transportation
Federal Highway Administration
1200 New Jersey Avenue, SE
Washington, DC 20590
Federal Highway Administration Research and Technology
Coordinating, Developing, and Delivering Highway Transportation Innovations
|This report is an archived publication and may contain dated technical, contact, and link information|
Publication Number: FHWA-HRT-10-038
Date: October 2010
Balancing Safety and Capacity in an Adaptive Signal Control System — Phase 1
2.0 SIGNAL TIMING AND SAFETY
Attributing safety effects to changes in signal timing and phasing parameters and separating these effects from those attributable to the intersection environment is not a simple exercise. For example, a crash pattern at an intersection might be attributed to poorly placed signal displays for an approach where drivers cannot discern the display until the last moment, or the crash pattern might be attributed to a poorly timed offset causing vehicles to stop unnecessarily. The complexity of adaptive signal control compounds the difficulty of this investigation because rather than remaining fixed, the timings change on a cycle-by-cycle basis.
Table 1 illustrates some of the hypothesized relationships between signal timing parameters and safety, as measured by traffic conflicts. For example, it may be found that rear-end crashes tend to be more common when shorter cycles and splits are used. If this finding were valid, then a logical conclusion would be to employ signal timing parameters that result in longer phase times (longer cycles and splits) to reduce the rate of rear-end crashes or conflicts. It is also likely that the correlation between safety and signal timing parameters will have to consider the level of traffic demand with respect to the value of the parameters.
Generally, rear-end crashes and conflicts were expected to be impacted the most heavily by changes in signal timing parameters, as shown in table 1. Therefore, it was important to identify measures from real-time detector output data that were correlated with rear-end crashes and conflicts. For example, higher lane occupancies and uniform speeds at concentrated times during the signal cycle on a coordinated approach indicate the presence of dense platoons, offsets that are working as intended, and, thus, less potential for rear-end crashes. Conversely, larger headways and less uniform lane occupancies during the signal cycle due to incorrect offsets might indicate higher potential for rear-end crashes. This and other similar relationships were investigated in this research and were used to develop a methodology to mathematically determine changes to signal timing parameters that balance safety and efficiency.
Similar explanatory reasoning can be applied to each of the cells in table 1 to determine a potential correlation between a signal timing parameter and a specific type of conflict or crash. For example, continuous changes in signal timing splits in an adaptive control system might generate conflicts between vehicles and pedestrians due to the fact that pedestrians and motorists typically expect consistency in signal timing intervals on a cycle-by-cycle basis. Shortening a left-turn split may induce a higher rate of left-turn crashes due to drivers making more risky maneuvers. Extending the main street split may induce red-light running on the side street. There are opportunities to reduce stops, delays, and blocking problems by changing the sequence of signal phases, including switching a left-turn phase between leading and lagging, skipping phases, or servicing a phase twice in a signal cycle. Changing the phase order may be particularly helpful to reduce blocking problems, for priority vehicle operations, or to support advanced traffic control strategies at closely spaced intersections such as freeway interchanges. The typical reason for not changing phase sequences is the belief that doing so violates driver expectancy and confuses pedestrians. In reality, there is little research to support these concerns. Altering phase sequences may have a positive impact on safety by reducing blocking and spillback between intersections and by decreasing driver frustration, which are both factors that contribute to crash potential. This research investigated these operational conditions and identified a methodology to detect and mitigate such occurrences in real time.
The literature analyzing safety performance of signal timing settings is limited because of the complexity of the problem due to the potential input variables, the infrequent and random nature of crashes, and the effects of regression to the mean. Causal linkages between potential input variables and crash frequency are not commonly found with strong correlations. Regression to the mean is a particularly challenging complication since locations with high levels of crashes in one reporting period may have a lower level of crashes in the following reporting period with no improvements to the operational or geometric parameters of a location.
Existing state-of-the-practice safety performance models for signal-controlled intersections predict crashes based on a variety of input variables typically related to the geometric characteristics of the facility and the traffic flows. A typical safety performance model is expressed as a nonlinear regression equation where the primary inputs that predict crash frequency are the crossing flows as follows:
Variables a, b, and c are parameters that are fit to individual datasets, which are nonnegative numbers typically in the range of [0.5, 1.5]. The coefficients b and c in most fitted models are ~0.5, so it is commonly asserted that crash frequency is driven by the square-root of the product of the crossing flows. Units of Vcross and Vmain are typically expressed in average annual daily traffic (AADT). Typical regression fitting performance is not particularly impressive due to the extremely rare nature of crashes and the myriad exogenous influences that lead to crashes not reflected in any controllable design parameters of an intersection. Good performance of a safety prediction function may result in R2 values of 0.45–0.5.
Past studies have most often analyzed CMFs in a one-by-one fashion. For example, if the base condition for a signalized intersection is to share the left lane with a through lane—a scenario that is becoming more and more rare since the safety effect is substantial—the addition of a left turn bay at a certain location might be predicted to reduce the crash rate by, say, 30 percent. So, a CMF for a left-turn bay would be expressed as a 0.7 multiplicative factor to the base model.
Other CMFs are applied similarly in a multiplicative fashion, so it is possible that a 30 percent reduction for the left-turn bay can be offset by, for example, a 25 percent increase due to on street parking. There have been no CMFs developed for common signal timing inputs such as cycle time, offsets, splits, phase sequence, or detector extension times.
To estimate the safety of various traffic facilities, including facilities that have not yet been built, research has focused on the establishment of safety performance functions that relate the number of crashes or crash rate to a number of operational (e.g., AADT, average speed) and non operational independent variables via a typically complex regression equation, including but not limited to AADT, occupancy, volume to capacity (V/C) ratios, and products of crossing volumes.(4,5) Calibration is then required to choose the equation parameters for the best statistical fit to the available data.
Studies performed by Gettman et al. identified research that was done on Bayesian methods and advanced statistical techniques (e.g., classification and regression trees) for revising crash estimates based on observations as a way to develop safety estimates for facilities with no crash data.(4,5) Other methods for combining crash rates and other measures into safety level of service measures or common indices based on one type of crash (e.g., property damage only) have also been proposed. These approaches all use macroscopic measurements of total flows rather than recording individual vehicle movements or events to develop safety level of service estimates. Despite the large body of safety modeling research, absolute numbers of crashes and crash rates are still difficult to predict accurately. This has led to increased interest in obtaining surrogate measures that reflect the safety of a facility or at least the increased probability of higher-than-average crash rates for a facility.
By definition, a conflict is an observable situation in which two or more road users approach each other in time and space to such an extent that there is risk of collision if their movements remain unchanged. The traffic conflicts technique is a methodology for field observers to identify conflict events at intersections by watching for strong braking and evasive maneuvers. Conflict methods have a long history of development, including research on topics such as recommended data collection methods, definitions of various types of conflicts, severity measures, how conflict measures are related to crash counts, how conflicts are related to specific crash types, standards for data collection, and standard definitions of conflict indices as used to compare the performance of multiple facilities.
The fact that the subjectivity of field observers introduces additional uncertainty into the collection of data on conflicts leads to a debate of the connection between conflict measures and crash predictions. Conflict studies are, however, still used to rank locations with respect to safety to identify construction upgrades. There is general consensus that higher rates of traffic conflicts can indicate lower levels of safety for a particular facility, given that conflicts generally result from a lack of or misunderstanding of communication between different road users.(4)
Tabulation of total numbers of traffic conflicts indicates frequency, one part of the safety issue. The other element of the safety issue is the severity of the conflicts that occur. The primary conflict severity measure that has been proposed is TTC.(7,8) Some researchers have indicated that TTC is the surrogate measure of safety, while others refute that lower TTC indicates higher severity of crashes, primarily because speed is not included in the measure.(8,9) That is to say that lower TTC certainly indicates a higher probability of collision but cannot be directly linked to the severity of the collision. Gettman et al. also state that others identify the deceleration rate (DR) as the primary indicator of severity instead of TTC.(5) Some of the common measures defining and characterizing a conflict are presented in table 2.
Table 2. Measures of conflict severity.(5)
Microscopic simulation is generally required for generating and collecting conflict severity statistics and/or other surrogate measures that require detailed information on vehicle acceleration, deceleration, position, etc. as a substitute for field studies. Simulation models have been built specifically for the simulation of a particular conflict type. Other models are based on varying approaches to the computation of conflicts.(5,6) One model by Fazio et al. contains a comprehensive treatment of conflict types and surrogate measures for both signalized and unsignalized intersections.(10) Special-purpose simulations are problematic in application since the level of detail and variety of modeling variables available to the user are typically compromised.
Some efforts prior to the FHWA SSAM project had focused on the modification of multipurpose traffic simulation models to include conflict statistics or other surrogates.(5) A brief overview of these simulation applications and their crash prediction indicators is extracted from the recent SSAM research and shown below.(5) They include the Helsinki Urban Traffic Simulation (HUTSIM); Transportation Analysis and Simulation System (TRANSIMS); Integrated Traffic Simulator (FRESIM), which is part of Corridor Simulation (CORSIM); Network Simulation (NETSIM), which is also now part of CORSIM; Traffic Experimental Analytical Simulation (TEXAS); Advanced Interactive Microscopic Simulator for Urban and Non-urban Networks (Aimsun); and INTEGRATION. A brief description of these models' crash prediction or safety indicator capabilities is as follows:
The FHWA SSAM project extended upon the approaches used in the preceding simulation models by combining microsimulation and automated conflict analysis to analyze the frequency and character of narrowly averted vehicle-to-vehicle collisions in the simulated traffic situation. The SSAM software application was developed to automate conflict analysis by directly postprocessing vehicle trajectory data from the simulation model. Researchers specified an open standard, universal vehicle trajectory data format designed to provide the location and dimension of each vehicle approximately every tenth of a second. The trajectory file format is currently supported as an export option by four traffic microsimulation models: VISSIM®, Aimsun, Quadstone Paramics, and TEXAS. It is hoped that in the coming years video processing technology will be capable of automatically extracting vehicle trajectory data adequate for SSAM processing from real-world sites. In addition, the approach could be applied to real-time analysis when IntelliDriveSM technology is ubiquitous.
There is limited quantitative research on surrogate measures for safety assessment. The main difficulty is illustrating the correlation between any proposed surrogates and crashes since crashes are rare events. The available literature is focused mainly on various aspects of traffic conflicts and related field studies for obtaining surrogate measures. Given the technical difficulty and cost of field studies, use of simulation models has been proposed, and some previous work has been done to develop specific models for simulating conflicts. The most notable surrogate measure of the severity of a conflict is TTC, although other surrogates such as PET and DR have been used to measure other characteristics of conflict situations. Only limited effort has been expended to modify or enhance existing general-purpose, microscopic simulations to obtain conflict or other surrogate measures for intersections and two-lane roads. The primary difficulty is defining a set of surrogate measures that: (1) have meaningful implications since they are extracted from simulations that were specifically designed to be crash free and (2) have reasonable connectivity to safety assessment of particular facilities using traditional measures (e.g., the frequency and severity of resulting crashes).
Except for the SSAM model, there is currently no calibrated tool that lends itself to modeling and predicting conflicts at signalized intersections to determine correlations between modifications to signal timing parameters and resulting intersection safety. Therefore, this research project applied the SSAM methodology and software to explore the various relationships between safety and signal timing parameters.