Most of the previous evaluation studies, either in the field or through microsimulation, used common traffic performance measures (PMs) related to travel times, number of stops, control delay, etc. to evaluate various traffic signal control strategies. Recent advancements in detection and controller data-logging capabilities have brought new types of signal performance measures. Two such measures known as Arrivals on Green (AOG) and Approach Delay (AD) have been recently used in a number of projects by other universities and state departments of transportation. Study that evaluated InSync through these high-resolution performance measures has shown that AOG as a single performance measure cannot be fully trusted when comparing various traffic signal alternatives, as it does not reflect efficiency of the allocated green time used by arriving platoon of vehicles. Findings also suggested that current used methodologies underestimate delays as they simplifie their computation. This study proposes development of a new high-resolution PMs – Average Arrivals on Green Ratio (AAOGR) that takes into account variability of cycle length and duration of green time, on a cycle-by-cycle basis (thus applicable for adaptive operations), and a discharge rate-based model for computing Approach Delay, implementable in the field. The findings show that AAOGR is a perfromance measure more applicable for comparative evaluations – from adaptive operations to signals from various corridors. Similarly, newly proposed Average Delay better estimates true average delay per vehicle, if measured in front of the appraoch detectors.

This study investigates the use of Basic Safety Messages (BSMs) as the only source of vehicular data for cycle-by-cycle queue length estimation.  The proposed approach is based on shockwave analysis.  The study also investigates the factors affecting the estimation accuracy.  Three algorithms are developed to measure: (1) residuals from previous cycle ( R _a), (2) maximum queue length, and end of cycle residual (R _b).  Simulation data collected for three coordinated intersections were used to test the proposed approach.  The results confirm that the queue length estimation is a probabilistic process affected by the stochastic nature of the traffic.  This probabilistic nature is defined by a queue formation coverage index (QI) that proved to significantly affect the queue length estimation accuracy.  Despite the results show no specific trend for the maximum queue length estimation accuracy over the different market penetrations, the estimation errors are between 0 and 33% which is acceptable.  For the residuals estimations, R _b algorithm shows overall acceptable results as compared to R_a  algorithm.  Yet, further testing with more residuals occurring in the actual data is required.  Overall, the results show that the proposed approach showed acceptable performance especially with a coverage index higher than 50%.  In addition, despite BSMs provide more information about the vehicle, the market penetration is still an issue.  In such limited data environments, the queue formation coverage index (QI) showed potential to help as an assessment tool to evaluate the obtained estimates.

The objective of this study is to develop a probabilistic method to evaluate the uncertainty of safety performance of the permissive right-turn phase at signalized intersections in Japan, where the traffic is left-hand drive. In this study, safety performance is estimated by occurring opportunity and severity of traffic conflicts caused by gap acceptance of right-turners. The occurring probability of traffic conflicts is estimated based on the length of unsaturated flow discharge interval of the opposing through traffic and the gap acceptance probability of right-turners. The severity of traffic conflicts is measured by Post Encroachment Time (PET) of the right-turning vehicle and the gap-ending opposing through vehicle at the conflict point. The uncertainty of safety performance is evaluated based on the probability distribution of the sum of 1/PET per cycle. Uncertainty of input variables is analyzed based on the field data collected at 9 intersections in Nagoya. The applicability of the proposed method is demonstrated by the use of Monte Carlo Simulation technique. Its effectiveness is validated by comparing the estimated traffic conflicts with the actual observed ones. Results indicate that the proposed method can provide more reliable results for safety estimation of the permissive right-turn phase. Unlike the traditional methods, the proposed method accounts for the randomness of traffic flow and driver behavior variables more comprehensively. It is thus more sensible to traffic conditions and driver behavior, and is capable of reflecting different levels of decision-makers’ tolerance for risk as well as the variety of traffic conditions.

Identifying detector malfunctions at signalized intersections has been addressed using a variety of methods over the past 40 years.  Most modern controllers have some basic heuristics for identifying detector faults that require too much workload and tuning and are rarely used.  The use of high-resolution data collected by traffic signal controllers has been developed and used for engineering related performance measures over the past ten years.  This data can also be used to develop maintenance related performance measures to help traffic signal system operators find and correct faulty or misconfigured equipment in a timely manner, returning the system to optimal operation and efficiency.  This paper proposes using high resolution data to develop statistical distributions of functioning detectors and compare those distributions with daily distributions during defined times of day to identify statistically significant deviations.  The Kolmogorov-Smirnov (KS) test is used to compare current and past vehicle detector operation, and is shown to effectively identify anomalies.  An algorithm coupling this KS test (for longer term anomaly detection) with a further test comparing the standard deviation in detections (for shorter term anomalies) is demonstrated to identify detection errors on two real-world traffic signal systems. This approach was applied to data collected from Elkhart County, Indiana, and correctly identified an un-operational loop amplifier affecting two detection channels.

Traffic signal system is one of key components in advanced transportation management system (ATMS) and traffic signal simulation plays a major role in estimating the performance of ATMS system. In this paper, we present a new concept of traffic signal simulation framework, referred to as "field-ready" hardware-in-the-loop signal simulation framework. With this new framework, Innovative control strategies can be hosted in a hardened single-board computer (SBC) and completely separated from traffic simulation engine. The SBC exchanges data with the software-in-the-loop signal emulators in traffic simulation via the National Transportation Communications for ITS Protocol (NTCIP). The advantages of this new simulation framework include: (1) the same SBC containing the control strategies can be directly connected to hardware NTCIP controllers in the field with little additional development efforts; (2) the SBC will not compromise the existing signal safety protections in signal cabinets. To demonstrate the potential of the new framework, a "field-ready" adaptive signal control prototype system are designed and evaluated within a simulation environment. The results of the case study revealed a significant reduction in traffic delay and vehicle stops. It is expected that this new traffic signal simulation framework will greatly facilitate the deployment of ATMS in the future.

In this paper, a control algorithm is presented that integrates connected vehicles in the feedback loop of traffic signal control, which results in highly flexible, signal-group based signalization and speed adaptation of vehicles. The method is based on Model Predictive Control and incorporates a mutual optimization of both traffic signal timings and vehicle trajectories. In light of emerging communication technology, cooperative vehicles are expected to deliver more detailed data about the current traffic flow compared to stationary detection. This data can be used to influence the signal timing. By capitalizing on the possibility of providing information to cooperative cars, a second means of influence is enabled: Information about future signal timing can be provided to the drivers and hence, further reductions in the number of stops and an increase of traffic flow at the beginning of the green time can be achieved. The complexity increases when both ways of influence are combined, which is often omitted in previous research. This combination is addressed in this paper by introducing an optimized signal control with an integrated speed advisory system. Another functionality of the presented algorithm is the possibility to adjust the predictability of signal timings to account for the reliability of speed advisory messages. A simulation study is carried out as a proof of concept and to evaluate the trade-off between optimality and predictability of the traffic signal control algorithm.

In this paper, we compare the performance of a few feedback traffic signal control policies that take into account real-time queue length measurements to dynamically change green time allocations among various phases. We specifically focus on decentralized control policies whose implementation at a given intersection requires information about the queue length associated only with the movements passing through that intersection. Within this class, we compare the performance of proportionally fair (PF) controller and a few variants of the max pressure (MP) controller. While most variants of MP controller require knowledge of turning ratios and saturated flow capacities for their implementation, the PF controller does not require this information. In spite of such a minimalist nature, simulation studies in a microscopic traffic simulator that we report in this paper, suggest that PF controller has at least the same throughput and lower average travel time in comparison to MP controllers. A closer look at the queue length time series shows a consistent phase shift between adjacent links under MP controllers, which could possibly be responsible for larger queue lengths and hence greater travel time.

This paper analyses the performance of centralised and distributed systems for urban traffic control with consideration of real time variations and uncertainties in traffic flow. The centralised control system is represented by the global brute force and genetic optimisation algorithms. The distributed systems considered herein include the linear quadratic regulator (TUC), demand-based and max-pressure controllers. It is known that the brute force approach takes enormous time to compute the global optimal solution and it is rarely used in general applications. Nevertheless, the brute force solution can be taken as a useful benchmark for evaluating the performance of each controller. Considering its plausibility and parsimony, the traffic dynamics is modelled by a cell transmission model. The control systems are applied to one-dimensional arterial and two-dimensional networks with different demand levels and distributions. The performances of the controllers are evaluated through a Monte Carlo stochastic simulation framework which incorporates various sources of variations that mimic the traffic variability observed in the real world. As expected, the centralised strategies generally outperform the distributed ones. However, the differences between the performances of the two systems are indeed not as significant as expected. It is found that most benefit gained from centralised control comes from its different setting of signal offsets. This suggests investigating calculation of offsets plays an important role in bridging the gap between the computationally demanding centralised systems and the parsimonious distributed systems.

Conventional Adaptive Traffic Signal Control (ATSC) systems typically use data from fixed traffic sensors (usually Inductive Loop Detectors (ILD)) to estimate the traffic flow parameters required for their signal timing optimization algorithms.  These parameters usually include traffic demand, approach speed, turning movement counts or ratios, and saturation flow rate. Obtaining these parameters using fixed point detectors requires multiple detectors on each approach, increasing the deployment, operating, and maintenance costs of these systems. Connected Vehicle (CV) technology enables individual connected vehicles to wirelessly broadcast their real-time travel data, which can be used in ATSC systems and to potentially reduce their dependency on fixed point detectors.

This paper presents an enhanced method for estimating lane saturation flow rates in real-time using CV data. The methodology has been evaluated using microsimulation on a multilane signalized arterial corridor for a range of approach geometries, market penetration rates of connected vehicles, and traffic demands. Regression analysis showed that there is a non-linear relationship between the market penetration rate of connected vehicles and the mean absolute relative error of the estimated saturation flow rate using the proposed methodology at each level of service. In general, estimation error increases as level of service (LOS) becomes worse and decreases as the CV level of market penetration (LMP) rate increases.  The maximum error was observed for LOS F and LMP = 10% but was only 6%.  These results suggest that the methodology is likely sufficiently accurate for practical applications across a wide range of geometries, traffic conditions, and CV LMPs. 

In this study, driving behavior was examined under two different signal phase-switching modes, namely green flashing mode and green countdown mode. Two typical at-grade intersections in Shanghai were selected for a case study. Video techniques were employed to obtain traffic data and capture drivers’ behaviors during the period of a signal switch from green to red. Both qualitative and quantitative analysis were conducted to identify drivers’ behaviors and adaptability, considering driver's psychology, running red lights and speed characteristics. Compared to green countdown mode, green flashing mode results in lower rates of red-light running and rushing, a higher rate of stopping, a lower speed/acceleration dispersion rate, lower average operating speed, and smaller speed fluctuation. These indicate that green flashing mode could induce more conservative maneuvers and hence enhance overall safety, while green countdown mode could cause more aggressive driving behaviors since drivers exactly know the end point of a green light. The downside of green flashing mode could be associated with the increased risk of car-following cases, due to unexpected stops or sudden speed drop-down of leading vehicles at intersections. The efforts were made to enrich the research findings of traffic signal control, providing references for the compilation of relevant norms and specifications and provide the basic for further development of ITS.

Keywords: Phase-Switching Modes; Driving Behavior; Green Flashing Mode; Countdown Mode

Driver-decisions in dilemma zones (DZs) during a traffic signal change interval plays a significant role in determining the level of safety at signalized intersections. The key objective of this study was to evaluate different DZ protection methods, and provide general guidelines on the strengths and weaknesses of each method. We present general guidelines and a software tool (DZ-Pro) that can assist engineers and analysts select the best DZ-protection method, based on a comparison of estimates of different methods’ predicted safety benefits, for a given set of traffic conditions. The guidelines indicate that the optimal multi-detector loop setup is different depending on traffic volume, and provide volume thresholds and a graph that can assist in designing a multi-loop system. The guidelines also indicate that green-termination and radar-based protection are superior to the multi-loop system, with the radar-based system providing the most protection. The DZ-Pro tool can be used to estimate the number of vehicles caught in DZ with each system.

With the rapid technological breakthroughs in the last years in the infrastructure-to-vehicle, vehicle-to-vehicle and vehicle-to-infrastructure wireless communications, a possibility of creating new concepts for traffic signal control emerged. Green Light Optimized Speed Advisory (GLOSA) approach uses signal control information and current position of the vehicle to recommend the appropriate speed for each vehicle in the traffic flow in order to reduce the number of stops on traffic signals. Recent research predominantly focused on environmental and performance measures resulting from GLOSA implementation. Purpose of this paper is to examine the effects that GLOSA has on vehicles arriving from un-signalized driveways or side approaches in terms of delay, capacity and surrogate safety measures. Highly calibrated and validated VISSIM simulation model of five intersection segment of 3500 South, urban corridor in Salt Lake City, Utah is used as a test environment. Different scenarios of various traffic loads and GLOSA activation have been defined and simulated. Outputs from VISSIM were used for evaluation of performance measures. The trajectory files were imported and processed into FHWA’s Surrogate Safety Assessment Model (SSAM) for evaluating the traffic safety effects of GLOSA by analysing the vehicular conflicts data. Results were statistically analysed and conclusions were made. Findings show that GLOSA has a significant impact on delay and very little impact on capacity and number of conflicts. Type of traffic signal control matters, as well as geometry and influence of the nearby signalized intersections. It has been proven again that GLOSA works much better for fixed-time signals than for actuated-coordinated control.

Recent advancements in high-resolution signal controller data logging and acquisition technologies have enabled development of fundamental performance measures that are effective at identifying split failures and rebalancing green times. However, from a system operator perspective it is important to scale these metrics system-wide for monitoring, triaging, and decision-making uses. This paper articulates the questions that concern a typical system operator, and the associated performance measures and tools that can be used to actively manage such systems macroscopically. To illustrate this methodology, the paper examines 70 signals across seven corridors over a six-month period. Approximately three billion high-resolution signal events including 217 million phase occurrences are analyzed, with 126 million of those phases occurring during the 0600–1900 time period. This paper proposes a scalable method to identify phase occurrences with insufficient green. Applying a previously-developed split failure performance measure for this purpose, over 106,000 split failures (0.08% percent) of the phase occurrences are estimated to have failures for the time of day period. These are displayed using a graphical visualization we called the Purdue Split Failure Ticker (PSFT). Subsequent sections of the paper recommend Pareto ranking, drill-down analysis by movement and hour, and corridor movement split summaries to identify where to prioritize resources for signal re-timing activities and geometric improvements.

In the paper, two stop bar vehicle detection technologies, wireless magnetometers and span wire cameras, are evaluated. Those technologies are considered to be practically suitable for large intersections, where span wire is generally required to achieve longer spans. To explore potential factors underlying variation in detection errors of those technologies, conditional inference trees were used. The results indicate that the wireless magnetometers are quite robust to various weather and environmental conditions if there is reliable communication between in-pavement magnetometers and the repeater and/or access point. The addition of a repeater at the test site significantly reduces false and stuck-on call errors potentially caused by the frequent passing of heavy trucks. In contrast, the span wire camera is more susceptible to various weather and environmental conditions, and wind appears to be an influential factor, especially for false call errors.

The object of this research was to develop guidelines for time-of-day use of permissive left-turn phasing, which can then be implemented using flashing yellow arrows (FYA). This required determining how the risk for left-turn crashes varied as traffic conditions varied during the course of a representative day. This was accomplished by developing statistical models which expressed how the risk of occurrence of a left-turn crash varied as a function of the left-turn demand, the opposing traffic volume, and a classification of the approach with respect to the opposing traffic speed limit, the type of left-turn protection, and whether or not opposing left-turn traffic could obstruct sight distance. The models were then embedded in a spreadsheet tool, which allows signal operators to enter, for a candidate intersection approach, existing turning movement counts and a classification of the approach with respect to speed limit, turn protection, and sight distance issues, and receive a prediction of how the risk of left-turn crash occurrence varies throughout the day, relative to a user-specified reference condition.

Turning movement count data, i.e., vehicle volumes broken down by movement, approach, and time periods, is the foundation of signal performance evaluations and a crucial component of a data-driven decision making process used by transportation agencies. In this paper the authors show how existing data available from intersections equipped with radar-based vehicle detection can be used to produce turning movement counts. A classification algorithm developed and discussed by the authors is capable of producing turning movement counts regardless of lane configuration and without the need for detection zones definition. The algorithm works by leveraging the power of data produced by vehicle detection systems that goes unused and never is communicated to the signal controller. The nature of the data collection process used makes the algorithm presented one that is independent of the controller type used. Results from the algorithm are promising as an average error of -0.26 vehicles during a 15 minute count period (absolute error of 2.31 vehicles) was obtained through the use of the algorithm. Furthermore, the application of the algorithm presented provides an opportunity beyond signal operations. Processed trajectory data and results from the algorithm could be used to break the boundary that often exists between operations, planning, and safety thus showing how a monitoring system that relies on the algorithm can help a traffic monitoring program meet the different, and sometimes competing, interest of agencies.

Rural, high-speed signalized intersections are associated with vehicle crashes due to dilemma zone problems. Dilemma zones (DZ) are areas where, at the onset of yellow, some drivers may decide to proceed, and some may decide to stop. This disagreement among drivers can lead to rear-end crashes and/or right-angle crashes. The research objectives of this work was to conduct a field comparison of a radar-based dilemma zone protection system using high-resolution data. Namely, we aimed to compare the frequency of red light runners (RLR) and percent arrivals in green (POG). In addition, it was also desired to shed more light into the value of using high-resolution data in such comparison, and how much data is needed.

Researchers found that one week of high-resolution data is sufficient to draw conclusions on the effectiveness of radar-based DZ protection system. A before-and-after comparison of red light running frequency indicated an up to 80 percent reduction in red light runners when a radar-based DZ protection system was used.

Video imaging technology has long been used for vehicle detection at signalized intersections. It provides many advantages over conventional inductive loops. Video imaging cameras are easier and safer to install and maintain; cause no damage to the pavement; and detection zones can be adjusted or redrawn when travel lanes are realigned or reassigned due to widening or re-marking of the pavement. Regardless of the advantages, video imaging cameras are generally susceptible to varying lighting conditions. In solving this issue, thermal imaging cameras produce images based on difference in thermal energy emitted by objects, not visible light. Thus, they are more robust to various lighting conditions. In this paper, both video imaging and thermal imaging technologies were evaluated for stop bar vehicle detection at signalized intersections. The detection errors were correlated with concurrent weather and environmental conditions. It shows that the performance of both video imaging camera and thermal imaging camera appear to be affected by wind and various weather events. In particular, uneven shade appears to be an issue for the video imaging camera, but not for the thermal imaging camera. The video imaging camera is more susceptible to false calls at the night, likely due to vehicle headlight. A higher chance of missed and stuck-on calls is experienced for the thermal imaging camera at the night, which is likely a result of reduced difference in temperature between vehicles and the pavement.

Agencies often find it difficult to justify investments in active traffic management. Historically, it has been a challenge to obtain data that would help make the case for those investments. While new data sources have emerged recently, there remains a poor understanding of the potential long term benefits from signal retiming using associated performance measures. This study presents a use case for an active traffic management strategy on a signalized corridor over a 5-year period, during which traffic volumes increased by approximately 36%, and offset optimization was performed every 2-3 years. Despite the considerable volume growth, arrivals on green were increased by more than 41%, and the percentage of vehicles arriving on green increased by 10%, a gain of 6 percentage points. Furthermore, drivers experienced an average of 5% reduction in travel time and travel time reliability costs after each optimization. This resulted in a total user benefit over the 5-year period of approximately $3.6 million. Agencies can utilize these strategies to quantitatively assess how traffic performance and signal timing degrade over time, in a manner similar to physical infrastructure assets. The results highlight the benefits and associated business case of adopting a long-term active traffic management strategy, based on data-driven performance monitoring and decision making.

Intersection crashes are a safety concern for many transportation agencies, and those related to red-light-running (RLR) vehicles are of particular interest. Many camera-based RLR detection systems are controversial with the public, and there is relatively little published literature on the methodologies. This study proposes a methodology that combines high resolution signal controller data with conventional stop bar loop detection to identify vehicles that enter the intersection after the start of red, when many of the most serious RLR crashes occur. The methodology is validated using on-site video collection at several locations, and the algorithm was refined to reduce the incidence of false RLR indications. One case study demonstrates that an increase in side street green split from 20% to 24% of cycle length is associated with a 34% reduction in daily RLR counts, and a reduction in the likelihood of RLR by a factor of 1.7 – a substantial safety improvement for minimal cost. Additionally, law enforcement and transportation agencies can utilize this technique to more efficiently manage and deploy safety resources, especially in cases where detailed crash histories are unknown or too infrequent.

As the driver population increases annually on the road, traffic engineers must find ways to maximize the efficiency of their traffic signal control system through use of emerging technologies. Different signal control strategies, including actuated, coordinated, and adaptive, have their own strengths and weaknesses; therefore it is necessary to comprehensively evaluate these three control modes to understand which strategy is most appropriate for the user. This research carries out a case study to evaluate ACS-Lite’s adaptive performance versus conventional coordinated actuated signal control, and Free, non-coordinated actuated control. The test was done along two congested arterial corridors around Disneyland in Anaheim, CA. During the three-week study period, the first week was on Free control, the second week was on time-base coordinated control, and the third week was on adaptive control. During the AM and PM peak hour, where the traffic demand was high and predictable, conventional coordinated time-of-day plans performed better in overall traffic mobility. However, when traffic demand is unpredictable, variable, and low volume, the results indicated that the adaptive traffic signal control is better suited to improving the efficiency of arterial intersections. Future research could be focusing on how to further improve adaptive signal control by adopting more coordinated features and how to fully utilize high-resolution data provided by Centracs Adaptive in order to improve the overall system efficiency.