<p dir="ltr">The emergence of time-dependent data provides researchers with unparalleled opportunities to investigate disaggregated levels of safety performance on roadway infrastructures. A disaggregated crash risk analysis uses both time-dependent data (e.g., hourly traffic, speed, weather conditions and signal controls) and fixed data (e.g., geometry) to estimate hourly crash probability. Despite abundant research on crash risk analysis, coordinated signalized intersections continue to require further investigation due to both the complexity of the safety problem and the relatively small number of past studies that investigated the risk factors of coordinated signalized intersections. This dissertation aimed to develop robust crash risk prediction models to better understand the risk factors of coordinated signalized intersections and to identify practical safety countermeasures. The crashes first were categorized into three types (same-direction, opposite-direction, and right-angle) within several crash-generating scenarios. The data needed were organized in hourly observations and included the following factors: road geometric features, traffic movement volumes, speeds, weather precipitation and temperature, and signal control settings. Assembling hourly observations for modeling crash risk was achieved by synchronizing and linking data sources organized at different time resolutions. Three different non-crash sampling strategies were applied to the following three statistical models (Conditional Logit, Firth Logit, and Mixed Logit) and two machine learning models (Random Forest and Penalized Support Vector Machine). Important risk factors, such as the presence of light rain, traffic volume, speed variability, and vehicle arrival pattern of downstream, were identified. The Firth Logit model was selected for implementation to signal coordination practice. This model turned out to be most robust based on its out-of-sample prediction performance and its inclusion of important risk factors. The implementation examples of the recommended crash risk model to building daily risk profiles and to estimating the safety benefits of improved coordination plans demonstrated the model’s practicality and usefulness in improving safety at coordinated signals by practicing engineers.</p>

The Flashing Yellow Arrow Left Turn signal is quickly becoming prominent in Central Florida as a new method of handling left turns at traffic signals. While the concept of a protected-permitted left turn is not groundbreaking, the departure from the typical display of a five-section signal head is, for this type of operation. The signal head introduced is a four-section head with a flashing yellow arrow between the yellow and green arrows. With this signal head quickly becoming the standard, there is a need to re-evaluate the operational characteristics of the left turning vehicle and advance the knowledge of the significant parameters that may affect the ability for a driver to make a left turn at a signalized intersection. With previous research into the behavioral and operational characteristics of the flashing yellow arrow conducted, there is more information becoming available about the differences between this signal and the previously accepted method of allowing left turns at an intersection. The protected-permitted signal is typically displayed at an intersection with up to two through lanes and generally a protected signal is installed when the number of through lanes increases above two unless specific criteria is met. With the advent of larger arterials and more traffic on the highway networks, the push to operate these intersections at their maximum efficiency has resulted in more of these protected-permitted signals being present at these larger intersections, including the flashing yellow arrow. The core of the research that follows is a comparative analysis of the operation and parameters that affect the left turn movement of the intersection with larger geometry to that of the smaller geometry. The significant parameters of the left turn movement were examined through means of collecting, organizing and analyzing just over 68 hours of field data. This research details the determining of the significant parameters based on the generation of a simulation model of the protected left turn using Synchro, a traffic simulation package, and regression models using field driven data to determine the significant parameters for predicting the number of left turns that can be made in the permitted phase under specific operating conditions. Intuitively, there is an expectation that a larger intersection will not allow for as many permitted lefts as a smaller intersection with all conditions remaining the same. The conclusions drawn from this analysis provide the framework to understanding the similarities and the differences that are encountered when the intersection geometry differs and help to more efficiently manage traffic at signalized intersections. The work of this field promises to enhance the operations of the left turning movement for traffic control devices. With an understanding of the statistical models generated, a broader base of knowledge is gained as to the significant parameters that affect a driver's ability to make the left turn. A discussion of the statistical differences and between the models generated from the small and large geometry intersections is critical to drive further research into standards being developed for the highway transportation network and the treatment of these large signalized intersections. The exploration of specific parameters to predict the number of permitted left turns will yield results as to if there is more to be considered with larger intersections moving forward as they become a standard sight on the roadway network.

Traffic signals are used at intersections to manage the flow of vehicles by allocating right-of-way in a timely manner for different users of the intersection. Traffic signals are therefore installed at an intersection to improve overall safety and to decrease vehicular average delay. However, the variation of driving speed in response to these signals causes an increase in fuel consumption and air emission levels. One solution to this problem is Eco-Cooperative Adaptive Cruise Control (Eco-CACC), which attempts to reduce vehicle fuel consumption and emission levels by optimizing driver behavior in the vicinity of a signalized intersection. Various Eco-CACC algorithms have been proposed by researchers to address this issue. With the help of vehicle-to-infrastructure (V2I) and vehicle-to-vehicle (V2V) communication, algorithms are being developed that utilize signal phasing and timing (SPaT) data together with queue information to optimize vehicle trajectories in the vicinity of signalized intersections. The research presented in this thesis constitutes the third phase of a project that entailed developing and evaluating an Eco-CACC system. Its main objective is to evaluate the benefits of the newly developed Eco-CACC algorithm that was proposed by the Center for Sustainable Mobility at the Virginia Tech Transportation Institute. This algorithm uses advanced signal information (SPaT) to compute the fuel-optimal trajectory of vehicles, and, then, send recommended speeds to drivers as an audio message or implement them directly into the subject vehicle. The objective of this study is to quantitatively quantify the fuel-efficiency of the Eco-CACC system in a real field environment. In addition, another goal of this study is to address the implementation issues and challenges with the field application of the Eco-CACC system. A dataset of 2112 trips were collected as part of this research effort using a 2014 Cadillac SRX equipped with a vehicle onboard unit for (V2V) and (V2I) communication. A total of 32 participants between the ages of 18 and 30 were randomly selected from one age group (18-30) with an equal number of males and females. The controlled experiment was conducted on the Virginia Smart Road facility during daylight hours for dry pavement conditions. The controlled field experiment included four different scenarios: normal driving, driving with red indication countdown information provided to drivers, driving with recommended speed information computed by the Eco-CACC system and delivered to drivers, and finally automated driving (automated Eco-CACC system). The controlled field experiment was conducted for four values of red indication offsets along an uphill and downhill approach. The collected data were compared with regard to fuel economy and travel time over a fixed distance upstream and downstream of the intersection (820 ft (250 m) upstream of the intersection to 590 ft (180 m) downstream for a total length of 1410 ft (430 m)). The results demonstrate that the Eco-CACC system is very efficient in reducing fuel consumption levels especially when driving downhill. The field data indicates that the automated scenario could produce fuel and travel time savings of 31% and 9% on average, respectively. In addition, the study demonstrates that driving with a red indication countdown and recommended speed information can produce fuel savings ranging from 4 to 21 percent with decreases in travel times ranging between 1 and 10 percent depending on the value of red indication offset and the direction. Split-split-plot design was used to analyze the data and test significant differences between the four scenarios with regards to fuel consumption and travel time. The analysis shows that the differences between normal driving and driving with either the manual or automated Eco-CACC systems are statistically significant for all the red indication offset values. / Master of Science

The dissertation studies traffic dispersion modeling in four parts. In the first part, the dissertation focuses on the Robertson platoon dispersion model which is the most widely used platoon dispersion model. The dissertation demonstrates the importance of the Yu and Van Aerde calibration procedure for the commonly accepted Robertson platoon dispersion model, which is implemented in the TRANSYT software. It demonstrates that the formulation results in an estimated downstream cyclic profile with a margin of error that increases as the size of the time step increases. In an attempt to address this shortcoming, the thesis proposes the use of three enhanced geometric distribution formulations that explicitly account for the time-step size within the modeling process. The proposed models are validated against field and simulated data. The second part focuses on implementation of the Robertson model inside the popular TRANSYT software. The dissertation first shows the importance of calibrating the recurrence platoon dispersion model. It is then demonstrated that the value of the travel time factor &#946; is critical in estimating appropriate signal-timing plans. Alternatively, the dissertation demonstrates that the value of the platoon dispersion factor &#945; does not significantly affect the estimated downstream cyclic flow profile; therefore, a unique value of &#945; provides the necessary precision. Unfortunately, the TRANSYT software only allows the user to calibrate the platoon dispersion factor but does not allow the user to calibrate the travel time factor. In an attempt to address this shortcoming, the document proposes a formulation using the basic properties of the recurrence relationship to enable the user to control the travel time factor indirectly by altering the link average travel time. In the third part of the dissertation, a more general study of platoon dispersion models is presented. The main objective of this part is to evaluate the effect of the underlying travel time distribution on the accuracy and efficiency of platoon dispersion models, through qualitative and quantitative analyses. Since the data used in this study are generated by the INTEGRATION microsimulator, the document first describes the ability of INTEGRATION in generating realistic traffic dispersion effects. The dissertation then uses the microsimulator generated data to evaluate the prediction precision and performance of seven different platoon dispersion models, as well as the effect of different traffic control characteristics on the important efficiency measures used in traffic engineering. The results demonstrate that in terms of prediction accuracy the resulting flow profiles from all the models are very close, and only the geometric distribution of travel times gives higher fit error than others. It also indicates that for all the models the prediction accuracy declines as the travel distance increases, with the flow profiles approaching normality. In terms of efficiency, the travel time distribution has minimum effect on the offset selection and resulting delay. The study also demonstrates that the efficiency is affected more by the distance of travel than the travel time distribution. Finally, in the fourth part of the dissertation, platoon dispersion is studied from a microscopic standpoint. From this perspective traffic dispersion is modeled as differences in desired speed selection, or speed variability. The dissertation first investigates the corresponding steady-state behavior of the car-following models used in popular commercially available traffic microsimulation software and classifies them based on their steady-state characteristics in the uncongested regime. It is illustrated that with one exception, INTEGRATION which uses the Van Aerde car-following model, all the software assume that the desired speed in the uncongested regime is insensitive to traffic conditions. The document then addresses the effect of speed variability on the steady-state characteristics of the car-following models. It is shown that speed variability has significant influence on the speed-at-capacity and alters the behavior of the model in the uncongested regime. A method is proposed to effectively consider the influence of speed variability in the calibration process in order to control the steady-state behavior of the model. Finally, the effectiveness and validity of the proposed method is demonstrated through an example application. / Ph. D.

Most U.S. urban traffic signal systems deploy multiple signal timing plans to account for daily variability of traffic demand (i.e. morning peak, midday, afternoon peak, off peak and night). Groups of signals (belonging to the one zone or section) along an urban arterial, usually operate in a coordinated manner. This essentially means that timing plans change at the same time for all the signals in the group, so as to facilitate vehicle progression of through a series of signals. Good traffic signal timing practices assume a certain level of monitoring and maintenance in order to guarantee that they are efficient in servicing current traffic conditions. / Includes bibliography. / Thesis (M.S.)--Florida Atlantic University, 2015 / FAU Electronic Theses and Dissertations Collection

This effort seeks to answer the question; can a transferable model be developed from easily obtainable, publicly available land-use, census, roadway, and network data for the use in safety performance functions? 474 stop-controlled minor legs were used as the training data set using ordinary least squares regression. A best-fit model of maximum independent variables, n=12 was chosen using an exhaustive approach using Mallow's Cp to select the model with least bias in the predictors. The results of the analysis revealed that the combination of variables from Washington, Ohio, and North Carolina did not have a strong relationship. The best-fit model incorporated functional class information of the major-leg, minor leg functional class information, longitudinal markings, access to a parking lot, and population density of census tract. Validation of the model demonstrated an average 59 percent error between the model estimated and actual AADT values for validation data set (n=54). Furthermore, separate models for each state revealed a lack of uniformity in the dependent variables, and more variance description of the state specific AADT.

This project examined cyclist red light running behavior using two data sets. Previous studies of cyclist compliance have investigated the tendencies of cyclists to run red lights on the whole by generalizing different maneuvers to their end outcome, running a red light. This project differentiates between the different types of red light running and focuses on the most egregious case, gap acceptance, which is when a cyclist runs a red light by accepting a gap in opposing traffic. Using video data, a mathematical model of cyclist red light running was developed for gap acceptance. Similar to other studies, this analysis utilized only information about the cyclist, intersection, and scenario that can be outwardly observed. This analysis found that the number of cyclists already waiting at the signal, the presence of a vehicle in the adjacent lane, and female sex were deterrents to red light running. Conversely, certain types of signal phasing, witnessing a violation, and lack of helmet increased the odds that a cyclist would run the red light. Interestingly, while women in general are less likely to run a red light, those who witnessed a violation were even more prone that men who had witnessed a violation to follow suit and run the red light themselves. It is likely that the differing socialization of women and men leads to different effects of witnessing a previous violator. The analysis also confirmed that a small subset of cyclists, similar to that found in the general population, are more prone to traffic violations. These cyclists are more willing to engage in multiple biking-related risk factors that include not wearing a helmet and running red lights. Although the model has definite explanatory power regarding decisions of cyclist compliance, much of the variance in the compliance choices of the sample is left unexplained. This points toward the influence of other, not outwardly observable variables on the decision to run a red light. Analysis of survey data from cyclists further confirms that individual characteristics not visible to the observer interact with intersection, scenario, and visible cyclist characteristics to result in a decision to comply (or not) with a traffic signal. Furthermore, cyclist characteristics, in general, and unobservable individual characteristics, specifically, play a larger role in compliance decisions as the number of compliance-inducing intersection traits (e.g. conflicting traffic volume) decrease. One such unobservable trait is the regard for the law by some cyclists, which becomes a more important determinant of compliance at simpler intersections. Cyclists were also shown to choose non-compliance if they questioned the validity of the red indication for them, as cyclists. The video and survey data have some comparable findings. For instance, the relationship of age to compliance was explored in both data analyses. Age was not found to be a significant predictor of non-compliance in the video data analysis while it was negatively correlated with stated non-compliance for two of the survey intersections. Gender, while having significant effects on non-compliance in the video dataset, did not emerge as an important factor in the stated non-compliance of survey takers. Helmet use had a consistent relationship with compliance between the video and survey datasets. Helmet use was positively associated with compliance in the video data and negatively associated with revealed non-compliance at two of the survey intersections. When coupled with the positive association between normlessness and stated willingness to run a red light, the relationship between helmet use and compliance solidifies the notion that a class of cyclists is more likely to consistently violate signals. It points towards a link between red light running and individuals who do not adhere to social norms and policies as strictly as others. Variables representing cyclists and motorists waiting at the signal were positively related to signal compliance in the video data. While an increased number of cyclists may be a physical deterrent to red light running, part of the influence on compliance that this variable and the variable representing the presence of a vehicle may be due to accountability of cyclists to other road users. This relationship, however, was not revealed in the stated non-compliance data from the survey. Efforts to increase cyclist compliance may not be worth a jurisdiction's resources since nearly 90% of cyclists in the video data were already compliant. If a problem intersection does warrant intervention, different methods of ensuring bicyclist compliance are warranted depending on the intersection characteristics. An alternative solution is to consider the applicability of traffic laws (originally designed for cars) to bicyclists. Creating separation in how laws affect motorists and cyclists might be a better solution for overly simple types of intersections where cyclists have fewer conflicts, better visibility, etc. than motorists. Education or other messaging aimed at cyclists about compliance is another strategy to increase compliance. Since cyclists appear to feel more justified in running red lights at low-volume, simple-looking intersections, it would probably be prudent to target messaging at these types of intersections. Many cyclists are deterred by high-volume and/or complicated looking intersections for safety reasons. Reminding cyclists of the potential dangers at other intersections may be a successful messaging strategy. Alternatively, reminding cyclists that it is still illegal to run a red light even if they feel safe doing so may be prudent. Additionally, messaging about the purpose of infrastructure such as bicycle-specific signals or lights that indicate detection at a signal may convince cyclists that stopping at the signal is in their best interest and that the wait will be minimal and/or warranted.

The 2009 edition of the Manual on Uniform Traffic Control Devices (MUTCD) recommends a pedestrian walking speed of 3.5 feet per second for use in the timing of pedestrian clearance intervals at signalized intersections (reduced from 4.0 feet per second in the 2003 edition). Jurisdictions across the state of Utah continue to maintain pedestrian walking speeds of 4.0 feet per second for normal intersections with guidance on engineering judgement for areas where a lower pedestrian walking speed should be considered. In 2018, it was decided that the current state guidance with regard to pedestrian walking speeds be evaluated for any needed changes, such as adopting the national guidance found in the 2009 MUTCD. To evaluate pedestrian walking speeds at signalized intersections, 15 sites throughout the state of Utah were studied, producing a total of 2,061 observations of pedestrian crossing events. These crossing events were evaluated to calculate walking speeds in relation to pedestrian demographics at each location. Evaluated demographics included pedestrian group size, gender, mobility status, age category, alertness, and potential distractions. Upon completion of data collection, a statistical analysis was conducted to determine mean and 15th percentile pedestrian walking speeds by demographic. The data collection procedure, data analysis, and limited recommendations for pedestrian start-up delay and pedestrian walking speeds as used in signal timing are discussed in this report. The data suggest that Utah continue to maintain its guidance of 4.0 feet per second walking speeds at most signalized intersections, while exercising engineering judgment at locations containing high pedestrian volumes or locations containing high percentages of elderly or disabled pedestrians.

The safety at half-signalized intersections in Portland, Oregon is analyzed in this thesis using 10 years of crash history and analysis of video that was collected at a subset of intersections. A half-signalized intersection has a standard red-yellow-green traffic signal for automobiles on the major road, a stop sign for motorists on the minor road, and a pedestrian signal with actuation for pedestrians and/or bicyclists on the minor road. Although prevalent in Canada, this type of intersection control is not typically found in the United States because the MUTCD explicitly prohibits its use. Half-signal use is limited mostly to two cities in the Pacific Northwest. In Portland, Oregon there are forty-seven intersections where half-signals are used but the last installation was in 1986; Seattle has over 100 intersections with half-signals and installs these in new locations where warranted. To explore the safety records of these intersections in Portland, crash data from 2002-2011 was analyzed. A total of 442 crashes over the ten-year period at half-signals were observed. Sixteen of these 442 crashes involved pedestrians. In the crashes involving pedestrians, significant differences were found between the approach street of the vehicle and whether the pedestrian or driver was at fault. In the crash error reports, it was found that significantly more of the crashes involving pedestrians were the fault of motorists departing from the minor road who collided with pedestrians crossing the major street. Further crash analysis at half-signals was performed by developing matched comparison groups of minor stop controlled and fully signalized intersections. Crash rates were 0.158 and 0.178 crashes per million entering vehicles for 3-leg and 4-leg half-signals and these rates did not differ significantly from the minor street stop controlled and signalized comparison groups. Results from the matched comparison showed that the half-signalized group had more rear-end crashes when compared with the minor stop controlled group. This was the only result that held significance when crash rates were considered. It was also observed that the minor stop controlled group had a higher proportion of angle crashes when compared with the half-signal group but this did not influence the crash severity. Pedestrian crashes were more prevalent in the half-signal group when compared with the fully-signalized group. Pedestrian volumes were not available which would be used to determine if this significant measure is a result of higher pedestrian use at half-signals. In addition to crash analysis, video was captured at five half-signalized intersections totaling 180 hours. Traffic volumes, pedestrian and bicycle volumes, and signal actuations were collected over a twenty-four hour period. Over this twenty-four hour period the five intersections averaged daily counts of 18613 vehicles on the major street, 591 vehicles on the minor street, 263 pedestrians crossing the major street, 285 pedestrians crossing the minor street, 52 bicycles on the major street, 37 bicycles on the minor street, and 126 signal actuations. Twenty-four hour observations from each of the intersections were used to study conflicts and compliance. No conflicts were observed that reflect the left-turning from the minor street pedestrian crashes that were identified in the crash history. Compliance of the half-signal by vehicles and pedestrians was comparable to compliance at fully-signalized intersections found in other studies with one exception. Across the intersections where video was collected, consisting of four 4-leg intersections and one 3-leg intersection, seven left turn on red violations were observed which had a significant impact on the time after red that red light violations were made. It is hypothesized that at half-signals vehicles on the major street make a left turn on the red signal very late into the red phase because there is not a risk of colliding with a vehicle traveling on the minor street since traffic volumes on the minor street are comparably low. The observed left turn on red violations did not put pedestrians at risk since by that point into the signal pedestrians were already clear of the intersection. Finally, a stop compliance logistic regression model was developed at four four-leg intersections to see what factors had an effect on minor street vehicle stop compliance. All 166 hours of video were used to observe vehicles that arrived at the half-signal during the pedestrian phase. The dependent variable collected was whether a vehicle came to an acceptable stop. Independent variables collected included the vehicle's queue position, if it was the peak school period, if there was a vehicle across the street on the minor road, if a vehicle was stopped at the signal on the major street, if a pedestrian was present when the vehicle arrived, and the movement that the vehicle made from the minor street. Independent variables used in the model included the vehicle's queue position, if a vehicle was stopped at the signal on the major street, if a pedestrian was present, and if the vehicle made a right turn at the signal. Pedestrian presence and right turning vehicles had a positive impact on stop compliance. Vehicles being further back in the queue and cars stopped at the signal on the major street had a negative impact on stop sign compliance. In the model, pedestrian presence had the largest positive impact on stop compliance. When pedestrians were present, a motorist on the minor street was four times more likely to stop at the sign.