High-Resolution Data-Based Methods for Arterial Traffic Volume Estimation

Abstract

Arterial traffic volume data (vehicular volume and pedestrian volume) are the key input for most urban traffic studies, but collecting region-wide traffic volume data is still challenging and time-consuming because only a few locations have been configured with detectors for automatic volume data collection due to cost. In order to improve the efficiency of traffic volume data collection, this dissertation focuses on developing methods using existing data sources to estimate real-time traffic volume. This dissertation has three major components: vehicular volume estimation at signalized intersections, pedestrian volume estimation at signalized midblock crosswalks, and pedestrian crossing volume estimation at signalized intersections.Vehicular volume data at signalized intersections is one of the most critical variables for most traffic studies such as signal retiming. However, collecting vehicular volume manually can be time-consuming and costly. In recent years, video-based sensor systems have been applied to signalized intersections for signal timing control. The detectors in video-based sensors generate large amounts of real-time high-resolution event-based data, including signal status and detection status data. The vehicle arrivals for each detection event is a stochastic process and have a relationship with the signal status and the detection duration (time occupancy). Therefore, a modified Dynamic Hidden Markov Model (DHMM) is proposed to estimate vehicular volume by modeling the vehicle arrivals using this high-resolution data collected at signalized intersections. The concept of an additional hidden state is introduced to make the vehicular volume finite by grouping volumes that have only a small probability of occurring into one hidden state. Additionally, a linear regression model is built to estimate the vehicular volume when the output of the DHMM is an additional hidden state. The resulting mean absolute percentage errors of the 15-minute estimated volume are 14.1%, 10.3%, and 10.5%, respectively, at three study locations in Tucson, Arizona. Similar to signalized intersections, midblock pedestrian signals are another important component of arterials. Pedestrian volume is essential for optimizing midblock pedestrian signals as well as for quantifying pedestrian exposure in safety analyses. However, previous methods of pedestrian volume collection require either time-consuming ground-truth data collection, or the purchase and maintenance of costly sensors in a large-scale application. Therefore, this study proposes a novel method for large-scale pedestrian estimation at midblock crosswalks using button-pushing and signal timing events. The pedestrian arrival is modeled as a Poisson process, and two submethods are developed to estimate pedestrian volume at one-stage and two-stage button-activated midblock crosswalks (BAMCs). To address the issue of missing cycles at two-stage BAMCs, all missing cycles that are identified by using the proposed paired cycle identification algorithm are accounted for and added by minimizing the error between estimation results of two stages. Eight days of the ground-truth pedestrian volume is manually collected from two study midblock crosswalks to evaluate the proposed methods. On average, 235 and 230 pedestrians per day were observed to cross the one-stage BAMC and two-stage BAMC, respectively. The resulting mean average error (MAE) of estimated pedestrian volume using a one-hour interval is 2.27 and 1.78 pedestrians/hour at two study locations, respectively. The evaluation results indicate that the proposed methods are promising for estimating pedestrian volume at midblock crosswalks using event-based data. A further sensitivity analysis of changing the estimation interval shows that the one-hour interval pedestrian volume estimation is recommended as having the least error. Pedestrian crossing volume at signalized intersections is also one of the most important variables used to retime and optimize the signal timing plans for traffic delay migration and traffic safety improvement. However, most of the existing studies only focus on long-term pedestrian volume estimation for planning purposes. To bridge the research gap, this study applied a Bayesian Additive Regression Trees (BART) model to estimate the short-term pedestrian crossing volume at signalized intersections equipped with pushbutton devices. Pedestrian-related traffic controller high-resolution data used as the time-dependent variables representing the temporal trend of pedestrian crossing volume in conjunction with employment data in surrounding buildings and transit trips are chosen as the inputs of the BART model. 70 signalized intersections from the Pima County region are selected to collect data for calibrating and validating the developed method. When compared with ground-truth data, the proposed method has an R-squared of 0.83, 0.81, and 0.71 for 60-min, 30-min, and 15-min intervals of pedestrian crossing volume, respectively. To further evaluate the performance of the proposed method, the proposed method is used in comparison to two traditional methods (stepwise linear regression and Random Forest). The comparison results show that the BART model is superior to the other two models for hourly pedestrian crossing volume estimation. The evaluation results show that the proposed method can accurately estimate pedestrian crossing volume and provide valuable information for signal retiming and other traffic studies.