Abdullah Kurkcu

Analysis of Traffic Crashes Involving Pedestrians Using Big Data: Investigation of Contributing Factors and Identification of Hotspots

This study aims to explore the potential of using big data in advancing the pedestrian risk analysis including the investigation of contributing factors and the hotspot identification. Massive amounts of data of Manhattan from a variety of sources were collected, integrated, and processed, including taxi trips, subway turnstile counts, traffic volumes, road network, land use, sociodemographic, and social media data. The whole study area was uniformly split into grid cells as the basic geographical units of analysis. The cell-structured framework makes it easy to incorporate rich and diversified data into risk analysis. The cost of each crash, weighted by injury severity, was assigned to the cells based on the relative distance to the crash site using a kernel density function. A tobit model was developed to relate grid-cell-specific contributing factors to crash costs that are left-censored at zero. The potential for safety improvement (PSI) that could be obtained by using the actual crash cost minus the cost of similar sites estimated by the tobit model was used as a measure to identify and rank pedestrian crash hotspots. The proposed hotspot identification method takes into account two important factors that are generally ignored, i.e., injury severity and effects of exposure indicators. Big data, on the one hand, enable more precise estimation of the effects of risk factors by providing richer data for modeling, and on the other hand, enable large-scale hotspot identification with higher resolution than conventional methods based on census tracts or traffic analysis zones.

Virtual sensors: Web-based real-time data collection methodology for transportation operation performance analysis

Recent advances in mobile networks and an increase in the number of GPS-equipped vehicles have led to exponential growth in real-time data generation. In the past decade, several online mapping and vehicle tracking services have made their data available to third-party users. This paper explores opportunities for use of real-time traffic data provided by online services and introduces a virtual sensor methodology for collecting, storing, and processing large volumes of network-level data. To assess the validity of the collected data with the proposed methodology, this paper compares these data with data from physical loop detectors and electronic toll tag readers. Statistical analyses show a strong correlation between the travel time measurements from infrastructure-based sensors and virtual sensors. A travel time reliability analysis is then conducted with the virtual sensor data methodology. The results are promising for future research and implementation.

Extended Implementation Method for Virtual Sensors: Web-Based Real-Time Transportation Data Collection and Analysis for Incident Management

Open data sources and social media data are gaining increasing attention as important information providers in transportation and incident management. In this paper, practical evidence for the emerging potential of online and open data sources is presented. The authors’ previous research on virtual sensors is combined and extended by integrating real-time incident information and social media network engagement. The fundamental contribution of this paper is the development of an extended virtual sensor framework to provide an automated travel time data collection method as incidents occur. In addition, social media data can be useful for more effective real-time incident response. The proposed framework can easily be modified and used to evaluate travel time effects of incidents on roadways and clearance times and to make use of social media data in obtaining time-critical incident-related information.

Data visualization tool for monitoring transit operation and performance

Using the automated vehicle location data combined with other technologies such as automated incident reporting, transit decision makers can now execute a variety of real-time strategies and performance evaluations. In this study, we show that it is possible to develop an easy to use but powerful web-based tool which acquires, stores, processes, and visualizes bus trajectory data. The developed web-based tool makes it easy for the end users to access stored data and to query it without any delay or external help. Moreover, the tool allows the users to conduct a series of data visualization and analysis operations demonstrating the potential of a such web-based tool for future applications.

Estimating Pedestrian Densities, Wait Times, and Flows with Wi-Fi and Bluetooth Sensors

Monitoring nonmotorized traffic is gaining more attention in the context of transportation studies. Most of the traditional pedestrian monitoring technologies focus on counting pedestrians passing through a fixed location in the network. It is thus not possible to anonymously track the movement of individuals or groups as they move outside each particular sensor’s range. Moreover, most agencies do not have continuous pedestrian counts mainly because of technological limitations. Wireless data collection technologies, however, can capture crowd dynamics by scanning mobile devices. Data collection that takes advantage of mobile devices has gained much interest in the transportation literature as a result of its low cost, ease of implementation, and richness of the captured data. In this paper, algorithms to filter and aggregate data collected by wireless sensors are investigated, as well as how to fuse additional data sources to improve the estimation of various pedestrian-based performance measures. Procedures to accurately filter the noise in the collected data and to find pedestrian flows, wait times, and counts with wireless sensors are presented. The developed methods are applied to a 2-month-long collection of public transportation terminal data carried out with the use of six sensors. Results point out that if the penetration rate of discoverable devices is known, then it is possible to accurately estimate the number of pedestrians, pedestrian flows, and average wait times in the detection zone of the developed sensors.

Exploring pedestrian Bluetooth and WiFi detection at public transportation terminals

This is an on-going study that explores the potential benefits of using pedestrian data for evaluation and enhancement of public transportation. The research team proposes the utilization of Bluetooth (BT) and WiFi technologies to estimate time-dependent origin-destination (OD) demands and station wait-times of transit bus and subway users. The detection approach and a complete system design are discussed in this paper. Preliminary results from multiple pilot field studies, that were conducted at some of the major New York City (NYC) public transportation facilities, are also presented. The main objective of this study is to inquire into the various ways this extensive transit rider data can be used and to establish a general framework through data-driven pedestrian modeling within transit stations that renders estimation of key parameters and strategic control of public transportation services possible.

Crowdsourcing Incident Information for Emergency Response using Open Data Sources in Smart Cities

Emergency events affect human security and safety as well as the integrity of the local infrastructure. Emergency response officials are required to make decisions using limited information and time. During emergency events, people post updates to social media networks, such as tweets, containing information about their status, help requests, incident reports, and other useful information. In this research project, the Latent Dirichlet Allocation (LDA) model is used to automatically classify incident-related tweets and incident types using Twitter data. Unlike the previous social media information models proposed in the related literature, the LDA is an unsupervised learning model which can be utilized directly without prior knowledge and preparation for data in order to save time during emergencies. Twitter data including messages and geolocation information during two recent events in New York City, the Chelsea explosion and Hurricane Sandy, are used as two case studies to test the accuracy of the LDA model for extracting incident-related tweets and labeling them by incident type. Results showed that the model could extract emergency events and classify them for both small and large-scale events, and the model’s hyper-parameters can be shared in a similar language environment to save model training time. Furthermore, the list of keywords generated by the model can be used as prior knowledge for emergency event classification and training of supervised classification models such as support vector machine and recurrent neural network.

Calibration and validation of large-scale traffic simulation networks: a case study

Abstract The availability, accuracy and relevance of real world input data are essential for developing a reliable traffic simulation model. Large-scale traffic simulation models, in particular, require data from many sources and in great detail. Though it is now possible to obtain detailed field data with the advent of new technologies such as GPS, cellular phones, RFIDs, it is still a challenge to gather all available data, especially traffic flow data, in the required spatial and temporal accuracy. The central theme of this paper is the calibration and validation (C&V) development of a large-scale traffic simulation model using data from multiple sources.

Application of Bayesian Stochastic Learning Automata for Modeling Lane Choice Behavior on High-Occupancy Toll Lanes on State Road 167

This paper investigates the learning behavior of users of State Road 167 high-occupancy toll lanes by use of toll transaction data collected over a 6-month period. The Bayesian stochastic learning algorithm theory was used to model drivers’ sequential lane choice decisions. Reward and penalty parameters were used to update users’ lane choice probabilities. The results showed that the effect of reward parameters that increased the probability of selection of an alternative after a satisfactory experience was more obvious than the effect of penalty parameters that decreased the probability of selection of an unfavorable choice. Low magnitudes of learning parameters might indicate strong habit formation of users. Moreover, the posterior distributions of learning parameters indicated that user perception heterogeneity existed when the outcomes of choices were evaluated. Finally, user familiarity was investigated with a subsample of less experienced users, and it was shown that the learning rates of more familiar users were lower than those of less familiar users.