Vehicle sensor data-based transportation research: Modeling, analysis, and management

Introduction

In the era of information and communications technology and big data, roadway traffic is being monitored by various sensors. Notably, sensors equipped in vehicles, as shown in Figure 1, are providing various Vehicle Sensor Data (VSD) including location-based data (LB-VSD) such as location, speed, and moving direction, and surrounding traffic data (ST-VSD) under Connected and Automated Vehicles (CAV) environment. LB-VSD is often referred to as floating car data or probe vehicle data due to the operation mode that the sensor-equipped vehicles travel on roads as regular vehicles. Passenger’s mobile phones and GPS on taxis/buses are continuously producing a gigantic amount of such data, which makes LB-VSD become widely available. In terms of making better use of them, one could trace driver/passenger’s route (or activity/trip chain); collect link or trip travel time; estimate traffic state; and model/optimize car-hailing services. Moreover, LB-VSD could provide data support for various traffic control and management systems. ST-VSD is provided by the on-board sensors that are able to monitor adjacent traffic conditions. It includes, but is not limited to, radar data, LIght Detection And Ranging (LIDAR) data, Controller Area Network (CAN) bus data, and data transmitted via Dedicated Short Range Communication (DSRC) or cellular network. The data provides opportunities to explicitly model traffic flow and design comfortable and environmentally friendly CAV driving strategies, or to develop advanced traffic management and control strategies in a CAV environment.

Figure 1. On-board vehicle sensors.

Developments in vehicle sensor data-based research

There is a need to utilize emerging VSD-based big data. Advanced tools and innovative ideas/applications utilizing VSD are emerging. They could deepen our understanding of traffic congestion and traveler behavior, as well as improve the efficiency and capability of traffic control and management. This Special Issue is thus focused on the recent advances in various transportation research areas including data quality, human behavior, transportation operations, and policy analysis by taking advantage of the widely available VSD-based data.

Data quality

This Special Issue starts off by providing an overview of the quality of VSD-based data that is commercially available in the market. Ahsani, Amin-Naseri, Knickerbocker, and Sharma (in press) investigated VSD-based data made available by INRIX and shed light on the quality of data, including coverage, speed bias, and congestion detection precision. The study concluded that: (i) INRIX data covers Interstate highways reliably and is more reliable during day time. In addition, there is a trend toward coverage expansion over the years; (ii) travel time information is trustworthy only when hundred percent real-time data are utilized. Travel time based on historical data, even partially, is not accurate; (iii) INRIX data does not perform well for congestion detection on segments shorter than 0.4 miles. For segments that are longer than 0.4 miles, the precision rate is 81.6% for identifying recurring congestion and 51.2% for non-recurring congestion.

Human behavior

A number of ways that VSD-based data can be potentially made use of are identified in this Special Issue. It can be used for human behavior modeling. In Chen, Zhou, and Li (in press), radar data are applied in predicting Red-Light-Running (RLR) behavior. The behavior prediction includes the vehicle’s stop-or-go choice and vehicle’s arrival time at the stop line. RLR is the major cause of collision at intersections leading to numerous fatalities and injuries. These costs can be effectively reduced by more accurate RLR prediction. In the past, RLR behavior is typically predicted utilizing data generated by loop detectors which is costly to maintain and provides inaccurate information. With radar data, RLR prediction can be greatly improved. The data is first translated into vehicle trajectory and then fed into a Bayesian network model for stop-or-go choice prediction. The prediction results are compared against those based on loop detector data and show at least 20% higher accuracy.

In Hu, Yuan, Zhu, Yang, and Xie (in press), an improvement on detour behavior modeling is made possible by VSD-based data. In the study, GPS trajectory data tagged by driver ID is collected both before and after a lane closure event on MoPac Expressway, Austin, TX. By taking advantage of this informative data, travelers’ detour behavior under the impact of work zone is revealed which is not feasible in the past when only fixed-location sensors are available. It is learned that detour pattern is intensively impacted by factors, including trip start time, length of trip, distance to the freeway of first choice and distance to the freeway of the second choice. The proposed method is also capable of identifying potential detour patterns preferred by the drivers. This information that was not available in the past can be crucial for decision-makers in order to be prepared for unexpected events.

Quantifying driving style differences between indoor simulators and on-road instrumented vehicles is made possible by VSD-based data. Simulators have always been a crucial tool for human behavior analysis. As a start, simulators are much more time efficient compared with conducting field tests. In addition, some hazardous scenarios are only feasible in simulation, considering the safety of the human subject. Furthermore, simulator results are more easily replicable while field tests are more difficult to replicate. However, there has always been doubts about how valid and representative simulator results are compared with real world. By collecting maneuver data from the same driver operating both an indoor vehicle simulator and an on-road instrumented vehicle, Qi, Guan, Li, Hounsell, and Stanton (in press) were able to answer the question of consistency between the two and discovered that human drivers are more aggressive in the simulator than in real world, as greater accelerations and decelerations are observed in simulators.

Operations

VSD-based data can be beneficial to traffic operations as well. As demonstrated in Kuang, Yan, Zhu, Tu, and Fan (in press), VSD-based data could be potentially used to help collision event duration prediction. The collision event duration prediction is important, since collisions typically leads to congestion. More accurate duration prediction could help traffic operator to provide more appropriate countermeasures, such as emergency response, traffic rerouting, and speed harmonization. The VSD-based data can capture various crash features including time of day, day of week, severity and number of vehicles involved in the crash. In the study, various features were then used to compare against historical records and generated event duration prediction by applying a two-level model consisting of a Bayesian network and a weighted KNN regression model. By comparing against conventional methods, the proposed method improved the accuracy rate by 3.8% and reduced misclassification rate by 13.8%.

Public transportation is one of the most important solutions to traffic congestion, fuel consumption, and pollution. Among various types of public transportation, public transit plays a major role. However, the uncertainty of when a bus will arrive has always been a challenge and casts anxieties on passengers which pushes away potential users of public transit. Dai, Ma, and Chen (in press) proposed a transit travel time prediction model that takes advantage of VSD-based data. The prediction model was enhanced by considering information obtained from link travel time and station dwell times. By using VSD-based data, a probabilistic model was developed to describe the interaction between buses approaching a bus bay. With more accurate estimation of link travel time and dwell time, transit travel time is predicted with much higher precision.

Policy

Finally, VSD-based data could be crucial for decision makers. In Zhai et al. (in press), VSD-based data are utilized to quantify operating mode distribution for light duty vehicles. Operating mode distribution is one of the most critical inputs for MOVES model which was developed by the United States Environmental Protection Agency and is one of the most well accepted methods for real-time microscopic traffic emissions estimation. It was discovered that VSD-based operating mode distribution is crucial to be adopted. The distribution is greatly affected by factors, including road type, average and vehicle type. Ignoring these factors may lead to errors of up to 20% in emissions estimation. The study also provided a collection of valuable data of the operating mode distribution in Beijing, China which could be made use of by future studies or decision makers.

Summary

The articles published in this Special Issue provide an overview of the state-of-the-art theoretical and practical developments on VSD-based data: VSD-based human/driver mobility characteristics and its implication on emissions estimation; driving behaviors have been investigated in terms of red-light running prediction, traveler responses to pre-planned road capacity reduction, and driving style differences between operating indoor driving simulators and on-road instrumented vehicles; methods have been developed for transit travel time prediction using VSD-based data; and finally, VSD-based data quality has been evaluated in terms of coverage and detection accuracy. It is clear that the potential is huge for VSD-based data to play a major role in future transportation planning, operations, and management. As the roll-out of Connected Vehicle technology, the coverage of VSD-based data is expected to expand, so does the impact of such data. It is our hope that the research published in this Special Issue will serve as a motivation for mining of VSD-based data and help various parties gain maximum benefits from their deployment of interest.

Acknowledgment

The coeditors of this Special Issue would like to thank anonymous referees for their hard work in meticulously reviewing the papers submitted for publication review. We also thank the Editor-in-Chief, Prof. Asad Khattak for his help during the initiation and review process of this special issue.