Elsevier

Graphical Models

Volume 106, November 2019, 101048
Graphical Models

Real-virtual consistent traffic flow interaction

https://doi.org/10.1016/j.gmod.2019.101048Get rights and content

Abstract

Traffic simulation has become an efficient tool, with the assistance of computer visualizing techniques, to solve traffic issues such as traffic congestion, network design, and similar problems. Properly controlling simulated traffic flow and modeling each vehicle’s irregular behaviors are key issues in the traffic simulation field. In this paper, we introduce real vehicle trajectories as a data-driven factor in simulated traffic situations to drive behaviors of other simulated vehicles. First, we train a driving model for each simulated vehicle using real traffic data that have a unique control strategy. Then, we fuse real trajectories driven vehicles with simulated trajectories driven vehicles to interact, guided by our learned traffic model, to accurately depict the reality of traffic flows. Compared with existing methods, traffic flows simulated using this method are more realistic and can preserve irregular characteristics of the real traffic flows.

Introduction

With economic development booming in many countries, the volume of vehicles has grown in kind, to the point where traffic congestion, network design, control signal optimization, and other issues have become highly challenging and urgent. Computer-assisted traffic simulation has become an efficient tool for exploring solutions to vehicle volume problems; however, a real traffic scenario is highly complicated, where traffic conditions result from multiple factors that change continuously. In most instances, during the process of driving, neighboring vehicles move constantly, road conditions change, vehicle behaviors are diversified, and drivers make different decisions, all of which cause driving behaviors to change frequently while the controlling parameters of vehicles change irregularly.

Advanced detection technologies such as a global positioning system (GPS), embedded sensors, and traffic monitors have contributed to researchers’ ability to present realistic scenarios in simulated environments. Traditional traffic simulation methods often use real trajectories to initialize the traffic state, and the evolution of traffic flows is then decided by a manually defined model. For example, the intelligent driver model [1] uses a defined math equation to calculate a vehicle’s acceleration according to its leading vehicle. Thus, the traffic flows simulated by this method are not as complex as real traffic flows because the simulations are not associated with real data and use a single model to guide all drivers’ behaviors.

In this paper, in an effort to simulate real traffic conditions as accurately as possible, we design a method where real trajectory data are introduced directly into the simulation process to maximize the characteristics of real-world traffic flows. Introducing real trajectories driven vehicles (RT-vehicles) to consistently interact with simulated trajectories driven vehicles (ST-vehicles) is our main contribution. More specifically, RT-vechiles are vehicles steered by personalized models trained by our approach while ST-vehicles are vehicles guided by the driving features (including positions, accelerations and velocities) we directly import from the dataset.The real trajectories come from real traffic flows; hence, we assume they have higher priority, which means the RT-vehicles’ behaviors cannot be changed. The main challenge we face is how to model each ST-vehicle’s behavior using RT-vehicles’ trajectories.

Our main idea is to use a data-driven model that controls the driving decision-making procedure according to the states of data [2]. Specifically, we aim to use a data-driven model [3] trained from real trajectories, generated simulated trajectories for each simulated vehicle to guide its behaviors given that different drivers have different driving habits. The trained model is based on genetic programming (GP) and includes two parts: car-following behavior and lane-changing behavior. In the simulation process, we implement a learned data-driven model to drive ST-vehicles’ interactions with RT-vehicles to present a more realistic traffic flow. Different from local control using existing simulation methods, we extract individual real trajectory data from traffic video and learn the personalized coordination strategies using GP to complete a real-simulated interaction traffic flow with global consistency.

The rest of this paper is organized as follows. Section 2 introduces related work on traffic simulation and the genetic algorithm. Section 3 provides a detailed description about the proposed method and GP. Experimental results are outlined in Section 4. Finally, Section 5 concludes this work and discusses future research directions.

Section snippets

Related work

Here, we briefly introduce previous work related to traffic simulation and learning algorithms.

Methodology

In this section, we present an overview of our approach and a detailed description of the model.

Simulation results

By applying the method described in Section 3, we set the initial number of generations to 20 with 20,000 individuals per generation. Then, the best fit of the learned data-driven model for the real data could be obtained. From our dataset, we used 152 car-following vehicles and 86 lane-changing vehicles. For each vehicle, we used two-thirds of its trajectories for training and one-third for testing. In the car-following process, we presented the average results in terms of acceleration,

Conclusion

In this study, we present a real-simulated interaction traffic simulation system where ST-vehicles can interact with a real trajectory under the assistance of a data-driven model. Our simulation results preserved the irregular and diverse characteristics of real traffic flows.

In potential automatic driving applications, ST-vehicles 21can play the role of automatic cars, and RT-vehicles can play the role of general cars. We hope the model proposed in this paper can simulate the interactions

Declaration of Competing interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests:

Acknowledgement

This work was supported in part by the National Natural Science Foundation of China under Grant 91748104, Grant U1811463, Grant 61632006, Grant 61425002, and Grant 61751203, in part by the National Key Research and Development Program of China under Grant 2018YFC0910506, in part by the Open Project Program of the State Key Lab of CAD&CG (Grant No. A1901), Zhejiang University, the Open Research Fund of Beijing Key Laboratory of Big Data Technology for Food Safety (Project No. BTBD-2018KF),

References (38)

  • M.J. Lighthill et al.

    On kinematic waves. ii. a theory of traffic flow on long crowded roads

    Proc. Royal Society of London A: Mathematical, Physical and Engineering Sciences

    (1955)
  • J. Lebacque

    A finite acceleration scheme for first order macroscopic traffic flow models

    Technical Report

    (1997)
  • J. Sewall et al.

    Continuum traffic simulation.

    Comput. Graph. Forum

    (2010)
  • L. Pipes et al.

    An operational analysis of traffic dynamics

    (1953)
  • E.W.M. Robert E. Chandler

    Traffic dynamics: studies in car following

    Oper. Res.

    (1958)
  • D. Gerlough

    Simulation of freeway traffic on a general-purpose discrete variable computer

    (1955)
  • M. Bando et al.

    Dynamical model of traffic congestion and numerical simulation

    Phys. Rev. E

    (1995)
  • M. Treiber et al.

    Microsimulations of freeway traffic including control measures

    at-Automatisierungstechnik Methoden und Anwendungen der Steuerungs-, Regelungs-und Informationstechnik

    (2001)
  • A. Kesting et al.

    Enhanced intelligent driver model to access the impact of driving strategies on traffic capacity

    arXiv preprint arXiv:0912.3613

    (2009)
  • Cited by (2)

    • Review on car-following sensor based and data-generation mapping for safety and traffic management and road map toward ITS

      2020, Vehicular Communications
      Citation Excerpt :

      A longitudinal predictive car-following model was proposed in [112]. A model that combined real-time and simulated car-following models using a traffic simulation system was developed in [113]. C) Data collected from car simulators

    • Virtual interaction algorithm of cultural heritage based on multi feature fusion

      2022, Journal of Computational Methods in Sciences and Engineering
    View full text