Car-following Data Research Articles

Automated Vehicle Identification Based on Car-Following Data With Machine Learning

Automated Vehicle Identification Based on Car-Following Data With Machine Learning

Understanding the traffic flow in different types of freeway tunnels based on car-following behaviors analysis

Tunnels are an engineering solution that has gained prominence for constructing freeways in mountainous regions. The length of the tunnels can vary, depending on the geological conditions, engineering requirements, and budget constraints. Car-following is the predominant driving behavior observed in tunnels, and understanding how drivers follow each other in different types of tunnels is crucial for ensuring smooth traffic flow and safety. Each type of tunnel environment can uniquely impact car-following behavior, which allows for targeted studies to optimize traffic management. In this research, natural driving data in freeway tunnels were collected through a driving experiment conducted on the Baomao Freeway in Chongqing, China. Then, the correlations and differences in car-following data between various tunnels and sections were analyzed. Finally, car-following models were developed considering various tunnel scenarios, and the influence of tunnel types on traffic flow was analyzed by simulation. The study revealed notable variations in car-following behavior across different types of tunnels, as well as within consecutive sections of the same tunnel. As tunnel length increased, the driving stability of following vehicles decreased, but the level of driving safety risk was not positively correlated with tunnel length. Significant vehicle trajectory oscillation was observed within the inner sections of long and extra-long tunnels, and a significant relationship between the acceleration of following vehicles and the location within the tunnel section was found. Additionally, the longer the tunnel, the greater the fluctuations in traffic flow, and the negative impact of the tunnel environment on traffic flow stability increased periodically downstream. These findings offer valuable insights for understanding and modeling car-following behavior in freeway tunnels, which ultimately facilitate traffic safety and mobility.

Promotion of driver compliance with V2V information in car-following tasks via multimodal display

The success of vehicle-to-vehicle (V2V) communication in improving the driving safety and efficiency primarily depends on driver compliance with the transmitted information. In this study, driver compliance with V2V information in simulated car-following tasks was measured and analyzed under different information display settings, such as baseline without information, visual information only, and multimodal information. Car-following data were collected from 30 participants in a driving simulation environment. The results revealed that V2V information could significantly reduce the driver's response time to the lead car’s speed change, resulting in an increase in the minimum time-to-collision when the lead car slowed down. Driver compliance was significantly higher under multimodal information settings than when using visual information alone. Moreover, driver compliance was higher when facing a decelerating lead car than when facing an accelerating lead car. No significant differences in driver compliance between the deceleration levels was observed; however, differences existed among different car-following preference groups. The findings of this study can be applied to V2V car-following models to improve the simulation accuracy for connected-vehicle traffic. Additionally, multimodal information presentation can provide a reference for V2V communication design.

An empirical investigation of driver car-following risk evolution using naturistic driving data and random parameters multinomial logit model with heterogeneity in means and variances

This study aims to address the questions of how driving risk evolves during car-following processes and what factors contribute to the underlying evolution patterns. An empirical study is conducted using real world car-following data collected in the Shanghai Naturalistic Driving Study (SH-NDS). The evolution of the driving risk induced by the dynamic coupling between the leading and following vehicles during the car-following process is characterized by how an instantaneous crash-risk measure - rear crash risk index (RCRI) - changes by time. A spectral clustering analysis is first conducted to classify the driving risk evolution of the observed car-following maneuvers, showing the existence of five distinctive risk evolution patterns in the car-following processes. In order to investigate the relationship between the identified driving risk evolution clusters and their contributing factors, a regression analysis employing a random parameter multinomial logit model with heterogeneity in means and variances is followed, revealing several significant contributing factors to the car-following risk evolution patterns, such as congestion level, driver’s ability to maintain stable headways, and vehicle deceleration. This study has provided important insights into driving risk from the new perspective of risk evolution patterns, which is expected to have significant implications for the future development of advanced traffic management and traveler information systems (ATMS/ATIS) strategies, advanced driver assistance systems (ADAS), and connected and autonomous vehicles (CAV).

Learning Car-Following Behaviors for a Connected Automated Vehicle System: An Improved Sequence-to-Sequence Deep Learning Model

Data-driven car-following modeling is of great significance to traffic behavior analysis and the development of connected automated vehicle (CAV) technology. The existing researches focus on reproducing the car-following process by capturing the behavior of the host vehicle using the information of its nearest preceding vehicle. While the other preceding vehicles may affect the host vehicle as well. To fill the gap above, this paper presents an improved sequence-to-sequence deep learning-based (ISDL) car-following model for a CAV system. Firstly, the kinematics information considering the multiple preceding vehicles are organized as the input characteristics. Secondly, an improved sequence-to-sequence deep learning framework is proposed by integrating an encoder with the bidirectional gated recurrent unit (GRU) neural network and a decoder using an attention-based GRU neural network in an end-to-end fashion. Finally, the car-following data with multiple preceding vehicles captured from the NGSIM dataset are employed to train and calibrate the proposed model. Experimental results indicate that the deep learning-based models’ performance in learning heterogeneous driving behavior can be enhanced by adding information about multiple preceding vehicles. In addition, the proposed ISDL model outperforms the benchmark car-following models in terms of the accuracy of the simulated speeds and simulated positions. Through tests on platoon simulation, the ISDL model is capable of reshaping the traffic oscillation phenomenon as well.

Modeling differential car-following behavior under normal and rainy conditions: A memory-based deep learning method with attention mechanism

In order to analyze and learn the difference in car-following behavior between normal and rainy days, we first collect car-following trajectory data of an urban elevated road on normal and rainy days by microwave radar and analyze the differences in speed, relative speed, acceleration, space headway, and time headway among data through statistics. Secondly, owing to the time-series characteristics of car-following data, we use the long short-term memory (LSTM) neural network optimized by attention mechanism (AM) and sparrow search algorithm (SSA) to learn the different car-following behaviors under different weather conditions and build corresponding models (ASL-Normal, ASL-Rain, where ASL stands for AM-SSA-LSTM), respectively. Finally, the simulation test shows that the mean square error (MSE) and reciprocal of time-to-collision (RTTC) of the ASL model are better than those of LSTM and intelligent diver model (IDM), which is closer to the real data. The ASL model can better learn different driving behaviors on normal and rainy days. However, it has a higher sensitivity to weather conditions from cross test on normal and rainy data-sets which need classification training or sample diversification processing. In the car-following platoon simulation, the stability performances of two models are excellent, which can describe the basic characteristics of traffic flow on normal and rainy days. Comparing with ASL-Rain model, the convergence time of ASL-Normal is shorter, reflecting that cautious driving behavior on rainy days will reduce traffic efficiency to a certain extent. However, ASL-Normal model produces a more severe and frequent traffic oscillation within a shorter period because of aggressive driving behavior on normal days.

Driver Identification Through Heterogeneity Modeling in Car-Following Sequences

Intra-driver and inter-driver heterogeneity has been confirmed to exist in human driving behaviors by many studies. This research proposes a driver identification method by modeling such heterogeneities in car following sequences. It is assumed that all drivers share a pool of driver states; under each state, a car-following data sequence obeys a specific probability distribution in feature space; each driver has his/her own probability distribution over the states, called driver profile, which characterize the intra-driver heterogeneity, while the difference between the driver profile of different drivers depicts the inter-driver heterogeneity. Thus, the driver profile can be used to distinguish a driver from others. Based on the assumption, a method of driver identification is proposed to take both intra- and inter-driver heterogeneity into consideration, and a method is developed to jointly learn parameters in behavioral feature extractor, driver states, and driver profiles. Experiments demonstrate the performance of the proposed method in driver identification on naturalistic car-following data: accuracy of 82.3% is achieved in an 8-driver experiment using 10 car-following sequences of duration 15 seconds for online inference. The potential of fast registration of new drivers is demonstrated and discussed.

Action-points in human driving and in SUMO

When following a vehicle, drivers change their acceleration at so called action-points (AP), and keep it constant in between them. By investigating a large data-set of car-following data, the state- and time-distributions of the APs is analyzed. In the state-space spanned by speed-difference and distance to the lead vehicle, this distribution of APs is mostly proportional to the distribution of all data-points, with small deviations from this. Therefore, the APs are not concentrated around certain thresholds as is claimed by psycho-physical car-following models.Instead, small distances indicate a slightly higher probability of finding an AP than is the case for large distances. A SUMO simulation with SUMO's implementation of the Wiedemann model confirms this view: the AP's of the Wiedemann model follow a completely different distribution than the empirical ones.

Driving strategy of connected and autonomous vehicles based on multiple preceding vehicles state estimation in mixed vehicular traffic

In the near future, connected and autonomous vehicles (CAVs) will share road space with human-driven vehicles (HVs). In this mixed vehicular traffic, effective following cooperation among multiple vehicles is an important basis for improving traffic efficiency and safety. However, CAVs are unable to communicate with HVs to acquire information. Therefore, how to obtain HV information and realize cooperative car-following has become an urgent problem for CAVs. This paper proposes a CAV driving strategy that considers multiple preceding vehicles, including HVs. The strategy first uses a large amount of real car-following data to build an upgraded Elman neural network (ENN) model optimized with the sparrow search algorithm (SSA), which is utilized to obtain HV information. Then, we combine the SSA-ENN with the classical car-following model and use a time-varying weighting model to analyze the impact of the different states of multiple preceding cars at various moments on the host car, so as to achieve car-following driving control. Numerical simulations are carried out, and the results show that the driving strategy can improve road capacity and suppress traffic oscillations. With the increase in CAV penetration, traffic efficiency, safety, and driving comfort are improved accordingly.

Finite Mixture of the Hidden Markov Model for Driving Style Analysis

Analyzing driving style is useful for developing intelligent vehicles. Previous studies usually consider the statistical features (e.g., the means and standard deviations of brake pressure) of the measured driving data or manually define the number of patterns divided by behavior semantics to characterize driving styles. In this paper, we propose a driving style analysis to describe the personalized driving styles from time-series driving data without specifying the levels in advance but by estimating them from the data. First, range, range rate, and acceleration are selected as three feature variables to describe car-following scenarios. Then, the car-following data are normalized to reduce the scale influence of different variables on the segmentation results. The hidden Markov model (HMM) and the finite mixture of the hidden Markov model (MHMM) are adopted to extract behavior semantics. Compared with the HMM, the MHMM can identify the heterogeneity of data and then provide more reasonable primitive driving patterns. Based on the results, this study uses the K-means clustering to label all the driving patterns semantically and identifies a total of 75 different driving patterns. We use the normalized frequency distributions to describe personalized driving behavior characteristics, and similarity evaluations of driving styles are applied using the Kolmogorov–Smirnov test. The proposed approach in this paper is useful for exploring the characteristics of driving habits.

An Application of Chaotic Time-series Analysis to Car-following Data at Capacity Bottleneck

An Application of Chaotic Time-series Analysis to Car-following Data at Capacity Bottleneck

Modeling Car-Following Heterogeneities by Considering Leader–Follower Compositions and Driving Style Differences

To better understand the behavioral heterogeneities of human-operated vehicles, the paper proposes a method to distinguish car-following behaviors in specific leader–follower contexts. Using the Next-Generation Simulation dataset, the car-following data are first classified into four leader–follower compositions, namely, truck–car, car–car, car–truck, and truck–truck. Based on the classified data, we calibrate the parameters of a few well-known car-following models, including Full Velocity Difference model, Intelligent Driver Model, and Gazis–Herman–Rothery model. Principal component analysis and clustering analysis are then applied to the calibrated parameters to discover the behavioral patterns and to find the probabilistic distributions of the parameters for the classified car-following (CCF) models. Simulation results show that compared with the unified car-following models, the estimation errors of calibrated CCF models are reduced by 20.79% to 49.05%, which indicates that the proposed method provides a more accurate description of car-following heterogeneities. The proposed framework could help highway traffic operators better know the traffic users.

Research on driving proneness in car-following behaviours based on multi-source real driving data

There is little research on the degrees of drivers’ short-term behaviours regarding driving safety. To solve this problem, this paper investigated the concept of driving proneness and evaluated the propensities of different drivers to engage in different operations for the following scenarios of urban traffic. From the real driving data of sixteen drivers on a city road, car-following data fragments were extracted and six key parameters were obtained: throttle percentage, change rate of throttle percentage, brake pressure, change rate of brake pressure, absolute value of steering angle and absolute value of steering angle speed. Symbolic Aggregate Approximation was used to reduce the dimensionality of the parameters. The input of the Hidden Markov Model-Viterbi was obtained by the use of statistical methods. The output of the model is the probability of the three proneness states of introversion, neutrality and extroversion, from which the proneness value of each driver was calculated. The weighted proneness value of each driver was obtained by the use of the entropy weight method to assign weights to each parameter. The operating characteristics of the drivers were also analysed and described. The method presented in this paper can provide accurate and real-time warning in network-driven environments.

Learning the Car-following Behavior of Drivers Using Maximum Entropy Deep Inverse Reinforcement Learning

The present study proposes a framework for learning the car-following behavior of drivers based on maximum entropy deep inverse reinforcement learning. The proposed framework enables learning the reward function, which is represented by a fully connected neural network, from driving data, including the speed of the driver’s vehicle, the distance to the leading vehicle, and the relative speed. Data from two field tests with 42 drivers are used. After clustering the participants into aggressive and conservative groups, the car-following data were used to train the proposed model, a fully connected neural network model, and a recurrent neural network model. Adopting the fivefold cross-validation method, the proposed model was proved to have the lowest root mean squared percentage error and modified Hausdorff distance among the different models, exhibiting superior ability for reproducing drivers’ car-following behaviors. Moreover, the proposed model captured the characteristics of different driving styles during car-following scenarios. The learned rewards and strategies were consistent with the demonstrations of the two groups. Inverse reinforcement learning can serve as a new tool to explain and model driving behavior, providing references for the development of human-like autonomous driving models.

Modeling Car-Following Behaviors and Driving Styles with Generative Adversarial Imitation Learning.

Building a human-like car-following model that can accurately simulate drivers’ car-following behaviors is helpful to the development of driving assistance systems and autonomous driving. Recent studies have shown the advantages of applying reinforcement learning methods in car-following modeling. However, a problem has remained where it is difficult to manually determine the reward function. This paper proposes a novel car-following model based on generative adversarial imitation learning. The proposed model can learn the strategy from drivers’ demonstrations without specifying the reward. Gated recurrent units was incorporated in the actor-critic network to enable the model to use historical information. Drivers’ car-following data collected by a test vehicle equipped with a millimeter-wave radar and controller area network acquisition card was used. The participants were divided into two driving styles by K-means with time-headway and time-headway when braking used as input features. Adopting five-fold cross-validation for model evaluation, the results show that the proposed model can reproduce drivers’ car-following trajectories and driving styles more accurately than the intelligent driver model and the recurrent neural network-based model, with the lowest average spacing error (19.40%) and speed validation error (5.57%), as well as the lowest Kullback-Leibler divergences of the two indicators used for driving style clustering.

A Car-Following Driver Model Capable of Retaining Naturalistic Driving Styles

The modeling of car-following behavior is an attractive research topic in traffic simulation and intelligent transportation. The driver plays an important role in car following but is ignored by most car-following models. This paper presents a novel car-following driver model, which can retain aspects of human driving styles. First, simulated car-following data are generated by using the speed control driver model and the real-world driving behavior data if the real-world car-following data are not available. Then, the car-following driver model is established by imitating human driving maneuver during real-world car following. This is accomplished by using a neural network-based learning control paradigm and car-following data. Finally, the FTP-72 driving cycle is borrowed as the speed profile of the leading vehicle for the model test. The driving style is quantitatively analyzed by AESD. The results show that the proposed car-following driver model is capable of retaining the naturalistic driving styles while well accomplishing the car-following task with the error of relative distance mostly less than 5 meters for every driving styles.

A Car-Following Data Collecting Method Based on Binocular Stereo Vision

This study proposes the design, implementation, and application of a new car-following data collecting method based on binocular stereo vision system. The car-following data consist of car-following distance, velocity, relative velocity and driving trajectory of a following vehicle. Specifically, triangulation principle is used to measure car-following distance. Robust locally weighted regression smoothing method is applied to automatically filter out the outliers caused by disturbance in real-world driving condition. Real-world experiments indicate that binocular cameras with 25mm focal length have advantages in traffic conditions with long car-following distance and high driving speed, and that binocular cameras with12mm focal length are suitable to measure on urban roads, where car-following distance is short and more driving conditions are needed to be observed. In 0-60m distance measuring range, it is verified that the mean absolute percentage error (MAPE) of binocular cameras with 25mm focal length is 3.2% and the MAPE of binocular cameras with 12mm focal length is 6.3%. The data collecting system helps to advance car following models due to its high accuracy and effective cost.

A Study on an Anthropomorphic Car-Following Strategy Framework of the Autonomous Coach in Mixed Traffic Flow

In this study, to explore the demand characteristics of autonomous coaches in mixed traffic flow, two sections of an expressway were selected for vehicle experiments. With personification as the goal, sensors of the ego-vehicle were used to collect naturalistic car-following behavior data of surrounding coaches. After analyzing the naturalistic car-following data, car-driving behavior characteristics of the coach drivers were acquired. The analysis results indicate that the overall car-following processes of coaches tend to be relatively stable, and most of the car-following processes are in the conditions of higher velocity, smaller acceleration, and smaller relative velocity than the front vehicle. The expected following spacing and time headway of the coach drivers increase with the increase of velocity. By analysis and comparison with existing safe following models, it is found that while the safety and the warning rate based on the car-following coach safety distance model are higher, the warning rate of the car-following time headway model at the relatively higher velocity is lower. The coach safety distance model established in this study considers the car-following risk and the acceptance rate of the drivers to the warning system. This model conforms to the driver’s car-following characteristics and fulfills the requirements of collision avoidance. Applying the safety distance model to the autonomous coach would effectively improve the anthropopathic behavior characteristics of the autonomous coach. Additionally, the abnormality caused by the interference of other vehicles in mixed traffic flow of autonomous and human vehicles is successfully avoided under the premise of ensuring the safe driving of the autonomous coach.

Car-following Behavior Analysis of Left-turn Vehicles at Signalized Intersections

In order to enrich the car-following theory of urban signalized intersections, and reveal the car-following characteristics of left turn at signalized intersections, the car-following behavior of left turn at signalized intersections is studied. The car-following data acquisition test which was based on high precision GPS was designed. And the car-following characteristics of left-turning vehicles at signalized intersections with different turning radii were analyzed. Based on which, the influence of radius on the car-following behavior was explained, and the New Full Velocity Difference (NFVD) model was developed. The genetic algorithm was used to calibrate the parameters of the NFVD model. The stability and accuracy of the calibrated model was further analyzed by using field data. The results showed that the average speed of the following car increases with the turning radius of the signalized intersection; the car-following speed which the highest frequency occurs under different turning radii tends to increase with the enlargement of turning radius; the larger the average headway distance between the car-following vehicles, the more intense of the driver’s response to the deceleration of the front vehicle. These findings could be used in traffic simulation and to make engineering decisions.

An improved car-following model with consideration of multiple preceding and following vehicles in a driver’s view

Common driving experience suggests that a driver’s car-following behavior is often affected by the preceding and following vehicles in the driver’s view. In this study, we propose an improved car-following model based on the full velocity difference (FVD) model by considering two preceding and the immediate following vehicles in a driver’s view. Car-following data extracted from real traffic situation is used to calibrate and verify both of the improved model and the FVD model. Our data analysis shows that the motion status of two preceding vehicles and one following vehicle in a driver’s view can influence traffic flow of following vehicles after the driver’s vehicle. We compare the performance of the improved model and the FVD model by simulating the starting and braking processes of the following vehicles. The numerical simulation study shows that the improved model fits the experimental data better than the FVD model by showing a shorter time of completing the starting and braking processes of the following vehicles.