Driving Behavior Data Research Articles

Denoising Recurrent Neural Networks for Classifying Crash-Related Events

With detailed sensor and visual data from automobiles, a data-driven model can learn to classify crash-related events during a drive. We propose a neural network model accepting time-series vehicle sensor data and forward-facing videos as input for learning classification of crash-related events and varying types of such events. To elaborate, a novel recurrent neural network structure is introduced, namely, denoising gated recurrent unit with decay, in order to deal with time-series automobile sensor data with missing value and noises. Our model detects crash and near-crash events based on a large set of time-series data collected from naturalistic driving behavior. Furthermore, the model classifies those events involving pedestrians, a vehicle in front, or a vehicle on either side. The effectiveness of our model is evaluated with more than two thousand 30-s clips from naturalistic driving behavior data. The results show that the model, including sensory encoder with denoising gated recurrent unit with decay, visual encoder, and attention mechanism, outperforms gated recurrent unit with decay, gated CNN, and other baselines not only in event classification and but also in event-type classification.

Mobile phone conversation distraction: Understanding differences in impact between simulator and naturalistic driving studies

A current issue within the driver distraction community centres around different findings regarding the impact of mobile phone conversation on driving found in driving simulators versus instrumented vehicles employed in real-world naturalistic driving studies (NDSs). This paper compares and contrasts the two types of studies and aims to provide reasons for the differences in findings that have been documented. A comprehensive review of literature and consultations with human factors experts highlighted that simulator studies tend to show degradation in driving performance, suggestive of increased crash risk as a result of mobile phone conversation. Whilst NDSs, at times, present data suggesting that mobile phone conversation distraction actually reduces crash risk. This study identifies that these differences may be attributed to behavioural hypotheses associated with driver self-regulation, arousal from cognitive loading, task displacement and gaze concentration – all of which need to be explicitly tested in future driving studies.Metric estimation and application was also revealed to be polarising results and the subsequent assessment of the crash risk. A common metric applied in this domain is the ‘Odds Ratio’, particularly prevalent in NDSs. This study presents a detailed investigation into the assumptions and application of the Odds Ratio which revealed the potential for over- and under-estimation of the metric depending on the core data and sampling assumptions. Furthermore, this research presents a comparative analysis of select driving simulator studies and an NDS considering only driving behaviour data as a means to consistently compare the findings of both methodologies. The findings from this investigation implores the need for greater consistency in the application of analysis methods and metrics across both simulator and NDSs. Improvements can yield a more robust platform to systematically compare and interpret data across both approaches, ultimately leading to enhanced planning and safety regarding mobile phone use while driving.

Effects of Highway Landscapes on Drivers’ Eye Movement Behavior and Emergency Reaction Time: A Driving Simulator Study

This study aims at investigating the effects of highway landscapes and alignments on drivers’ eye movement behavior and emergency reaction time, based on a driving simulator experiment. In this study, four simulation scenarios are evaluated including open space, semiopen space, semiclosed space, and enclosed space landscapes on highways in Yunnan Province, China. Twenty-four experienced drivers participated in a 6-kilometer driving experiment in each landscape scenario. Each subject was required to drive at 80 km/h in the scenarios and the driving behavior data were collected. Three different data analysis methods were employed: (1) descriptive analysis of the characteristics of drivers’ visual fixation area; (2) statistical tests of emergency reaction time with drivers’ demographic characteristics, highway landscapes, and alignments; and (3) multiple linear regression analysis of emergency reaction time, highway landscapes, and alignments. The results show that emergency reaction time is significantly influenced by highway landscapes and alignments, and the multiple linear regression model built in this experiment could accurately predict drivers’ emergency reaction time in different highway landscapes and alignments.

An analysis of the relationship between driver characteristics and driving safety using structural equation models

Traffic crashes pose a serious threat to global health and have negative impacts on social and economic development. Driving safety could be significantly improved if drivers obtained effective training based on practical suggestions from studies of the relationship between driver characteristics and driving safety. In this study, driver characteristics are analyzed with a focus on illegal driving behaviors, and a new method is developed to reduce driver error. However, driver characteristics are complex, and the associated interrelationships among variables are not easily identified. A structural equation model (SEM) is adopted in this paper to capture the complex relationships among related variables. This model can simultaneously address the complex relationships among endogenous and exogenous variables. We investigate five illegal driving behaviors, including illegal passing, illegal lane changes, speeding, running red lights and distracted driving, based on a driving simulator. Distracted driving is chosen as an example to construct the SEM. We group 13 variables into four latent variables in the model. In this paper, 44 participants (average age ± S.D. = 33 ± 12.8 years, range: 19–55 years; average driving experience ± S.D. = 10 ± 8.8 years, range: 2–30 years) were recruited for the experiment, driving behavior data were collected via a driving simulator and driving attitude data were collected using a questionnaire. Our findings reveal that the main factor that influences illegal driving actions is driving attitude and the main factor that influences distracted driver performance is basic driver characteristics. Moreover, the driver training level is the most significant factor that negatively affects the basic driver characteristics (factor load = −0.91), and anger with slow driving is the main factor that influences the driver attitude (factor load = 0.90). According to the SEM results, the relationship between driver characteristics and driving safety is analyzed. The model results suggest that driver training lessons must be implemented and traffic issues must be urgently solved. In addition, our results suggest that extra training sessions should be required of young females and drivers with bad attitudes and serious road rage to improve driving safety. Although the results are based on distracted driving, the method presented in this paper is suitable for other analyses of driving behavior.

A hierarchical machine learning classification approach for secondary task identification from observed driving behavior data

According to NHTSA, more than 3477 people (including 551 non-occupants) were killed and 391,000 were injured due to distraction-related crashes in 2015. The distracted driving epidemic has long been under research to identify its impact on driving behavior. There have been a few attempts to detect drivers’ engagement in secondary tasks from observed driving behavior. Yet, to the authors’ knowledge, not much effort has been directed to identify the types of secondary tasks from driving behavior parameters. This study proposes a bi-level hierarchical classification methodology using machine learning to identify the different types of secondary tasks drivers are engaged in using their driving behavior parameters. At the first level, drivers’ engagement in secondary tasks is detected, while at the second level, the distinct types of secondary tasks are identified. Comparative evaluation is performed between nine ensemble tree classification methods to identify three types of secondary tasks (hand-held cellphone calling, cellphone texting, and interaction with an adjacent passenger). The inputs to the models are five driving behavior parameters (speed, longitudinal acceleration, lateral acceleration, pedal position, and yaw rate) along with their standard deviations. The results showed that the overall secondary task detection accuracy ranged from 66% to 96%, except for the Decision Tree that was able to detect engagement in secondary tasks with a high accuracy of 99.8%. For the identification of secondary tasks types, the overall accuracy ranged from 55% to 79%, with the highest accuracy of 82.2% achieved by the Random Forest method. The findings of the paper show the proposed methodology promising to (1) characterize drivers’ engagement in unlawful secondary tasks (such as texting) as a counter measure to prevent crashes, and (2) alert drivers to pay attention back to the main driving task when risky changes to their driving behavior take place.

Advancing usage‐based insurance – a contextual driving risk modelling and analysis approach

Most researchers and insurers evaluate driver's risks solely based on user global positioning system (GPS) trajectories. A new approach is proposed that collects individualised driving behaviour data from smartphone GPS module, combined with geographical network information and dynamic traffic conditions, to identify driving risk factors and evaluate driving behaviours in various contexts. The multi-source data reveals real world activity patterns as to when, where and how an individual driver performs, from which performance measurements on both the trip and driver level are defined to measure the risks. In addition, the relationship between the defined driving performance measurements and accident rate can be examined and verified by combining the crash history of the drivers. Numeric analysis on the trip level demonstrates that driving behaviour is context-sensitive, and the principal component analysis performed at the driver level shows that pedal operation and driving speed are the two most important performance measurements to characterise an individual's driving pattern. The subsequent correlation analysis of crash history and driving performances verify both pedal operation and driving speed are significantly related to more at-fault accidents, which validates the modelling and analysis efforts. The findings of their study further existing knowledge and provide foundations for advanced pay-as-you-drive-and-you-save insurance pricing.

Classification of Distracted Driving Based on Visual Features and Behavior Data using a Random Forest Method

This research is to explore the relationship between a driver’s visual features and driving behaviors of distracted driving, and a random forest (RF) method is developed to classify driving behaviors and improve the accuracy of detecting distracted driving. Drivers were required to complete four distraction tasks while they followed simulated vehicles in the experiment. In data analysis, the features of distracted driving behaviors are first described, and the visual data are classified into three distraction levels based on the AttenD algorithm. Based on the collected data, this paper shows the relationship between visual features and driving behavior. Significant differences are discovered between different distraction tasks and distraction levels. Additionally, driving behavior data is used to build an RF model to classify distracted driving into three levels. Results demonstrate that this model is feasible to capture the classification of distraction and its accuracy for each distraction task is over 90%. Areas under receiver operating characteristic curve calculated through error-correcting output codes are mainly around 0.9, indicating good reliability. With this classification method, distraction levels could be classified with vehicle operation characteristics. The model established by this method could detect distractions in actual driving through the detection of driving behavior without the need of eye tracking systems.

Uncertainty Analysis of Rear-End Collision Risk Based on Car-Following Driving Simulation Experiments

Rear-end collisions are one of the most common types of accidents, and the importance of examining rear-end collisions has been demonstrated by numerous accidents analysis researches. Over the past decades, lots of models have been built to describe driving behaviour during car following to better understand the cause of collisions. However, it is necessary to consider individual difference in car-following modelling while it seems to be ignored in most of previous models. In this study, a rear-end collision avoidance behaviour model considering drivers’ individual differences was developed based on a common deceleration pattern extracted from driving behaviour data, which were collected in a car-following driving simulation experiment. Parameters of variables in the model were calibrated by liner regression and Monte Carlo method was adopted in model simulation for uncertainty analysis. Simulation results confirmed the effectiveness of this model by comparing them to the experiment data and the influence of driving speed and headway distance on the rear-end collision risk was indicated as well. The thresholds for driving speed and headway distance were 18 m/s and 15 m, respectively. An obvious increase of collision risk was observed according to the simulation results.

Design of an experimental protocol to examine medication non-adherence among young drivers diagnosed with ADHD: A driving simulator study.

The diagnosis of ADHD among teens and young adults has been associated with a higher likelihood of motor vehicle crashes. Some studies suggest a beneficial effect of ADHD medication but the exact efficacy is still being debated. Further, medication adherence, which is low in this age group, can further reduce effectiveness. Our long-term objective is to reduce unsafe driving among drivers with ADHD by detecting medication non-adherence through driver behavior modeling and monitoring. As a first step, we developed the described lab study protocol to obtain reliable driver behavior data that will then be used to design and train behavior models built through machine learning. This experimental study protocol was developed to systematically compare driving behaviors under two medication conditions (before and after intake of medication) among young adults with ADHD and a control group of non-ADHD. A driving simulator was used to examine driving behaviors and interactions with traffic. The primary outcome was speed management for two comparisons (ADHD vs. non-ADHD and before vs. after medication), and secondary objectives involved understanding differences among the participants utilizing self-reported surveys about ADHD symptoms, drivers' knowledge, and perception about safety. The study protocol was designed to maximize participant safety and efficiency of data collection, as multiple measures were collected over two 2-h study visits. The sampled ADHD drivers were demographically and psychosocially similar but clinically different from the non-ADHD group. Overall, this protocol was effective in participant recruitment and retention, allowed staggered data collection, and can be incorporated in a subsequent clinical trial that examines the efficacy of a machine-learning based driver monitoring intervention.

Recognizing driving styles based on topic models

With the explosion of information in our current era, senders of information increasingly need to target their messages to recipients. However, messages within transportation systems, including traffic information and commercial advertisements, tend to be transmitted to all drivers indiscriminately. This is because the information providers (such as other vehicles, roads, facilities, buildings etc.), can hardly recognize the variations within drivers, who should be treated differently as information recipients. As a result of the rapid development of data collection technologies and machine learning techniques in recent years, extraction and recognition of drivers’ unique driving style from actual driving behaviour data become possible. In this paper, two kinds of topic models are investigated: mLDA and mHLDA, to discover distinguishable driving style information with hidden structure from the real-world driving behaviour data. The results show that the proposed models can successfully recognize the differences between driving styles. The study is of great value for providing deep insight into the underlying structure of driving styles and can effectively support the recognition of drivers with different driving styles.

운전자 친화 적응형 순항제어 알고리즘에 관한 연구

Adaptive cruise control is one of the popular features of the advanced driver assistance system(ADAS). It is an extended form of conventional cruise control, and it has been considered as a contributing factor in driving safety and string stability. Recently, human behavior in the automotive industry has become more significant. This paper proposed driver-friendly adaptive cruise control(DF-ACC), and it demonstrated the deduction method control parameter of a current driver. For the design of the DF-ACC algorithm, actual driver behavior data with a driving simulator based on the HIL system were collected and analyzed. Then, driving data were separated in a stable and transient zone. The driver’s desired distance and acceleration as control parameters could be estimated from the driving data in each zone. Lastly, this paper compared real driver driving and DF-ACC simulation data.

Toward Intelligent Vehicle Intrusion Detection Using the Neural Knowledge DNA

ABSTRACTIn this paper, we propose a novel intrusion detection approach using past driving experience and the neural knowledge DNA for in-vehicle information system security. The neural knowledge DNA is a novel knowledge representation method designed to support discovering, storing, reusing, improving, and sharing knowledge among machines and computing systems. We examine our approach for classifying malicious vehicle control commands based on learning from past valid driving behavior data on a simulator.

Analysis of lane change behavior with driver behavior data on expressway

Analysis of lane change behavior with driver behavior data on expressway

Driving behavior recognition using EEG data from a simulated car-following experiment

Driving behavior recognition is the foundation of driver assistance systems, with potential applications in automated driving systems. Most prevailing studies have used subjective questionnaire data and objective driving data to classify driving behaviors, while few studies have used physiological signals such as electroencephalography (EEG) to gather data. To bridge this gap, this paper proposes a two-layer learning method for driving behavior recognition using EEG data. A simulated car-following driving experiment was designed and conducted to simultaneously collect data on the driving behaviors and EEG data of drivers. The proposed learning method consists of two layers. In Layer I, two-dimensional driving behavior features representing driving style and stability were selected and extracted from raw driving behavior data using K-means and support vector machine recursive feature elimination. Five groups of driving behaviors were classified based on these two-dimensional driving behavior features. In Layer II, the classification results from Layer I were utilized as inputs to generate a k-Nearest-Neighbor classifier identifying driving behavior groups using EEG data. Using independent component analysis, a fast Fourier transformation, and linear discriminant analysis sequentially, the raw EEG signals were processed to extract two core EEG features. Classifier performance was enhanced using the adaptive synthetic sampling approach. A leave-one-subject-out cross validation was conducted. The results showed that the average classification accuracy for all tested traffic states was 69.5% and the highest accuracy reached 83.5%, suggesting a significant correlation between EEG patterns and car-following behavior.

Visualization of Driving Behavior Based on Hidden Feature Extraction by Using Deep Learning

In this paper, we propose a visualization method for driving behavior that helps people to recognize distinctive driving behavior patterns in continuous driving behavior data. Driving behavior can be measured using various types of sensors connected to a control area network. The measured multi-dimensional time series data are called driving behavior data. In many cases, each dimension of the time series data is not independent of each other in a statistical sense. For example, accelerator opening rate and longitudinal acceleration are mutually dependent. We hypothesize that only a small number of hidden features that are essential for driving behavior are generating the multivariate driving behavior data. Thus, extracting essential hidden features from measured redundant driving behavior data is a problem to be solved to develop an effective visualization method for driving behavior. In this paper, we propose using deep sparse autoencoder (DSAE) to extract hidden features for visualization of driving behavior. Based on the DSAE, we propose a visualization method called a driving color map by mapping the extracted 3-D hidden feature to the red green blue (RGB) color space. A driving color map is produced by placing the colors in the corresponding positions on the map. The subjective experiment shows that feature extraction method based on the DSAE is effective for visualization. In addition, its performance is also evaluated numerically by using pattern recognition method. We also provide examples of applications that use driving color maps in practical problems. In summary, it is shown the driving color map based on DSAE facilitates better visualization of driving behavior.

Naturalistic Driver Behaviour in Response to the Multi-Sensory Experience of Rear-End Collisions

The purpose of this study was to quantify naturalistic driver behaviour during real-world rear-end collisions. Rear-end collisions from the 100-car naturalistic driving database were reviewed and behaviour of drivers in the struck vehicles (i.e., lead vehicles) were analyzed. Results indicate that rear-ended drivers disengage their foot from the brake pedal and then reapply within average perception and response intervals (0.90 seconds; 0.15 SD) and braking intensities (0.50 g; 0.27 SD) consistent with visual detection of immediate collision hazards, despite the added complexity of a multi-sensory collision experience. Together with previous research, these data suggest that perception and response durations and braking intensities are scaled to the severity of the detected hazard. This research has applications to forensic investigations of collisions in which the inclusion of driver behaviour data is a requisite to quantifying impact dynamics and the analysis of collision avoidance potential in real-world conditions.

Drivers’ social-work relationships as antecedents of unsafe driving: A social network perspective

In order to reduce road accidents rates, studies around the globe have attempted to shed light on the antecedents for unsafe road behaviors. The aim of the current research is to contribute to this literature by offering a new organizational antecedent of driver's unsafe behavior: The driver's relationships with his or her peers, as reflected in three types of social networks: negative relationships network, friendship networks and advice networks (safety consulting). We hypothesized that a driver's position in negative relationship networks, friendship networks, and advice networks will predict unsafe driving. Additionally, we hypothesized the existence of mutual influences among the driver’s positions in these various networks, and suggested that the driver's positions interact to predict unsafe driving behaviors. The research included 83 professional drivers from four different organizations. Driving behavior data were gathered via the IVDR (In-Vehicle Data Recorder) system, installed in every truck to measure and record the driver's behavior. The findings indicated that the drivers' position in the team networks predicts safe driving behavior: Centrality in negative relationship networks is positively related to unsafe driving, and centrality in friendship networks is negatively related to unsafe driving, while centrality in advice networks is not related to unsafe driving. Furthermore, we found an interaction effect between negative network centrality and centrality in friendship networks. The relation between negative networks and unsafe behavior is weaker when high levels of friendship network centrality exist. The implications will be presented in the Discussion section.

The value of vehicle telematics data in insurance risk selection processes

The advent of the Internet of Things enables companies to collect an increasing amount of sensor generated data which creates plenty of new business opportunities. This study investigates how this sensor data can improve the risk selection process in an insurance company. More specifically, several risk assessment models based on three different data mining techniques are augmented with driving behaviour data collected from In-Vehicle Data Recorders. This study proves that including standard telematics variables significantly improves the risk assessment of customers. As a result, insurers will be better able to tailor their products to the customers' risk profile. Moreover, this research illustrates the importance of including industry knowledge, combined with data expertise, in the variable creation process. Especially when a regulator forces the use of easily interpretable data mining techniques, expert-based telematics variables are able to improve the risk assessment model in addition to the standard telematics variables. Further, the results suggest that if a manager wants to implement Usage-Based-Insurances, Pay-As-You-Drive related variables are most valuable to tailor the premium to the risk. Finally, the study illustrates that this new type of telematics-based insurance product can quickly be implemented since three months of data is already sufficient to obtain the best risk estimations.

Safety impacts of signalized lane merge control at highway work zones

ABSTRACTLane closures due to highway work zones present many challenges to the goal of ensuring smooth traffic operations and a safe environment for both drivers and workers. Late merge behavior at a work zone closure is a dangerous behavior that impacts the traffic conflicts upstream of work zone closures. This paper analyzes the safety impacts of using a signalized lane control strategy at the work zone merge points. To achieve the objective of this research, a field study has been conducted at a highway work zone to collect traffic and driver behavior data, and a two-stage, simulation-based approach is used to analyze the safety impacts of implementing a signalized lane merge control strategy at the studied work zone. In the first stage, micro-simulation models are developed and calibrated based on field data to generate vehicle trajectories. In the second stage, the U.S. Federal Highway Administration’s Surrogate Safety Assessment Model is employed to identify potential conflicts under different traffic conditions. The paper concludes that a proposed signal control device could significantly reduce lane-change conflicts at work zone merge points. In addition, recommendations on the signal cycle length and timing splits are provided.

Stochastic Forecasting of Vehicle Dynamics Using Sequential Monte Carlo Simulation

Predictive control is a popular approach for further improving the efficiency and performance of vehicular systems enabling intelligent systems behavior appropriate to the driving situation. To calculate such control strategies, the future vehicle dynamics or subsequent states have to be predicted. We introduce a stochastic framework based on an explanatory model and stochastic processes to predict future vehicle dynamics with road network data. The distributions of the future states are approximated using sequential Monte Carlo simulation. The proposed approach enables stochastic forecasts incorporating uncertain driver behavior and available road data. Parameter inference is shown for exemplary real-drive test data, and predictive performance is evaluated using commonly used reference models. The results show that the explanatory model provides more specific information than time-series models do, still considering the uncertainty in the driver's behavior or the situation. The framework can be applied with predictive control algorithms enabling intelligent control of vehicular systems. Furthermore, the framework or parts of it may be usable for other applications like predicting behavior of traffic participants or general characterization of driver behavior.