A framework for evaluating aggressive driving behaviors based on in-vehicle driving records

https://doi.org/10.1016/j.trf.2017.11.021Get rights and content

Highlights

  • This study developed a framework to cluster drivers’ behavior using driving records.

  • The framework was applied to large-scale data from real driving environment.

  • Representative driving patterns were identified on the cluster map.

  • The cluster map can be used as a reference in evaluating other driver’s behavior.

Abstract

Driving behavior is how drivers respond to actual driving environments and a major factor for road traffic safety. Recent advances in in-vehicle sensors facilitate continuous monitoring of driving behaviors; large-scale driving data have been accumulated. This study develops a framework to evaluate large-scale driving records and to establish clusters that can be used to identify potentially aggressive driving behaviors. The framework employs three steps of data analytic methods: abrupt change detection to extract meaningful driving events from raw data, feature extraction using an auto-encoder, and two-level clustering. This framework is applied to real driving data that were obtained from 43 taxis in Korean metropolitan cities. The application shows that the framework can characterize driving patterns from large-scale driving records and identify clusters with high potential for aggressive driving. The findings imply that the outcome clusters represent the norm of driving behavior and thus can be used as a reference in diagnosing other drivers’ behavior.

Introduction

Driving behavior, which is reflected in how drivers control speed and steer their vehicles, is a major interest in traffic safety research. Driving behavior widely varies across individuals; differences of driving behavior have significant effects on roadway traffic. Especially, aggressive driving behavior has been reported as the main cause of traffic safety problems (NHTSA, 1998). In support of this, there have been many studies that have revealed the causal relation between frequency of aggressive driving behavior and crash occurrence (Boyce and Geller, 2002, Evans, 2004, Klauer et al., 2009). To enhance traffic safety, therefore, it is effective to identify aggressive driving behaviors and provide drivers with proper feedback to adjust their driving (Toledo, Musicant, & Lotan, 2008).

Earlier studies about aggressive driving behavior have been separated largely into two groups: (i) finding influential factors on aggressive driving occurrences (Evans and Wasielewski, 1983, Cooper, 1990, Gregersen and Berg, 1994, Shinar, 1998, Shinar and Compton, 2004) and (ii) identifying and defining the categories of aggressive driving behavior (Tasca, 2000, AAA Foundation for Traffic Safety, 2009, Abou-Zeid et al., 2011). The essential part of driving behavior analysis is driving data collection. However, the driving data used in these studies were collected from questionnaires, site investigations, or laboratory simulations, which may differ from real driving environments and result in biased outcomes.

Recent advance in in-vehicle sensors allow a real driving data collection by continuously recording driving behavior during whole trips. The collected data include vehicle’s operating conditions such as speed, trip distance, brake signal and vehicle locations. Since the inception of this so-called driving recorder, they were first installed in commercial vehicles that are generally exposed to higher risk of crash occurrences and severe injuries as they are driven longer distances and carry greater weights than other vehicles. Hence, there have been some efforts to give feedback to drivers about driving behavior by collecting and analyzing their driving records (Misener et al., 2008, Toledo et al., 2008, Wouters and Bos, 2000). In addition to these studies, some recent studies have attempted to characterize various types of aggressive driving behaviors and to develop their criteria (Oh et al., 2009, Murphey et al., 2009, Li et al., 2014, Carboni and Bogorny, 2015). However, the criteria developed in previous studies were based on simple threshold values of recorders or often developed with lack of quantitative evidence although driving behavior should be viewed in multidimensional perspectives. This could cause false-positive or true-negative detections. To systematically identify aggressive driving behaviors, there are needs for developing more comprehensive criteria that can reflect multidimensional aspects of driving behavior.

This paper aims to develop a framework to evaluate in-vehicle driving record data and, thus, to identify driving behaviors with high potential to indicate aggressive driving. The developed framework is based on abrupt change detection and a two-level clustering approach. In the following section, the procedure to identify abrupt changes in driving and to cluster driving events is described (Section 2). This procedure is then applied to real in-vehicle driving record data that were collected from taxis in Korean metropolitan cities (Section 3). The clusters established in Section 3 are used to diagnose other drivers’ driving records and, consequently, to provide information on potential aggressive driving that may lead to risk of crashes (Section 4). The findings and implications are discussed in Section 5.

Section snippets

Framework for clustering driving behaviors

This study applied a sequential procedure for clustering driving record data to identify aggressive driving behaviors. The procedure is composed of three steps – (i) abrupt change detection; (ii) feature extraction; and (iii) two-level clustering. In-vehicle driving recorder was designed to record driving conditions continuously over time. Thus, driving record data included information from the time periods when driving characteristics were not particularly pronounced. Driving characteristics

Data description

In Korea, it was legislated in 2011 that all commercial vehicles (e.g. trucks, buses and taxis) must install a digital tachograph (DTG), which is a type of in-vehicle driving recorder, for monitoring safety (Traffic Safety Act Article 55 in Korea). The DTG device imports driving records from On-Board Diagnostic Systems (OBD-II) terminal and stores them in the Secure Digital (SD) memory card. The stored data in the memory are periodically downloaded from the memory and transmitted to the server

Driving event detection

Event time periods that have high potential for the observation of aggressive driving behaviors were identified using ACD. The change point is identified when the change point score becomes greater than a threshold value. In this study, the threshold was set at 500, which was the top 5% of all scores. In computing this score, the size of the time window was set at 15 s because the time needed to complete a single driving event is 15 s (Zhang, Zhao, & Rong, 2014). Then, the change point score

Diagnosis of driving behaviors using the established clusters

The cluster map, identified as an outcome of the process, represents the norm of driving events by taxi drivers. Thus, the map can be used as a reference to evaluate how a driver behavior deviate from the norm. In this section, we obtained driving records from four different taxi drivers and diagnosed their driving patterns based on the cluster map. Driving records of the four taxi drivers were filtered using ACD and driving events with abrupt changes were identified. Then, the events were

Conclusions

This paper has developed a framework to evaluate driving behaviors using in-vehicle driving record data. Data of this kind are becoming more available as in-vehicle sensor technologies advance and spread widely. The framework is composed of abrupt change detection, which extracts driving events with sudden changes in driving behaviors, and clustering, which categorizes driving events using an unsupervised learning approach. The case study applying the framework to real data from taxis shows

Acknowledgements

This research was supported by the National Research Foundation of Korea grant funded by the Korea government (MSIP) (NRF-2014R1A2A1A11052725).

References (29)

  • J. Vesanto

    SOM-based data visualization methods

    Intelligent Data Analysis

    (1999)
  • P.I. Wouters et al.

    Traffic accident reduction by monitoring driver behaviour with in-car data recorders

    Accident Analysis & Prevention

    (2000)
  • X. Zhang et al.

    A study of individual characteristics of driving behavior based on hidden markov model

    Sensors & Transducers

    (2014)
  • AAA Foundation for Traffic Safety

    Aggressive driving: Research update

    (2009)
  • Cited by (43)

    • Optimized structure learning of Bayesian Network for investigating causation of vehicles’ on-road crashes

      2022, Reliability Engineering and System Safety
      Citation Excerpt :

      Thus, investigating the causal relationships between the contributing factors and vehicles’ crashes is of importance towards the remediation of crashes and the enhancement of road transportation reliability. Previous studies have focused on the contributing factors from various aspects in recent years, such as vehicles’ kinetic characteristics (e.g. position, velocity, etc., extracted from trajectory) [15,55,69,72], driving behaviors (e.g. improper driving maneuver, aggressive driving, etc.) [4,44,71], surrounding environment (e.g., weather, road condition, etc.) [29,48,75,82], human factors (e.g. physiology, psychology, social background, etc.) [3,5,25,26], transportation infrastructure [8],[83], etc. Since a vehicles’ crash can be regarded as a systematic failure which usually results from the effect of multiple factors rather than only a single mistake [62], many researchers have attempted to synthesize the factors from multiple aspects when investigating a vehicles’ crash [12,28,79].

    • Benchmarking the driver acceleration impact on vehicle energy consumption and CO<inf>2</inf> emissions

      2022, Transportation Research Part D: Transport and Environment
      Citation Excerpt :

      Finally, Section 6 draws general conclusions and thoughts for future work. The characterization of the driving style has been tackled in the literature through the analysis of different experimentally acquired features (Li et al., 2017) (Van Ly et al., 2013) (Tawfeek and El-Basyouny, 2019) (Zhao et al., 2020) (Yan et al., 2019) (Lee and Jang, 2019) (Feng et al., 2017) (Eboli et al., 2016). A complete review of literature on the topic can be found in (Makridis et al., 2022) and references therein.

    • Modeling aggressive driving behavior based on graph construction

      2022, Transportation Research Part C: Emerging Technologies
      Citation Excerpt :

      On the other hand, many researchers have attempted to model aggressive driving behavior from varied methods (Hong, Dong, Zheng, & Wei, 2011). For example, Lee and Jang (2019) developed a framework to evaluate large-scale driving records and to establish clusters to identify potentially aggressive driving behaviors. Some of those research always use instantaneous, mean, and Standard deviation (S.D.) as model inputs (Songchitruksa & Tarko, 2006; Davis, 2008; Saunier & Sayed, 2008; Kluger, Smith, Park, & Dailey, 2016).

    View all citing articles on Scopus
    View full text