Detecting Powered-Two-Wheeler incidents from high resolution naturalistic data Eleni I. Vlahogianni ⇑ , George Yannis, John C. Golias National Technical University of Athens, 5 Iroon Polytechniou Str, Zografou Campus, 15773 Athens, Greece article info Article history: Received 29 October 2012 Received in revised form 9 October 2013 Accepted 7 November 2013 Keywords: Powered Two Wheelers Driving behavior Naturalistic experiments Microscopic incident analysis Outlier detection abstract During risky conditions, Powered-Two-Wheeler (PTW) drivers often alter their behavior from a regular driving pattern to an irregular chain of driving actions by braking, changing the throttle pressure, maneuvering and so on, or combinations of the above. However, both the actual and perceived thresholds of regular and irregular driving behavior differ among PTW drivers. A simple and flexible methodology is proposed in order to define PTW driving profiles by distinguishing between regular and irregular PTW driving behaviors using high resolution naturalistic data. ‘‘Irregularities’’ in driving behavior are consistently expressed as outlying values in the dataset of driving parameters. The detected irregularities are those that diverge from the centroid of the jointly considered driving parameters. These irregularities may be considered to define critical driving situations (incidents) that are fur- ther associated to typical driving events. Results indicate that the joint consideration of variables which are directly connected to the mechanical characteristics of PTW, such as front and rear brake activation, wheel speed, throttle and steering, are adequate to distin- guish the regular from irregular PTW driving behavior. Ó 2013 Elsevier Ltd. All rights reserved. 1. Introduction Understanding Powered-Two-Wheelers (PTWs) driving behavior and interactions with the rest of the traffic is critical to the design of efficient accident countermeasures and is, hence, essential (Yannis, Golias, & Papadimitriou, 2005). An efficient manner to understand PTW driving behavior – given the improvements and innovations of modern technology – is through constant monitoring and analysis of PTW driver’s actions during driving. Until recently, PTW accident risk has been largely studied through macroscopic and in-depth data analyses (Dupont et al., 2009; Thomas et al., 2005; Yannis, Papadimitriou, Dupont, & Martensen, 2010), as well as through behavior analyses such as questionnaire based surveys, guided discussions, video-based methods or simulators (Engstrom, Johansson, & Ostlund, 2005; Haque, Chin, & Huang, 2010; Huang & Abdel-Aty, 2010; Savolainen & Mannering, 2007). These analyses are inherently destined to qualitatively assess on the factors that in- crease accident risk mainly from a social point of view, without being able to extract accurate and detailed information on the manner PTW drivers behave on the road and especially before, during and after critical driving situations (incidents). A new and efficient way to understand PTW driving behavior is by creating a least intrusive–restrictive (naturalistic) – environment to monitor and record drivers’ actions on the road by employing advanced sensor technologies. Recently, a number of such attempts have been conducted in Europe, the US and Australia to understand driver’s behavior. Some pre- vailing efforts are described in large scale projects such as the 100-Car study (NHTSA, 2006), SHRP 2 Naturalistic Driving Study (SHRP2 2011) and Euro-FOT (Csepinszky & Benmimoun, 2010). Several findings on commercial vehicle driver’s 1369-8478/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.trf.2013.11.002 ⇑ Corresponding author. Tel.: +30 2107721369; fax: +30 2107721454. E-mail addresses: elenivl@central.ntua.gr (E.I. Vlahogianni), geyannis@central.ntua.gr (G. Yannis), igolias@central.ntua.gr (J.C. Golias). Transportation Research Part F 22 (2014) 86–95 Contents lists available at ScienceDirect Transportation Research Part F journal homepage: www.elsevier.com/locate/trf