Guillaume Saint Pierre

Using statistical models to characterize eco-driving style with an aggregated indicator

This paper presents the construction of an aggregated indicator of a fuel-efficient driving style, in order to construct an efficient Ecological Driving Assistance System (EDAS). Such an eco-index can be used to detect eco-driving behaviour, but also to give to the driver useful advices to help him improving his driving efficiency without deteriorating safety. The logistic regression is used to model our experimental dataset of twenty subjects driving twice the same route: normally or following the golden rules of eco-driving. Depending on some driving indicators, the estimated probability of being an eco-driver is used as an eco-index to characterize that driving pattern. This work show how such a simple aggregated indicator, related to driving dynamics rather than fuel consumption, can be useful for driver monitoring and information. Two models, from the simplest to the most complicated, are compared, and their performances analysed.

Explaining Senior Drivers’ Road Near Misses Using Both Self-reported and Automatic Collected Data

This paper reports on part of the French MEDOC Project that was supported by Foundation MAIF, Paris, France [grant number RP4-F14140]. The main goal of this project was to understand near miss situations using both objective and subjective methods for data collection: the vehicle dynamics and registered driver’s actions; and the self-reported conditions that led to the occurrence and self-reported road users’ actions to avoid an accident. The collected data allowed for comparing reported ones. The project involved a total of 154 drivers. However, this paper just addresses the group of 27 older drivers. Although older drivers are generally considered safe and cautious drivers, age-related perceptual and cognitive declines might have an impact on self-perception of their own abilities and behavior leading some of them to be under or over-estimators, which will influence their perception of any sudden event. Thus, this study is important to compare their self-reports with the collected data from sensors.

Estimation of space-speed profiles: A functional approach using smoothing splines

Individual space-speed profiles are very informative to study drivers behavior and their road usage. Due to uncertainties in the measurements of position and speed from sensors, there is need of use a smoothing procedure in order to estimate these space-speed profiles. The purpose of this study is to propose a smoothing method based on a functional approach. This approach take inspiration from Functional Data Analysis, a statistical domain that has developed recently. So, we define and study some properties of the functional space of space speed profiles. Then, due to the correspondence between speed and position (and implicitly time), we prove that a change of study area (distance vs time) is more appropriate and leads to a smoothing problem under monotonicity constraint. An estimation method of monotone smoothing spline is used and tested with simulated data.