Modeling and Detecting Aggressiveness from Driving Signals
The development of advanced driver assistance systems (ADASs) will be a crucial element in the construction of future intelligent transportation systems with the objective of reducing the number of traffic accidents and their subsequent fatalities. Specifically, driving behaviors could be monitored online to determine the crash risk and provide warning information to the driver via their ADAS. In this paper, we focus on aggressiveness as one of the potential causes of traffic accidents. We demonstrate that aggressiveness can be detected by monitoring external driving signals such as lateral and longitudinal accelerations and speed. We model aggressiveness as a linear filter operating on these signals, thus scaling their probability distribution functions and modifying their mean value, standard deviation, and dynamic range. Next, we proceed to validate this model via an experiment, conducted under real driving conditions, involving ten different drivers, traveling a route with five different types of road sections, subject to both smooth and aggressive behaviors. The obtained results confirm the validity of the model of aggressiveness. In addition, they show the generality of this model and its applicability to specific driving signals (speed, longitudinal, and lateral accelerations), every single driver, and every road type. Finally, we build a classifier capable of detecting aggressive behavior from the driving signal. This classifier achieves a success rate up to 92%.