Automatic Classification of Driving Conditions for the Detection of Driver-Induced Steering Oscillation

We proposed an approach of automatically identifying two different driving conditions ? driving on a straight, and cornering, respectively, by a cognitively distracted human driver in TORCS environment. The cognitive distraction of the driver results in driver-induced steering oscillations. In order to detect these steering oscillations ? e.g., by analysing the magnitude of power spectra of lateral acceleration ? it is crucial to automatically distinguish the driving condition so that variable threshold ? corresponding to these different driving conditions ? could be applied. Our experimental results indicate that a specific low pass filter implemented as a sliding window averaging on the discrete sampled values of lateral acceleration identifies the driving conditions adequately.