Identifying critical incidents in naturalistic driving data: Experiences from a promoting real life observation for gaining understanding of road user behaviour in Europe small-scale field trial

Abstract

The methodology of naturalistic driving observation aspires to observe the driver and his environment while driving in natural driving settings. It is of great importance in research on road safety as this method of observing road users eliminates the disadvantages of traditional methods like simulator studies or interviews. However, it produces vast such amounts of data and challenges data reduction and data analysis. Therefore automatic methods for filtering critical incidents based on thresholds for numerical data are often applied to select the data to be analysed. This study reports a small-scale field trial in Valencia, Spain, which was conducted within the promoting real life observation for gaining understanding of road user behaviour in Europe project. The analysis of the numerical data using thresholds resulted in a great number of false alarms and did not identify safety-critical sequences. In contrast, video analysis revealed a number of critical events that had not been previously detected using the numerical parameters. The study conveyed the importance of continuous video recording in these kinds of studies and showed that the methodology of data reduction for naturalistic driving studies requires further development in order to be able to capture all the relevant incidents automatically. © The Institution of Engineering and Technology 2013.