Analysis of Drivers´ Responses under Hazardous Situations in Vehicle Traffic

This study focused on the analysis of drivers´ reactions under hazardous scenarios in vehicle traffic. Driving behavior signals were utilized to detect a chain of changes in driver status and to retrieve incidents from a large real-world driving database obtained from the center for integrated acoustic information research (CIAIR). All the existing 25 potentially hazardous scenes in the database were hand-labeled and categorized. A new feature, based on joint-histograms of these behavioral signals and their dynamics was proposed and utilized to indicate anomalies in driving behavior. Brake pedal force-based method attained a true positive (TP) rate of 100% for a false positive (FP) rate of 4.5%, concerning the detection of 17 scenes where drivers slammed on the brakes. Results stressed the relevance of individuality in drivers´ reactions for this retrieval. In 11 of the 25 hand-labeled scenes, drivers reacted verbally. Scenes where high-energy words were present were adequately retrieved by the speech-based detection, which achieved a TP rate of 54% (6 scenes), for a FP rate of 6.4%. In addition, the proposed integration method, which combined brake force and speech signals, was satisfactory in boosting the detection of the most subjectively dangerous situations.