Abstract
The objective of this study was to monitor driver behaviour based on basic vehicle sensors and output from an aftermarket advanced driver assistance system, in order to discern the role of the driver in safety-critical events (SCEs). The goal was to quantify whether the driver was distracted based on these sensors, to discriminate between events in which driver error played a role and those that were caused by external factors. A passenger vehicle was instrumented with a commercially available mono-camera system with ADAS functionality and an in-vehicle software platform. This platform extracts information from the vehicle CAN and the mono-camera system and processes and stores it with modular algorithms. The driver safety performance indicator (DSPI) estimates whether the driver is distracted, based on his/her lane-keeping or car-following performance. First, the distraction algorithms were validated in a driving simulator experiment with 18 drivers which included distraction tasks and SCEs (sudden braking of a preceding vehicle, a construction section and an off-ramp). The DSPI correctly predicted distraction in 79% of the car-following detections and in 85% of the lane-keeping detections on highway sections. In an off-ramp condition, only 72% of the predictions were correct, making the algorithm less applicable to strongly curved roads. Secondly, the instrumented vehicle demonstrated the ability to log driver-based SCEs based on approximately 10 hours of driving. A total of 84 SCEs were detected, predominantly on highways and rural roads. This resulted in 9.0 SCEs per hour, of which the majority (80%) were due to distraction detections. Around 30% of the lane-keeping distraction detections were false positives due to irregular lane markings, while around 7% of the car-following distraction detections were false positives due to incorrect object detection. The results of these newly developed algorithms may be applicable to reduce the number of false positive warnings of lane keeping assist (LKA) and forward collision warning (FCW) systems. Furthermore, they allow for logging SCEs in which the driver was distracted. The developed in-vehicle software architecture allows for low-cost upscaling to enable large fleets to be instrumented to detect safety-critical driver behaviour.
KEYWORDS – driver behaviour, distraction, safety-critical events, driving simulator, event logging