Fatigue Driving Detection of Bus Drivers under Idling Conditions Using An Improved YOLOv5 Algorithm

In order to detect fatigue of bus drivers under idling conditions, an improved YOLOv5 target detection algorithm combining with a counting logic method on complete time series is proposed. Driving simulation experiments were conducted to capture driver's facial region under the condition of wearing masks and without masks. BiFPN network structure is used to retain multi-scale feature information, which makes the prediction network more sensitive to targets of different sizes. Then, number of blink and yawn were counted according to the logical method from the perspective of complete time series. Results show that the method can be effective for both wearing masks and without masks situations, proved by a counting accuracy of blink above 95% and up to one missing counting of yawn with single frame detection time at 0.01 s. This research can be helpful for bus driver fatigue detection in connected vehicles.