Investigation of the environmental effects of driver distraction and detection using object detection algorithms

Traffic accidents are one of the most important issues threatening human life in developed and developing countries. It is seen that the largest share of the fault rates in traffic accidents belongs to the drivers. When the causes of driver errors are examined, driver distraction comes to the fore. Driver distraction has a significant impact on environmental pollution as well as traffic safety. Within the scope of this thesis, the effects of driver distraction on traffic safety and environmental pollution are discussed separately. Mobile phone use, which is the most common action exhibited by distracted drivers, and its effects were evaluated at 23 intersections in Erzurum province. When drivers' mobile phone usage status is examined, it is seen that there is an annual delay of 291,37 hours, 1,472 kg CO2, 0,44 kg NOx emissions and 261,6 liters of fuel consumption. In order to prevent or partially prevent driver distraction, object detection algorithms were used within the scope of the thesis. To detect driver distraction, YOLOv7 and YOLOv8 models belonging to the YOLO architecture were trained and the detection process was carried out. The priority levels of action classes determined for model training were determined using the Analytic Hierarchy Process. A new data set was created to train YOLO models and a total of 39,996 images reflecting different driving characteristics were included in the study. According to the mAP 0.5:0.95 criterion metric, an accuracy rate of 91.17% for the YOLOv7 model and 92.03% for the YOLOv8 model was obtained, demonstrating successful performance. In addition, within the scope of Intelligent Transportation Systems, it is aimed to eliminate the negativities in terms of traffic safety and environmental pollution by considering the detection and driver warning approach based on real-time driving.