Using YOLOv5 Object Detection and a Raspberry Pi to Improve the Safety of Drivers

Kryzsko Commons Ballroom, Winona, Minnesota

This research proposes using YOLOv5 object detection in vehicles to detect possible obstructions to the driver using a Raspberry Pi. Because the detections are made in real time, the YOLOv5 nano model is used, which is a smaller model that sacrifices some accuracy for higher speed. The obstructions accounted for are vehicles, emergency vehicles, pedestrians, bicyclists, animals, motorcycles, and traffic lights. A dataset of images mainly taken from dashcam footage were used in this study, as it closely simulates the environment the model will be used in. Overall, we found that this did improve the safety of drivers in terms of the evaluation metric using mean average precision.