ROAD SAFETY MEASURES BY POTHOLES AND TRAFFIC SIGNS DETECTION USING DEEP LEARNING

Abstract

The modern transportation industry depends on road safety together with maintenance practices because they play vital roles because autonomous vehicles become more prevalent. Driving safety for vehicles alongside their occupants remains threatened by potholes while traffic signs preserve operational security within the highways. The implemented solution implements YOLOv8 as its main technology because it proves effective for advanced object detection requirements. Real-time operation of this system uses cameras as inputs to precisely detect potholes and locate their positions and also recognizes multiple traffic signs.
The research investigates deep learning through YOLOv8 model implementation to accomplish pothole detection with real-time capabilities alongside traffic signs and traffic light recognition which resolves road performance and safety issues. The custom dataset obtained its mean Average Precision (mAP@0.5) of 0.991 through Roboflow augmentation while each class precision reached between 0.985 and 0.994 in diverse road condition training. The system unites CIoU loss for bounding box regression with objectness scoring because its deployment is designed within a Streamlit-based application. The study demonstrates that the model outperforms conventional and YOLO-based previous versions through extensive detection capabilities in different lighting situations and weather conditions. Scalable road monitoring automation becomes feasible through this system even though it does not perform optimally in severe weather conditions and detecting tiny objects.