This document proposes using vehicle-mounted cameras on public vehicles like garbage trucks to detect blurred road markings as a way to efficiently monitor road infrastructure deterioration. An object detection approach is used that predicts the location, size, class and confidence of different types of road markings using a YOLO network with VGG16 features trained on a new dataset from drive recorder videos. Evaluation shows the approach detects white lines with 51.7% mAP but struggles with color lines and marks. Future work to improve detection accuracy includes expanding the dataset and exploring semi-supervised learning.