Unsigned Road Incidents

This repository contains the dataset for the paper entitled “Detecting Unsigned Physical Road Incidents from Driver-view Images”, released in IEEE Transactions on Intelligent Vehicles. This image dataset contains over 12.500 images of unsigned (without warning signs) road incidents spread out over 8 classes (fire, animal on road, road collapse, snow, rain, treefall, landslides, and vehicle crashes). It consists of a zip file which contains images from the Geograph project and their attribution, as well as a SQLite database which contains URLs to images harvested from Google, Bing, and Flickr. URLs marked as '0' have since disappeared from the web. Unfortunately, due to copyright concerns we cannot make these images available. However, reproduction can be facilitated by directly reaching out, in which case we will make a full copy of the dataset available. Please contact the authors directly if you are interested in using the full dataset. We make this dataset available for non-commercial purposes only. The preferred citation for this dataset is as follows: Levering, A., Tomko, M., Tuia, D. and Khoshelham, K., 2020. Detecting unsigned physical road incidents from driver-view images. IEEE Transactions on Intelligent Vehicles, 6(1), pp.24-33.