RSDD Computer Vision Project

Description

Potholes are a common problem in damaged roads, where people get stumbled, vehicles get damaged, and drivers lose control over their cars. The maintenance of roads is a costly necessity that developing countries' authorities often struggle to deliver in time. The present dataset was collected to develop a prioritisation system that combines deep learning models and traditional computer vision techniques to automate the analysis of road irregularities reported by citizens [1]. Although the images in the dataset come from different sources (e.g. web scrapping), we attribute the authorship of part of the images to the well-known RDD2020 dataset [2]. For those labelled images, we enhanced the original annotations by relabelling and focusing on three categories: crocodile cracks, single cracks and potholes. In addition, we iteratively filtered bad samples and improved the annotations because of the challenges to label cracks. As a consequence, the resulting object detection models have let us discriminate better road damages by severity, in contrast to just detecting potholes. Here are a few use cases for this dataset:

1. Automated Road Inspection: Researchers working on transportation and mobility issues could use the RSDD dataset to train models and embed them inside UAVs and road quality survey vehicles to make road damage inspection more efficient. Most cities in the world inspect failures in person and track them through mere paperwork. Developing an edge-ai based system could help register failures in longer roads, which in turn could help improve road damage maintenance prioritisation.

2. Road Maintenance and Repair: Municipalities and public works departments could use the RSDD dataset to train models to analyze road conditions in the server side, helping to prioritize repair activities by identifying more serious damages such as crocodile cracks and potholes.

3. Autonomous Vehicles: Developers in the autonomous vehicle industry could utilize the RSDD dataset to enhance the situational awareness capability of their vehicles. By recognizing road damage, AI systems could make more informed navigation decisions, enhancing safety and efficiency.

[1] E. Salcedo, M. Jaber, and J. Requena Carrión, “A novel road maintenance prioritisation system based on computer vision and crowdsourced reporting,” Journal of Sensor and Actuator Networks, vol. 11, no. 1, p. 15, 2022. doi:10.3390/jsan11010015

[2] D. Arya, H. Maeda, S. K. Ghosh, D. Toshniwal, and Y. Sekimoto, “RDD2020: An annotated image dataset for Automatic Road Damage Detection Using Deep Learning,” Data in Brief, vol. 36, p. 107133, 2021. doi:10.1016/j.dib.2021.107133

Citation

If you use this dataset in your research, please cite the following work:

@article{Salcedo_Jaber_Requena Carrión_2022, title={A novel road maintenance prioritisation system based on computer vision and crowdsourced reporting}, volume={11}, DOI={10.3390/jsan11010015}, number={1}, journal={Journal of Sensor and Actuator Networks}, author={Salcedo, Edwin and Jaber, Mona and Requena Carrión, Jesús}, year={2022}, pages={15}}

You are also encourage to explore the original dataset provided by D. Arya et al.