Automated On-Vehicle Road Defect Data Collection and Detection.

This paper proposes a pipeline for the automated on-vehicle data collection, filtering, and classification of road surface defects. The proposed pipeline provides a flexible framework that allows for the integration of a variety of systems. The pipelines flexibly allow for various sensors such as camera, 3D camera and lidar; computational resources such as on-vehicle edge computing or cloud computing; data transfer such as 5G or on-site upload; and data storage. The pipeline was tested using an edge computer on board a contracted road sweeping vehicle with an image taken every 10 s with image processing and evaluation occurring between. Post installation, the pipeline required no input from the driver of the sweeper vehicle besides turning on the road sweeper. The data was transferred via WiFi as the road sweeper was pulling up at the end of its shift. During operation around 21k road, defects were identified with over 90% of these images containing road defects.