Automated Detection of Pavement Defects Using Computer Vision

Abstract

Road managers are obliged by law to regularly monitor the condition of road pavements as part of road inspections. Visual inspections provide basic information on the condition of the road and regular assessments are the basis for planning maintenance and repairs. These inspections are usually carried out from a dedicated car and recorded manually by an operator or done by special sophisticated and very costly devices with cameras and various sensors. Inspections are done in defined periods based on road class and type of inspection. This paper presents a pilot test of a new method of monitoring pavement defects based on visual inspection by an autonomous vehicle-mounted system with automatic real-time evaluation performed by this device. The device processes the video recordings and uses deep neural networks for the detection and classification of pavement defects. The resulting metadata and location are immediately sent from this device to the cloud infrastructure. All the data are GDPR safe by design, no images or videos leave the device. The detection is not meant to be as precise as detection made by special diagnostic cars, it is used to do instant community-based monitoring of significant damages on the road network and hence serves as a pre-selection tool to provide road administrators valuable data on where detailed inspection or diagnostics is needed. In addition to the pavement condition, other parameters related to road objects and equipment can also be evaluated.