Quality and accuracy for your road asset management
We are continually enhancing RoadAI's capabilities through ongoing testing, training and calibration.
Trusted by hundreds of government public works and road maintenance organizations globally
Continuous testing and improvement
We are continually enhancing and refining RoadAI's image recognition and computer vision capabilities through ongoing training and calibration.
Our training dataset comprises tens of thousands of images and undergoes weekly review and supplementation to rectify any potential errors in object and defect recognition, classification, and anonymization. Similar procedures are applied to our validation dataset, which consists of thousands of images. Our models are updated with new capabilities (e.g. new defect and asset labels) and training data multiple times per year. Before integrating them into RoadAI, our computer vision models undergo rigorous testing using various methods.
Over 2 650 000 miles of roads analyzed in 2023
Unparalelled repeatability
How does RoadAI measure up in accuracy and repeatability? How is the data captured and used?
Recent repeatability study reveals that with the chosen repeatability metric, RoadAI was significantly more repeatable than humans (103%). Additionally, RoadAI data from one app compared with data from the other showed half as much variation as data from one human surveyor compared with the other. We were really interested to discover as well that RoadAI datasets matched to H2 better than H1 matched to H2, further highlighting the variations between humans. Discover more in this study.
RoadAI client stories
