This technology is a device where a processor automatically determines road condition information when a road image is input into an AI algorithm stored in memory. The algorithm is trained using virtual road data generated based on actual road measurement data.
With the surge in traffic volume and vehicles, road degradation such as cracks and potholes has become frequent, impacting driver safety and vehicle fuel efficiency, making road condition assessment critical. However, professional equipment for quantitative road evaluation is expensive and not widely used, and even with AI, there has been a limitation in securing enough labeled road images for training.
This technology selects a portion of actual road flatness measurements to calculate sigma values, creates noise data, and passes it through a predetermined transfer function to obtain 2D virtual road information, which is then converted into 3D virtual road data. It expands training data by generating multiple sets of the same road surface with varying light incidence angles. Applicable to routine pavement inspections by road authorities, vehicle-mounted camera-based surface monitoring, and prioritizing pothole repairs, it enables the mass production of training images with automatically labeled flatness levels, allowing for the construction of real-time surface evaluation models without expensive equipment or manual labeling.
This invention was developed with support from the Ministry of Land, Infrastructure and Transport's project for developing non-contact safety diagnosis technology based on unmanned survey vehicles for real-time dynamic bridge behavior prediction.
N/A
