Quantification method of concrete pavement diseases based on rotating box annotation

As a vital component of road transportation, the condition of concrete pavement has a significant impact on the quality of road access, with direct implications for traffic safety and ride comfort. Remarkable efforts have been made to develop pavement disease detection. However, the horizontal box algorithm, which is currently used in intelligent quantification of pavement diseases, has inherent limitations. It is unable to accurately quantify the dimensions of the diseases and is prone to repeatedly detecting the same disease in successive frames of images. To address these issues, an unmanned aerial vehicle-based concrete pavement disease dataset with rotated-box annotations has been established. Subsequently, the impact of the dataset split and network architecture on the accuracy of model detection was validated. Comparative experiments were conducted to assess the quantification of disease dimensions by model detection and manual detection. The results demonstrate that the relative error of disease size quantification is less than 5.42%. Furthermore, impact of the target detection model on the tracking outcome is evaluated using the enhanced model. The findings demonstrate that the precision of the disease enumeration reaches 90%, representing a 44% improvement over the baseline model. The model exhibits robust performance, providing a dependable foundation for subsequent quantification of concrete pavement diseases.