Research on Intelligent Extraction and Visualization Methods for Lane-Level Road Defects

Abstract. To enhance the efficiency and visualization level of road maintenance work, this paper proposes a lane-level road defect visualization method based on multi-source data fusion. Traditional visualization methods usually only display single data or overall road conditions, which are difficult to meet the needs of intuitive presentation of complex road operation situations. To address this, this paper combines multi-source data such as Beidou GPS data, road inspection images, defect detection results, lane line information, and camera calibration to construct a complete multi-source data fusion visualization framework. Firstly, by introducing the Polar R-CNN network model to efficiently extract lane line information, and using the improved YOLOv8 model for object detection of road defects; secondly, in order to obtain the morphological features of road defects, this paper proposes an image segmentation method based on anchor box cropping and improved Otsu threshold algorithm, which effectively enhances the extraction effect of crack texture details; then, inverse perspective mapping (IPM) is used to transform the inclined images into orthographic images to achieve accurate mapping of the spatial positions of objects. The experimental results show that this method performs well in lane line detection, defect shape extraction, and spatial positioning, and can accurately visualize the display of different types of defects in multiple lanes, providing an intuitive and efficient decision support tool for road maintenance departments. The visualization scheme proposed in this paper not only enhances the interpretability and interactivity of data but also provides an important reference for the development of future intelligent road inspection systems.