Efficiency Improvement of Blurring Construction Panoramic Inspection Based on Improved NAFNet

Abstract

Daily inspections are essential to ensure safety, progress, and quality on construction sites. Traditional manual inspections, which are labor-intensive, inefficient, and inconsistent in monitoring quality, are increasingly being supplemented by panoramic inspections. Panoramic inspections offer a broader scope and faster execution with fewer technical demands. However, challenges like motion blur can significantly impair the effectiveness of subsequent visual tasks. This study proposed a comprehensive method for correcting motion-blurred panoramic inspection images at construction sites, aiming to enhance the efficiency of vision-based tasks such as object detection and localization. The method comprised three stages: first, randomly generating motion blur to create clear-blurred image pairs; second, training an improved NAFNet deblurring model to reduce motion blur and enhance image clarity; and finally, applying this model to a construction project in China, resulting in improved accuracy in object detection and Simultaneous Localization and Mapping (SLAM) feature point matching. This approach not only advances the efficiency of indoor inspections for subsequent object detection and localization tasks but also contributes to the development of automated construction inspection techniques.".