Linear target change detection from a single image based on three-dimensional real scene

Abstract

Change detection is a critical component in the field of remote sensing, with significant implications for resource management and land monitoring. Currently, most conventional methods for remote sensing change detection often rely on qualitative monitoring, which usually requires data collection from the entire scene over multiple time periods. In this paper, we propose a method that can be computationally intensive and lacks reusability, especially when dealing with large datasets. We use a novel methodology that leverages the texture features and geometric structure information derived from three-dimensional (3D) real scenes. By establishing a two-dimensional (2D)–3D geometric relationship between a single observational image and the corresponding 3D scene, we can obtain more accurate positional information for the image. This relationship allows us to transfer the depth information from the 3D model to the observational image, thereby facilitating precise geometric change measurements for specific planar targets. Experimental results indicate that our approach enables millimetre-level change detection of minuscule targets based on a single image. Compared with conventional methods, our technique offers enhanced efficiency and reusability, making it a valuable tool for the fine-grained change detection of small targets based on 3D real scene.