Robust Estimation of 3-D Line Segments from Satellite Images for Model Building and Change Detection

This paper addresses how to automatically match and reconstruct line segments from multiple satellite images and detect changes in man-made structures by using reconstructed 3-D line segments. We propose a new approach to line reconstruction and use wireframe models to better estimate the actual 3-D geometry by using the incidence relations between reconstructed line segments. Our change detection method is widely applicable as man-made structures are the main focus of most change detection applications. Detection of structural changes in images of the same scene has become a key issue due to a large number of applications in computer vision. Important applications of change detection are video surveillance, urban planning, agricultural analysis and military intelligence (Huertas & Nevatia 1998; Radke et al. 2005). The core problem of change detection is to identify sets of features (pixels, regions, lines, etc.) that are significantly different than previous images of the same scene. Some of the early work on change detection focuses on intensity values of pixels and regions. The main drawback of these methods is their high likelihood to create false alarms in cases where pixel values are affected by viewpoint, illumination, seasonal and atmospheric changes. This is the reason why intensity (pixel and region) based change detection algorithms such as simple differencing methods (Bruzzone & Prieto 2002) and background modeling methods (Stauffer & Grimson 1999) fail in such scenarios. The notion of “important change” varies according to the focus of the application. Generally the main focus is on man-made structures, especially buildings. Considering that most man-made structures consist of 3-D line segments, a change detection system based on modeling the objects in a scene using 3-D line segments is promising because it may provide more accurate and reliable results. In this study we have been interested in developing a line segment based change detection method. This rarely studied approach requires attention to develop more efficient and accurate algorithms that can be generalized to complex problems. In order to make our change detection approach more efficient and accurate, we propose novel techniques for line