Detection of geometric change in railway curves caused by earthquakes, using high-resolution stereo satellite images

Abstract

Rapid change detection of railway curves is crucial for the damage assessment of unexpected disasters such as earthquakes. Detection of the geometric change in railway curves involves the measurement of the shape change by comparing the pre- and post-seismic railway curves. This paper proposes an approach for change detection in railway curves by the use of a pre-seismic digital topographic map and post-seismic stereo IKONOS images. The proposed approach consists of four main components, they are as follows: (1) The geometric parameters of the pre-seismic railway curves are first estimated based on the digitized points from the digital topographic map by the use of least squares adjustment. (2) The feature points on the post-seismic railway curves are then extracted from the high-resolution stereo satellite imagery, and the ground coordinates of these points are further refined by the use of rational function polynomial coefficients bundle adjustment. (3) Five change detection models—the shift model, the scale and shift model, the affine model, the second-order polynomial model, and the similarity model—are introduced to describe the geometric change between the pre- and post-seismic railway curves. Based on this, an objective function is formulated to minimize the squared distances between the transformed points from the post-seismic points refined from the high-resolution stereo imagery, based on the change detection model and the pre-seismic railway curves. An empirical experiment in change detection in railway curves as a result of the 2008 Wenchuan earthquake in Dujiangyan City, China, was conducted. The experimental results showed the feasibility of the proposed geometric change detection method, they are as follows: (1) The root-mean-square error of the geometric parameter estimation of the pre-seismic railway curve was 0.068 m, which showed that the proposed least squares approach could recover the geometric curve of the pre-event railway with a centimeter-level accuracy by the use of the digital topographic map. (2) The result of the refinement of the ground control points extracted from the pre-seismic high-resolution satellite imagery by the use of the four bias compensation models showed that the second-order polynomial model achieved the highest geo-positioning accuracy of 0.788 and 0.701 m in the x- and y-directions. This result indicated that the second-order polynomial model was the best bias compensation model for improving the accuracy of ground points extracted from the post-event stereo image pairs. (3) A comparison of the five change detection models, in terms of displacement between the pre- and post-seismic railway curves, showed that the affine model obtained the highest accuracy of 0.318 m. This result showed that the geometric change in railway curves due to earthquakes could be detected with a decimeter-level accuracy by the use of the proposed change detection approach, based on a pre-seismic digital topographic map and post-seismic IKONOS stereo images.