No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Reconstructing 3D land surface from a sequence of aerial images
Abstract
This paper proposes a method for reconstructing a 3D surface landscape from an aerial image sequence captured by a single noncalibrated camera. Reconstructing a 3D surface landscape is more difficult than constructing a landscape of buildings or objects in a room because of the lack of available information about camera parameters, the need for mosaicking of 3D surface elements, and the introduction of nonrigid objects. Therefore, conventional methods are not directly applicable. In order to solve these problems, we apply socalled 2-Dimensional Continuous Dynamic Programming (2DCDP) to obtain full pixel trajectories between successive image frames in a sequence of aerial images. Then we apply Tomasi-Kanade Factorization to the full pixel trajectories to reconstruct the 3D surface. We also develop a mosaicking technique for connecting all of the partially reconstructed surfaces. The experimental results show that our proposed method is very promising for reconstructing 3D surfaces, including a forest, a mountain, a lake and several houses. We conduct experiments to compare our method against a SIFT-based method using two sets of data, namely, artificial and real image sequence data.
In this paper we propose extensions to the match propa- gation algorithm which is a technique for computing quasi- dense point correspondences between two views. The ex- tensions make the match propagation applicable for wide baseline matching, i.e., for cases where the camera pose can vary a lot between the views. Our r st extension is to use a local afne model for the geometric transformation between the images. The estimate of the local transformation is ob- tained from afne covariant interest regions which are used as seed matches. The second extension is to use the second order intensity moments to adapt the current estimate of the local afne transformation during the propagation. This al- lows a single seed match to propagate into regions where the local transformation between the views differs from the initial one. The experiments with real data show that the proposed techniques improve both the quality and coverage of the quasi-dense disparity map.
In this paper, we show how relative D reconstruction from point correspondences of multiple uncalibrated images can be achieved through reference points. The original contributions with respect to related works in the field are mainly a direct global method for relative 3D reconstruction, and a geometrical method to select a correct set of reference points among all image points. Experimental results from both simulated and real image sequences are presented, and robustness of the method and reconstruction precision of the results are discussed.