Extraction of digital elevation models from satellite stereo images through stereo matching based on epipolarity and scene geometry Hae-Yeoun Lee a, * , Taejung Kim b , Wonkyu Park c , Heung Kyu Lee a a Department of EECS, Korea Advanced Institute of Science and Technology, 373-1, Kusung-dong, Yusung-gu, Taejon 305-701, South Korea b Department of Geoinformatic Engineering, Inha University, 253 Yonghyun-dong, Nam-gu, Incheon, 402-751, South Korea c SaTReCi, 461-26, Jeonmin-dong, Yusung-gu, Taejon 305-811, South Korea Received 9 January 2002; received in revised form 7 May 2003; accepted 9 May 2003 Abstract This paper addresses the problem of generating digital elevation models from satellite images taken by linear pushbroom cameras. Since there exist unique geometric properties for linear pushbroom images, we argue that the conventional DEM generation schemes developed for perspective images are not suitable for satellite images. Using the geometric properties of linear pushbroom images, we design a new matching strategy optimized for linear pushbroom image in three aspects: conjugate search method, correlation patch design and match sequence determination. We will discuss in what aspect conventional approaches and our new approach differ and show how performance has improved by hiring proper techniques. A series of experiments using SPOT panchromatic stereo pairs showed that our approach outperformed conventional approaches in terms of accuracy and processing time. q 2003 Elsevier B.V. All rights reserved. Keywords: Stereo matching; Digital elevation model; Epipolar geometry 1. Introduction A Digital Elevation Model (DEM) is digital data in which each point represents latitude, longitude and height (or easting, northing and height). Extraction of accurate DEMs is important for map generation, three-dimensional GIS, cell planning, environmental monitoring and so on. Among many possible sources that can be used for DEM generation, satellite images have potentially several advan- tages [1]. A single scene normally covers large geographic extents than aerial photos. It is easier to acquire images periodically. Satellite images are naturally digital data, which makes the automation more feasible. DEM generation requires many processing steps such as camera modeling, stereo matching, editing and interp- olation. All these steps contribute to the quality of DEMs, among which stereo matching is crucial to the accuracy and completeness of a DEM. Stereo matching is a process of finding conjugate points in a stereo image pair. There have been many publications regarding stereo matching techniques for various applications [2]. For stereo matching satellite images taken by linear pushbroom cameras, the use of pyramidal matching [1,3–5], approximation of linear pushbroom images as perspective images [6], the use of least squares correlation matching [7] have been reported among others. However, automated DEM generation from satellite images is still being considered difficult, if not impossible. It normally takes several hours of computation to generate a DEM, besides the time required for operators’ measure- ments. Resulting DEMs often, unfortunately, have accuracy and completeness problems and longer time is required to manually correct errors. We suspect these problems are due to lack of proper understanding of satellite image geometry. Most of existing technologies for DEM generation are originally developed for aerial photos (taken by perspective cameras) and modified for satellite images (usually taken by linear pushbroom cameras) with or without acknowledging the difference between them. In this paper, we introduce a new DEM generation technique for satellite images taken by linear pushbroom cameras. Since there exist unique geometric properties for linear pushbroom images, we argue that the conventional DEM generation schemes developed for perspective images 0262-8856/03/$ - see front matter q 2003 Elsevier B.V. All rights reserved. doi:10.1016/S0262-8856(03)00092-1 Image and Vision Computing 21 (2003) 789–796 www.elsevier.com/locate/imavis * Corresponding author. Tel.: þ 82-42-365-7532; fax: þ82-42-365-7549. E-mail addresses: hytoiy@satreci.co.kr (H.-Y. Lee), tjkim@satrec. kaist.ac.kr (T. Kim).