An optical flow-based terrain extraction framework of VHR optical satellite stereo images

Abstract

Stereo vision has attracted much interest in computer vision and remote sensing. Large-scale terrain extraction from optical satellites has achieved various applications due to the continuous development of optical satellite technology in the last forty years. This study analyzes the literature sources published in the last decade on stereo vision, optical flow, and remote sensing and finds that applications of the optical flow method in very high resolution (VHR) optical satellite stereo matching are inadequate. Therefore, it proposes a comprehensive optical flow-based terrain extraction framework from optical satellite stereo images. Experiments are conducted using stereo images from four different optical satellites, and the results are compared to the pertinent algorithms of ENVI, ERDAS, and ASP software. The qualitative and quantitative analysis demonstrates that the proposed method has significantly superior matching effect and robustness with a higher accuracy level compared with other approaches, especially for mountainous areas. In addition, the root mean square errors (RMSEs) of elevation on four datasets are 1.228 m, 2.159 m, 3.150 m, and 5.763 m, indicating the feasibility of the proposed method for terrain extraction from optical satellite stereo images.

Keywords

• Dense optical flow
• Stereo vision
• Optical satellite stereo images
• Epipolar images