Volume IV-2/W5
ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., IV-2/W5, 479-484, 2019
https://doi.org/10.5194/isprs-annals-IV-2-W5-479-2019
© Author(s) 2019. This work is distributed under
the Creative Commons Attribution 4.0 License.
ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., IV-2/W5, 479-484, 2019
https://doi.org/10.5194/isprs-annals-IV-2-W5-479-2019
© Author(s) 2019. This work is distributed under
the Creative Commons Attribution 4.0 License.

  29 May 2019

29 May 2019

MOUNTAINOUS REMOTE SENSING IMAGES REGISTRATION BASED ON IMPROVED OPTICAL FLOW ESTIMATION

R. Feng1, X. Li2, and H. Shen1,3,4 R. Feng et al.
  • 1School of Resource and Environmental Sciences, Wuhan University, P. R. China
  • 2School of Remote Sensing and Information Engineering, Wuhan University, P. R. China
  • 3Collaborative Innovation Center of Geospatial Technology, Wuhan University, P. R. China
  • 4Key Laboratory of Geographic Information System, Ministry of Education, Wuhan University, P. R. China

Keywords: Displacement Modification, Laplacian of Gaussian, Mountainous Remote Sensing Image, Optical Flow, Registration

Abstract. Mountainous remote sensing images registration is more complicated than in other areas as geometric distortion caused by topographic relief, which could not be precisely achieved via constructing local mapping functions in the feature-based framework. Optical flow algorithm estimating motion of consecutive frames in computer vision pixel by pixel is introduced for mountainous remote sensing images registration. However, it is sensitive to land cover changes that are inevitable for remote sensing image, resulting in incorrect displacement. To address this problem, we proposed an improved optical flow estimation concentrated on post-processing, namely displacement modification. First of all, the Laplacian of Gaussian (LoG) algorithm is employed to detect the abnormal value in color map of displacement. Then, the abnormal displacement is recalculated in the interpolation surface constructed by the rest accurate displacements. Following the successful coordinate transformation and resampling, the registration outcome is generated. Experiments demonstrated that the proposed method is insensitive in changeable region of mountainous remote sensing image, generating precise registration, outperforming the other local transformation model estimation methods in both visual judgment and quantitative evaluation.