Fast epipolar resampling of trinocular linear scanners images using Chandrayaan-1 TMC dataset

Abstract

In remote sensing linear scanners are widely used for acquiring stereo images as they provide data with very high resolution. A trinocular stereo view increases the accuracy of 3D modeling as it gives an additional view to compensate occlusions. However data in triplet mode by linear scanners is challenged by data processing steps which lead to a new research area for remote sensing society. In various earth and planetary missions such as ALOS PRISM, Chang'e-1, Pléiades and Chandrayaan-1 a trinocular stereo sensor was used to capture data in triplet mode. The sensor used is a linear pushbroom scanner with 3 linear array sensors looking in three directions of Aft, Nadir and Fore, which captures trinocular stereo images in along track mode. This exploration has enabled availability of large science planetary data publically but some parameters are not available as RPC parameters. Without these parameters for generation of digital elevation model (DEM), the images are required to be resampled to the epipolar geometry and then matched using area based methods. So that corresponding conjugate points can be determined, using which, parallax and finally the elevation can be estimated. In case of linear scanners, the epipolar geometry is quite challenging and it is not as straightforward as for a perspective (pin hole) frame camera. The third view gives an additional complexity to the system. These three aforestated points increase the difficulty of DEM generation for planetary missions. In this paper based on the triplet of images from linear scanner the geometrical relations are established between the three images and a method is proposed for fast epipolar resampling of triplet of images.