A GPU Based Image Matching Approach for DEM Generation using Stereo Imagery (IEEE)
Derivation of height information from stereo images is a trivial problem in the field of photogrammetry, remote sensing and computer vision. The algorithms for generation of height information from the stereo imagery in the form of Digital Elevation Model (DEM) primarily involve computationally complex image matching procedures. In this paper, we propose a Graphic Processing Units (GPU) based method that significantly improves the computational performance of the hierarchical image matching technique used for DEM generation. We have compared our results against single and multithreaded CPU based implementations. Experiments performed on Chandrayaan-1 Terrain Mapping Camera stereo image datasets using the Nvidia Tesla M2070 GPU show that the new method can perform as fast as 138 times compared to single threaded implementation and 5 times compared to a multithreaded implementation running on a multi CPU system.
Paper available at IEEE.