Parallel realization of SIFT feature abstraction based on GPU using Tegra K1

The fast and steady feature extraction and key point matching are key factors to SAR (synthetic aperture radar) image automatic registration algorithm. The real-time processing of the algorithm in SAR is one of the challenges that need to be solved urgently. This paper aims at improving the efficiency of the description of key point in SIFT algorithm. We implemented the rapid calculation of the feature extraction section in SIFT (scale invariant feature transform) based on embedded NVIDIA GPU-Tegra K1 by fully taking the advantages of multi-cores of GPU in parallel computing, floating point calculation, memory management, etc. And the speed ratio of the accelerated calculation of the SIFT feature reaches 5.5.