Research On Image Matching Navigation Algorithm Based On GPU Parallel Computing And OpenCV

Because of its own unique high accuracy, anti-jamming and autonomous navigation and other powerful capabilities, inertial / image matching aided navigation system has become a hot research direction. It is the target of the image matching aided navigation technology, so the accuracy and real-time performance is the key factor of the image matching algorithm. In this paper, we have carried out a very deep research on them.First of all, taking into account the ORB algorithm in other aspects are very outstanding, but only lacking of good support for scale invariance. This paper has proposed an improved ORB algorithm, which has introduced the concept of scale space and image pyramid into ORB algorithm. The experimental results show that the proposed algorithm has all the advantages of ORB algorithm, and also can be compared with the performance of SIFT in terms of translation invariance, rotation invariance, scale invariance, affine transformation and so on.Secondly, in order to further improve the matching accuracy of ORB feature points, this paper proposes a pre filtering strategy which is based on Hash bidirectional pre filtering, so as to optimize the RANSAC algorithm to achieve a very high matching accuracy and computational efficiency. More importantly, this paper introduces the GPU and CUDA parallel computing into the image matching aided navigation system, so that the traditional image matching algorithm can achieve over 20 times faster than before.Finally, this paper summarizes and implements a set of image matching navigation algorithm based on OpenCV, CUDA parallel computing and improved ORB algorithm. And the simulation experiment is carried out. The experimental results show that the proposed algorithm has a high accuracy and a robustness, and can control image matching time under milliseconds, so as to meet the requirements of accuracy, real-time and robustness in image matching navigation system.