Multi-scale adaptive corner detection and feature matching algorithm for UUV task target image

Abstract

In order to ensure the collected images can be matched exactly when UUV operates autonomously underwater. An improved Harris-FAST algorithm is proposed in this paper. The algorithm takes full advantages of FAST algorithm and Harris algorithm, which makes the threshold value with self-adaptive and the detected corner in multi-scale and omnibearingly. It can overcome the corner information loss and position offset caused by corner detection in single-scale and optimize the non-maxima suppression. After completing the corner detection, using feature descriptor to generate eigenvector and complete the corner matching and elimination of mismatching based on Euclidean distance method. UUV simulation test system of binocular vision is built in this paper, and underwater images are captured. Through the contrast test, it can verify the validity that the proposed algorithm is used in underwater images.