SURF-Based Image Matching Method for Landing on Small Celestial Bodies

Abstract

In deep space exploration missions, one of the main methods used to achieve accurate landing of small celestial body by detectors is the terrain-matching navigation method based on optical images. Image matching technology is the key technology of this method. This paper proposed a small celestial image matching method based on SURF (Speeded Up Robust Features) to improve the adhesion accuracy of small celestial bodies. Firstly, we used the SURF feature detector to perform feature point detection on the surface image of the target celestial body then, the feature points are matched by the fast nearest neighbor search method. And mismatches are eliminated with NNDR and RANSAC. Finally, under the influence of image rotation, Gaussian noise, etc., the matching results of the algorithm were simulated and analyzed. The simulation results demonstrate that the proposed method has good robustness in complex environment of small celestial and high matching accuracy. It can provide effective landmark information and attitude information for subsequent visual navigation. Introduction The scientific significance of small celestial exploration is very significant, whether it is the study of the formation and evolution of the solar system, the origin and evolution of life, or the defense against foreign celestial bodies in the future. Exploration of small celestial bodies are gradually developed to the current detection methods of landing, etc.[1] During the attachment process, the detector relies on identifying the keypoint in the image captured by the optical camera for autonomous optical navigation. Among them, the detection and matching of image features play a crucial role[2,3]. The extracted keypoint must have high uniqueness, for example, the edge of the crater on the surface of the small celestial body, the edge of the groove, etc. What’s more, the keypoint need to have scale invariance and rotation invariance. And it should have good adaptability to light changes.[4] The image feature matching process is mainly divided into three steps: keypoint detection, feature descriptor generation and feature matching. At present, the keypoint detection algorithms mainly include SIFT, SURF, ORB, KAZE, etc. SIFT (Scale-invariant feature transform) has a good image matching effect with different image scales, different brightness and different rotation, and the application range is very wide[5,6]. SURF(Speeded Up Robust Features) has the invariance of translation, scaling and rotation. At the same time, it is also relatively robust to illumination, affine and projection variability[7]. As the requirements for keypoint matching speed increase, Edward Rosten et al.[8] proposed the FAST algorithm in 2006. Then the ORB[9] and the BRISK[10] algorithm were generated on the basis of FAST. In terms of speed, ORB is the fastest among them, followed by BRISK, and the slowest is SIFT. In terms of feature extraction, especially in the context of small celestial attachment, ORB is not suitable for working in this context because it does not have scale invariance. BRISK is fast, but not as robust as SIFT and SURF, and SURF is faster than SIFT. Therefore, this paper chooses SURF algorithm as feature detector[11,12,13]. The surface atmosphere and geological environment of small celestial bodies are complex, especially the difference of illumination. At the same time, the detector landing needs to complete image matching in high speed. In this paper, SURF is used for image keypoint detection, and then International Conference on Modeling, Analysis, Simulation Technologies and Applications (MASTA 2019) Copyright © 2019, the Authors. Published by Atlantis Press. This is an open access article under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/). Advances in Intelligent Systems Research, volume 168