Research On Seafloor Image Registration Based On Narigation Information

Seafloor image alignment is of great significance to the exploration and research of deep-sea resources,and has been a hot spot and focus of research in recent years.The main steps of submarine image feature point alignment are:feature point extraction,describing feature points and matching feature points.However,due to the scattering of suspended particles in the water and the influence of artificial light source,the image quality of deep-sea seafloor is seriously reduced.Besides,the seafloor landform texture is concentrated in a certain area,and most of the similar textures are found.Therefore,it is difficult for current research to obtain the matching of feature points with high accuracy and uniform distribution.Fortunately,the robot is equipped with a navigation system that takes images and records geographic coordinates,which can be used to estimate changes in position between the images.Therefore,navigation information is taken as a priori information in this paper to improve the matching accuracy of feature points and homogenize the distribution of feature points.aiming at the deep sea seabed images captured by robot navigation.The specific work is as follows:(1)Aiming at the problem that the accuracy of feature points matching in the sequential and non-sequential submarine images captured by robot navigation is low,the accuracy of the algorithm is improved from two aspects of feature point extraction and feature point matching.Feature point description: the traditional algorithm uses the straight statistical graph method to calculate the main direction,the main direction obtained by this method has large error,which affects the accuracy of feature point description.To solve this problem,this paper uses navigation information to estimate the rotation Angle between sequence and non-sequence images,and takes the rotation Angle as the main direction to make the main direction more accurate,so as to improve the identification ability of feature point description,and finally achieve the goal of improving the accuracy of feature point matching.For the matching of feature points: Random sampling consistency algorithm is a more classical method of feature point matching.Its main idea is to eliminate the wrong matching point pairs by estimating model parameters.The algorithm has a great dependence on the proportion of the correct matching point pairs in the data set.For the data set with a large proportion of error points,the homography matrix obtained by RANSAC algorithm is not of high precision.In this paper,the top 85% feature point pairs in the original data set are selected first to form a new data set.Then,the homography matrix is obtained by using the new data set.Experimental results show that the algorithm effectively solves the problem of low precision calculation of main direction,improves the recognition ability of descriptor,and improves the accuracy of feature point matching.By optimizing RANSAC algorithm,the accuracy of feature point matching is further improved,and the number of feature points is increased.(2)Aiming at the problem of uneven distribution of feature points extracted from deep seabed image sequences obtained by robot navigation,the algorithm in this paper achieves uniform distribution of feature points by dividing the grid to extract feature points and selecting high-quality feature points based on the idea of quadtree.Meanwhile,in order to reduce the influence of texture similarity on feature point matching,the characteristics of sequence images taken by using the robot as the prior information according to the navigation information,the search space of feature point matching is reduced.and the local region matching was used to replace the global matching,thus,the matching accuracy of feature points can be improved.Experimental results show that this algorithm can effectively solve the problem of uneven distribution of feature points,and on the basis of homogenization of feature points,the matching algorithm based on navigation information can reduce the influence of similar texture on feature point matching.Finally,the uniform distribution of feature points is obtained,and the high precision matching is carried out to improve the accuracy of unisexual matrix calculation.(3)An interactive image registration system is developed.The main modules of the system are image reading module,feature point matching module,outer point elimination module and inner point saving module.The feature point matching module provides a variety of classical feature point matching algorithms and the matching algorithm proposed in this paper.and the elimination of external points module provides the optimized RANSAC algorithm proposed in this paper,and provides a variety of transformation model choices.The experimental results show that the system has high reliability and practicability.