A method of fundamental matrix estimation based on nsga-ii and improved quasi-affine transform

作     者：Fan, Yi-Kai Liu, Shi-Jian Kong, Ling-Ping Pan, Jeng-Shyang 

作者机构：College of Computer Science and Mathematics Fujian University of Technology Fuzhou350118 China VSB-Technical University of Ostrava Czech Republic Fujian Provincial Key Laboratory of Big Data Mining and Applications Fujian University of Technology Fuzhou350118 China College of Computer Science and Engineering Shandong University of Science and Technology Qingdao266590 China Department of Information Management Chaoyang University of Technology Taiwan 

出 版 物：《Journal of Network Intelligence》 (J. Network Intell.)

年 卷 期：2021年第6卷第2期

页      面：313-327页

核心收录：

基　　金：7. Acknowledgment. This work is supported by the Collaborative Education Program of Industry and Education of Ministry of Education (No. 201901052008)  Scientific Research Project of Fujian Education Department (No. JAT190069  No. JK2017029) and Scientific Research Project of Fujian University of Technology (No. XF-X19017) 

主　　题：Efficiency 

摘      要：The fundamental matrix is the mainstream solution to computer vision problems such as 3D reconstruction, real-time location and map building. Accuracy and efficiency are two main measurement indexes in fundamental matrix estimation. When the accuracy is not enough, it often needs to be corrected through back-end optimization and other costly ways, and low efficiency will affect the real-time performance of the system. In order to solve this problem, this paper proposes a new estimation method of fundamental matrix based on improved quasi affine transformation. Specifically, based on the QUATRE algorithm, this method first proposes a population cooperation method based on a specific gene-chromosome pattern. Secondly, combining the advantages of NSGA-II in solving multi-objective problems, the Pareto dominance relationship of the population chromosome was firstly calculated according to the objective function of mean polarity distance and internal points in the way of NSGA-II, and the crowding degree was also calculated. The operations of population initialization, mutation and crossover in the discrete solution space represented by the homogeneous coordinate system are redefined. The selection operation is then performed according to the elite policy of NSGA-II. In addition, a confidence-based method to determine the number of iterations is proposed to accelerate the algorithm. Experimental results show that the proposed method can effectively eliminate noise and mismatching, and is superior to the current mainstream methods in accuracy and efficiency, and can effectively solve the problem of fundamental matrix estimation. © 2021 Global Research Online. All rights reserved.