基于局内点分布的图优化方法  

Graph Optimization Method Based on Inlier Distribution

作　　者：翁秀玲 王云峰[1] 吴炜[1] 郭东辉[1] WENG Xiuling;WANG Yunfeng;WU Wei;GUO Donghui(School of Information Science and Engineering,Xiamen University,Xiamen 361005,China)

机构地区：[1]厦门大学信息科学与技术学院,福建厦门361005

出　　处：《厦门大学学报（自然科学版）》2018年第2期265-270,共6页Journal of Xiamen University：Natural Science

基　　金：国家自然科学基金(61274133)

摘　　要：COP-SLAM(closed-form online pose-chain simultaneous locatization and mapping)是一种轻量型图优化算法,可以实时优化.但当局内特征点集中位于图像的特定区域时,其信息矩阵不能很好地表征里程计精度,影响实时优化效果.基于局内特征点的分布,引入局内点集的面积表示局内点的分布,提出了一种视觉里程计的后端图优化方法.实验结果表明,该优化方法有效地降低了相机绝对轨迹误差.Closed-form online pose-chain simultaneous locatization and mapping(COP-SLAM)is a lightweight graph optimization algorithm that meets real-time constraints.However,it cannot be a good measure of visual odometry accuracy when inlier points are mainly distributed on the specific region of the image,affecting the real-time optimization results.A general approach to back-end graph optimization of visual odometry is proposed based on inlier distribution.The area of inlier set is introduced to represent inlier distribution.Experimental results have demonstrated that the method can effectively reduce absolute trajectory errors.

关 键 词：图优化 信息矩阵 局内点 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]