摘要
在基于视觉的同时定位和地图构建系统中通过回环检测判别机器人是否到达过先前的位置,可有效消除位姿估计造成的累积误差。随着数据量的增大,现有方法会导致网络模型泛化性能降低。为实现可持续回环检测,提出一种网络权重参数敏感度分析视觉回环检测方法,将残差神经网络与广义均值池化相结合来构建轻量化特征提取网络;设计了可变相似感知区域,将可变滑动窗口和相似度矩阵相结合来提取三重样本;提出的网络权重参数敏感度分析方法降低了网络模型的灾难性遗忘。与典型方法MAC相比,所提方法在Nordland等数据集的召回率提高了42%左右。
In visual-based simultaneous localization and mapping,the loop-closure detection might determine whether the robot reached the previous positions,so that the accumulated errors caused by pose estimation might be effectively eliminated.With the increase of data volume,the generalization performance of network model will be reduced by the existing methods.In order to realize sustainable loopback detection,a visual loop detection method was proposed herein based on sensitivity analysis of network weight parameters.A lightweight feature extraction network was constructed by combining residual neural network and generalized mean pooling.The variable similarity sensing areas were designed,and the variable sliding windows were combined with the similarity matrix to extract three samples.The proposed sensitivity analysis method of network weight parameters reduced the catastrophic forgetting of network models.Compared with the typical method MAC,the recall rate of the proposed method in Nordland and other data sets is improved by about 42%.
作者
沈晔湖
李欢
张大庆
苗洋
赵冲
蒋全胜
SHEN Yehu;LI Huan;ZHANG Daqing;MIAO Yang;ZHAO Chong;JIANG Quansheng(School of Mechanical Engineering,Suzhou University of Science and Technology,Suzhou,Jiangsu,215009;Department of Computer Science and Technology,Xiamen University,Xiamen,Fujian,361005)
出处
《中国机械工程》
EI
CAS
CSCD
北大核心
2024年第7期1212-1221,共10页
China Mechanical Engineering
基金
国家自然科学基金(51975394)
江苏省自然科学基金(BK20211336)
江苏省研究生科研创新计划(092290148)。
关键词
回环检测
终身学习
对比学习
同时定位和地图构建
网络权重参数敏感度分析
loop-closure detection
lifelong learning
comparative learning
simultaneous localization and mapping
sensitivity analysis of network weight parameters
