摘要
回环检测是视觉SLAM中不可或缺的一部分,正确检测出回环能有效减小移动机器人在定位和建图过程中产生的累计位置漂移问题。SLAM回环检测目前主要有词袋模型方法,具有良好的实时性,但人工特征具有对光照变化非常敏感等问题。针对这些问题,提出了深度学习和回环检测结合的方法,将卷积神经网络模型应用于SLAM回环检测中,并且通过模型对RGB-D图像进行特征提取,提出了两种图像融合方法,通过对比判断最佳措施;进而通过特征匹配来判断回环。与现有方法相比,本研究提出的方法达到了具有更高精确性和实时性的效果。
Loop detection is an indispensable part of visual SLAM,and correct detection of loop can effectively reduce the problem of accumulated pose drift in the positioning and mapping process of mobile robot.At present,SLAM loop detection mainly has the method of word bag model,which has good real-time performance.But the artificial feature is very sensitive to illumination change.To solve these problems.This paper proposed a method combining deep learning and loop detection to apply the convolution neural network model to SLAM loop detection,the features of RGB-D images through the model,and the paper proposed two image fusion methods to determine the best measures through comparison.Then the loop is judged by feature matching.Compared with existing methods,the method proposed in this paper achieves higher accuracy and real-time performance.
作者
李若楠
许钢
郭芮
邢广鑫
江娟娟
Li Ruonan;Xu Gang;Guo Rui;Xing Guangxin;Jiang Juanjuan(Key Laboratory of Detection Technology and Energy-saving device,Anhui Polytechnic University,Wuhu,Anhui 241000,China)
出处
《黑龙江工业学院学报（综合版）》
2019年第11期42-48,共7页
Journal of Heilongjiang University of Technology(Comprehensive Edition)
基金
安徽高校自然科学研究重点项目(编号:KJ2018A0111)
检测技术与节能装置安徽省重点实验室开放基金(编号:2017070503B026-A01)
