Triplet loss based metric learning for closed loop detection in VSLAM system

作者：

Highlights：

• A triplet selection strategy has been constructed to speed up the convergence.

• The Softplus is applied so that the loss in the negative axis will be soft margin.

• The margin constructed for different classes has been proposed.

• Metric learning has been proposed for closed loop detection.

摘要

•A triplet selection strategy has been constructed to speed up the convergence.•The Softplus is applied so that the loss in the negative axis will be soft margin.•The margin constructed for different classes has been proposed.•Metric learning has been proposed for closed loop detection.

论文关键词：Metric learning,Triplet loss,Feature vector,Euclidean distance,Loop closure detection

论文评审过程：Received 15 December 2020, Revised 7 May 2021, Accepted 20 July 2021, Available online 24 July 2021, Version of Record 28 July 2021.

论文官网地址：https://doi.org/10.1016/j.eswa.2021.115646