Hierarchical visual localization for visually impaired people using multimodal images

作者：

Highlights：

• Dual Desc extracts both global and local descriptors from multimodal images.

• Attentive global descriptors surpass SOTA and robustify image retrieval.

• Infrared and depth-based geometric verification refines retrievals.

• Hierarchical visual localization pipeline performs well in real-world scenarios.

摘要

•Dual Desc extracts both global and local descriptors from multimodal images.•Attentive global descriptors surpass SOTA and robustify image retrieval.•Infrared and depth-based geometric verification refines retrievals.•Hierarchical visual localization pipeline performs well in real-world scenarios.

论文关键词：Deep image descriptor,Place recognition,Geometric validation,Sequence matching,RGB-D-IR image

论文评审过程：Received 4 March 2020, Revised 19 June 2020, Accepted 10 July 2020, Available online 25 July 2020, Version of Record 16 August 2020.

论文官网地址：https://doi.org/10.1016/j.eswa.2020.113743