Visual simultaneous localization and mapping (vSLAM) algorithm based on improved Vision Transformer semantic segmentation in dynamic scenes

Chen, Mengyuan; Guo, Hangrong; Qian, Runbang; Gong, Guangqiang; Cheng, Hao

doi:https://doi.org/10.5194/ms-15-1-2024

Articles | Volume 15, issue 1

https://doi.org/10.5194/ms-15-1-2024

© Author(s) 2024. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/ms-15-1-2024

© Author(s) 2024. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 15, issue 1

Research article

|

03 Jan 2024

Research article |

| 03 Jan 2024

Visual simultaneous localization and mapping (vSLAM) algorithm based on improved Vision Transformer semantic segmentation in dynamic scenes

Mengyuan Chen, Hangrong Guo, Runbang Qian, Guangqiang Gong, and Hao Cheng

Model code and software

VSLAM Algorithm Based on Improved Vision Transformer Semantic Segmentation in Dynamic Scenes Mengyuan Chen, Hangrong Guo, Runbang Qian, Guangqiang Gong, and Hao Cheng https://github.com/Oneghr/VTD-SLAM

Video supplement

VSLAM Algorithm Based on Improved Vision Transformer Semantic Segmentation in Dynamic Scenes Mengyuan Chen, Hangrong Guo, Runbang Qian, Guangqiang Gong, and Hao Cheng https://github.com/Oneghr/VTD-SLAM

Short summary

The proposed VTD-SLAM algorithm has the following advantages. (1) A multiclass feature enhancement and multiclass feature guidance semantic segmentation network (MSNET) is proposed to improve the semantic segmentation ability of dynamic objects. (2) Optimal neighbor pixel matching is used to complete the image of the deleted region. Our method can effectively solve the influence of dynamic objects, so that the SLAM system can operate effectively.