Coal mine inspection robots have great potential in improving the efficiency and safety of mine operations. However, challenges such as low light, weak textures, and dynamic obstacles make traditional visual SLAM methods inadequate. This paper proposes SemViSLAM, a visual SLAM model that integrates semantic information fusion to address positioning and environmental perception issues in coal mine inspections. By combining semantic segmentation, image enhancement, and loop detection modules, SemViSLAM optimizes feature matching and map construction, significantly improving the system’s accuracy and robustness. Experimental results demonstrate that SemViSLAM outperforms traditional methods like ORB-SLAM3 in coal mine environments, with a 15% improvement in positioning accuracy and a 10% increase in loop detection accuracy, especially in low-light and dynamic obstacle conditions. This work provides a robust positioning solution for coal mine inspection robots and advances the application of SLAM technology in complex industrial environments.
Shi et al. (Mon,) studied this question.