A novel terrain semantic map building framework for field robots
Haoran Zhang, Botao Zhang, Yawen Li, Tao Hong, Yun Chen, Qiang LuPurpose
To address the limitations of existing terrain mapping methods that ignore temporal misalignment, adopt non-adaptive pixel-grid projection and only provide coarse traversability information, this paper aims to propose Terrain Semantic Map Building Framework (TSM-BF) to construct fine-grained terrain semantic maps for more reliable and intelligent navigation of field robots in complex outdoor environments.
Design/methodology/approach
A TSM-BF is proposed, integrating depth camera and 3D LiDAR data for terrain perception. The framework introduces a timestamp cropping mechanism to handle temporal misalignment between semantic and depth image streams, and an adaptive Pixel-Grid Match strategy to achieve accurate semantic projection onto grid maps.
Findings
Experimental results demonstrate that the terrain semantic maps generated by TSM-BF provide comprehensive environmental details. The proposed timestamp cropping mechanism and adaptive projection modules effectively improve mapping accuracy and robustness, while maintaining real-time performance for resource-constrained mobile robots.
Originality/value
TSM-BF realizes the integrated design of temporal synchronization and adaptive semantic projection, which overcomes the limitations of coarse-grained traversability representation and sensor temporal misalignment in existing methods. This work provides a practical and deployable terrain semantic mapping solution for autonomous field robots.