Intelligent agricultural machinery navigation and pose estimation in plantation/nursery environments integrating improved RANSAC and lightweight YOLOv11

In response to the problems of low efficiency in traditional forestry surveying and insufficient real-time perception and navigation capabilities of intelligent agricultural machinery in complex forest environments, this study proposes a forestry crop pose estimation and intelligent agricultural machinery collaborative navigation method that integrates improved Random Sample Consensus (RANSAC) algorithm and lightweight visual detection. By improving the RANSAC algorithm, automated measurement of tree diameter at breast height and 3D pose estimation can be achieved, and high-precision offline point cloud maps can be constructed; On this basis, a lightweight online visual perception module is integrated to achieve real-time detection and localization of dynamic obstacles. A complete navigation system architecture has been developed, which deeply integrates offline 3D reconstruction and online real-time perception. In the field testing of intelligent agricultural machinery navigation, the fusion strategy reduces the maximum lateral tracking deviation from 0.85 m when relying only on the global path on challenging paths containing Unmapped obstacles to 0.28 m, effectively improving the navigation accuracy, obstacle avoidance ability, and operational safety of agricultural machinery in complex forest environments. This study provides a systematic solution for autonomous navigation of forestry intelligent agricultural machinery through the collaborative technology of offline 3D maps and online visual systems, which has strong practical application value.