Navigation Line Extraction Method for Feed-Pushing Robots Based on Machine Vision
Lingmin Zhang, Meng Li, Wenbiao Chen · 2025
To ensure the autonomous navigation of feed-pushing robots in pasture environments and to mitigate the limitations of traditional navigation methods affected by environmental variations, this paper proposes a machine vision-based navigation line extraction method. The aim is to precisely guide the robot by capturing environmental information and performing image processing. The method utilizes the YOLOv8 detection model to identify cattle pens, extracting feature points from the detection bounding boxes. Based on these feature points and fitting techniques, the cattle pen lines on both sides of the robot are obtained, which in turn are used to compute the robot's navigation path. An improved RANSAC algorithm is used to fit the navigation lin. Experimental results indicate that YOLOv8-based cattle pen detection model achieves an accuracy of 84.9%, demonstrating its good adaptability in pasture environments. The improved RANSAC line fitting algorithm shows strong anti-interference capability in complex environments, with a fitting accuracy of 83.1% and an average deviation of 1.56°, outperforming the Least Squares method. The experimental results validate the feasibility of this method in effectively extracting the robot's navigation line in complex pasture environments.