Real-Time Tree Row End Detection: A Machine Vision-Based Approach to Improve Navigation of Electric Vehicles in Vineyards

This study presents a novel approach for real-time tree row end detection in vineyards using a machine vision-based system leveraging the YOLOv8 object detection model. Addressing the limitations of GNSS technology in environments where signal obstruction is prevalent, this research aims to enhance the precision and reliability of autonomous navigation systems for electric vehicles (EVs) in agricultural settings. The developed system utilizes a dataset created from images captured using an HD camera mounted on the EV, encompassing various weather and canopy conditions to train the YOLOv8 model effectively. Real-time field experiments demonstrated the system's effectiveness, showing that the ‘End of row’ flag is accurately generated as the EV approaches the end of the vineyard row, with the initial distance measurement consistently falling within the 2.5-3 m range. This allows the EV to stop within the next 2–3 seconds at an optimal distance of approximately 1 m from the row end. This research highlights the potential of integrating advanced machine vision techniques with agricultural machinery, providing a robust alternative to traditional navigation aids in precision agriculture.