The necessity to enhance the efficiency and safety of human labour is leading to the increasing integration of robots into everyday life. Specifically, in agricultural environments, robots capable of monitoring plant health, counting fruit, harvesting, and pruning plants can provide valuable support in cooperation with human workers. However, in vineyards, due to the repetitive structure of the rows and, most significantly, the seasonal changes throughout the year, LiDAR-based and vision-based navigation and localization become extremely challenging. Existing solutions based on expensive sensing systems, such as RTK-GPS, seek to address this aspect; nevertheless, their high cost limits usability and scalability. The aim of this thesis is to overcome this limitation by relying exclusively on vision, thereby significantly reducing the overall system cost.