Local Positioning of an Agricultural Mobile Robot Based on a Vision System

Determining coordinates is an important component task for developing a control algorithm for mobile agricultural robots. At the moment, this problem is solved by using global positioning satellite systems. However, satellite connection have not effective forever and everywhere. Then, for agricultural land processing, the navigation systems with local positioning could be applied. One of such systems is the visual positioning system based on landmarks. For a long time, its implementation was hampered by the imperfection of systems for recognizing objects in images. However, today neural networks perform computer vision tasks successfully. The purpose of this research was to develop and verify a mathematical model for a positioning system based on optical data of terrain images with beacons. As a result, mathematical formulas were developed to determine the position of the robot in a rectangular coordinate system, and their verification was carried out. The general requirements for the optical system were also worked out and experiments were carried out on real equipment. Ultimately, the performance of the proposed algorithm was proven by conducting an experiment using a real robot with installed camera on it.