Autonomous Navigation System for Agricultural Vehicles Using Machine Vision Recognition and Localization of Panoramic Landmarks

Abstract

This paper develops a low-cost image real-time positioning system platform for agricultural vehicle localization system and its deployment and field-testing in rice paddy. The system enables affordable precision agriculture in rice production by providing the vehicle's position in the paddy without the use of expensive RTK-GPS (Real-Time Kinematic GPS) positioning system. We designed specific artificial landmarks to form four markers at the corners of the farmland. The system uses the camera panorama image to detect a specific marker., then uses the angle of the marker in the image to calculate the current position. The results of static positioning experiments show that in the field of view of 30 ${\mathrm{m} \times 20 ~\mathrm{m}}$, the maximum error of X and $\mathbf{Y}$ axis is 20 cm, the average error of X axis is 9 cm and the average error of $\mathbf{Y}$ axis is 7 cm, This study is actually installed on a rice transplanter for dynamic positioning verification. The system can solve the problems of unstable and interrupted satellite system leading to loss of control of agricultural vehicles and high cost of RTK-GPS system. This study has been experimentally verified to be able to localize within a range of $\mathbf{50} \mathbf{~m} \times \mathbf{50}\mathbf{~m}$.