Artificial Landmark Positioning System Using Omnidirectional Vision for Agricultural Vehicle Navigation

Abstract

The positioning of agricultural vehicle for autonomous navigation systems is important. We proposed a novel simple and inexpensive positioning system utilized by four artificial landmarks and an omni directional vision sensor. The red intensity pixels not less than threshold were extracted as a small area in the omni directional image and the center of the gravity was calculated as one landmark position. The direction angles between the neighboring landmarks were calculated after detecting all landmarks. Location of the vehicle was estimated by the center of gravity of four intersections formed by four arcs according to geometric transformation based on four direction angles of four landmarks with camera. In the field experiment, the sensor was fixed on the ground 50 m × 50 m square area in the sunshine to get clear images and examined the accuracy of the position estimation algorithm. Another experiment was done to test practical application with vehicle in a straight-line driving. The field experimental results showed that the mean absolute error of the four direction angles was about 1 to 2 degree and the RMS error in distance was about 24.51 cm. The vehicle practical experimental results showed that the real-time distance errors were mostly between about 3 cm to 10 cm. The RMS error was about 8.34 cm. This positioning system is a simple, feasible and prospective system for agricultural vehicle navigation.