Application of visual inertia fusion technology in rice transplanter operation

Automatic navigation technology for agricultural machines is booming, but the majority of existing research is based on satellite positioning technology. In response to the limited information acquisition capabilities of satellite positioning technology and the insufficient intelligence of agricultural machinery, this study conducted visual/inertial integrated navigation research on riding rice transplanters, using deep learning networks and machine vision to extract guidance routes in paddy fields. The inertial measurement unit (IMU) as well as the vision sensors are jointly calibrated with the attitude and navigation parameters of the rice transplanter as states, and an extended Kalman filter is designed to fuse the information from the outputs of vision and inertial guidance. We designed a PID controller to simulate and compensate for the side-slip conditions that occur in the work of agricultural machines. A real-time kinematic global positioning system (RTK-GNSS) is used to calculate the lateral deviation of the navigation in the field experiment. The result shows the lateral deviation of navigation was 0.044 ± 0.040 m for the straight section and −0.088 ± 0.143 m for the turning section in a paddy field.