Robust Ego-Velocity Estimate Method Based on GNSS and Stereo Visual Sensor Fusion in Variable-Rate Application for Agricultural Vehicles

A stable ego-velocity estimate method by integrating a low-cost single global navigation satellite system (GNSS) and stereo visual sensor is proposed in this study, to address the problem of vehicle velocity measurement noise in variable-rate tasking for agricultural production. The proposed system uses a Kalman filter to reduce sensor noise for raw velocity data, and then a fusion algorithm based on variance weights was designed to integrate GNSS- and stereo visual-based velocities. Field experiments were designed and conducted to evaluate the accuracy of the raw velocity measurement, filtering, and fusion effects under low and high GNSS signal environment scenarios with vehicle motion speeds of 0.9 m/s, respectively. The experimental results proved that the root mean square error (RMSE) after smoothing and fusion was decreased to 4.66 and 5.42 cm/s. By comparing with the raw GNSS velocity in low and high GNSS signal environments, RMSE was reduced by 68.02% and 26.06%, respectively. These results indicate that, the proposed fusion method can yield an accurate estimation of ego-velocity that satisfies the requirements of variable-rate tasking for agricultural production.