Time Offset Estimation for Visual-Inertial Navigation Systems in Semi-Structured Orchard Environments

Aiming at the problem of low accuracy of time offset estimation in the process of vision-inertial sensor data fusion for robot navigation system in semi-structured orchard environment, which leads to poor positioning accuracy of orchard robots, this paper proposes an on-line estimation method of sensor information fusion time offset by combining point and line features. The method first establishes the velocity model of line features based on the semi-structured environment characteristics of the orchard, then carries out inverse projection based on the above velocity model, establishes the reprojection error of the line features for the fusion time offset, and finally establishes a nonlinear optimal solution objective function for the combined pointline time offset. Experiments on public datasets show that, compared with optimization methods that only consider point features, the time offset estimation accuracy of the method proposed in this paper is improved by at least 20 % and at most 77.78 %. Experimental results in actual orchard greenhouses show that, compared with point feature methods, the accuracy of the method proposed in this paper is improved by 63.11 %. This method can provide more accurate multi-sensor information fusion technology support for the positioning and navigation of greenhouse plant protection robots.