An Orchard Dense Map Construction Method Based on Improved ORB-SLAM3

To solve the problem of small number of image feature points matching, feature points loss and sparse point cloud caused by the complex lighting conditions in the orchard environment in hilly and mountain. An improved ORB-SLAM3 orchard dense map construction method was proposed in this paper. This method proposes an adaptive threshold FAST corner extraction method to the tracking thread of ORB-SLAM3. According to the average pixel of the current image under different illumination, the corresponding threshold is solved, which increases the number of feature points extracted under different conditions and improves the robustness of the method. Then, the line feature extraction method is added to the tracking thread, and the point-line fusion feature tracking is carried out to enrich the feature information in the environment. Finally, on the basis of the original method, a dense mapping thread is added to construct a three-dimensional dense map of the orchard, which solves the problem of sparse point cloud information in the original method. The experimental result shows that, compared with the original ORB-SLAM3, the average matching number of feature points in this method increases by $12.67 \%, 14.61 \%$ and 31.36 % under normal illumination, weak illumination and rainy weather. The method proposed in this paper has richer point cloud information and greatly restores scene information, which can provide support for orchard robot navigation path planning.