Objectives Individual tree localization is an important issue in the field of forestry remote sensing. Monocular visual simultaneous localization and mapping (SLAM) algorithm is a significant tool for outdoor spatial localization and mapping, which solved the problem of inability to locate due to missing global navigation satellite system signal caused by tree canopy occlusion in forest resource survey. However, the existing monocular vision SLAM algorithm cannot achieve direct localization of standing trees in sample plots. To solve this problem, we proposed the individual tree (Indi-tree) SLAM algorithm based on monocular vision SLAM algorithm.
Methods First, camera pose estimation was performed using the direct sparse odometry SLAM algorithm and the actual displacement of the camera was applied to restore the map scale. Afterwards, according to the principle of edge detection technique in depth image, the position of standing trees in sample plots was determined. Finally, the camera coordinates and tree coordinates were converted based on the relationship between the camera and standing trees during the image acquisition process, ultimately achieving the direct positioning of standing trees within the sample plots.
Results In this experiment, three square sample plots with a side length of 40 m were scanned by the camera to verify the accuracy of the Indi-tree SLAM algorithm. The experimental results indicated that the root mean square error of the sample wood coordinates calculated by the proposed single-wood positioning algorithm was 0.44 m along both x and y axis, and the average positioning error is 6.3%.
Conclusions This study achieves direct positioning of standing trees in the sample plots, greatly reducing the measurement time of forest structure parameters and offering an accurate and efficient feasible solution for forest resource inventory.
DownLoad: