| Reliable forest resource information is needed to assess the forest development status and design management plans for forest maintenance and conservation.Forest observation equipment is an important means to obtain forest resource information.In this research,a portable RGB-D SLAM tree measurement system is constructed for observing standing tree measurement factors(i.e.tree position,DBH and height)and forest field sample inventories.The new construction system has the following characteristics:(1)it can be installed on a smartphone platform,so the equipment is portable and cheap;(2)it is based on the online RGB-D SLAM technology,so the equipment can perform real-time relative positioning under forests,so that observers can observe tree measurement factors and the like during the movement process;(3)it is designed based on augmented reality,gesture interaction and other technologies for simple observation operation procedures and lightweight algorithms;(4)it presents the observation results to the observer in the form of augmented reality scenes in real time,which not only realizes what is seen and obtained,but also monitors the quality of the estimated results and effectively avoids possible gross errors;(5)it eliminates indoor work and provides three different data management forms(SQLite database management,document management,and three-dimensional structured management)based on SLAM system output data and survey results.The key works are as follows:(1)In the estimation of tree position and DBH,this research first constructed a method for determining the breast height of tree based on the operation process and augmented reality;and then,an algorithm was designed to estimate the DBH and position of tree using the depth image and the camera pose,which are obtained during the scanning of the breast height of tree.The test results in the filed samples showed that the DBH estimations had a 0.28 cm BIAS and a 1.00 cm root mean square error(RMSE).(2)In tree height estimation,this research built a tree height estimation algorithm based on the coordinate of a point at the ground diameter obtained during the observation process,and the pose and the tree top pixel coordinate while observing the treetop.The test results in the field samples showed that the tree height estimations had a 0.08 m BIAS and a 0.65 m RMSE;and when the tree height was lower than 25 m,the estimations was more reliable.(3)In the forest plot survey,the accuracy of the pose obtained by the SLAM system directly affects the accuracy of estimation results such as tree position.In this research,an online trunk-based backend was designed to accurately estimate tree position and correct pose drift in real time.Specifically,a trunk-based loop closure detection algorithm was designed for detecting whether an earlier observed tree is re-observed to provide nodes and constraints for tree position graph optimization;the provided nodes and constraints were used to build and optimize the tree position graph and then correct the current pose based on the optimized globally consistent tree position graph.The test scheme in this paper attempts to evaluate the pose estimation result indirectly by the accuracy of the tree position estimation.And the test results of the plots showed that the distance mean between the estimated and reference tree positions was 0.134 m and the max distance was 0.32 m.The tree position estimations had-0.06 m,0.001 m,and-0.036 m BIAS in the x-axis,y-axis,and z-axis directions;and had 0.085 m,0.086 m,and 0.078 m RMSEs in the tree directions.(4)The forest plot survey system was constructed based on the tree position,DBH and tree height observation methods and the trunk-based SLAM backend designed in this research.In this system,the method of constructing the plot coordinate system was also designed to describe the tree positions in the plot;the scanning path of the plot was designed to optimize the tree positions using the designed trunk-based SLAM backend;the algorithm of slope and aspect estimation was designed;three different data management forms were provided.The test results of the plots showed that the average DBH estimations had a-0.05 cm BIAS and a 0.05 cm RMSE;the average tree height estimations had a 0.01 m BIAS and a0.03 m RMSE;the volume estimations had a-0.2021 m~3/hm~2 BIAS and a 9.0690 m~3/hm~2 RMSE;the cross-sectional area estimations had a-0.1015 m~2/hm~2 BIAS and a 0.1897 m~2/hm~2 RMSE;the stem density estimations had a-1.25/hm~2 BIAS and a 31.25/hm~2 RMSE;the slope estimations had a-0.12°BIAS and a 0.29°RMSE;the aspect estimations had a-1.30°BIAS and a 14.73°RMSE.The aspect estimations had a large RMSE due to the estimated pose errors of the SLAM system,but the aspect measurements were still unbiased as a whole.In short,the portable RGB-D SLAM tree measurement system constructed in this research not only has a series of advantageous features,but also can obtain higher accuracy measurement values.Obviously,this system is a potential solution for observing trees and investigating forest field samples. |
