Research On Forest Land SLAM Mapping Based On Lidar Inertial Navigation Fusion

The forest resources survey is significant to the national economy,carbon neutralization,ecological improvement,and climate regulation.With the development of lidar synchronous map construction and mapping(SLAM),knapsack lidar equipment,as an advanced forestry survey method,has been applied to the construction of forest point cloud maps.The efficiency of this investigation method is much higher than the traditional manual data acquisition method and higher than the stand-by lidar scanner.However,the terrain of the forest land mapping environment is different from that of an ordinary urban environment.The existing synchronous map construction and positioning algorithms can not give full play to the performance of backpack lidar equipment in a forest land environment.The synchronous map construction and positioning technology of laser inertial navigation fusion make up this problem.The SLAM algorithm of laser inertial navigation fusion can directly fuse the original laser feature information and the measured value of the inertial unit through the tight coupling method.It has the advantages of solid robustness and high mapping accuracy.The angular velocity and acceleration data of inertial measurement unit(IMU)are applied to the feature extraction and inter-frame matching of lidar to generate lidar odometer and IMU odometer to further optimize the mapping accuracy in the back-end optimization.However,the factors of the woodland environment are mainly trunk,crown,and ground,and there are few point clouds with more plane features such as buildings.The woodland point cloud map aims to extract key forest structure parameters,such as DBH,tree height,and single tree position.Under the slam front-end of existing feature matching,increasing the proportion of trunk features is conducive to improving the matching of a trunk point cloud.To better extract the key forest structure parameters,we need to improve the construction of tree trunks in the map to avoid the impact of tree crowns and ground points on the mapping accuracy.Therefore,this study improved the SLAM algorithm for trunk characteristics,ground point filtering,and crown point filtering.We reduce the proportion of ground points and crown points in a single frame point cloud and improve the proportion of trunk features to improve the mapping accuracy of forest land laser inertial navigation fusion slam.Finally,we use the root mean square error(RMSE)and mean absolute error(MAE)of the single wood position to verify the mapping accuracy.Our primary research contents and experiments are as follows:(1)We propose a forest land SLAM algorithm based on arc radius to extract features.Most of the forest trunk point clouds are columnar.At the same time,the existing SLAM algorithm uses radian to extract the characteristics of the point cloud,and many ground points and crown points are selected,which reduces the mapping and positioning accuracy of the trunk point cloud.This study uses the arc radius as the feature extraction method.Most of the features are distributed on the trunk,and most of the ground points are removed by using the attitude information of IMU to improve the accuracy of inter-frame matching and loop optimization.The experimental results show that the improved algorithm improves robustness and accuracy in forest land samples.RMSE and Mae are 0.67 and 0.41 m,respectively,reducing 0.67 and 0.56 m,which effectively improves the accuracy of mapping and positioning.(2)A ground point filtering algorithm based on local point cloud fitting is proposed.When using knapsack lidar equipment to walk and measure forest land,some laser beams will hit the ground,resulting in many similar ground point clouds,which is not conducive to the construction of forest land point clouds.In this study,the single frame LIDAR point cloud is divided into ordered point cloud sub-regions,and all point clouds are reduced in dimension.Finally,straight-line fitting is carried out in each sub-region to judge the ground and nonground points,and the nonground points are used for mapping.The experimental results show that RMSE and Mae are 0.39 and 0.36 M,which are reduced by 0.12 M,respectively,and the accuracy of forest land mapping and positioning is improved.(3)A forest land SLAM algorithm is proposed to extract tree stem points based on a depth map.The scanning distance of lidar will generally exceed the range of forest land to be constructed,and only the point cloud constituting the loop is a complete tree point cloud that can extract parameters.In this study,as long as the mapping object is the trunk,crown,and distant noise in the loop,it interferes with the mapping.Therefore,this paper studies the conversion of a three-dimensional LIDAR point cloud into a two-dimensional depth map.It extracts tree stem points in the depth map.It then converts them into a three-dimensional point cloud to filter out tree crown points,and distant interference point clouds improve the mapping and positioning accuracy.The experimental results show that the method proposed in this study can effectively filter out crown points and distant interference points.RMSE and Mae are 1.03 and 0.62 M,reducing 0.01 and 0.1M,respectively,to improve forest land mapping and positioning accuracy.In addition,this study lacks the experimental comparison of forest areas with more tree species and sharp slope changes.The follow-up plan is to conduct further refinement experiments for more tree species and different slopes and quantitatively evaluate the algorithm’s performance.