Study Of Forest TLS Point Cloud Data Automatic Registration Algorithm

TLS(Terrestrial laser scanning)has been widely applied to analyze the 3D structure of single tree at the level of plot for forest inventories,and provide more high-precision ground verification data for large-scale remote sensing technologies such as ALS(airborne laser scanning).The work style of TLS can be divided in single and multiple.In comparison,multi-scan can obtain more complete information of the plot and the single tree detection and structure parameter extraction accuracy is much higher than single-scan,so multi-scan are more widely used in plot surveys.Classic registration algorithms are mostly based on flat surface which cannot be directly applied to forest scenes and the traditional manual setting of target for point cloud data registration is less efficient.In order to solve the problem that the current point cloud automatic registration algorithm is not suitable for forest plot scenes,this paper proposes an automatic registration algorithm of forest TLS point cloud data without any artificial reference objects.The main research contents and experimental results are shown as follows:(1)In order to solve the problem that the current ground point filtering algorithm is well applied to ALS point cloud data but not suitable for TLS point cloud data,an octree-based ground point filtering algorithm is proposed in this paper.The experimental results show that the algorithm makes full use of the octree feature of the halving of the scale of voxels layer by layer,and the ground point can be automatically filtered with high accuracy only by setting the resolution.(2)In order to solve the problem that the low recognition accuracy and efficiency of the widely used trunk point recognition algorithm,a trunk point recognition algorithm based on voxel projection density is proposed in this paper.The experimental results show that the algorithm can realize trunk points recognition voxel by voxel which improves the efficiency of trunk point recognition,and the noise rate of trunk point recognition results is relatively small,That is,the accuracy of the trunk point recognition is improved while improving the efficiency.(3)Based on accurately identified ground and trunk point cloud data,this paper proposes an automatic registration algorithm for forest scenes TLS point cloud data.The experimental results show that using the position and DBH of the single tree can quickly search for the corresponding feature points,and achieve the coarse registration of the two station point cloud data.With the matching trunk and ground data,the ICP algorithm can be used to achieve fine registration of the two station point cloud data,and finally achieve high-precision automatic registration of forest point cloud data.In summary,this paper divides the registration process into two steps:coarse registration and fine registration,which gradually achieve automatic high-precision registration of forest TLS point cloud data.Based on the high-precision ground point and trunk point extraction,the single tree position and DBH(diameter at breast height)can be used to quickly and accurately acquire the matched points of the base station and the registration station,and achieve coarse registration of point cloud data.The ICP algorithm can achieve high-precision automatic registration of forest TLS point cloud data on the basis of coarse registration.