Triangular Mesh Reconstruction Of Crop Leaf Point Cloud Data

With the development of 3D data collection technology,the collection of crop leaf point cloud data becomes easier and easier.The point cloud data is composed of disordered point sets,which is not conducive to the direct phenotype analysis of crops,and they need to be reconstructed into triangular meshes.Due to the limitation of 3D data acquisition equipment and the influence of the surrounding environment,the collected point cloud data of crop leaves often has the phenomenon of data redundancy and data loss.The traditional methods will produce multi-level and nonmanifold surfaces,even with unsmooth boundaries when reconstructing triangular meshes.Therefore,it is necessary to explore more effective triangular mesh reconstruction techniques for crop leaf point cloud data.In this paper,a 3D reconstruction framework based on crop leaf-based triangular mesh template matching deformation is proposed.It first extracts the skeleton information of the crop leaf point cloud model,and then determines the sampling points on the leaf point cloud through the skeleton,and performs template matching on the feature histogram descriptor for each model construction point through the sampling points,and obtains the crop leaf point cloud.The triangular mesh template being most similar to the input crop leaf point data can be found,and finally is deformed to obtain the reconstruction model of the triangular mesh.Our main contributions can be summarized as follows:1.A method of crop leaf point cloud skeleton extraction based on adaptive weighting operator is proposed.It calculates the skeleton constraint point set of crop leaf point cloud by constructing an adaptive weighting operator,and introduces the main curve to fit the skeleton constraint point set to obtain the optimized leaf point cloud skeleton.The point cloud integrity weight used in this method can improve the skeleton extraction effect of missing leaves in the point cloud.Comparing the skeleton extracted by using the complete leaf point cloud and manually removing some points to form the missing leaf point cloud,the ratio of Hausdorff distance to leaf length is lower than 1.3%.2.A method for retrieving crop leaf 3D model based on point-to-point feature histogram is proposed.It firstly determines the location of the model sampling points through the skeleton of the crop leaves,then constructs the point-to-feature histogram of each sampling point,and synthesizes all the point-to-feature histograms of the model to construct the distance matrix of the model.The shape-distance distribution curve of the computational model is used as a descriptor to conduct the match,and finally a triangular mesh template that is most similar to the crop leaf point cloud data is obtained,which provides a high-quality template for the subsequent mesh deformation reconstruction.3.A 3D reconstruction method based on the deformation of crop leaf triangular mesh template is proposed.It builds the stiffness energy function of the mesh deformation by accumulating the stiffness deviation of each element,and uses the position information of the point cloud data of the crop leaves as a constraint to realize the rigid deformation of the mesh template,and obtain the final triangular mesh of the crop leaves.A large number of experimental results show that this method can avoid the generation of multi-level,nonmanifold and non-smooth triangular mesh surfaces during reconstruction,and achieves better reconstruction results than some traditional methods.