Method of fruit tree canopy leaf reconstruction based on point cloud

Abstract: Accurate three-dimensional canopy structure of fruit trees is an important carrier for the study of functional structure model, and the canopy leaf is an important part of the fruit tree canopy structure. The spatial distribution and morphological structure of leaf canopy play an important role in fruit yield and quality. Compared with field crops, fruit trees have high branches, complex branches and luxuriant branches, so the rapid and accurate reconstruction of tree canopy is the focus of current research. With the development of laser measurement technology, 3D (three-dimensional) laser scanner is widely used in 3D reconstruction of plants because of its high precision, wide collection range, high efficiency and non contact. An accurate and automatic canopy leaf reconstruction method of fruit tree is presented based on point cloud. Firstly, using FARO Laser Scanner Focus3D to obtain dense point cloud of fruit tree canopy leaf, and according to global characteristics (the point cloud density distribution meets the ellipsoid layered characteristics) and local characteristics (different organs and organs of different parts have different point cloud density) of the leafy trees point cloud, this paper proposes a point cloud density shrinkage method with ellipsoidal layer to realize the separation of leaf point cloud. The model's parameters (the leaf width and the radius of main branches) are obtained by manual measurement. And experiment shows that point cloud segmentation effect is the best when the relation coefficient is 0.75 between the density threshold and the average density of the point cloud. After the point cloud density calculation, high-density point cloud on each leaf is gathered in the veins, and high-density point cloud of branches is gathered in the intersection of branch and branch bulge so as to realize the segmentation between organs. Then we calculate leaf characteristic parameter by the principal component analysis algorithm of the neighboring point cloud, in which the K mean algorithm is used to simplify the leaf point cloud, and the principal component analysis is used to calculate the normal vector of the leaf point cloud. Then, after removing the leaf point cloud, the branches point cloud is contracted into a connected skeleton by the Laplacian shrinkage algorithm which is a classical point cloud shrinkage algorithm, so as to realize the automatic reconstruction of canopy leaf combined leaf template. Finally, with C++ and Point Cloud Library (PCL), the automatic reconstruction system of point cloud of canopy leaf is developed on the PlantCAD development platform, and the method is validated by different types of fruit trees (Fuji apple tree, Wanglin apple tree, Maogu citrus tree). The results show that the recognition accuracy of leaf is higher than 90%, the correct rate of leaf area index is more than 95%, the number of leaf inclination deviation within 5 degrees is more than 90% of the total leaf, the efficiency is more than 7 times that of the artificial, and there is less artificial participation in the whole process. It gets better effect of visualization, and strong sense of reality and canopy leaf reconstruction precision, which provides effective technical support for the research on the photosynthesis, pruning and training of the tree as well as experimental simulations in the later period. Nevertheless, this method depends on the density of point cloud data, and has poor robustness to non ellipsoidal tree and severe noise point cloud (acquiring under windless weather). In the future, we will study the adjacent relation of the leaf point cloud, and make the algorithm adaptive to different tree structure and sparse canopy leaf point cloud.