基于点云的果树冠层叶片重建方法

Method of fruit tree canopy leaf reconstruction based on point cloud

精确的果树三维冠层结构是农业科研人员进行功能结构模型研究的重要载体,该文提出一种快速、精确、自动的果树冠层叶片重建方法。首先根据带叶果树点云的局部和全局特征,建立椭球分层的点云密度收缩方法实现器官点云分离,然后利用邻近传播主成分分析算法实现叶片特征参数的求解,利用Laplacian收缩算法实现冠层骨架点的连通,从而实现冠层叶片的快速自动重建。最后利用C++及Point Cloud Library(PCL)点云库,开发果树叶片点云冠层自动重建系统,对苹果树、柑橘树等不同类型果树进行算法验证,结果表明该方法能够正确识别出的叶片数占冠层总叶片数的90%以上,叶面积指数的正确率大于95%,叶片倾角偏离5?以内的叶片数占总叶片数的90%以上。该方法得到了较好的可视化效果和叶冠三维重建精度,可为后期树体冠层内光合作用的研究、整形修剪、农业仿真试验等提供参考。

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 Focus3 D 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 PlantC AD 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.