Research On 3D Plant Modeling Based On Scattered Point Cloud

With the development of computer technology and the popularity of computer graphics,applications related to plant models are more common,including virtual reality,computer games,movies TV animation and gardens design system.With the emergence of 2D image acquisition and 3D plant point cloud acquisition technology,how to combine early graphic-based plant models’ methods with modern 3D point cloud models and human-computer interaction technology to efficiently and conveniently complete plant models and how to combine the theory of plant models with practical applications have become research hotspots for scholars at home and abroad.However,due to the diversity of plants’ own species,the complexity of their own structures and the severity of self-occlusion,plants’ reconstructions are complex and challenging.In this thesis,the whole plant is taken as the research object,and the initial plant model obtained by scanning the whole plant using the portable 3D scanner Kinect has a series of problems such as noise and adhesion between the leaves,In order to solve these problems,the following research was carried out:(1)The original 3D plant point cloud model is obtained by 360° scanning of the original plant using the depth camera Kinect.The K-Means algorithm and the heterochromatic point algorithm are used to de-noise the outliers and the heterochromatic points in the initial 3D plant point cloud model.And the de-noised 3D plant point cloud model is simplified by a point-encrypted inverse algorithm.(2)A spherical seed filling algorithm is proposed to cluster the de-noised and simplified 3D plant point cloud model,so that each leaf after clustering becomes a separate class.Then the tangential midpoint algorithm is proposed to obtain the initial skeleton of the 3D plant point cloud model.The initial skeleton is combined with the cubic B-spline algorithm and smoothed by adjusting the control nodes to obtain the skeleton of the 3D plant point cloud model.(3)A relatively complete leaf is selected from the original plant as a leaf model and a 2D image of the leaf model is obtained,and a coordinate midpoint algorithm is proposed on the obtained 2D image to obtain information such as a skeleton,a length and a width of the 2D leaf.According to the skeleton,the length and the width of the 2D leaf,proposing a 3D leaf drawing algorithm based on the ellipse to obtain the 3D leaf template,and the skeleton of the 3D leaf template is obtained according to the proposed tangent midpoint algorithm and the cubic B-spline algorithm.Finally,the skeleton of the 3D plant point cloud model is fitted by adjusting the position of the 3D leaf template and the skeleton control node to achieve the reconstruction of the entire 3D plant model.(4)Based on the above theory and method,the system of 3D plant model is designed and implemented in combination with OpenGL graphics library,Qt interface library and OpenSceneGraph graphics library.The system is mainly divided into the de-noising and simplification module of the 3D plant point cloud model,the clustering module of the 3D plant point cloud model,the skeleton extraction module of the 3D plant point cloud model,the reconstruction module of the 3D leaf template,the fitting adjustment module of the 3D plant point cloud model based the 3D leaf template and the skeleton of the 3D plant point cloud model.Finally,constructing a user-friendly interface to realize the function of reconstructing the 3D plant model.