Image-based Tree 3D Modeling And Tree Parameter Extraction

Trees are an important part of the ecosystem and play an important role in maintaining carbon and oxygen balance and maintaining biodiversity.In virtual reality and forestry mapping,the three-dimensional model of trees has always been the focus of research.Due to the variety of trees and the large differences in morphology and structure,virtual plant modeling has become a hot issue in computer graphics research.At present,the construction methods of tree 3D models are based on sketches,rule-based,image-based methods,and so on.The image-based tree modeling method has the characteristics of strong sense of reality.This paper studies the 3D model of trees based on the image method to obtain a three-dimensional model of the tree with similar appearance.The model can be applied in virtual forestry,virtual reality and other scenarios.At the same time,with the continuous development of photogrammetry technology,trees can be measured by a single photo.This paper studies the camera imaging model,and quickly obtains the tree measurement factor based on the image measurement method to reduce the labor intensity of field investigation.Improve the efficiency of forest resource surveys.The research contents of this thesis include foreground extraction of tree image,extraction of tree trunk and canopy,image correction of camera tilting,establishment of tree 3D model and extraction of tree parameters based on photogrammetry.Firstly,the problem of complex and irregular structure of tree image is proposed.Close-Form and K-means tree foreground extraction method,which can calculate the foreground and background of trees under a small number of user tags.Then use GMM and region growing algorithm to segment the canopy and trunk image.The method introduces the measure of cosine similarity and makes full use of GMM clustering probability information.Then order to preserve the linear features of the image,the homography matrix is derived to perform perspective transformation on the image,and the oblique photographic image is corrected to the front view image.Then,the three-dimensional model building method of trees is studied.In order to obtain more realistic tree skeleton information,the tree skeleton is added by sketch method,and the tree trunk and secondary branch topological structure are extracted,and the skeleton topological relationship and the crown convex hull are established to be similar to the real tree shape.3D model of the trees.Finally,the image-based tree parameter extraction method is studied.The commonly used tree-measuring factor and leaf area index are selected to study.According to the camera imaging model,the conversion relationship between the pixel coordinates of the oblique image and the real space coordinate is obtained,which is the accuracy of the verification method.The collected 35 tree samples were tested experimentally.The average absolute error of the three tree-measuring factors was 0.23m,0.17m,and 0.97cm,and the average relative accuracy was 5.74%.,6.53%,8.62%.In order to reflect the growth and growth of the trees,the density parameters of the canopy were defined.The tree factor and the density of the canopy were used as inputs.The BP neural model was used for regression and the leaf area index was inverted.The experimental results show that compared with the BP neural model model with input only tree factor,the correlation coefficient of the BP model with increased dendritic density as input is improved,and the mean square error is reduced.The mean square error of the test set reaches 0.24,and the correlation coefficient reaches 0.77.The image-based tree measurement method proposed in this paper meets the accuracy requirements of forestry surveys,and multiple parameters can be obtained in one acquisition,which has certain applicability.Because the tree growth has regional consistency,the leaf area index inversion model proposed in this paper has certain reference.For urban street trees and plantations,the leaf area index can be inverted according to the tree measurement factor,which is convenient for quantitative analysis of plant communities.