Three-dimensional Reconstruction Of Soybean Plants Based On Point Clouds

With the rapid development of digital agriculture,three dimensional reconstruction technology of crops has been applied more and more widely in agriculture.As the simplest element,point can record the three-dimensional information on the surface of the object,and then express the shape of the object more accurately,so that the object can be analyzed and processed quickly.Therefore,it is particularly necessary to study the technology of 3D point cloud processing and reconstruction.Three-dimensional reconstruction technology can be more rapid and direct to observe the physiological and biochemical problems of crops,and to solve them.Nowadays,there are few studies on the three-dimensional reconstruction technology of soybean plants,especially for the research of the reconstruction of soybean plants.At present,most of the instruments used in 3D reconstruction are most expensive or large,and some of the methods for point cloud processing are complicated or imprecise.In order to solve these problems,this paper takes the soybean plant as the research object,taking the cheap and portable Kinect as the test instrument,studies the method of point cloud data preprocessing,denoising,and the method of fast three-dimensional reconstruction of point cloud data,in order to realize the selection of better methods and the high quality reconstruction,and get better method reconstruction.It is convenient for other researchers to analyze the indicators of soybean plants and carry out more intuitive and convenient research.The main research work and conclusions are as follows:(1)In order to preprocess a large number of three dimensional soybean point clouds,theobtained point cloud data contains a large amount of redundant data.Redundant data not only occupy most of the storage space,but also reduce the efficiency of processing.For the redundant point cloud,the point cloud is streamlined by the octree method,and the point cloud data is reduced by setting different thresholds.Through the experiment,it can be obtained that the shape characteristics of the point cloud data can still be retained effectively when the reduction rate is 95.78%.(2)For some outlier cloud data of soybean point cloud data,and the problem of uneven surface of point cloud data will affect the subsequent reconstruction effect,a denoising smoothing algorithm based on Adaptive Density Clustering and bilateral filtering is studied.First,the method uses the density clustering method to remove some point cloud data which is far away from the target point cloud,and then uses a bilateral filtering method to smooth the point cloud data.The reduced threshold value increases from 10 to 80 and the de-noising smoothing time decreases from 100.765 s to 1.655 s,which can effectively remove the outlier noise of the point cloud data and smooth the surface of the data.(3)Three dimensional reconstruction of soybean point cloud is realized by Bowyer-Watson algorithm in three-dimensional Delaunay triangulation,and the effect of reconstruction is analyzed and compared.Three dimensional reconstruction of original point cloud and point cloud data with different processing degree is carried out.The experimental results show that the reduced point cloud data can improve the efficiency of 3D reconstruction,and the de-noising smoothing is better than that of the point cloud without noise smoothing.The point cloud data processed by point cloudsimplification and denoising and smoothing can effectively reconstruct three dimensional reconstruction under the condition of keeping soybean morphological characteristics well.The high quality reconstruction of soybean plants can provide a basis for the reconstruction of crops in the future,and provide an important theoretical basis for the study of physiological and biochemical indexes of soybean plants.