Design Of Panoramic 3D Reconstruction System And Phenotypic Measurement Research For Cabbage Leaves

Cabbage,as one of the important vegetable crops,has a high nutritional value.The leaves are the main organ for photosynthesis and growth and development of the body.Whether in production or new variety breeding,leaf phenotype contains key information,and the quantitative acquisition of its phenotypic traits is of great significance.To address the problems of incomplete reconstruction and low efficiency of traditional 3D reconstruction methods,as well as the difficulty of efficient and accurate manual collection of crop phenotypic information,this paper designs a panoramic 3D reconstruction system for cabbage leaves,quantitatively extracts the reconstructed leaf 3D model phenotypic parameters and evaluates the accuracy and error of the extracted phenotypic parameters.The specific research contents and results of this paper are as follows:(1)A platform for automatic acquisition of panoramic image sequences was developed independently.The platform mainly consists of two industrial color cameras,a speedadjustable turntable that drives the rotation of the cameras with stepper motors as the main body,and an electric control part that controls the motor parameters.In order to establish a three-dimensional model that can completely reflect the cabbage blade,a transparent glass plate carrier is placed in the middle of the platform,which is able to collect the panoramic image information in the upper and lower directions of the blade.According to the number of image frames required for leaf point cloud reconstruction,the appropriate camera photo interval and turntable rotation speed are adjusted,and the acquired leaf image sequences are automatically stored into the computer.(2)Structure From Motion(SFM)method is used to reconstruct the 3D point cloud of the blade image sequence,and a point cloud pre-processing method is designed.First,the original point cloud of the blade is filtered by direct-pass filtering and statistical filtering to remove most of the background and noise point clouds,then it is downsampled to reduce the number of point clouds to improve the efficiency of subsequent point cloud processing;finally,the point clouds of the blade generated in both directions are scaled based on Principal Component Analysis(PCA)to make their Finally,the leaf point clouds generated in both directions are scaled based on Principal Component Analysis(PCA)to make them the same size.(3)We designed a leaf point cloud alignment stitching method and leaf phenotype parameter extraction method to obtain the leaf panoramic 3D point cloud model and analyzed the accuracy and error of the reconstructed model.First,markers are placed when the images are taken,and the 3D point cloud model is generated together with the blade.The 4-Points Congruent Sets(4PCS)and PCA methods are used for initial alignment of the marker point clouds,and the results show that the 4PCS algorithm works better.Secondly,the Iterative Closest Point(ICP)algorithm is used to finely align the marker point cloud after the initial alignment,and the transformation matrix is obtained.Since the relative positions of the blade and the markers are constant,the transformation matrix is applied to the blade point cloud to complete the alignment of the blade point cloud and obtain the final blade panoramic 3D point cloud model.Finally,coordinate transformation is performed on the generated leaf point cloud model to generate the point cloud minimum enclosing box and obtain the leaf length,leaf width,leaf height and leaf minimum enclosing box volume,which have a significant linear relationship with the corresponding manually measured true values,and the fitted coefficients of determination R2 are 0.9198,0.9213,0.9667 and 0.9672,respectively.the results show that the obtained cabbage leaf 3D point cloud model has high measurement accuracy.The method proposed in this paper can not only reconstruct the panoramic 3D point cloud model of cabbage leaves with good visual effect,but also extract the phenotypic information with high accuracy.In addition,the applicability verification results of the 3D reconstruction system in this paper show that it can also be used to reconstruct the panoramic 3D point cloud models of other small and medium-sized objects with high data acquisition efficiency and low cost,which can provide a reference for 3D reconstruction-related work.