Design And Experiment Of Corn Plant 3D Reconstruction System Based On Consumer-grade Depth Camera

At present,China is in a critical period of agricultural modernization,and the threedimensional digital model of crops is an important foundation for the development of agriculture in the direction of modernization.Nowadays,there are few schemes for 3D reconstruction of crop plants,and most of them are based on expensive professional 3D scanning equipment,which is difficult to achieve universal application.Due to the complex surface geometric characteristics of crop plants,the target corn plants are extracted during the process of 3D reconstruction of crop plants.Unsatisfactory point cloud effect,inaccurate recognition of outliers in the process of noise reduction smoothing,low noise point removal rate and transition noise reduction,poor registration effect between multi-view local point clouds of corn plants,and corresponding points are far away Problems with local registration weight being too large.In response to these problems,this study took corn plants at the flowering and kernel stage as the test object.First,a set of acquisition devices based on consumer-grade depth cameras and automatic control turntables were developed to automatically acquire data;and then a new method based on cone surface fitting was proposedThe ICP registration algorithm realizes the fast and high-quality registration of the point cloud;finally,the Delaunay triangulation algorithm based on the Bowyer-Watson method is inserted point by point to triangulate the point cloud.The corn plant 3D reconstruction system based on consumer-grade depth cameras proposed in this research provides a theoretical and experimental basis for low-cost,fast and efficient reconstruction of corn plant 3D models,which is conducive to the popularization and application of corn plant 3D reconstruction in the field of agricultural information technology..The main research contents and conclusions are as follows:(a)Developed a set of fully automatic corn plant depth data collection system without markers.The acquisition system is based on the consumer-grade depth camera Kinect V2 to realize simple and high-accuracy data acquisition of experimental corn plants.The results show that the error between the center of the flowerpot bottom and the rotor center of the turntable stepper motor is within 1.5mm.Excluding the stabilization delay time,the system only needs55.3s to collect 10 sets of data,and the average time for one rotation and data collection is 5.7s.High-speed acquisition.(b)A straight-through filtering algorithm for corn plant point cloud extraction based on limit point coordinates is proposed.This algorithm uses the determination of the limit point value to delete the environment point through the filter,which reduces the complexity of distinguishing the target plant point cloud data from the environmental point cloud,improves the extraction speed of the target point cloud,and the redundant point cloud removal rate reaches 95.51%.On the basis of ensuring that all the point clouds of the target corn plants are retained,the environmental point clouds and noise points are effectively removed.(c)A local conical surface fitting-ICP algorithm with interval registration method is proposed.Through conical surface fitting of the local flowerpot wall point cloud,the flowerpot rotation axis is obtained as the turntable rotation axis,and the point cloud is coarsely registered and rotated according to the rotation angle.The use of interval registration for ICP fine registration effectively reduces the accumulation of errors in the registration of 10-angle point clouds.Experiments show that after coarse registration,the average distance of the corresponding point error of the point cloud is 0.0067 m,the full-angle registration takes 57.32 s,and the average distance difference of the corresponding points is 0.001981 m,which has high registration efficiency and accuracy.(d)The Delaunay triangulation algorithm of corn plant point cloud based on BowyerWatson method is selected.Aiming at the characteristic that the corn plant point cloud is a disordered point cloud and the stem part is a closed model,the Delaunay triangulation algorithm inserted point by point by the Bowyer-Watson method is selected to encapsulate the point cloud by a triangular mesh.The results show that the relative error of the key parts of the maize plant3 D model reconstructed by this research is controlled within 11.51%,and the morphological data has good reliability,and the reconstructed model has a reasonable structure ratio..