Three-dimensional Reconstruction And Character Extraction Of Corn Plants Based On Kinect Sensor

Corn is an important food crop and an important source of feed,with high economic value.Maize plant height and stem thickness reflect plant growth speed and robustness to a certain extent,and leaves have significant effects on maize yield and disease resistance.Therefore,accurate acquisition of maize phenotypic parameters is of great significance for understanding crop growth,crop yield estimation,disease resistance detection,and breeding.At present,crop information collection has problems such as lack of collected data information,low authenticity,expensive collection equipment and inconvenient carrying,etc.,which is not suitable for agricultural promotion;corn leaf segmentation damages the stalk structure,and results in corn leaves with stalk information,Affecting the accuracy of leaf segmentation;most of the surveys of crop traits still use traditional manual measurement methods,which have problems such as strong subjectivity,cumbersome measurement process,lossy,and discontinuity measurement.In view of the above problems,this study is based on the Kinect sensor to study the methods of maize plant three-dimensional reconstruction,point cloud feature segmentation and phenotypic parameter extraction.The main research contents and results are as follows:(1)Point cloud acquisition and 3D reconstruction of corn plants.Aiming at the problem of high equipment price and low accuracy,comprehensively considering the data acquisition effect and price of equipment,this study uses the three-dimensional scanning equipment Kinect to collect corn point cloud data.The corn plant is scanned by the turntable method.During the scanning process,the Kinect is fixed and the corn is placed in the center of the rotating table.Each time the rotating table rotates at a fixed angle of 36°,Kinect obtains the current view angle data of the corn plant.The turntable rotates 9 times to obtain completeness.Corn data.Remove background information from the collected corn point cloud,use statistical filtering to denoise,and use manual registration and ICP algorithm to perform coarse and fine registration respectively on the denoised point cloud data.At this time,the complete point cloud data is obtained,and the registered point cloud data is simplified by various methods,and the optimal simplification method is selected to use the grid method to simplify the point cloud.(2)Ellipse fitting and segmentation of maize leaves.Aiming at the problem that the current research on plant point cloud segmentation methods is not targeted and the segmentation accuracy is low,this research proposes a new corn leaf segmentation method.To facilitate subsequent point cloud processing,rotate and translate the simplified complete corn point cloud so that the main direction of the corn is consistent with the direction of the stem.Using the corn point cloud coordinate information,the straight-through filtering method is used to remove the corn flowerpot point cloud.After removing the flowerpot,the least square method is used to fit the ellipse of the stalk in the plant to extract the stalk point cloud.Then filter the non-stalk point cloud to remove the noise and extract a single leaf based on the region growth segmentation algorithm.The corn leaf segmentation method effectively solves the problems of over-segmentation and insufficient segmentation caused by direct leaf segmentation of the point cloud.(3)The measurement and precision analysis of phenotypic parameters of maize plants.Due to the low efficiency of manual measurement,the continuity of plants cannot be measured.In this study,an algorithm was used to obtain maize phenotypic data.Use point cloud maximum value traversal,ellipse fitting equation,and region growth segmentation results to obtain the plant height,stem thickness,and leaf number of the corn 3D model,use triangle patching to obtain the corn leaf area and perimeter,and use plane fitting Obtain the leaf angle and curling degree of corn plants.By comparing the measured value of the algorithm with the manually measured value,the accuracy of the algorithm to measure the plant height is 97.622%,the average relative error of the long axis of the stalk is 9.46%,and the average relative error of the short axis of the corn stalk is 11.17%.The area accuracy is 95.577%.A series of experiments have proved that this method can accurately construct the three-dimensional shape of corn,segment corn leaves and extract crop phenotypic parameters non-destructively and accurately,thereby providing methods and higher precision and larger data for corn phenotyping research.Support and guide actual crop production at the same time.