Study On 3D Reconstruction And Growth Measurement Methods For Leafy Crops Based On Depth Camera

Non-destructive crop measurement based on 3D machine vision has great value for crop growth and health research.The technique shows advantages in accuracy,objectiveness,automation,and reflecting crop spatial morphology.This paper studied on plant 3D reconstruction and growth measurement methods,using Kinect v2 as major equipment and leafy crops as main objects.The main contents and conclusions are as follows:(1)This paper proposed a 3D reconstruction platform and the corresponding method for non-destructive plant growth measurement,on the basis of Kinect performace analysis.Kinect was used to measure the potted crop which is on a truntable that turned intermittently.Multiview crop 3D point clouds were obtained automatically by a series of steps including repeated sampling,poing cloud preprocessing,turntable recognition and background segmentation,marker detection,point cloud registration,and outlier removal.The method shows good accuracy,robustness and usability,and low cost.(2)A multiview point clouds denoising method was proposed in this paper to reduce noise for crop point clouds and solve the layered problem of leaf point clouds.Noise reduction was performed by eliminating the interferences among point clouds,based on an iterative process.The method can preserve local features for point clouds while reducing noise,and it can fix the layered leaf point clouds effectively.The method shows good robustness and can preserve topology relationships for point clouds,but it has a time-consuming problem.Qualitative and quantitative experiments with simulated and practically obtained data showed that,the proposed method has overall advantages than TSDF(truncated signed distance function)and MLS(moving least squares),while the existing low noise can be further reduced by MLS.(3)This paper achieved the automatic measurement of relative plant height,(two kinds of)absolute plant height,total leaf area,projected leaf area and volume,based on the steps including crop point cloud segementation,denoising,trigulation and tetrahedralization,and pot feature recognition.Fine 3D point clouds and mesh models of crops can also be obtained in the process.The batch experiment for butter lettuce showed that:the plant height and projected leaf area measured by the proposed method both have good linear relationships with reference values,while the total leaf area and volume both have power function relationships with reference values,and all these data have good fitting results;besides,the total leaf area and volume also have power function relationships with biomass.(4)DM(Data Matrix)code was used in this paper for crop identification.After recovering the distorted DM code image,crop information can be read,and the point clouds captured in different times for the same crop can be aligned.The programmed software "Crop 3D reconstruction and growth information management system" integrated all the methods used by this paper,with the functions of crop data acquisition,processing and management,and dynamic growth visualization and parameter customization.The continuous measurement experiment for multiple kinds of crops showed that,all the obtained growth parameters can well reflect the growth danamics of crops.The plant height measurement shows favourable accuracy but the accuracy of projected leaf area is not so good,while the measurement accuracy of total leaf area,volume and biomass still remains untested.The established prior models have certain effects on adjusting most growth parameters,but are invalid for a small number of crops.In addition,the obtained 3D point clouds and mesh models also well reflected the danamic changes of crop morphology.The automatic crop 3D reconstruction and growth parameter measurement system proposed in this paper shows favourable accuracy,integration level,usability and universality,and the cost is low,thereby demonstrating good prospects for application.