| With the development of information technology,precision planting systems that overcome time and space constraints are developing rapidly.The precision planting system can formulate corresponding planting plans by analyzing plant information,improving the utilization rate of water and fertilizer,and crop yields.Crop monitoring is the foundation of precision agriculture,and visionbased 3D reconstruction is an effective way to accurately obtain plant information.However,plants have complex morphology and structure,and the branches shield each other,which affects the effect of three-dimensional reconstruction of plants.Obtaining three-dimensional models of plants quickly and accurately is a hot and difficult point in recent years in the field of computer and precision planting.In this paper,how to quickly and efficiently obtain plant 3D models and improve the quality of reconstructed plant models is the main research goal,analyze the current problems in plant 3D reconstruction and common technical methods,and propose a research on plant 3D reconstruction based on point cloud data.The framework and feasible methods are researched on key technologies such as point cloud data acquisition and point cloud complementation of plants.Based on the point cloud generation model,an orchard planting system combined with virtual reality is constructed to realize remote planting.The main work of this paper is as follows:1.Determine the overall process from point cloud to model.First,use the algorithm of motion recovery structure to recover the plant point cloud data,and the recovered plant point cloud is incomplete.Then use the deep learning algorithm to complement the point cloud to obtain more complete point cloud information.After the complete point cloud is obtained,mesh reconstruction is performed to obtain a more accurate three-dimensional model of the plant.2.Obtain the plant point cloud.Taking wolfberry as an example,we carried out corresponding image acquisition and three-dimensional reconstruction of multi-view images of wolfberry plants,explored different image acquisition methods,and determined a collection method suitable for wolfberry plants.The dense point cloud obtained by this method is of higher quality.It is the basis for subsequent reconstruction work.3.Aiming at the incompleteness of the point cloud in the 3D reconstruction of the image sequence,a method of filling the shape of the plant point cloud based on deep learning is adopted.Combine reinforcement learning and generative confrontation network to solve the problem of lack of plant point cloud information.4.Designed a virtual reality orchard planting system.The reconstructed plant model is dynamically displayed to the user through virtual reality technology.Each plant can display detailed information,allowing the user to accurately understand the growth status of the crop and remotely control the orchard equipment,To help users to plant accurately. |
