| 摘要: |
| 针对人工测量作物表型结构参数不够准确的问题,提出基于双目视觉的作物表型参数提取系统. 利用鸡毛菜的叶面积和平均叶倾角两个重要表型参数,通过采集鸡毛菜原始RGB图和深度图,将原始信息合成为颜色点云后进行预处理. 采用超体素聚类的分割算法将每片叶片从作物点云中分离,并改进贪婪投影三角剖分算法,获得最佳表面重建效果,实现网格模型的颜色渲染. 在VTK库中完成网格模型的优化,获得真实感较强的鸡毛菜网格模型.在网格模型中实现对两个参数的提取,并与人工测量值比较. 对于第四组大于4 cm的叶片,自动提取的叶倾角平均绝对误差小于5.5°,验证了自动化无损监测鸡毛菜作物的可行性. |
| 关键词: 农业工程 双目视觉 表型参数 立体相机 三维分割 表面重建 点云库 |
| DOI：10.20192/j.cnki.JSHNU(NS).2024.04.004 |
| 分类号:TP391.41 |
| 基金项目:上海市自然科学基金（20ZR1440500）；上海市青年科技英才扬帆计划（18YFl411000）；上海师范大学一般科研项目（SK202123） |
|
| Point cloud based phenotypic parameter extraction of Chinese white cabbage |
|
LIU Xiangpeng1, LU Wei1, WANG Danning1, ZHENG Jiafeng2, PENG Yulin1, AN Kang1
|
|
1.College of Information, Mechanical and Electrical Engineering, Shanghai Normal University, Shanghai 201418, China;2.School of Mechanical Engineering, Shanghai Jiao Tong University, Shanghai 200240, China
|
| Abstract: |
| A phenotype parameter extraction system was proposed to replace the inaccurate manual measurement. Leaf area and average leaf inclination angle were utilized by collecting the original RGB image and the deep-seated image of Chinese white cabbage. The data was then synthesized and passed into a color point cloud for preprocessing. Segmentation based on hypervoxel clustering was applied to separate each leaf from the crop point cloud, and greedy projection triangulation was adopted to obtain the best surface reconstruction. As a result, color rendering for the mesh model was realized, for which the optimization was completed in the VTK library to obtain a more realistic model. The extraction of two parameters was realized in the grid model, which was compared with the manual solution. For the 4th group of leaves larger than 4 cm, the absolute error of the average inclination angle was less than 5.5°, which verified the feasibility of automated non-destructive monitoring for Chinese white cabbage. |
| Key words: agricultural engineering binocular vision phenotypic parameter stereo camera 3D segmentation surface reconstruction point cloud library |
