基于计算机视觉的玉米果穗三维重建方法  

Three Dimensional Reconstruction of Maize Ear Based on Computer Vision

作　　者：王传宇[1] 郭新宇[1] 吴升[1] 肖伯祥[1] 杜建军[1] 

机构地区：[1]北京农业信息技术研究中心,北京100097

出　　处：《农业机械学报》2014年第9期274-279,253,共7页Transactions of the Chinese Society for Agricultural Machinery

基　　金：'十二五'国家科技支撑计划资助项目(2012BAD35B01);农业部行业科技计划资助项目(201203026);北京市农林科学院自主创新专项资助项目(KJCX201104011)

摘　　要：提出了一种快速、准确、自动化的基于计算机视觉的玉米果穗三维重建方法。以一定角度间隔旋转果穗获取各视角下的图像,通过双目立体视觉技术重建各视角下的玉米果穗表面点云,计算重投影误差去除点云中的外点,寻找两相邻图像的对应匹配点,并由匹配点确定果穗表面点云中三维配准点的集合,计算两相邻视角下配准点的旋转矩阵与平移向量,采用RANSAC方法检验配准模型的一致性。依次对各视角下的点云配准拼接获得整个果穗表面点云,进行冗余点去除、网格简化、纹理贴图等后处理,获得最终果穗三维造型。实验结果表明:重建模型的体积与实测值不存在显著性差异,所述方法能够满足玉米果穗三维重建的需求。An approach for rapid, accurate and automatic 3D reconstruction of maize ear based on computer vision was presented. Firstly, we rotate the maize ear in a proper angle interval to acquire images in different views, and then calculate out points cloud of maize ear surface with binocular stereovision. Secondly, we eliminate outliers according to the threshold of reprojection error, find out 2D matching points in two neighboring images, determine the 3D matching points set of points clouds of maize ear surfaces by the 2D matching points, calculate the rotation matrix and translation vector of the matching points between two neighboring views, and test the consistency of the 3D registration model by RANSAC method. Finally, by rotating and translating each point cloud of different views to stitch the whole maize ear surface, eliminating the redundant points, simplifying the mesh, and mapping the texture, we get the final 3D shape of maize ear. The experiment results show that the volume of the 3D reconstruction maize ear has no significant difference from the measurement value, and the method proposed can meet the need of 3D reconstruction of maize ear.

关 键 词：玉米果穗 三维重建 双目立体视觉 3D 匹配 SIFT 算法 RANSAC 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]