基于多源信息融合的果树冠层三维点云拼接方法研究  

Three Dimensional Point Cloud Splicing of Tree Canopy Based on Multi-Source Camera

作　　者：周薇[1,2] 马晓丹[1,3] 张丽娇[1] 郭彩玲[1] 刘刚(指导)[1] 

机构地区：[1]中国农业大学现代精细农业系统集成研究教育部重点实验室,农业部农业信息获取技术重点实验室,北京100083 [2]福建省特种设备检验研究院,福州350008 [3]黑龙江八一农垦大学信息技术学院,大庆163319

出　　处：《光学学报》2014年第12期185-192,共8页Acta Optica Sinica

基　　金：国家自然科学基金(31371532);中央高校基本科研业务费专项资金(2013YJ008)

摘　　要：构建了基于彩色相机和光学混合探测(PMD)相机的多源视觉系统,旨在建立具有真彩色信息的果树冠层三维点云模型,为果树的剪枝、疏花疏果和采摘等果园管理提供技术支持。针对PMD相机获取的目标场景三维点云,结合PMD相机的幅度图像和密度聚类算法提取有效点,利用前期研究的图像配准方法得到多源图像之间的坐标转换关系,完成了果树冠层多源信息融合。通过主成分分析法得到较好的初始位置,再采用最近点迭代算法,实现两组三维点云之间的拼接。对自然场景下的开花期和坐果期的果树冠层三维点云拼接方法进行了实验验证,结果表明多视角三维点云拼接误差为2.62cm,可以较好地弥补单个角度下拍摄造成的数据缺失,实现了果树冠层完整的三维显示。In order to guide the pruning, flower thinning and harvesting of fruit trees in orchard, a novel vision system which combines a color-camera system with a photo mixing detector （PMD）-camera is constructed. For the three-dimensional coordinate information of target scene acquired by the PMD camera, effective point cloud combining PMD amplitude image with density-based spatial clustering of applications with noise （DBSCAN） algorithm is extracted. Multi-source information fusion is completed with the result of image registration in the previous studies. Primary component analysis algorithm （PCA） is used to get the initial state of the point cloud at different locations, which is called prealignment. Accurate splice between two point clouds is realized by the iterative closest point （ICP） algorithm based on the least square method to get the optimal matching. Coordinate transformations are obtained by singular value decomposition （SVD） after prealigment and accurate splice. Several groups of experiments are used for verification, which show the average error of multi-view point cloud splicing reaches 2.62 cm and can better make up full three dimensional display of apple tree canopy without missing data than a single angle shot.

关 键 词：机器视觉 果树冠层 信息融合 三维点云拼接 最近点迭代法 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]