对光照变化鲁棒的温室作物立体视觉成像算法研究

3D imaging and scanning of greenhouse plants can not only provide intelligent greenhouse control systems with direct visual cue, but also contribute to developing of high-throughput plant phenotyping. For the purpose of tackling the problems such as unstable illumination environment, complicated 3D structure and insignificant texture of plants, and meanwhile satisfying the needs of facility agriculture on cost, efficacy, and real-time performance, this proposal studies stereo matching algorithms that fit for unstable greenhouse illumination condition and complicated plant structures. This research also place interest on designing a reliable 3D point cloud generation method and carrying out plant organ analysis that follows. This research will explore the following innovative scientific topics: first, study illumination-robust stereo matching method for greenhouse plants with frequent depth variations; second, design sub-pixel accuracy point cloud enhancement technique to improve the accuracy of plant organ recognition based on point clouds. This research can serve greenhouse cultivation of our country; can also help to raise the output and quality of agricultural products, and finally will promote the development of 3D digitization and virtual visualization technology.

对温室作物进行三维成像与扫描不仅能为智能温室控制系统提供直观的控制依据以改进栽培，还对推进高通量植物表型学的发展具有重要作用。为了应对温室中不稳定的光环境、作物复杂的三维结构与并不显著的纹理特征等问题；也同时为了满足设施农业对成本、效率、实时性方面的要求，本课题依靠立体视觉成像技术这一类前沿的三维数字化成像方法，研究适用于复杂作物结构和温室生长环境的立体视觉匹配算法，并设计可靠的点云生成方法与作物器官分析方法。课题将探索以下具有创新性的科学问题：首先，研究具有光照变化鲁棒性、并适用于深度频繁跃变的作物表面立体视觉匹配算法；其次，研究次像素级精度的点云改善技术以提高作物器官识别的准确性。本课题的研究服务于我国温室栽培技术，有助于提升设施农产品品质和产量；也将推动三维数字化成像和虚拟可视化技术的进步和发展。

本基金项目基于计算机图像图形技术对作物冠层点云进行成像和自动表型算法的研究，其最终目标是实现真正的智能设施农业，减少育种环节中繁琐的人工劳动。本项目完成了5套作物点云成像平台的建设、设计了一种温室作物点云成像算法和一种植物点云单叶片分割和分析算法。形成了多套常见作物品种（如：番茄、绿萝）的叶片器官数据库。完成了一套温室作物的三维点云表型分析可视化软件，该软件与基于Kinect阵列的点云采集硬件平台所配套，具有叶片分割、叶面积计算、叶倾角计算等功能。发表高水平论文7篇，其中第一作者SCI论文5篇，第一作者EI国际会议论文1篇，通信作者EI国际会议论文1篇。获得一项发明专利授权，完成一项发明专利的申请。参加了两次国内学术会议CCC2017与CCDC2017，参加一次国际学术会议WCAM2019。在本项目资助下，共指导本科毕业设计22项，指导硕士生8人。