BonnBeetClouds3D

Agricultural production is facing severe challenges
in the next decades induced by climate change and the need
for sustainability, reducing its impact on the environment.
Advancement in field management through non-chemical weed-
ing by robots in combination with monitoring of crops by
autonomous unmanned aerial vehicles (UAVs) and breeding
of novel and more resilient crop varieties are helpful to
address these challenges. The analysis of plant traits is called
phenotyping, and is an essential activity in plant breeding, it
however involves a great amount of manual labor. With this
paper, we address the problem of automatic fine-grained organ-
level geometric analysis needed for precision phenotyping.
However, the availability of real-world data for such fine-
grained perception tasks in this domain is relatively scarce
compared to other domains such as autonomous driving. To
work towards closing this gap, we propose a novel dataset
that was acquired using UAVs capturing high-resolution im-
ages of a real breeding trial. This has the big advantage of
containing a multitude of plant varieties, leading to a great
morphological and appearance diversity covered by our dataset.
This enables the development of approaches for autonomous
phenotyping that generalize well to different varieties. Based
on overlapping high-resolution images from multiple viewing
angles, we compute photogrammetric dense point clouds via
bundle adjustment that capture the geometric structure of
the plants. We provide detailed and accurate point-wise labels
for individual plants, individual leaves, salient points on the
leaves such as the tip and the base. Additionally we include
measurements of phenotypic traits performed by experts from
the German Federal Plant Variety Office (”Bundessortenamt)
on the real plants, allowing to evaluate approaches not only
on segmentation and keypoint detection, but also directly
on the downstream tasks. The provided labeled point clouds
enable fine-grained plant analysis and opens the door for
further progress in the development of automatic phenotyping
approaches, but also enable further research in closely related
application areas such as surface reconstruction, point cloud
completion, and semantic interpretation of point clouds.

未来数十年间，气候变化与农业可持续发展的需求正给农业生产带来严峻挑战，同时还需降低农业生产对环境的负面影响。通过机器人开展非化学除草、结合自主无人机（Unmanned Aerial Vehicles, UAVs）对作物进行监测，以及培育新型高抗逆作物品种等田间管理技术的进步，有助于应对上述挑战。对植物性状的分析称为表型组学（phenotyping），是植物育种中的核心环节，但该过程往往需要耗费大量人力。本文针对精准表型组学所需的自动细粒度器官级几何分析问题展开研究。然而，相较于自动驾驶等其他领域，该领域内适用于此类细粒度感知任务的真实世界数据集相对匮乏。为填补这一研究空白，我们构建了一套全新的数据集：该数据集通过无人机采集真实育种试验田的高分辨率图像生成。其显著优势在于涵盖了多种植物品种，因此数据集具备丰富的形态学与外观多样性，能够支撑开发可泛化至不同品种的自主表型分析方法。基于多视角重叠的高分辨率图像，我们通过光束法平差（bundle adjustment）计算得到摄影测量密集点云，以完整捕捉植株的几何结构。我们为单株植物、单张叶片以及叶片上的关键点（如叶尖与叶基）提供了精细准确的逐点标注。此外，数据集还包含了由德国联邦植物品种办公室（Bundessortenamt）的专家对真实植株开展的表型性状测量结果，使得相关研究不仅可在分割与关键点检测任务中开展评估，还可直接针对下游任务进行验证。本次发布的带标注点云数据集既支持细粒度植物分析，为自动表型分析方法的发展带来进一步突破空间，也为表面重建、点云补全以及点云语义解释等密切相关的应用领域的研究提供了重要支撑。