WO2016074154A1 - 植物扫描与重建方法 - Google Patents

植物扫描与重建方法 Download PDF

Info

Publication number
WO2016074154A1
WO2016074154A1 PCT/CN2014/090820 CN2014090820W WO2016074154A1 WO 2016074154 A1 WO2016074154 A1 WO 2016074154A1 CN 2014090820 W CN2014090820 W CN 2014090820W WO 2016074154 A1 WO2016074154 A1 WO 2016074154A1
Authority
WO
WIPO (PCT)
Prior art keywords
leaf
plant
skeleton
point cloud
leaves
Prior art date
Application number
PCT/CN2014/090820
Other languages
English (en)
French (fr)
Inventor
黄惠
尹康学
Original Assignee
中国科学院深圳先进技术研究院
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 中国科学院深圳先进技术研究院 filed Critical 中国科学院深圳先进技术研究院
Priority to PCT/CN2014/090820 priority Critical patent/WO2016074154A1/zh
Publication of WO2016074154A1 publication Critical patent/WO2016074154A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects

Definitions

  • the invention relates to the field of three-dimensional plant modeling technology, in particular to a method of plant scanning and reconstruction.
  • 3D plant modeling is an important and widely used research topic. In the game design and development, the quality of the plant model in the scene directly affects the game's realism and user experience. In the field of botany, three-dimensional plant modeling can be used to study plant growth and behavior in different physical environments. In agriculture, three-dimensional plant models contribute to the study of pest control and fertilization methods.
  • Embodiments of the present invention provide a plant scanning and reconstruction method for reducing user interaction and improving scanning and reconstruction accuracy, and the method includes:
  • the point cloud data of each leaf is reconstructed into a surface model, and the reconstruction result of the stem is obtained by Poisson reconstruction algorithm;
  • the point cloud data of each leaf is reconstructed into a surface model, including:
  • the slice is divided into two parts: the blade and the petiole;
  • a curvature-based quadratic distance minimization method is used to fit a NURBS (Non-Uniform Rational B-Splines) curve to each slice, the blade is fitted with a non-closed curve, and the petiole fitting is closed. curve;
  • NURBS Non-Uniform Rational B-Splines
  • the plant is scanned in its entirety using a hand-held structured light 3D scanner and each leaf of the plant is scanned separately.
  • the optimal location for all NURBS control points is optimized, including:
  • E data is the accumulation of the closest distance from the point in the point cloud to the NURBS curve
  • E smooth is the roughness of the curve formed by the control points of the same identification ID on different NUBRS curves
  • E bound is the accumulation of the nearest distance from the non-closed NURBS endpoint to the blade point cloud boundary
  • E round is the perimeter area ratio of each closed NURBS curve
  • ⁇ , ⁇ , ⁇ are constant.
  • all of the reconstructed leaves and stems are aligned with the overall scan data of the plant, including:
  • the control object is changed from the skeleton to the control point uniformly sampled by the WLOP algorithm, and the non-rigid registration is performed again.
  • the skeletal drive blade and stem deformation are used for non-rigid registration, including: optimally solving the skeleton point rotation and translation transformation.
  • the optimization solves the skeleton point rotation and the translation transformation, including:
  • the BFGS algorithm is used to minimize the objective function to obtain the optimal transformation defined on the skeleton, and the optimal transformation is applied to the blade and stem model, and the objective function is defined as the deformed blade to the point cloud.
  • control object is changed from a skeleton to a control point uniformly sampled by the WLOP algorithm, and the non-rigid registration is performed again, including:
  • the Laplacian smoothing operator is moved from the skeleton to a grid of control points interconnected.
  • the point cloud data aligned with all the leaves is merged together, including:
  • the point cloud data aligned with all the leaves is merged together, including:
  • the end of the leaf skeleton is deformed to a point on the stem closest to the end of the leaf skeleton, and the deformation of the skeleton is used to drive the deformation of the leaf to achieve the connection of the stem and the leaf; if there is no stem, the tip of the petiole is automatically detected from the other petiole The distance, if the distance is less than the threshold, the deformation of the skeleton is used to drive the deformation of the leaves to achieve the connection of the leaves.
  • each leaf is separately scanned and reconstructed, and can be scanned into each part of the plant, thereby overcoming the problem that the blade cannot be scanned by mutual occlusion, and accurate and complete point cloud data can be obtained, thereby obtaining good results.
  • Reconstruction accuracy; the reconstruction process for each leaf is completely automatic and does not require interaction, which greatly reduces the user's workload.
  • a hand-held structured light 3D scanner to complete plant scanning will significantly reduce costs compared to the use of a CT scanner or a high precision laser scanner.
  • FIG. 1 is a flow chart of a method for plant scanning and reconstruction in an embodiment of the present invention
  • FIG. 2 is a schematic diagram of a process of reconstructing a leaf according to an embodiment of the present invention.
  • FIG. 1 is a flow chart of a method of plant scanning and reconstruction in an embodiment of the present invention. As shown in FIG. 1, the method for plant scanning and reconstruction in the embodiment of the present invention may include:
  • Step 101 Perform an overall scan on the plant to obtain overall scan data of the plant
  • Step 102 Perform a separate scan on each leaf of the plant to obtain point cloud data of each leaf;
  • Step 103 reconstruct point cloud data of each leaf into a surface model, and obtain a reconstruction result of the stem by a Poisson reconstruction algorithm
  • Step 104 Align all the reconstructed leaves and stems with the overall scan data of the plant;
  • Step 105 Fusion the point cloud data of all the leaves to obtain a whole plant model.
  • the method of plant scanning and reconstruction in the embodiment of the present invention includes four steps of scanning, reconstruction, registration and fusion.
  • the whole plant is scanned first, then all the leaves can be cut and scanned separately.
  • a reconstruction algorithm specifically designed for the leaves is used in the embodiment of the invention to reconstruct the individually scanned leaves.
  • the registration process which is to align the reconstructed leaf model to the point cloud.
  • the fusion process is about merging the subsequent leaves into a plant model.
  • the scanning process in the embodiment of the present invention is divided into two steps: first, the whole plant is scanned once, then all the leaves are cut and separately scanned. This is a clever divide and conquer strategy. Because the leaves of plants often occlude each other, complete point cloud data is not available for an overall scan. Without complete data, the effect of reconstruction will naturally not be good. Therefore, it is necessary to scan all the blades separately.
  • the strategy of separately scanning and reconstructing each blade that is, the divide and conquer strategy, can improve the scanning and reconstruction accuracy, and the accuracy obtained can even exceed the effect obtained by an expensive CT scanner.
  • FIG. 2 is a schematic diagram of a process of reconstructing a leaf according to an embodiment of the present invention. As shown in FIG. 2, reconstructing the point cloud data of each leaf into a surface model may include:
  • Step 201 Extracting a skeleton of a leaf by using an L1-median algorithm
  • Step 202 Perform vertical sectioning on the point cloud of the leaf along the skeleton
  • Step 203 Divide the slice into two parts of the blade and the petiole according to the aspect ratio of the point cloud slice;
  • Step 204 Using a curvature-based quadratic distance minimization method, fitting a non-uniform rational B-spline NURBS curve to each slice, the blade fitting a non-closed curve, and the petiole fitting a closed curve;
  • Step 205 Optimize the optimal position of all NURBS control points
  • Step 206 Connect all the slice shapes of the leaves to obtain the shape of the leaves.
  • optimally solving the optimal position of all NURBS control points may include:
  • x is the position of the control point where the solution is required;
  • E data is the accumulation of the closest distance from the point in the point cloud to the NURBS curve;
  • E smooth is the roughness of the curve formed by the control points of the same identification ID on different NUBRS curves;
  • E bound is the accumulation of the nearest distance from the non-closed NURBS endpoint to the blade point cloud boundary,
  • E round is the perimeter area ratio of each closed NURBS curve; ⁇ , ⁇ , ⁇ are constant.
  • the NURBS curve defines all the slice shapes of the leaves, and connecting them gives the shape of the leaves.
  • the purpose of the registration is to align all of the reconstructed leaves and stems with the overall scan data of the plant, accurately aligning the individually scanned leaves to the overall point cloud of the plant. Registration can be divided into two steps, rigid registration, and non-rigid registration:
  • the skeleton is used to drive the blade and stem deformation, and the first step of non-rigid registration is performed. Specifically, the skeleton point rotation and translation transformation can be optimized.
  • the BFGS algorithm can be used to minimize the objective function to find the best transformation defined on the skeleton and apply it to the blade and stalk model.
  • the objective function is defined as the distance between the deformed blade and the point cloud + A Laplacian smoothing operator that defines one on the skeleton.
  • the control object can be changed from the skeleton to the control point uniformly sampled by the WLOP algorithm, and the non-rigid registration is performed again.
  • the Laplacian smoothing operator can be moved from the skeleton to a grid in which the control points are connected to each other. Further detail registration can be obtained using the same optimization method as the first step of non-rigid registration.
  • the rigid ICP algorithm and the non-rigid ICP algorithm can also be used for registration.
  • the whole plant model can be obtained by fusing together the point clouds obtained after the registration. Fusion needs to solve two problems: avoiding the intersection of leaves and connecting stems and leaves.
  • the end of the leaf skeleton can be deformed to the point on the stem closest to it, and the deformation of the skeleton is used to drive the deformation of the leaf to achieve the purpose of connecting the stems and leaves. If there is no stem, the distance between the ends of the petiole from the other petiole is automatically detected. If the distance is less than a certain threshold, they are connected by deformation in the same way, that is, the deformation of the leaf is driven by the deformation of the skeleton to achieve the leaf connection.
  • each leaf is separately scanned and reconstructed, and each part of the plant can be scanned, which overcomes the problem that the blade cannot be scanned by mutual occlusion, and accurate and complete point cloud data can be obtained.
  • the reconstruction process for each leaf is completely automatic, no interaction is required, which greatly reduces the user workload. It has been experimentally proved that the embodiment of the present invention has a very good effect on the highly missing point cloud reconstruction.
  • a hand-held structured light 3D scanner to complete plant scanning will significantly reduce costs compared to the use of a CT scanner or a high precision laser scanner.
  • the plant scanning and reconstruction method in the embodiments of the present invention can be applied not only to plants with only leaves, but also to scanning and reconstruction of other objects, such as scanning and reconstruction of flowers, and even scanning and reconstruction of human bodies.
  • embodiments of the present invention can be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or a combination of software and hardware. Moreover, the invention can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) including computer usable program code.
  • computer-usable storage media including but not limited to disk storage, CD-ROM, optical storage, etc.
  • the computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture comprising the instruction device.
  • the apparatus implements the functions specified in one or more blocks of a flow or a flow and/or block diagram of the flowchart.
  • These computer program instructions can also be loaded onto a computer or other programmable data processing device such that a series of operational steps are performed on a computer or other programmable device to produce computer-implemented processing for execution on a computer or other programmable device.
  • the instructions provide steps for implementing the functions specified in one or more of the flow or in a block or blocks of a flow diagram.

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Computer Graphics (AREA)
  • Geometry (AREA)
  • Software Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Image Processing (AREA)

Abstract

一种植物扫描与重建方法，该方法包括：对植物进行整体扫描，获得植物的整体扫描数据；对植物的每个叶子进行单独扫描，获得每个叶子的点云数据；将每个叶子的点云数据重建为曲面模型，通过泊松重建算法得到茎杆的重建结果；将重建后的所有叶子和茎杆与植物的整体扫描数据对齐；将所有叶子对齐后的点云数据融合到一起，得到整棵植物模型。其中，对每个叶子的重建过程是完全自动的，不需要交互，这样就大大减少了用户工作量；对每个叶片单独扫描并重建的策略，即分治策略，可以提高扫描和重建精度。

Description

植物扫描与重建方法 技术领域
本发明涉及三维植物建模技术领域，尤其涉及植物扫描与重建方法。
背景技术
三维植物建模是一个重要且应用广泛的研究课题。游戏设计与开发中，场景中植物模型质量的高低，直接影响游戏的真实感和用户体验。在植物学领域，三维植物建模可以用于研究植物的生长和在不同物理环境下的行为。在农业上，三维植物模型有助于病虫害防治和施肥方法的研究。
由于植物结构复杂，直接让美工来进行与实物相符的精确建模是困难的。通过三维扫描仪扫描植物的三维点云数据，并通过重建算法来获得真实的植物模型相比手工建模更为快捷可行。然而，由于植物叶片的相互遮挡，想获得植物的完整扫描数据是很困难的。并且，由于叶片是二维曲面，叶梗和茎秆是三维柱体，传统的曲面重建算法很难直接适用于这种情况。因此，要解决的问题为两个方面：1、如何获取植物完整的三维点云数据；2、如何快速精确地自动将获得的点云数据重建为曲面模型。
植物的完整扫描与重建是一个很有挑战性的课题。来自日本的研究人员Takashi Ijiri提出了使用CT扫描设备来获取花的三维数据，并交互式地重建出曲面模型。但是CT扫描设备或者高精度激光扫描仪等十分昂贵，用它来扫描和重建植物模型的成本太高，一般用户难以承受。并且他们的重建方法不能很好地处理复杂的叶片相互之间的遮挡，因此需要大量的用户交互，并不便于使用。还有一些研究人员提出了从三维点云场景中提取树木的骨架来重建树木的方法。但是这种方法只用于粗糙地重建树枝和树干，没有很好的机制来处理缺失数据和噪声数据，并不能精确重建出植物叶片。
发明内容
本发明实施例提供一种植物扫描与重建方法，用以减少用户交互，提高扫描和重建精度，该方法包括：
对植物进行整体扫描，获得植物的整体扫描数据；
对植物的每个叶子进行单独扫描，获得每个叶子的点云数据；
将每个叶子的点云数据重建为曲面模型，通过泊松重建算法得到茎杆的重建结果；
将重建后的所有叶子和茎杆与植物的整体扫描数据对齐；
将所有叶子对齐后的点云数据融合到一起，得到整棵植物模型；
其中，将每个叶子的点云数据重建为曲面模型，包括：
利用L1-中值算法提取叶子的骨架；
沿着骨架对叶子的点云做垂直切片；
根据点云切片的纵横比将切片分为叶片和叶柄两部分；
采用基于曲率的二次距离极小化方法，对每个切片拟合一个NURBS(Non-Uniform Rational B-Splines，非统一有理B样条)曲线，叶片拟合一个非闭合曲线，叶柄拟合闭合曲线；
优化求解所有NURBS控制点的最佳位置；
连接叶子的所有切片形状，得到叶子的形状。
一个实施例中，采用手持式结构光3D扫描仪，对植物进行整体扫描，以及，对植物的每个叶子进行单独扫描。
一个实施例中，优化求解所有NURBS控制点的最佳位置，包括：
通过BFGS算法极小化如下目标函数，得到NURBS控制点的最佳位置：
f(x)＝Edata(x)+αEsmooth(x)+βEbound(x)+γEround(x)；
其中，x为要求解的控制点位置；Edata为所有点云中的点到NURBS曲线最近距离的累加；Esmooth为不同NUBRS曲线上相同标识ID的控制点连成的曲线的不光滑度；Ebound为非闭合NURBS端点到叶片点云边界处的最近距离的累加，Eround为每个闭合NURBS曲线的周长面积比；α、β、γ为常数。
一个实施例中，将重建后的所有叶子和茎杆与植物的整体扫描数据对齐，包括：
为每个叶子或茎秆模型定义多对到植物点云的对应点，求出所述多对对应点定义的刚性变换，并以求出的刚性变换对每个叶片或茎秆进行变换；
利用骨架驱动叶片和茎秆变形，进行非刚性配准；
将控制对象由骨架变为用WLOP算法均匀采样的控制点，再次进行非刚性配准。
一个实施例中，利用骨架驱动叶片和茎秆变形，进行非刚性配准，包括：优化求解骨架点旋转和平移变换。
一个实施例中，优化求解骨架点旋转和平移变换，包括：
采用BFGS算法极小化目标函数，以求得定义在骨架上的最佳变换，并将所述最佳变换应用到叶片和茎秆模型上，所述目标函数定义为变形后叶片到点云之间的距离+定义在骨架上的一个的拉普拉斯光滑算子。
一个实施例中，将控制对象由骨架变为用WLOP算法均匀采样的控制点，再次进行非刚性配准，包括：
将拉普拉斯光滑算子由骨架移到控制点相互连接成的网格上。
一个实施例中，将所有叶子对齐后的点云数据融合到一起，包括：
通过检测三角形相交求出两片相交叶子的相交轮廓，然后搜索出使轮廓长度下降最快的位移方向，并通过移动轮廓附近的控制点来驱动叶子变形。
一个实施例中，将所有叶子对齐后的点云数据融合到一起，包括：
如果植物有茎，将叶子骨架末端变形到与叶子骨架末端距离最近的茎上的点，并利用骨架的变形驱动叶子的变形来达到茎叶相连；若没有茎，自动检测叶柄的末端距其它叶柄的距离，若距离小于阈值，则利用骨架的变形驱动叶子的变形来达到叶子相连。
本发明实施例中，对每个叶子单独扫描并重建，可以扫描到植物的每一处，克服了因为叶片相互遮挡扫描不到的难题，可以得到精确且完整的点云数据，从而获得良好的重建精度；其中对每个叶子的重建过程是完全自动的，不需要交互，这样大大减少了用户工作量。
进一步的，采用手持式结构光3D扫描仪来完成植物扫描，相比使用CT扫描仪或高精度激光扫描仪的方案，将大幅降低成本。
附图说明
为了更清楚地说明本发明实施例中的技术方案，下面将对实施例描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。在附图中：
图1为本发明实施例中植物扫描与重建方法的流程图；
图2为本发明实施例中叶子的重建过程示意图。
具体实施方式
为使本发明实施例的目的、技术方案和优点更加清楚明白，下面结合附图对本发明实施例做进一步详细说明。在此，本发明的示意性实施例及其说明用于解释本发明，但并不作为对本发明的限定。
图1为本发明实施例中植物扫描与重建方法的流程图。如图1所示，本发明实施例中植物扫描与重建方法可以包括：
步骤101、对植物进行整体扫描，获得植物的整体扫描数据；
步骤102、对植物的每个叶子进行单独扫描，获得每个叶子的点云数据；
步骤103、将每个叶子的点云数据重建为曲面模型，通过泊松重建算法得到茎杆的重建结果；
步骤104、将重建后的所有叶子和茎杆与植物的整体扫描数据对齐；
步骤105、将所有叶子对齐后的点云数据融合到一起，得到整棵植物模型。
由图1可以得知，本发明实施例中植物扫描与重建方法包括扫描、重建、配准及融合4个步骤。扫描过程中，首先对植物整体进行扫描，然后可以剪下所有叶子，并对它们进行单独扫描。重建过程，在本发明实施例中使用一种专门为叶子设计的重建算法来重建单独扫描的叶子。配准过程，即将重建的叶子模型对齐到点云上。融合过程，即将对其后的叶子合并为一个植物模型。
为了能充分暴露被遮挡的部分，本发明实施例中的扫描过程分为两步：首先对整棵植物做一次扫描，然后剪下所有叶片并对它们进行单独扫描。这是一种巧妙的分治策略。因为植物的叶片往往相互遮挡，一次整体扫描时得不到完整的点云数据。没有完整的数据，重建的效果自然也不会好。因此，需要再次对所有叶片进行单独扫描。对每个叶片单独扫描并重建的策略，即分治策略，可以提高扫描和重建精度，获得的精度甚至可以超过通过昂贵CT扫描仪得到的效果。
针对于传统方案成本高的缺点，可以用手持式结构光3D扫描仪对植物进行扫描。手持式结构光3D扫描仪的成本现在已经降到了千元级，如微软公司的Kinect。因此，采用手持式结构光3D扫描仪，对植物进行整体扫描，以及，对植物的每个叶子进行单独扫描，可以为用户节省大量资金。
为了减少用户交互，本发明实施例中对每个叶子的重建过程是完全自动的，不需要交互，这样就大大减少了用户工作量。图2为本发明实施例中叶子的重建过程示意图。如图2所示，将每个叶子的点云数据重建为曲面模型，可以包括：
步骤201、利用L1-中值算法提取叶子的骨架；
步骤202、沿着骨架对叶子的点云做垂直切片；
步骤203、根据点云切片的纵横比将切片分为叶片和叶柄两部分；
步骤204、采用基于曲率的二次距离极小化方法，对每个切片拟合一个非统一有理B样条NURBS曲线，叶片拟合一个非闭合曲线，叶柄拟合闭合曲线；
步骤205、优化求解所有NURBS控制点的最佳位置；
步骤206、连接叶子的所有切片形状，得到叶子的形状。
具体实施时，优化求解所有NURBS控制点的最佳位置，可以包括：
通过BFGS算法极小化如下目标函数，得到NURBS控制点的最佳位置：
f(x)＝Edata(x)+αEsmooth(x)+βEbound(x)+γEround(x)；
其中，x为要求解的控制点位置；Edata为所有点云中的点到NURBS曲线最近距离的累加；Esmooth为不同NUBRS曲线上相同标识ID的控制点连成的曲线的不光滑度；Ebound为非闭合NURBS端点到叶片点云边界处的最近距离的累加，Eround为每个闭合NURBS曲线的周长面积比；α、β、γ为常数。NURBS曲线定义了叶子的所有切片形状，连接它们就得到了叶子的形状。
实施例中，也可以采用曲面拟合算法等其它算法进行单个叶子的重建。
具体实施时，配准的目的是将重建后的所有叶子和茎秆与植物的整体扫描数据对齐，精确地将单独扫描的叶子对齐到了植物的整体点云上。配准可以分为两步，即刚性配准，和非刚性配准：
为每个叶片或者茎秆模型定义多对(例如三对)从它到植物点云的对应点，求出这多对对应点定义的刚性变换，并以此对每个叶片或者茎秆进行变换，来达到刚性配准的目的。
利用骨架驱动叶片和茎秆变形，并以此进行第一步非刚性配准。具体的，可以优化求解骨架点旋转和平移变换。可以用BFGS算法极小化目标函数，以求得定义在骨架上的最佳变换，并将其应用到叶片和茎秆模型上，该目标函数定义为变形后叶片到点云之间的距离+定义在骨架上的一个的拉普拉斯光滑算子。
在骨架驱动的配准完成之后，可以将控制对象由骨架变为用WLOP算法均匀采样的控制点，再次进行非刚性配准。具体的，可以将拉普拉斯光滑算子由骨架移到控制点相互连接成的网格上。用与第一步非刚性配准同样的优化方法可以得到进一步的细节配准。
实施例中，也可以采用刚性ICP算法和非刚性ICP算法进行配准。
具体实施时，融合即将配准后得到的点云融合到一起即可以得到整棵植物模型。融合需要解决两个问题：避免叶子相交，茎叶相连。
通过检测三角形相交来求出两片相交叶子的相交轮廓，然后搜索出使轮廓长度下降最快的位移方向，并通过移动轮廓附近的控制点来驱动叶子变形可以规避相交。
如果植物有茎，可以将叶子骨架末端变形到距其最近的茎上的点，并利用骨架的变形来驱动叶子的变形来达到茎叶相连的目的。若没有茎，自动检测叶柄的末端距其它叶柄的距离，若距离小于一定阈值，则以同样的方式通过变形使它们相连，即利用骨架的变形驱动叶子的变形来达到叶子相连。
综上所述，本发明实施例中，对每个叶子单独扫描并重建，可以扫描到植物的每一处，克服了因为叶片相互遮挡扫描不到的难题，可以得到精确且完整的点云数据，从而获得良好的重建精度；其中对每个叶子的重建过程是完全自动的，不需要交互，这样大大减少了用户工作量。经过试验证明，本发明实施例对高度缺失的点云重建效果非常好。
进一步的，采用手持式结构光3D扫描仪来完成植物扫描，相比使用CT扫描仪或高精度激光扫描仪的方案，将大幅降低成本。
本发明实施例中的植物扫描与重建方法不仅可以应用于仅有叶子的植物上，也可以应用于其它对象的扫描与重建，例如花朵的扫描与重建、甚至人体的扫描与重建。
本领域内的技术人员应明白，本发明的实施例可提供为方法、系统、或计算机程序产品。因此，本发明可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且，本发明可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本发明是参照根据本发明实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中，使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。
以上所述的具体实施例，对本发明的目的、技术方案和有益效果进行了进一步详细说明，所应理解的是，以上所述仅为本发明的具体实施例而已，并不用于限定本发明的保护范围，凡在本发明的精神和原则之内，所做的任何修改、等同替换、改进等，均应包含在本发明的保护范围之内。

Claims (9)

  1. 一种植物扫描与重建方法，其特征在于，包括：
    对植物进行整体扫描，获得植物的整体扫描数据；
    对植物的每个叶子进行单独扫描，获得每个叶子的点云数据；
    将每个叶子的点云数据重建为曲面模型，通过泊松重建算法得到茎杆的重建结果；
    将重建后的所有叶子和茎杆与植物的整体扫描数据对齐；
    将所有叶子对齐后的点云数据融合到一起，得到整棵植物模型；
    其中，将每个叶子的点云数据重建为曲面模型，包括：
    利用L1-中值算法提取叶子的骨架；
    沿着骨架对叶子的点云做垂直切片；
    根据点云切片的纵横比将切片分为叶片和叶柄两部分；
    采用基于曲率的二次距离极小化方法，对每个切片拟合一个非统一有理B样条NURBS曲线，叶片拟合一个非闭合曲线，叶柄拟合闭合曲线；
    优化求解所有NURBS控制点的最佳位置；
    连接叶子的所有切片形状，得到叶子的形状。
  2. 如权利要求1所述的方法，其特征在于，采用手持式结构光3D扫描仪，对植物进行整体扫描，以及，对植物的每个叶子进行单独扫描。
  3. 如权利要求1所述的方法，其特征在于，优化求解所有NURBS控制点的最佳位置，包括：
    通过BFGS算法极小化如下目标函数，得到NURBS控制点的最佳位置：
    f(x)＝Edata(x)+αEsmooth(x)+βEbound(x)+γEround(x)；
    其中，x为要求解的控制点位置；Edata为所有点云中的点到NURBS曲线最近距离的累加；Esmooth为不同NUBRS曲线上相同标识ID的控制点连成的曲线的不光滑度；Ebound为非闭合NURBS端点到叶片点云边界处的最近距离的累加，Eround为每个闭合NURBS曲线的周长面积比；α、β、γ为常数。
  4. 如权利要求1所述的方法，其特征在于，将重建后的所有叶子和茎杆与植物的整体扫描数据对齐，包括：
    为每个叶子或茎秆模型定义多对到植物点云的对应点，求出所述多对对应点定义的刚性变换，并以求出的刚性变换对每个叶片或茎秆进行变换；
    利用骨架驱动叶片和茎秆变形，进行非刚性配准；
    将控制对象由骨架变为用WLOP算法均匀采样的控制点，再次进行非刚性配准。
  5. 如权利要求4所述的方法，其特征在于，利用骨架驱动叶片和茎秆变形，进行非刚性配准，包括：优化求解骨架点旋转和平移变换。
  6. 如权利要求5所述的方法，其特征在于，优化求解骨架点旋转和平移变换，包括：
    采用BFGS算法极小化目标函数，以求得定义在骨架上的最佳变换，并将所述最佳变换应用到叶片和茎秆模型上，所述目标函数定义为变形后叶片到点云之间的距离+定义在骨架上的一个的拉普拉斯光滑算子。
  7. 如权利要求6所述的方法，其特征在于，将控制对象由骨架变为用WLOP算法均匀采样的控制点，再次进行非刚性配准，包括：
    将拉普拉斯光滑算子由骨架移到控制点相互连接成的网格上。
  8. 如权利要求1所述的方法，其特征在于，将所有叶子对齐后的点云数据融合到一起，包括：
    通过检测三角形相交求出两片相交叶子的相交轮廓，然后搜索出使轮廓长度下降最快的位移方向，并通过移动轮廓附近的控制点来驱动叶子变形。
  9. 如权利要求1所述的方法，其特征在于，将所有叶子对齐后的点云数据融合到一起，包括：
    如果植物有茎，将叶子骨架末端变形到与叶子骨架末端距离最近的茎上的点，并利用骨架的变形驱动叶子的变形来达到茎叶相连；若没有茎，自动检测叶柄的末端距其它叶柄的距离，若距离小于阈值，则利用骨架的变形驱动叶子的变形来达到叶子相连。
PCT/CN2014/090820 2014-11-11 2014-11-11 植物扫描与重建方法 WO2016074154A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2014/090820 WO2016074154A1 (zh) 2014-11-11 2014-11-11 植物扫描与重建方法

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2014/090820 WO2016074154A1 (zh) 2014-11-11 2014-11-11 植物扫描与重建方法

Publications (1)

Publication Number Publication Date
WO2016074154A1 true WO2016074154A1 (zh) 2016-05-19

Family

ID=55953560

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2014/090820 WO2016074154A1 (zh) 2014-11-11 2014-11-11 植物扫描与重建方法

Country Status (1)

Country Link
WO (1) WO2016074154A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111814666A (zh) * 2020-07-07 2020-10-23 华中农业大学 一种复杂林分下的单木参数提取方法、系统、介质及设备

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8004517B1 (en) * 2005-06-24 2011-08-23 Geomagic, Inc. Methods, apparatus and computer program products that model three-dimensional surface structures
CN102163342A (zh) * 2011-04-07 2011-08-24 北京农业信息技术研究中心 基于多尺度测量数据的果树形态结构三维重建方法
CN102184564A (zh) * 2011-04-13 2011-09-14 北京农业信息技术研究中心 基于双尺度三维数字化数据的设施园艺植物三维重建方法
CN103279980A (zh) * 2013-05-08 2013-09-04 西安理工大学 基于点云数据的树叶建模方法
CN104408765A (zh) * 2014-11-11 2015-03-11 中国科学院深圳先进技术研究院 植物扫描与重建方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8004517B1 (en) * 2005-06-24 2011-08-23 Geomagic, Inc. Methods, apparatus and computer program products that model three-dimensional surface structures
CN102163342A (zh) * 2011-04-07 2011-08-24 北京农业信息技术研究中心 基于多尺度测量数据的果树形态结构三维重建方法
CN102184564A (zh) * 2011-04-13 2011-09-14 北京农业信息技术研究中心 基于双尺度三维数字化数据的设施园艺植物三维重建方法
CN103279980A (zh) * 2013-05-08 2013-09-04 西安理工大学 基于点云数据的树叶建模方法
CN104408765A (zh) * 2014-11-11 2015-03-11 中国科学院深圳先进技术研究院 植物扫描与重建方法

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
HUANG, HUI ET AL.: "L1-Medial Skeleton of Point Cloud", JOURNALACM TRANSACTIONS ON GRAPHICS-SIGGRAPH 2013 CONFERENCE PROCEEDINGS, vol. 32, no. 4, 31 July 2013 (2013-07-31) *
SUN, ZHIHUI ET AL.: "Surfaces Reconstruction of Plant Leaves Based on Point Cloud Data", TRANSACTIONS OF THE CHINESE SOCIETY OF AGRICULTURAL ENGINEERING, vol. 28, no. 3, 29 February 2012 (2012-02-29), pages 184 - 190 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111814666A (zh) * 2020-07-07 2020-10-23 华中农业大学 一种复杂林分下的单木参数提取方法、系统、介质及设备

Similar Documents

Publication Publication Date Title
Chaivivatrakul et al. Automatic morphological trait characterization for corn plants via 3D holographic reconstruction
CN107808415B (zh) 基于机器视觉的鞋底边缘轨迹及涂胶位姿提取方法
Yin et al. Morfit: interactive surface reconstruction from incomplete point clouds with curve-driven topology and geometry control.
CN102163342B (zh) 基于多尺度测量数据的果树形态结构三维重建方法
CN104408765B (zh) 植物扫描与重建方法
Yin et al. Full 3D plant reconstruction via intrusive acquisition
Medeiros et al. Modeling dormant fruit trees for agricultural automation
WO2015149302A1 (zh) 基于点云与数据驱动的树木模型重建方法
CN107330281B (zh) 全自动骨折钢板模型个性化重构方法
US20120113116A1 (en) Method and apparatus for preparing image representative data
CN109242951A (zh) 一种脸部实时三维重建方法
CN109859099A (zh) 基于sfm点云深度的盆栽玉米杂草快速去除方法
CN106777721B (zh) 基于有限元模型的种植体结构优化方法
Pirk et al. Modeling plant life in computer graphics
WO2006097926A2 (en) Methods and systems for creating and altering cad models
WO2016074154A1 (zh) 植物扫描与重建方法
Brophy et al. A method for global non-rigid registration of multiple thin structures
Xie et al. Geometric modeling of Rosa roxburghii fruit based on three-dimensional point cloud reconstruction
Xing et al. Highly parallel algorithms for visual-perception-guided surface remeshing
Lin et al. Segmentation of crop organs through region growing in 3D space
CN114549795A (zh) 一种鞋楦曲面的参数化重构方法、系统、介质及设备
Gibbs et al. Three-dimensional reconstruction of plant shoots from multiple images using an active vision system
Beardsley et al. Editable parametric dense foliage from 3D capture
CN112131744A (zh) 一种基于实测参数及实体特征的舒茶早鲜叶片的建模方法
Chaudhury et al. Geometry reconstruction of plants

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 14906054

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 25.08.2017)

122 Ep: pct application non-entry in european phase

Ref document number: 14906054

Country of ref document: EP

Kind code of ref document: A1