WO2020114321A1 - 点云去噪方法、图像处理设备及具有存储功能的装置 - Google Patents

点云去噪方法、图像处理设备及具有存储功能的装置 Download PDF

Info

Publication number
WO2020114321A1
WO2020114321A1 PCT/CN2019/121762 CN2019121762W WO2020114321A1 WO 2020114321 A1 WO2020114321 A1 WO 2020114321A1 CN 2019121762 W CN2019121762 W CN 2019121762W WO 2020114321 A1 WO2020114321 A1 WO 2020114321A1
Authority
WO
WIPO (PCT)
Prior art keywords
point cloud
data
point
points
original
Prior art date
Application number
PCT/CN2019/121762
Other languages
English (en)
French (fr)
Inventor
张智胜
祁春超
Original Assignee
深圳市华讯方舟太赫兹科技有限公司
华讯方舟科技有限公司
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 深圳市华讯方舟太赫兹科技有限公司, 华讯方舟科技有限公司 filed Critical 深圳市华讯方舟太赫兹科技有限公司
Publication of WO2020114321A1 publication Critical patent/WO2020114321A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/70—Denoising; Smoothing
    • 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
    • G06T17/20—Finite element generation, e.g. wire-frame surface description, tesselation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10028—Range image; Depth image; 3D point clouds

Definitions

  • the present application relates to the field of image processing technology, in particular to a point cloud denoising method, image processing equipment, and a device with a storage function.
  • the three-dimensional point cloud data model has become an emerging digital media, which has been widely used in industrial manufacturing, architectural design, product display, medicine, e-commerce, etc.
  • some floating noise points will be formed on the surface of the scanning point cloud. These floating noise points belong to data redundancy and will lead to a reduction in the efficiency of subsequent data processing.
  • This application mainly provides a point cloud denoising method, image processing equipment, and a device with a storage function, which can remove floating noise points on the surface.
  • a technical solution adopted by the present application is to provide a point cloud denoising method, which includes: acquiring original point cloud data; identifying a reference surface in the original point cloud data, the reference surface including the reference surface The first point cloud; obtain the neighborhood points of the data points in the first point cloud to obtain the denoised point cloud.
  • an image processing device which is characterized by comprising: a communication circuit and a processor connected to each other; the communication circuit is used to obtain original point cloud data; processing The device is used to identify the reference surface in the original point cloud data.
  • the reference surface includes the first point cloud that constitutes the reference surface, and obtains the neighborhood points of the data points in the first point cloud to obtain the denoised point cloud.
  • another technical solution adopted by the present application is to provide a device with a storage function that stores instructions, and when the instructions are executed, the point cloud denoising method described above is implemented.
  • the beneficial effects of the present application are: different from the situation in the prior art.
  • the original point cloud data is obtained; the reference surface in the original point cloud data is identified, and the reference surface includes the first component constituting the reference surface Point cloud; Obtain the neighborhood point of the data point in the first point cloud to get the denoised point cloud.
  • the neighbor points of the reference surface can be easily obtained, and the non-neighbor points can be determined as noise points, so that the composition of the neighbor points of the reference surface can be obtained
  • the effect of removing noise points floating on the surface of the point cloud can be achieved, making the contour characteristics of the surface of the object more obvious, which helps to improve the efficiency of subsequent data processing.
  • FIG. 1 is a schematic flowchart of an embodiment of a method for denoising a point cloud according to the present application
  • FIG. 2 is a schematic flowchart of step S12 in FIG. 1;
  • FIG. 3 is a schematic flowchart of step S13 in FIG. 1;
  • FIG. 4 is a schematic flowchart of step S131 in FIG. 3;
  • FIG. 5 is a schematic diagram of the positional relationship between neighborhood points and noise points and the first data point
  • FIG. 6 is a schematic diagram of the original point cloud data without denoising
  • FIG. 7 is a schematic diagram of point cloud data after denoising using the point cloud denoising method of the present application.
  • FIG. 8 is a schematic structural diagram of an embodiment of an image processing apparatus of the present application.
  • FIG. 9 is a schematic structural diagram of an embodiment of a device with a storage function according to the present application.
  • an embodiment of a method for denoising a point cloud according to the present application includes:
  • a point cloud is a large number of data points expressing the surface of an object, and each data point can be expressed in three-dimensional coordinates.
  • the original point cloud data includes at least the position coordinates of a plurality of data points expressing the contour of the target 3D image.
  • the 3D point cloud data of the target can be directly scanned by the 3D scanning device, or it can be obtained from the storage device or other devices in advance.
  • the point cloud data of the stored three-dimensional image is not specifically limited here.
  • the point cloud is an approximate expression of the surface of the object, when the surface of the object characterized by the original point cloud data is identified, it can be easily judged which floating noise points are in the point cloud data.
  • the original point cloud data is a set of data points obtained by scanning the target object with a three-dimensional scanning device.
  • the original point cloud data is an approximate expression of the surface of the target object, which can characterize the three-dimensional surface of the target object.
  • an image processing algorithm can be used to identify a reference surface, which is the identified three-dimensional surface that is similar to the target object.
  • the reference surface includes a first point cloud constituting the reference surface, and a positional relationship between data points in the first point cloud, the data points in the first point cloud are close to the original point cloud data data point.
  • step S12 includes:
  • S121 Use the surface fitting method to perform grid reconstruction on the original point cloud data.
  • the surface fitting method can use implicit surface reconstruction methods such as Poisson Surface Reconstruction (Possion Surface Reconstruction) or SSD (Smooth Signed Distance History Surface Reconstruction) to fit a better reference surface.
  • implicit surface reconstruction methods such as Poisson Surface Reconstruction (Possion Surface Reconstruction) or SSD (Smooth Signed Distance History Surface Reconstruction) to fit a better reference surface.
  • the surface fitting may also use other methods.
  • the original point cloud data is subjected to surface fitting using a Poisson grid reconstruction method to obtain a grid surface.
  • the Poisson grid reconstruction method adopts the implicit fitting method
  • the implicit equation represented by the surface information described by the original point cloud data is obtained by solving the Poisson equation, and by isosurface extraction of the equation, it can be Obtain a surface model with geometric entity information, that is, the reference surface.
  • the model reconstructed by Poisson mesh reconstruction method has good geometric surface characteristics and detail characteristics.
  • the reference surface includes a first point cloud obtained by surface fitting from the original point cloud data, and the data points in the first point cloud are data points that are fitted close to the real surface of the target object, which is different from the original point cloud data.
  • the mesh surface after the mesh reconstruction is composed of the connection relationship between the point cloud and the data points in the point cloud. Therefore, the reference surface includes the first point cloud obtained by surface fitting from the original point cloud data, and also includes the connection relationship of each data point in the first point cloud.
  • the mesh surface after the mesh reconstruction is a three-dimensional triangular mesh surface
  • the mesh surface includes triangle vertices constituting the mesh and the connection relationship of the triangle vertices.
  • the top points of the triangles forming the grid constitute the first point cloud.
  • each data point in the first point cloud that constitutes the reference surface can be obtained from the original point cloud data by means of distance judgment, etc.
  • step S13 specifically includes:
  • S131 Calculate the distance between the first data point in the first point cloud and the data point to be identified in the original point cloud data.
  • the data point to be identified is a data point in the original point cloud data.
  • each data point can be used as the first data point, and compared with each data point to be identified in the original point cloud data, Calculate the distance between the two. Since the first data point and the data point to be recognized are both three-dimensional space points, during the calculation, the three-dimensional space coordinates of the two (such as Euclidean space coordinates) can be obtained, and then the Euclidean distance of the two can be calculated using the Euclidean space coordinates of the two. .
  • step S131 when searching for neighborhood points for the data points in the first point cloud, the data points that have been searched for may be marked as searched by means of labeling, so as to subsequently perform neighborhood point search on other data points.
  • step S131 includes:
  • S1311 Select unmarked data points in the first point cloud as the first data points.
  • the data points in the first point cloud may be marked data points and unmarked data points, respectively, marked data points are data points that have searched for neighbor points, and unmarked data points are unprocessed neighbors
  • marked data points are data points that have searched for neighbor points
  • unmarked data points are unprocessed neighbors
  • the data point of the point search, the first data point is the unmarked data point.
  • the selection of the first data point may be random, or may be selected according to a certain preset order, for example, data storage order.
  • S1312 Obtain the position coordinates of the first data point in the first point cloud and the position coordinates of the data point to be identified.
  • the first data point is a randomly selected data point in the first point cloud
  • the position coordinates of the first data point and the second data point are three-dimensional space coordinates, such as Euclidean space coordinates.
  • the position coordinates of the data point may also be other types of coordinates, such as spherical space coordinates.
  • S1313 Calculate the distance between the data point to be identified and the first data point, and mark the first data point as found.
  • the position coordinates (x 1 , y 1 , z 1 ) of the first data point A in the first point cloud and the position coordinates (x 2 , y 2 ) of the data point B to be identified are obtained.
  • z 2 you can use the following formula
  • the Euclidean distance between A and B is calculated, and the Euclidean distance is the distance between the data point B to be identified and the first data point A.
  • the first data point can be marked as searched, so as to subsequently perform neighborhood point search on other data points in the first point cloud.
  • the preset distance is a preset distance threshold for identifying floating noise points.
  • the specific value of the preset distance can be set according to the recognition accuracy requirement. The higher the recognition accuracy requirement, the smaller the preset distance.
  • step S133 If it is less, the following step S133 is executed, otherwise, step S134 is executed.
  • S133 Mark the data point to be identified as a neighbor point.
  • the data point to be identified is marked as a noise point (floating noise point), otherwise the data point to be identified is The neighborhood point marked as the first data point.
  • the distance between the data point B to be identified and the first data point A is less than the preset distance R
  • the data point B to be identified is marked as a neighborhood point
  • the data point C to be identified and the first data is marked as a noise point.
  • S135 Determine whether there are unmarked data points in the first point cloud.
  • step S136 as follows.
  • step S133 or S134 After each step S133 or S134 is executed, some data points in the original point cloud data will be marked as neighborhood points or noise points. At this time, it is necessary to determine whether the denoising process can be ended, and the first Whether there are unmarked data points in the point cloud, if there are, it means that there are some data points in the first point cloud that have not been searched for neighborhood points.
  • step S131 you can return to step S131 to select unmarked data points for neighborhood point search To mark the data point to be identified until all data points in the first point cloud are marked as searched, that is, all neighboring points of the first point cloud in the original point cloud data have been obtained
  • all neighboring points of the first point cloud in the original point cloud data are extracted, which can form a denoised point cloud.
  • the noise point may not need to be marked, but only needs to mark the neighboring point.
  • all the marked data points in the original point cloud data are extracted to form a denoised point cloud.
  • the reference surface in the original point cloud data when acquiring the original point cloud data, the reference surface in the original point cloud data is identified, the reference surface includes the first point cloud constituting the reference surface, and the neighborhood points of the data points in the first point cloud are obtained to obtain Point cloud after denoising. Therefore, after identifying the reference surface represented by the point cloud data in this application, the neighbor points of the reference surface can be easily obtained, and the non-neighbor points can be determined as noise points, so that all the neighbor points of the reference surface can be obtained The composed denoised point cloud can further achieve the effect of removing noise points floating on the surface of the point cloud, which makes the contour characteristics of the characterized object surface more obvious, and helps to improve the efficiency of subsequent data processing.
  • the original point cloud data has floating noise points M, which are processed by the point cloud denoising method of the present application, such as The noise points floating on the surface of the point cloud after denoising shown in FIG. 7 are removed. It can be seen from this that the point cloud denoising method of the present application can effectively remove noise points floating on the surface of the point cloud, making the contour characteristics of the object surface characterized by the point cloud more obvious.
  • an embodiment of the image processing apparatus 20 of the present application includes: a communication circuit 201 and a processor 202 connected to each other.
  • the communication circuit 201 is used to send and receive data, and is an interface for the image processing device 20 to communicate with other devices. Specifically, the communication circuit 201 is used to obtain original point cloud data.
  • the original point cloud data includes at least the position coordinates of a plurality of data points expressing the contour of the target 3D image.
  • the 3D point cloud data of the target can be directly scanned by the 3D scanning device, or it can be obtained from the storage device or other devices in advance.
  • the point cloud data of the stored three-dimensional image is not specifically limited here.
  • the processor 202 controls the operation of the communication device, and the processor 202 may also be called a CPU (Central Processing Unit, central processing unit).
  • the processor 202 may be an integrated circuit chip with signal processing capabilities.
  • the processor 202 may also be a general-purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components .
  • the general-purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
  • the processor 202 is used to execute a program to implement the method as provided in an embodiment of the point cloud denoising method of the present application.
  • the processor 202 is specifically used to identify the reference surface in the original point cloud data, the reference surface includes the first point cloud that constitutes the reference surface, and obtains the neighborhood points of the data points in the first point cloud to obtain the denoised Point cloud.
  • the processor 202 is also used to perform mesh reconstruction on the original point cloud data using the surface fitting method, and the mesh surface after the mesh reconstruction is used as a reference surface, which includes surface fitting from the original point cloud data.
  • First point cloud the surface fitting may use implicit surface fitting methods such as Poisson surface fitting or SSD smooth surface fitting to obtain a better fitting effect.
  • the processor 202 is also used to calculate the distance between the first data point in the first point cloud and the data point to be identified in the original point cloud data, determine whether the distance is less than the preset distance, and when the distance is less than the preset distance, The data points to be identified are marked as neighborhood points, and all the neighborhood points in the original point cloud data are extracted to form a denoised point cloud.
  • the image processing device 20 may further include other components such as a memory (not shown), which is not specifically limited here.
  • the image processing device in this embodiment may be a mobile terminal, a fixed terminal, a server, a three-dimensional scanner, a security inspection instrument, etc., or an integrated independent component, such as an image processing chip.
  • the original point cloud data is obtained, the reference surface in the original point cloud data is identified, and the neighboring points of the data points constituting the reference surface are obtained to obtain a denoised point cloud. Therefore, after identifying the reference surface represented by the point cloud data in this application, the neighbor points of the reference surface can be easily obtained, and the non-neighbor points can be determined as noise points, so that all the neighbor points of the reference surface can be obtained.
  • the denoised point cloud composed can further achieve the effect of removing noise points floating on the surface of the point cloud, making the surface contour of the object characterized by the point cloud more obvious, and helping to improve the efficiency of subsequent data processing.
  • an instruction 301 is stored inside the device with a storage function 30, and when the instruction 301 is executed, it is implemented as provided in an embodiment of the point cloud denoising method of the present application Methods.
  • the device 30 with a storage function may be a portable storage medium such as a U disk, an optical disk, or may be a terminal, a server, or an integrated independent component, such as an image processing chip.
  • the original point cloud data is acquired, the reference surface in the original point cloud data is identified, and the neighboring points of the data points constituting the reference surface are obtained to obtain denoising
  • the neighboring points of the reference surface can be easily obtained, and the non-neighboring points can be determined as noise points, so that all neighbors of the reference surface can be obtained
  • the denoised point cloud composed of domain points can further achieve the effect of removing noise points floating on the surface of the point cloud, making the surface contour of the object characterized by the point cloud more obvious, which helps to improve the efficiency of subsequent data processing.

Landscapes

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

Abstract

本申请公开一种点云去噪方法、图像处理设备及具有存储功能的装置，该方法包括：获取原始点云数据；识别原始点云数据中的基准表面，该基准表面包括组成该基准表面的第一点云；获取第一点云中数据点的邻域点，以得到去噪后的点云。由此，本申请能够达到去除点云表面漂浮的噪声点的效果，使得物体表面的轮廓特征更明显，有助于提高后续数据处理的效率。

Description

点云去噪方法、图像处理设备及具有存储功能的装置 技术领域
本申请涉及图像处理技术领域，特别是涉及一种点云去噪方法、图像处理设备及具有存储功能的装置。
背景技术
随着三维扫描技术的发展，使三维点云数据模型成为一种新兴数字媒体，在工业制造、建筑设计、产品展示、医学、电子商务等方面有着广泛的应用。然而，有时由于设备精度以及扫描噪声的影响，会在扫描点云的表面形成一些漂浮的噪声点，这些漂浮噪声点属于数据冗余，会导致后续数据处理的效率降低。
技术解决方案
本申请主要提供一种点云去噪方法、图像处理设备及具有存储功能的装置，能够去除表面的漂浮噪声点。
为解决上述技术问题，本申请采用的一个技术方案是：提供一种点云去噪方法，包括：获取原始点云数据；识别原始点云数据中的基准表面，该基准表面包括组成该基准表面的第一点云；获取第一点云中数据点的邻域点，以得到去噪后的点云。
为解决上述技术问题，本申请采用的另一个技术方案是：提供一种图像处理设备，其特征在于，包括：相互连接的通信电路和处理器；该通信电路用以获取原始点云数据；处理器用于识别原始点云数据中的基准表面，该基准表面包括组成该基准表面的第一点云，并获取第一点云中数据点的邻域点，以得到去噪后的点云。
为解决上述技术问题，本申请采用的又一个技术方案是：提供一种具有存储功能的装置，存储有指令，该指令被执行时实现如上所述的点云去噪方法。
本申请的有益效果是：区别于现有技术的情况，本申请的实施例中，在获取原始点云数据；识别原始点云数据中的基准表面，该基准表面包括组成该基准表面的第一点云；获取第一点云中数据点的邻域点，以得到去噪后的点云。由此，本申请识别点云数据表征的基准表面后，可以很容易地获取基准表面的 邻域点，而非邻域点即可以判定为噪声点，从而可以得到该基准表面的邻域点组成的去噪后的点云，进而能够达到去除点云表面漂浮的噪声点的效果，使得物体表面的轮廓特征更明显，有助于提高后续数据处理的效率。
附图说明
图1是本申请点云去噪方法一实施例的流程示意图；
图2是图1中步骤S12的具体流程示意图；
图3是图1中步骤S13的具体流程示意图；
图4是图3中步骤S131的具体流程示意图；
图5是邻域点及噪声点与第一数据点之间的位置关系示意图；
图6是未进行去噪的原始点云数据的示意图；
图7是利用本申请点云去噪方法进行去噪后的点云数据的示意图；
图8是本申请图像处理设备一实施例的结构示意图；
图9是本申请具有存储功能的装置一实施例的结构示意图。
本发明的实施方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅是本申请的一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
如图1所示，本申请点云去噪方法一实施例包括：
S11：获取原始点云数据。
点云是表达物体表面的大量数据点，每个数据点均可以用三维坐标表示。其中，该原始点云数据至少包括表达目标三维图像轮廓的多个数据点的位置坐标，可以直接利用三维扫描设备扫描得到目标的三维点云数据，也可以是从存储设备或其他设备中获取预先存储的三维图像的点云数据，此处不做具体限定。
由于设备精度以及扫描噪声的影响，会在扫描点云的表面形成一些漂浮的噪声点，该原始点云数据中可能存在漂浮的噪声点，这些漂浮噪声点属于数据冗余，会导致后续数据处理的效率降低。
S12：识别原始点云数据中的基准表面，该基准表面包括组成该基准表面的第一点云。
由于点云是物体表面的近似表达，当识别出原始点云数据表征的物体的表面后，即可以很容易地判断点云数据中哪些是漂浮的噪声点。
具体地，在一个应用例中，原始点云数据是通过三维扫描设备扫描目标物体后得到的数据点集合，该原始点云数据是目标物体表面的近似表达，其可以表征目标物体的三维表面。虽然目标物体的三维表面难以完全还原，但可以利用图像处理算法识别出一基准表面，该基准表面即为识别出的近似于目标物体的三维表面。其中，该基准表面包括组成该基准表面的第一点云，及该第一点云中数据点之间的位置关系，该第一点云中的数据点是接近于原始点云数据中的部分数据点。
可选地，由于原始点云数据中通常没有显式的几何拓扑关系，数据点间没有明确的位置关系信息，可以利用曲面拟合建立三维网格，该建立的三维网格即为基准表面。具体地，如图2所示，步骤S12包括：
S121：利用曲面拟合方法在原始点云数据中进行网格重建。
其中，曲面拟合方法可以采用泊松表面重建(Possion Surface Reconstruction)或者SSD(Smooth Signed Distance Surface Reconstruction)等隐式曲面重建方法，以拟合出较好的基准表面。当然，在其他实施例中，该曲面拟合也可以采用其他方法。
具体地，在一个应用例中，获取原始点云数据后，利用泊松网格重建方法对该原始点云数据进行曲面拟合，可以得到网格曲面。由于泊松网格重建方法采取隐性拟合的方式，通过求解泊松方程来取得原始点云数据所描述的表面信息代表的隐性方程，并通过对该方程进行等值面提取，从而可以得到具有几何实体信息的表面模型，即该基准表面。泊松网格重建方法重建出的模型具有良好的几何表面特性和细节特性。其中，该基准表面包括由原始点云数据进行曲面拟合得到的第一点云，该第一点云中的数据点是拟合出的接近目标物体真实表面的数据点，区别于原始点云数据。
S122：将网格重建后的网格曲面作为基准表面。
其中，网格重建后的网格曲面由点云和点云中的数据点之间的连接关系构成的。因此，该基准表面包括由原始点云数据进行曲面拟合得到的第一点云，还包括该第一点云中各数据点的连接关系。
例如，该网格重建后的网格曲面是三维三角网格曲面时，该网格曲面包括构成网格的三角形顶点以及该三角形顶点的连接关系。该构成网格的三角形顶 点组成该第一点云。
S13：获取第一点云中数据点的邻域点，以得到去噪后的点云。
具体地，在一个应用例中，通过曲面拟合的方法识别该基准表面后，即可以通过距离判断等方式，从原始点云数据中获取组成该基准表面的第一点云中每个数据点的邻域点，从而得到由所有邻域点组成的去噪后的点云。
进一步地，如图3所示，步骤S13具体包括：
S131：计算第一点云中第一数据点与原始点云数据中的待识别数据点的距离。
其中，该待识别数据点是原始点云数据中的数据点。
具体地，在一个应用例中，第一点云中通常具有多个数据点，每个数据点均可以作为该第一数据点，与原始点云数据中的每个待识别数据点进行比较，计算二者之间的距离。由于第一数据点和待识别数据点均为三维空间点，计算时，可以获取二者的三维空间坐标(如欧式空间坐标)，然后利用二者的欧式空间坐标即可以计算二者的欧式距离。
可选地，对该第一点云中的数据点查找邻域点时，可以通过标记的方式，将已经查找过的数据点标记为已查找，以便后续对其他数据点进行邻域点查找。具体如图4所示，步骤S131包括：
S1311：选择第一点云中的未标记数据点作为第一数据点。
其中，该第一点云中的数据点可以分别已标记数据点和未标记数据点，已标记数据点即为已经查找过邻域点的数据点，而未标记数据点即为未进行邻域点查找的数据点，该第一数据点即为未标记数据点。
选择该第一数据点可以是随机的，也可以是按照某种预设顺序选取，例如数据存储顺序。
S1312：获取第一点云中的第一数据点的位置坐标和待识别数据点的位置坐标。
在一个应用例中，该第一数据点是第一点云中随机选择的数据点，该第一数据点和该第二数据点的位置坐标为三维空间坐标，如欧式空间坐标。当然，在其他实施例中，数据点的位置坐标也可以是其他类型的坐标，如球形空间坐标。
S1313：计算待识别数据点与第一数据点之间的距离，并将该第一数据点标记为已查找。
具体地，结合图5所示，获取第一点云中的第一数据点A的位置坐标(x 1，y 1，z 1)以及待识别数据点B的位置坐标(x 2，y 2，z 2)后，可以利用如下公式
Figure PCTCN2019121762-appb-000001
计算得到A和B之间的欧式距离，该欧式距离即为该待识别数据点B和第一数据点A之间的距离。每计算完一个第一数据点与所有待识别数据点的距离后，即可以将该第一数据点标记为已查找，以便后续对第一点云中的其他数据点进行邻域点查找。
S132：判断该距离是否小于预设距离。
该预设距离是预先设定的识别漂浮的噪声点的距离阈值。该预设距离的具体取值可以根据识别精度需求设置，识别精度需求越高，该预设距离越小。
若小于，则执行如下步骤S133，否则，执行步骤S134。
S133：将待识别数据点标记为邻域点。
S134：将该待识别数据点标记为噪声点。
具体地，当待识别数据点与第一数据点之间的距离大于该预设距离时，该待识别数据点即被标记为噪声点(漂浮的噪声点)，否则该待识别数据点即被标记为第一数据点的邻域点。
例如图5中，待识别数据点B与第一数据点A之间的距离小于该预设距离R，该待识别数据点B被标记为邻域点，而待识别数据点C与第一数据点A之间的距离大于预设距离R，该待识别数据点C被标记为噪声点。
S135：判断第一点云中是否存在未标记数据点。
若存在，则返回执行步骤S131，否则执行如下步骤S136。
S136：提取原始点云数据中的所有邻域点，组成去噪后的点云。
具体地，每次执行完步骤S133或S134后，原始点云数据中即有部分数据点会被标记为邻域点或噪声点，此时，需要判断是否可以结束去噪过程，可以判断第一点云中是否存在未标记数据点，若存在，则表明第一点云中还有部分数据点未进行邻域点查找，此时可以返回执行步骤S131，选择未标记数据点进行邻域点查找，以对该待识别数据点进行标记，直到第一点云中所有数据点均被标记为已查找，即已经得到该原始点云数据中第一点云的所有邻域点，此时可以结束去噪过程，提取原始点云数据中第一点云的所有邻域点，即可以组成去噪后的点云。
当然，在其他实施例中，该噪声点也可以不需要进行标记，只需要标记邻 域点即可，最终提取原始点云数据中所有已标记的数据点即可以组成去噪后的点云。
本实施例中，在获取原始点云数据，识别原始点云数据中的基准表面，该基准表面包括组成基准表面的第一点云，获取第一点云中数据点的邻域点，以得到去噪后的点云。由此，本申请识别点云数据表征的基准表面后，可以很容易地获取基准表面的邻域点，而非邻域点即可以判定为噪声点，从而可以得到该基准表面的所有邻域点组成的去噪后的点云，进而能够达到去除点云表面漂浮的噪声点的效果，使得表征的物体表面的轮廓特征更明显，有助于提高后续数据处理的效率。
结合图6和图7所示，如图6所示的原始点云数据中，可以明显看出该原始点云数据存在漂浮的噪声点M，经过本申请的点云去噪方法处理后，如图7所示去噪后的点云表面漂浮的噪声点被去除。由此可以看出，本申请的点云去噪方法能够有效去除点云表面漂浮的噪声点，使得该点云表征的物体表面的轮廓特征更明显。
如图8所示，本申请图像处理设备20一实施例包括：相互连接的通信电路201和处理器202。
其中，该通信电路201用于发送和接收数据，是图像处理设备20与其他设备进行通信的接口。具体地，该通信电路201用于获取原始点云数据。
其中，该原始点云数据至少包括表达目标三维图像轮廓的多个数据点的位置坐标，可以直接利用三维扫描设备扫描得到目标的三维点云数据，也可以是从存储设备或其他设备中获取预先存储的三维图像的点云数据，此处不做具体限定。
处理器202控制通信设备的操作，处理器202还可以称为CPU(Central Processing Unit，中央处理单元)。处理器202可能是一种集成电路芯片，具有信号的处理能力。处理器202还可以是通用处理器、数字信号处理器(DSP)、专用集成电路(ASIC)、现成可编程门阵列(FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
具体地，该处理器202用于执行程序以实现如本申请点云去噪方法一实施例所提供的方法。
该处理器202具体用于识别原始点云数据中的基准表面，该基准表面包括 组成基准表面的第一点云，并获取第一点云中数据点的邻域点，以得到去噪后的点云。
处理器202还用于利用曲面拟合方法在原始点云数据中进行网格重建，并将网格重建后的网格曲面作为基准表面，该基准表面包括由原始点云数据进行曲面拟合得到的第一点云。其中，该曲面拟合可以采用泊松表面拟合或SSD平滑表面拟合等隐式曲面拟合方法，以得到较好的拟合效果。
处理器202还用于计算第一点云中第一数据点与原始点云数据中的待识别数据点的距离，判断该距离是否小于预设距离，并在该距离小于预设距离时，将待识别数据点标记为邻域点，并提取原始点云数据中的所有邻域点，组成去噪后的点云。
当然，在其他实施例中，该图像处理设备20还可以包括存储器(图未示)等其他部件，此处不做具体限定。
本实施例中的图像处理设备可以是移动终端、固定终端、服务器、三维扫描仪、安检仪等，也可以是集成的独立部件，例如图像处理芯片。
本实施例中，在获取原始点云数据，识别原始点云数据中的基准表面，获取组成基准表面的数据点的邻域点，以得到去噪后的点云。由此，本申请识别点云数据表征的基准表面后，可以很容易地获取基准表面的邻域点，而非邻域点即可以判定为噪声点，从而可以得到该基准表面的所有邻域点组成的去噪后的点云，进而能够达到去除点云表面漂浮的噪声点的效果，使得点云表征的物体表面轮廓更明显，有助于提高后续数据处理的效率。
如图9所示，本申请具有存储功能的装置一实施例中，具有存储功能的装置30内部存储有指令301，该指令301被执行时实现如本申请点云去噪方法一实施例所提供的方法。
其中，具有存储功能的装置30可以是便携式存储介质如U盘、光盘，也可以是终端、服务器或集成的独立部件，例如图像处理芯片等。
本实施例中，具有存储功能的装置中的指令被执行时，在获取原始点云数据，识别原始点云数据中的基准表面，获取组成基准表面的数据点的邻域点，以得到去噪后的点云，从而在识别点云数据表征的基准表面后，可以很容易地获取基准表面的邻域点，而非邻域点即可以判定为噪声点，从而可以得到该基准表面的所有邻域点组成的去噪后的点云，进而能够达到去除点云表面漂浮的噪声点的效果，使得点云表征的物体表面轮廓更明显，有助于提高后续数据处 理的效率。
以上所述仅为本申请的实施方式，并非因此限制本申请的专利范围，凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本申请的专利保护范围内。

Claims (10)

  1. 一种点云去噪方法，其特征在于，包括：
    获取原始点云数据；
    识别所述原始点云数据中的基准表面，所述基准表面包括组成所述基准表面的第一点云；
    获取所述第一点云中数据点的邻域点，以得到去噪后的点云。
  2. 根据权利要求1所述的方法，其特征在于，所述识别所述原始点云数据中的基准表面包括：
    利用曲面拟合方法在所述原始点云数据中进行网格重建；
    将所述网格重建后的网格曲面作为所述基准表面，所述基准表面包括由所述原始点云数据进行曲面拟合得到的所述第一点云。
  3. 根据权利要求2所述的方法，其特征在于，所述获取所述第一点云中数据点的邻域点，以得到去噪后的点云包括：
    计算所述第一点云中第一数据点与所述原始点云数据中的待识别数据点的距离；
    判断所述距离是否小于预设距离；
    若小于，则将所述待识别数据点标记为所述邻域点；
    提取所述原始点云数据中的所有所述邻域点，组成所述去噪后的点云。
  4. 根据权利要求3所述的方法，其特征在于，
    所述计算所述第一点云中第一数据点与所述原始点云数据中的待识别数据点的距离包括：
    选择所述第一点云中的未标记数据点作为所述第一数据点；
    获取所述第一点云中的第一数据点的位置坐标和所述待识别数据点的位置坐标；
    计算所述待识别数据点与所述第一数据点之间的距离，并将所述第一数据点标记为已查找；
    所述提取所述原始点云数据中的所有所述邻域点之前，还包括：
    判断所述第一点云中是否存在未标记数据点；
    若存在，则返回执行所述选择所述第一点云中的未标记数据点作为所述第一数据点的步骤。
  5. 根据权利要求3所述的方法，其特征在于，所述判断所述距离是否小于预设距离之后，还包括：
    若所述距离大于所述预设距离，则将所述待识别数据点标记为噪声点。
  6. 根据权利要求2所述的方法，其特征在于，所述利用曲面拟合方法在所述原始点云数据中进行网格重建包括：
    利用隐式曲面拟合方法对所述原始点云数据进行曲面拟合，以得到所述网格曲面。
  7. 一种图像处理设备，其特征在于，相互连接的通信电路和处理器；
    所述通信电路用于获取原始点云数据；
    所述处理器用于识别所述原始点云数据中的基准表面，所述基准表面包括组成所述基准表面的第一点云，并获取所述第一点云中数据点的邻域点，以得到去噪后的点云。
  8. 根据权利要求7所述的图像处理设备，其特征在于，所述处理器还用于利用曲面拟合方法在所述原始点云数据中进行网格重建，并将所述网格重建后的网格曲面作为所述基准表面，所述基准表面包括由所述原始点云数据进行曲面拟合得到的所述第一点云。
  9. 根据权利要求8所述的图像处理设备，其特征在于，所述处理器还用于计算所述第一点云中第一数据点与所述原始点云数据中的待识别数据点的距离，判断所述距离是否小于预设距离，并在所述距离小于所述预设距离时，将所述待识别数据点标记为所述邻域点，提取所述原始点云数据中的所有所述邻域点，组成所述去噪后的点云。
  10. 一种具有存储功能的装置，存储有指令，其特征在于，所述指令被执行时实现如权利要求1-6任一项所述的点云去噪方法。
PCT/CN2019/121762 2018-12-03 2019-11-28 点云去噪方法、图像处理设备及具有存储功能的装置 WO2020114321A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811466082.3 2018-12-03
CN201811466082.3A CN109767391A (zh) 2018-12-03 2018-12-03 点云去噪方法、图像处理设备及具有存储功能的装置

Publications (1)

Publication Number Publication Date
WO2020114321A1 true WO2020114321A1 (zh) 2020-06-11

Family

ID=66450358

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/121762 WO2020114321A1 (zh) 2018-12-03 2019-11-28 点云去噪方法、图像处理设备及具有存储功能的装置

Country Status (2)

Country Link
CN (1) CN109767391A (zh)
WO (1) WO2020114321A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113538264A (zh) * 2021-06-30 2021-10-22 深圳大学 一种点云数据的去噪方法、装置及存储介质

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109767391A (zh) * 2018-12-03 2019-05-17 深圳市华讯方舟太赫兹科技有限公司 点云去噪方法、图像处理设备及具有存储功能的装置
CN110515054B (zh) * 2019-08-23 2021-07-23 斯坦德机器人（深圳）有限公司 滤波方法和装置、电子设备、计算机存储介质
CN114240779B (zh) * 2021-12-15 2024-03-19 成都飞机工业（集团）有限责任公司 一种点云去噪方法、装置、设备及存储介质
CN117934858B (zh) * 2024-03-21 2024-06-04 之江实验室 一种点云的处理方法、装置、存储介质及电子设备

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2003027961A2 (en) * 2001-09-24 2003-04-03 Raindrop Geomagic, Inc. Surfaces reconstruction from data point sets
CN105096268A (zh) * 2015-07-13 2015-11-25 西北农林科技大学 一种点云去噪平滑方法
CN105574929A (zh) * 2015-12-15 2016-05-11 电子科技大学 一种基于地面LiDAR点云数据的单株植被三维建模方法
CN109767391A (zh) * 2018-12-03 2019-05-17 深圳市华讯方舟太赫兹科技有限公司 点云去噪方法、图像处理设备及具有存储功能的装置

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101127031B (zh) * 2006-08-18 2011-05-04 鸿富锦精密工业（深圳）有限公司 点云数据均值过滤系统及方法
CN102663237B (zh) * 2012-03-21 2014-12-17 武汉大学 基于网格分块与移动最小二乘的点云数据全自动滤波方法
CN102881047B (zh) * 2012-08-01 2015-04-15 桂林电子科技大学 一种自动的非封闭隐式曲面重建方法
CN106017472B (zh) * 2016-05-17 2019-01-25 成都通甲优博科技有限责任公司 全局路线规划方法、全局路线规划系统及无人机
CN106548520A (zh) * 2016-11-16 2017-03-29 湖南拓视觉信息技术有限公司 一种点云数据去噪的方法和系统

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2003027961A2 (en) * 2001-09-24 2003-04-03 Raindrop Geomagic, Inc. Surfaces reconstruction from data point sets
CN105096268A (zh) * 2015-07-13 2015-11-25 西北农林科技大学 一种点云去噪平滑方法
CN105574929A (zh) * 2015-12-15 2016-05-11 电子科技大学 一种基于地面LiDAR点云数据的单株植被三维建模方法
CN109767391A (zh) * 2018-12-03 2019-05-17 深圳市华讯方舟太赫兹科技有限公司 点云去噪方法、图像处理设备及具有存储功能的装置

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113538264A (zh) * 2021-06-30 2021-10-22 深圳大学 一种点云数据的去噪方法、装置及存储介质

Also Published As

Publication number Publication date
CN109767391A (zh) 2019-05-17

Similar Documents

Publication Publication Date Title
WO2020114321A1 (zh) 点云去噪方法、图像处理设备及具有存储功能的装置
US10510148B2 (en) Systems and methods for block based edgel detection with false edge elimination
WO2018120038A1 (zh) 一种目标检测的方法及装置
CN110992326B (zh) 一种qfn芯片引脚图像快速倾斜校正方法
JP4926116B2 (ja) 画像検査装置
JP2019032830A (ja) 対象オブジェクトをつかみ取るための把持姿勢を検出するシステム及び方法
WO2019169635A1 (en) Object recognition
CN104616278B (zh) 三维点云兴趣点检测方法和系统
US9886619B2 (en) Fingerprint template based on fuzzy feature point information and fingerprint identification method
CN107749071B (zh) 大畸变棋盘格图像角点检测方法及装置
US11189032B2 (en) Method and apparatus for extracting a satellite image-based building footprint
JP2020148625A (ja) 画像処理装置、画像処理方法、及び画像処理プログラム
JP6338429B2 (ja) 被写体検出装置、被写体検出方法及びプログラム
JP7121132B2 (ja) 画像処理方法、装置及び電子機器
CN115131521A (zh) 根据结构化点云生成对象表面的三维表示
CN113516619A (zh) 一种基于图像处理技术的产品表面瑕疵识别方法
CN117274246B (zh) 一种焊盘识别方法、计算机设备以及存储介质
CN114626118A (zh) 建筑物室内模型生成方法及装置
US9792675B1 (en) Object recognition using morphologically-processed images
WO2024016686A1 (zh) 角点检测的方法和装置
CN116434071A (zh) 一种规整化建筑掩膜的确定方法、确定装置、设备及介质
CN116342585A (zh) 一种产品缺陷检测方法、装置、设备及存储介质
CN111144489B (zh) 匹配对滤除方法、装置、电子设备与存储介质
CN113870190A (zh) 竖直线条检测方法、装置、设备及存储介质
CN109815791B (zh) 基于血管的身份识别方法和装置

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: 19893517

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19893517

Country of ref document: EP

Kind code of ref document: A1