WO2017113260A1 - 一种三维点云模型重建方法及装置 - Google Patents
一种三维点云模型重建方法及装置 Download PDFInfo
- Publication number
- WO2017113260A1 WO2017113260A1 PCT/CN2015/100005 CN2015100005W WO2017113260A1 WO 2017113260 A1 WO2017113260 A1 WO 2017113260A1 CN 2015100005 W CN2015100005 W CN 2015100005W WO 2017113260 A1 WO2017113260 A1 WO 2017113260A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- point
- skeleton
- points
- dimensional
- normal vector
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims abstract description 69
- 239000013598 vector Substances 0.000 claims abstract description 64
- 238000005070 sampling Methods 0.000 claims abstract description 47
- 238000005457 optimization Methods 0.000 claims abstract description 46
- 230000000295 complement effect Effects 0.000 claims abstract description 14
- 230000003044 adaptive effect Effects 0.000 claims abstract description 12
- 238000009826 distribution Methods 0.000 claims description 22
- 230000006870 function Effects 0.000 claims description 20
- 230000014759 maintenance of location Effects 0.000 claims description 11
- 238000004458 analytical method Methods 0.000 claims description 5
- 238000009499 grossing Methods 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 5
- 230000002146 bilateral effect Effects 0.000 claims description 3
- 238000000513 principal component analysis Methods 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 claims 4
- 238000004422 calculation algorithm Methods 0.000 description 23
- 238000010586 diagram Methods 0.000 description 11
- 230000000694 effects Effects 0.000 description 10
- 238000000605 extraction Methods 0.000 description 8
- 238000012545 processing Methods 0.000 description 7
- 238000004590 computer program Methods 0.000 description 6
- 230000008901 benefit Effects 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 239000011159 matrix material Substances 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 230000005484 gravity Effects 0.000 description 3
- 238000003860 storage Methods 0.000 description 3
- 235000011449 Rosa Nutrition 0.000 description 2
- 230000008602 contraction Effects 0.000 description 2
- 230000001419 dependent effect Effects 0.000 description 2
- 239000006185 dispersion Substances 0.000 description 2
- 238000006073 displacement reaction Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 238000011960 computer-aided design Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000000265 homogenisation Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000009877 rendering Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T13/00—Animation
- G06T13/20—3D [Three Dimensional] animation
- G06T13/40—3D [Three Dimensional] animation of characters, e.g. humans, animals or virtual beings
-
- 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
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2210/00—Indexing scheme for image generation or computer graphics
- G06T2210/56—Particle system, point based geometry or rendering
Definitions
- the invention relates to an image processing technology, in particular to a method and a device for reconstructing a three-dimensional point cloud model.
- the so-called point cloud model generally refers to a collection of three-dimensional coordinate points on the surface of the object by the scanning light emitted by the three-dimensional scanning device onto the surface of the object to be measured, receiving the reflected light.
- the so-called point cloud 3D reconstruction refers to complementing and reconstructing the grid data representing the source model based on a certain point cloud model data, which is convenient for computer rendering and user interaction. How to directly get the practical point cloud model directly from scanning point cloud data is still a problem.
- the difficulty of point cloud data processing is that point cloud data is generally scattered, with a lot of missing, noise and external points.
- point cloud completion is a recognized ill-conditioned problem. All methods use the existing information to infer unknown information. There is no way to ensure that the region and source model that are complemented by the algorithm are consistent.
- the prior art point cloud completion method is to fit and fill the missing information with the quadric surface according to the local information of the point cloud model. This method can fill a small area of the hole well, but can not handle large areas of the missing. Considering more global information is an important way to improve existing completion methods.
- An embodiment of the present invention provides a method for reconstructing a three-dimensional point cloud model, including:
- the present invention also provides a three-dimensional point cloud model reconstruction apparatus, including:
- Scanning module scanning model obtains input point set
- Memory storing program instructions
- the processor is coupled to the scan module and the memory, executes program instructions in the memory, and processes the input point set as follows:
- FIG. 1 is a flowchart of a method for reconstructing a three-dimensional point cloud model provided by the present invention
- FIG. 2 is a schematic structural diagram of a three-dimensional point cloud model reconstruction apparatus provided by the present invention.
- FIG. 3 is a comparison diagram of a point cloud expression of the present invention and a point cloud expression of the prior art
- Figure 5 is a schematic view of an embodiment of the present invention.
- FIG. 6 is a comparison diagram of a point cloud expression of the present invention and a point cloud expression of the prior art
- FIG. 7 is a comparison diagram of a point cloud expression of the present invention and a point cloud expression of the prior art
- FIG. 8 is a comparison diagram of a point cloud expression of the present invention and a point cloud expression of the prior art
- Figure 9 is a schematic view of an embodiment of the present invention.
- FIG. 10 is a flowchart of a technical solution in an embodiment of the present invention.
- FIG. 11 is a schematic diagram of surface point inward shifting and two neighborhoods in an embodiment of the present invention.
- An embodiment of the present invention provides a method for reconstructing a three-dimensional point cloud model, as shown in FIG. 1 , which is a flowchart of a method for reconstructing a three-dimensional point cloud model provided by the present invention, including:
- Step S101 sampling the input point set and WLPO optimization to generate an initial surface point set, and copying the initial surface point set as an initial position of the internal skeleton point set, and establishing a correspondence relationship between the surface point and the skeleton point;
- Step S102 moving the points in the inner skeleton point set inward in the opposite direction of the normal vector to generate an internal point
- Step S103 using an adaptive anisotropic neighborhood as a regular term to solve the optimization problem of the internal point, and generating a skeleton point;
- Step S104 performing optimization and complementing the initial surface point set by using the skeleton point to generate an optimized surface point
- Step S105 Perform reconstruction of the three-dimensional point cloud model according to the corresponding relationship between the skeleton point, the surface point, and the surface point and the skeleton point.
- the present invention also provides a three-dimensional point cloud model reconstruction device, as shown in FIG. 2, which is a schematic structural diagram of a three-dimensional point cloud model reconstruction device 20 provided by the present invention, including:
- the scanning module 321 scans the model to obtain an input point set
- a memory 323 storing program instructions
- the processor 323 is connected to the scan module and the memory, executes program instructions in the memory, processes the input point set as follows, and outputs a model:
- the invention mainly proposes a new point cloud expression mode, which can be used to solve the problem of denoising and filling holes of the point cloud model.
- the corresponding computer 3D point cloud data can be quickly obtained.
- This data is often accompanied by a large number of missing and noise (as shown in the black dot on the right in Figure 3).
- the goal of our method is to automatically obtain the set of line segments as shown in the left figure.
- One end of the line segment is the optimized surface point, and the other end is the skeleton point that can express the global topology information of the object.
- the main innovation of this method is to uniformly process the local surface and semi-global skeleton information of the 3D model, that is, extract the skeleton information according to the surface information of the model, and use the extracted skeleton information to complete the missing information, and finally jointly optimize the two dualities. information.
- the skeleton extracted by the present invention is not a traditional one-dimensional curve skeleton, but a mixed skeleton having a one-dimensional curve and a two-dimensional sheet. Compared with the two, the one-dimensional curve skeleton is more abstract, and the combined skeleton is more expressive.
- the method produces a one-dimensional curve, and in other regions, a two-dimensional slice is produced.
- a targeted strategy can be used when the point cloud is completed. Therefore, how to extract a combined skeleton from a point cloud is one of the key points of the present invention.
- the present invention proposes a new point cloud expression method, which relates to point cloud denoising optimization, point cloud skeleton extraction, point cloud completion and reconstruction in the prior art, and the present invention is directed to the prior art.
- Point cloud denoising optimization, point cloud skeleton extraction, point cloud completion and reconstruction are improved.
- the best methods for the three directions in this prior art are described below, as well as the differences from the present method.
- Locally Optimal Projection is the projection (downsampling) of a scattered point cloud into a new set of points that better express the potential model of the object.
- it can also streamline and homogenize the point cloud, and at the same time facilitate the calculation of point cloud normal vectors and topologies.
- the advantages of this algorithm are: parameterless, robust, efficient, and do not require the use of point normal vector information.
- the principle of the algorithm is as follows:
- the point set P is randomly sampled from the origin set C.
- is a quadratic paradigm, Is a fast-decreasing weight function that defines the range of neighborhood search.
- h is Where d bb is the diagonal of the 3D model binding cube, and m is the number of points in C.
- E 1 is equivalent to a kind of "gravity", pulling each point x i in P k to its corresponding center position of the set of neighbors in C, in order to let P k Eliminate the noise while expressing the original model implied in P as much as possible.
- E 2 is equivalent to a "repulsive force” that causes points in P k to repel each other. The closer the distance between the two points, the greater the repulsive force, thus achieving the goal of global homogenization.
- the two forces in P k are minimized, that is, the two forces are relatively balanced, the resulting set of projection points P is more compact and uniform.
- WLOP Weighted LOP
- sampling point p i does not have enough input points c j as the basis for data fitting, use our newly defined volume holding item, that is, the "volume" of the sampling point p i and its surrounding neighboring point p j should be exhausted. May be consistent.
- volume holding item that is, the "volume" of the sampling point p i and its surrounding neighboring point p j should be exhausted. May be consistent.
- each sampling point corresponds to a skeleton point one by one.
- the specific algorithm will be introduced later.
- the algorithm of this scheme does not hardly classify the sampling point P into two different cases, but adds a weight function to mix the data fitting term and the volume holding term. That is, if the sampling point p i is closer to the input data point and the surrounding volume changes greatly, the weight of the data fitting term is larger; if the sampling point p i is far from the input data point and the surrounding volume changes are relatively uniform, then The weight retention item has a larger weight. Therefore, WLOP can only keep the surface smooth on the existing data, and this program adds a volume retention (smoothing) item, so that it can be fully optimized where there is no data, as shown in Figure 4, where a It is the input point cloud, and b is the result of mesh reconstruction directly on the input point cloud. c is the result of WLOP, and d is the corresponding mesh reconstruction. e is the result of our method, and f is the corresponding mesh reconstruction. As you can see, using our method, we can get better optimization where the data is missing.
- the present invention proposes to project "gravity” and “repulsive force” in WLOP onto the hair vector and the tangent plane of the model surface, respectively.
- “gravity” and “repulsive force” are two major advantages to this:
- “gravitational force” only acts on the normal vector direction of the surface point (the surface point is basically synonymous with the sampling point), that is, the surface point can only follow its “gravitational force”
- the normal vector moves in the forward and reverse directions.
- the sampling point may go out of the boundary of the input model, that is, go to the area without input point support, making the hole possible.
- connection information of surface points and skeleton points to correct the normal vectors of sample points without data support.
- the specific algorithm will be introduced later. Through the use of this technology, even in areas where data is seriously missing, the sampling points can well complete the task of filling holes with the aid of skeleton points.
- the core of the L1 central skeleton method is the local L_1 median with a regular term, which can be summarized by the following optimization formula:
- the starting point is a set of points that are randomly sampled from Q, and the number of points
- X becomes a regular, "skeleton point set" at the center of the input model, as shown in Figure 5.
- the distribution of points in space is generally described in terms of dispersion or intensity.
- the distribution of the skeleton points is relatively dispersed in each dimension.
- the skeleton main part the skeleton points are continuously concentrated and distributed on one-dimensional straight lines.
- the effect of the distance weight function ⁇ is the same as the localized L 1 median value, in order to consider the local distribution of the sampling points rather than the weight distribution.
- x i is a row vector and C i is a 3 ⁇ 3 matrix.
- C i is a 3 ⁇ 3 matrix.
- the corresponding feature vector form an orthogonal frame, which is the three main directions of the point set.
- the magnitude of the feature value corresponding to a main direction indicates the intensity of the point distribution in the main direction, which may also be referred to as energy in the direction. If the size of the three eigenvalues is similar, the three main directions are similar in intensity, that is, the distribution of the point set in the three-dimensional space is more dispersed. Conversely, if the largest eigenvalue is much larger than the other two eigenvalues, the point is concentrated in a certain direction.
- ⁇ i ⁇ i ⁇ I is a balance parameter used to control the balance between the energy of the source input point and the repulsive force between the sample points.
- the above is an introduction to the L1-center skeleton.
- the present invention inherits its idea in the point cloud skeleton calculation method, but has the following differences:
- point cloud skeleton extraction An important application of point cloud skeleton extraction is point cloud data completion. In the case of serious data loss, the point cloud skeleton can help ensure that the reconstructed data can maintain the original topology of the model. It has been observed that the use of only one-dimensional curved skeleton is not conducive to the completion of three-dimensional shapes. For a non-cylindrical part of the object, it is more convenient to use a two-dimensional curved skeleton. As shown in the figure below, the one-dimensional curve skeleton is of great help in the complementation of cylindrical objects. However, the original one-dimensional curve plus the two-dimensional curved hybrid skeleton of the method can better express the non-cylindrical object parts, and thus the complementing effect is better.
- Figure 6 The effect of different types of point cloud skeletons on point cloud completion.
- (a) is to perform mesh reconstruction directly on the input point cloud.
- (b) and (c) calculate the one-dimensional curve skeleton using the L1 point cloud center skeleton and the ROSA point cloud skeleton algorithm, respectively, and use it for point cloud completion.
- (d) is the effect of point cloud completion by the hybrid point cloud skeleton calculated by our method.
- this scheme makes two improvements to the L1-center skeleton:
- the surface points are initially moved inward to obtain a set of points inside the object.
- the main idea of moving the surface point is to let the surface point go along the opposite direction of its normal vector (ie, toward the inside of the object) until it stops at the point of internal displacement of the object.
- One is to avoid the problem that the calculated skeleton of the existing method may be located outside the object. Once the point cloud skeleton is outside the object, it will have a catastrophic effect on the point cloud completion and destroy the overall topology of the model.
- the problem of neighborhood ambiguity is solved by gradually increasing the neighborhood in the L1-center skeleton, which greatly saves the calculation time of the algorithm and the time for the user to adjust the algorithm parameters (the initial neighborhood and the neighborhood growth rate).
- the point distribution weight function in the weighted regular term of the L1-center skeleton It was remodeled.
- the main idea is to replace the weighted regularity of the L1-central skeleton with a regularity of the opposite sex.
- the weighted regular item controls the size of the "repulsive force" of the regular item by the distribution of local points, and the regularity of our anisotropic is based on the distribution of internal points and the distribution of the inner and outer points. Go to each local main direction.
- This type of traditional implicit surface reconstruction algorithm does a good job of fitting localized regions of data, but in areas where global data is largely missing, only the output mesh is guaranteed to be sealed.
- Another method is to use the point cloud skeleton to complete the missing data, to ensure that the global topology of the object is maintained, and then use the traditional method for surface reconstruction.
- the method of the present invention belongs to the second type of method with the L1-center skeleton and the ROSA skeleton mentioned above. As shown in the following two figures, especially for heavily missing or flaky data, our method utilizes the semi-global information of the skeleton to better restore the overall topology of the object.
- (b) in Fig. 7 is the result of the Poisson reconstruction algorithm on (a) the input point cloud.
- (c) (d) and (e) are the skeleton points produced by our method, the optimized surface points and the results of Poisson reconstruction on the surface points.
- (a) a picture of an object.
- (b) WLOP optimized point cloud.
- (c) Optimization results of our algorithm.
- (d) Possion reconstructs the results on (b).
- a new point cloud expression - depth point Each surface of the depth point has a one-to-one correspondence with a skeleton point. It is an expression that combines the local surface and semi-global volume information of an object.
- a method for extracting a hybrid point cloud skeleton combining a one-dimensional curve and a two-dimensional thin surface is a key technique for constructing depth points.
- a point cloud optimization method that uses WLOP to improve WLOP, and information can be supplemented and enhanced in areas where no data is input.
- FIG. 10 is a flowchart of a technical solution of the present invention.
- Figure 9 is a physical example in which we obtain an input scattered point cloud (a) by scanning the model; then randomly downsample a few points from the point cloud data of (a) and optimize it to denoise using the traditional WLOP algorithm. (b); Next, we calculate the skeleton point (c) by the technique of surface point internal displacement and internal point contraction; then, we use the skeleton point to further optimize and complement the surface sampling points (d); Get a complete set of depth points (e) we proposed. The specific implementation details are described step by step below.
- Point cloud preprocessing ie point cloud down sampling and WLOP optimization.
- For input point sets We perform random sampling to get a point set It is smoothed and optimized by the traditional WLOP algorithm introduced above. Then, we get the internal skeleton point set by copying the surface point set P The initial position. At this point, the two sets of points are identical, and in the subsequent algorithm, there is no addition, deletion or change of the point number of the two point sets, thereby naturally ensuring that each surface point and its corresponding skeleton point are always maintained.
- One-to-one correspondence One-to-one correspondence.
- a depth point is composed of a pair of points ⁇ p i , q i >, wherein the distribution of p i and q i is located on the surface of the model and the inner skeleton. a complete set of depth points The following characteristics should be met:
- the surface point P is situated on the implicit surface of the model and is evenly distributed;
- - skeleton point set Q constitutes a mixed skeleton of a model consisting of a one-dimensional curve and a two-dimensional thin surface
- the surface points move inward to form internal points. That is, the surface point set underwear which is the copy of the initial position of the skeleton point, and uses the normal vector information of the point cloud to make the point set Perform a preliminary internal shift to obtain an internal point set inside the object.
- point shifting is to let the point go in the opposite direction of its normal vector (ie, toward the inside of the object) until it stops at the inward point of the object.
- the condition that the point stops moving inward in the neighborhood Q i of the point, the angle with the maximum normal vector of the point is less than a threshold, ie Among them, the default value of ⁇ is 45°.
- the moving step t of each point is set to r/2.
- FIG. 11 is a schematic diagram of surface point inward shifting and two neighborhoods in an embodiment of the present invention.
- the shape of the neighborhood is a thin and long ellipsoid; in the plane of the class, the shape of the neighborhood is a large and flat ellipsoid; at the edge of the endpoint and the plane, adjacent
- the shape of the domain is a small ellipsoid, avoiding excessive contraction of the boundary and maintaining the overall shape.
- PCA principal component analysis on the neighborhood Q i of q i to obtain the three principal axes of the neighborhood ellipsoid.
- second section Is a regular item, defined as: among them Is any two orthogonal vectors perpendicular to the tangent plane of the normal vector.
- the third item is the shape (volume) retention term, defined as: among them Is the average volume thickness of the current point.
- the volume retention item can be replaced as a volume minimization, ie
- the analytical expression among them The neighboring point of the i' sampling point.
- a new way of expressing point clouds depth points. Each surface of the depth point has a one-to-one correspondence with a skeleton point. It is an expression that combines the local surface and semi-global volume information of an object. It can be applied to point cloud completion and reconstruction, feature extraction, point cloud deformation, point cloud registration and so on.
- a method for extracting a hybrid point cloud skeleton combining a one-dimensional curve and a two-dimensional thin surface is a key technique for constructing a depth point.
- the hybrid skeleton is more abundant than the traditional one-dimensional curved skeleton, and can be applied to point cloud completion, three-dimensional matching retrieval, three-dimensional animation and the like.
- a point cloud optimization method that uses the depth point to improve the WLOP algorithm.
- the traditional WLOP method can only perform denoising and smoothing in areas with input data, and this method can complement and enhance data in areas where no data is input.
- a technical concept for joint optimization of surface points and skeleton points in the process of forming depth points is dependent on the relationship information such as the surface neighborhood provided by the optimized surface points.
- the complement and enhancement of the surface points depends on the volume and internal and external orientation provided by the formed skeleton points. The two complement each other and can be continuously iterated and optimized.
- the present invention proposes a point cloud skeleton expression method which is more conducive to point cloud completion, and combines the concept of depth points to make the best point cloud available.
- the noise method WLOP can perform point cloud completion while denoising.
- the three-dimensional reconstruction based on the complementary and optimized point cloud is more suitable for the large number of missing data than the direct 3D reconstruction.
- the reconstructed model is more correct in topology.
- 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
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Geometry (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computer Graphics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Processing Or Creating Images (AREA)
- Image Generation (AREA)
- Image Processing (AREA)
Abstract
Description
Claims (24)
- 一种三维点云模型重建方法,其特征在于,所述的方法包括:1)对输入点集进行采样及WLPO优化生成初始表面点集,并复制所述初始表面点集作为内部骨架点集的初始位置,建立表面点与骨架点的对应关系;2)将所述内部骨架点集中的点沿其法向量的反方向内移生成内部点;3)用自适应的各项异性邻域作为规整项求解所述内部点的最优化问题,生成骨架点;4)利用所述骨架点对所述初始表面点集进行优化补全生成优化后的表面点;5)根据所述的骨架点、表面点及表面点与骨架点的对应关系进行三维点云模型的重建。
- 如权利要求4所述的三维点云模型重建方法,其特征在于,所述的将各采样点沿其法向量的反方向移动还包括:各采样点的首次移动步长t为r/2,其中,r为每个采样点与最近邻点的平均距离;各采样点后续的移动步长为其邻点在上一次迭代的移动步长的平均,用的是动态邻域Qi={qi′|||qi′-qi||<σqr}
- 一种三维点云模型重建装置,其特征在于,所述的装置包括:扫描模块,扫描模型获得输入点集;存储器,存储程序指令;处理器,与所述扫描模块和存储器相连接,执行存储器中的程序指令,按如下步骤对输入点集进行处理:1)对输入点集进行采样及WLPO优化生成初始表面点集,并复制所述初始表面点集作为内部骨架点集的初始位置,建立表面点与骨架点的对应关系;2)将所述内部骨架点集中的点沿其法向量的反方向内移生成内部点;3)用自适应的各项异性邻域作为规整项求解所述内部点的最优化问题,生成骨架点;4)利用所述骨架点对所述初始表面点集进行优化补全生成优化后的表面点;5)根据所述的骨架点、表面点及表面点与骨架点的对应关系进行三维点云模型的重建。
- 如权利要求16所述的三维点云模型重建装置,其特征在于,所述的将各采样点沿其法向量的反方向移动还包括:各采样点的首次移动步长t为r/2,其中,r为每个采样点与最近邻点的平均距离;各采样点后续的移动步长为其邻点在上一次迭代的移动步长的平均,用的是动态邻域Qi={qi′|||qi′-qi||<σqr}
Priority Applications (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201580001258.6A CN107223268B (zh) | 2015-12-30 | 2015-12-30 | 一种三维点云模型重建方法及装置 |
PCT/CN2015/100005 WO2017113260A1 (zh) | 2015-12-30 | 2015-12-30 | 一种三维点云模型重建方法及装置 |
US15/389,959 US10089781B2 (en) | 2015-12-30 | 2016-12-23 | Three-dimensional point cloud model reconstruction method, computer readable storage medium and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/CN2015/100005 WO2017113260A1 (zh) | 2015-12-30 | 2015-12-30 | 一种三维点云模型重建方法及装置 |
Related Child Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US15/389,959 Continuation US10089781B2 (en) | 2015-12-30 | 2016-12-23 | Three-dimensional point cloud model reconstruction method, computer readable storage medium and device |
Publications (2)
Publication Number | Publication Date |
---|---|
WO2017113260A1 true WO2017113260A1 (zh) | 2017-07-06 |
WO2017113260A8 WO2017113260A8 (zh) | 2017-08-10 |
Family
ID=59224208
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2015/100005 WO2017113260A1 (zh) | 2015-12-30 | 2015-12-30 | 一种三维点云模型重建方法及装置 |
Country Status (3)
Country | Link |
---|---|
US (1) | US10089781B2 (zh) |
CN (1) | CN107223268B (zh) |
WO (1) | WO2017113260A1 (zh) |
Families Citing this family (51)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2017031718A1 (zh) * | 2015-08-26 | 2017-03-02 | 中国科学院深圳先进技术研究院 | 弹性物体变形运动的建模方法 |
US10474927B2 (en) * | 2015-09-03 | 2019-11-12 | Stc. Unm | Accelerated precomputation of reduced deformable models |
US10559079B2 (en) * | 2017-05-19 | 2020-02-11 | Uih America, Inc. | System and method for image reconstruction |
EP3467789A1 (en) * | 2017-10-06 | 2019-04-10 | Thomson Licensing | A method and apparatus for reconstructing a point cloud representing a 3d object |
US11244502B2 (en) * | 2017-11-29 | 2022-02-08 | Adobe Inc. | Generating 3D structures using genetic programming to satisfy functional and geometric constraints |
CN108022221A (zh) * | 2017-12-06 | 2018-05-11 | 河北工业大学 | 一种各项异性扩散滤波的三维散乱点云平滑去噪方法 |
CN108460733B (zh) * | 2018-01-31 | 2022-04-22 | 北京大学深圳研究生院 | 一种逐步精细化的图像去噪方法及*** |
CN108492329B (zh) * | 2018-03-19 | 2022-04-12 | 北京航空航天大学 | 一种三维重建点云精度和完整度评价方法 |
CN110390683B (zh) * | 2018-04-17 | 2022-08-12 | 河海大学 | 一种基于密集点云的古城墙三维裂缝检测方法 |
CN110570511B (zh) * | 2018-06-06 | 2023-04-28 | 阿里巴巴集团控股有限公司 | 点云数据的处理方法、装置、***和存储介质 |
CN109147038B (zh) * | 2018-08-21 | 2023-02-07 | 北京工业大学 | 基于三维点云处理的管道三维建模方法 |
CN109389672B (zh) * | 2018-09-29 | 2023-03-21 | 广东工业大学 | 一种反光工件点云的处理方法 |
CN109816705B (zh) * | 2019-01-03 | 2022-10-11 | 西安财经大学 | 一种缺失颅骨的特征点配准方法 |
CN109887009B (zh) * | 2019-01-24 | 2022-12-09 | 西北大学 | 一种点云局部匹配方法 |
CN110009726B (zh) * | 2019-03-08 | 2022-09-30 | 浙江中海达空间信息技术有限公司 | 一种根据平面基元间的结构关系对点云提取平面的方法 |
CN110246092B (zh) * | 2019-05-02 | 2022-09-02 | 江西理工大学 | 一种顾及邻域点均值距离和斜率的三维激光点云去噪方法 |
US11671580B2 (en) * | 2019-05-14 | 2023-06-06 | Board Of Supervisors Of Louisiana State University And Agricultural And Mechanical College | System and method for reconstructing 3-D shapes of objects from reflection images |
CN112602121B (zh) * | 2019-08-01 | 2022-06-24 | 西门子(中国)有限公司 | 重构点云模型的方法、装置和*** |
US10825243B1 (en) | 2019-08-15 | 2020-11-03 | Autodesk, Inc. | Three-dimensional (3D) model creation and incremental model refinement from laser scans |
CN110796694A (zh) * | 2019-10-13 | 2020-02-14 | 西北农林科技大学 | 一种基于KinectV2的果实三维点云实时获取方法 |
CN111091614B (zh) * | 2019-11-04 | 2023-09-15 | 国网内蒙古东部电力有限公司检修分公司 | 一种变电站三维实景模型建立方法及*** |
CN111402393A (zh) * | 2019-12-06 | 2020-07-10 | 温州大学 | 参数曲面模拟点云生成方法 |
CN111080684B (zh) * | 2019-12-12 | 2023-11-17 | 哈尔滨工程大学 | 一种点邻域尺度差异描述的点云配准方法 |
CN114981845A (zh) * | 2020-01-19 | 2022-08-30 | Oppo广东移动通信有限公司 | 图像扫描方法及装置、设备、存储介质 |
CN111105381B (zh) * | 2020-02-07 | 2023-03-31 | 武汉玄景科技有限公司 | 一种基于球面模型的稠密点云平滑方法 |
CN111353416B (zh) * | 2020-02-26 | 2023-07-07 | 广东温氏种猪科技有限公司 | 基于牲畜三维测量的姿态检测方法、***及存储介质 |
CN111354090B (zh) * | 2020-03-06 | 2023-06-13 | 贝壳技术有限公司 | 一种自识别的三维点云模型孔洞修补方法及装置 |
CN111383355B (zh) * | 2020-04-03 | 2023-06-27 | 如你所视(北京)科技有限公司 | 三维点云补全方法、装置和计算机可读存储介质 |
CN111540040B (zh) * | 2020-04-20 | 2022-12-27 | 上海曼恒数字技术股份有限公司 | 一种基于点云数据进行模型构建方法、装置、存储介质 |
CN111583391B (zh) * | 2020-04-29 | 2023-07-14 | 北京深测科技有限公司 | 一种物体三维重建方法和*** |
CN111583392B (zh) * | 2020-04-29 | 2023-07-14 | 北京深测科技有限公司 | 一种物体三维重建方法和*** |
CN111739134B (zh) * | 2020-07-30 | 2023-05-26 | 腾讯科技(深圳)有限公司 | 虚拟角色的模型处理方法、装置及可读存储介质 |
CN112348959B (zh) * | 2020-11-23 | 2024-02-13 | 杭州师范大学 | 一种基于深度学习的自适应扰动点云上采样方法 |
CN112465832B (zh) * | 2020-11-25 | 2024-04-16 | 重庆大学 | 一种基于双目视觉的单面树木点云骨架线提取方法及*** |
CN112396567B (zh) * | 2020-11-26 | 2023-06-06 | 河北工业大学 | 基于法向修正与位置滤波两步法的散乱点云去噪方法 |
CN112802178B (zh) * | 2021-01-15 | 2024-06-11 | 浙江华云信息科技有限公司 | 一种基于泊松曲面重建算法修复三维模型缺失信息的方法 |
CN112767551B (zh) * | 2021-01-18 | 2022-08-09 | 贝壳找房(北京)科技有限公司 | 三维模型构建方法、装置以及电子设备、存储介质 |
US11210793B1 (en) | 2021-02-12 | 2021-12-28 | Optum Technology, Inc. | Mixed reality object detection |
US11164391B1 (en) | 2021-02-12 | 2021-11-02 | Optum Technology, Inc. | Mixed reality object detection |
CN113424220B (zh) * | 2021-03-30 | 2024-03-01 | 商汤国际私人有限公司 | 生成点云补全网络及点云数据的处理 |
CN113516756B (zh) * | 2021-05-12 | 2022-03-22 | 华南农业大学 | 一种基于三维重构牲畜体点云的标准姿态鉴定方法 |
CN113295142B (zh) * | 2021-05-14 | 2023-02-21 | 上海大学 | 一种基于faro扫描仪和点云的地形扫描分析方法和装置 |
CN113722876B (zh) * | 2021-07-02 | 2023-06-02 | 上海工程技术大学 | 一种用于三维模型的自适应迭代蜂窝网格化计算方法 |
CN113542756B (zh) * | 2021-07-04 | 2023-02-10 | 西北工业大学 | 基于平面拟合中心坐标投影的点云空间可伸缩编码几何重构方法 |
CN113706686B (zh) * | 2021-07-09 | 2023-07-21 | 苏州浪潮智能科技有限公司 | 一种三维点云重建结果补全方法及相关组件 |
CN113409452B (zh) * | 2021-07-12 | 2023-01-03 | 深圳大学 | 一种三维线条的生成方法、存储介质及*** |
CN113855233B (zh) * | 2021-11-01 | 2023-05-05 | 杭州柳叶刀机器人有限公司 | 手术范围确定方法、装置、电子设备及存储介质 |
CN114241018B (zh) * | 2021-12-09 | 2023-05-09 | 电子科技大学 | 一种牙齿点云配准方法、***及可读存储介质 |
CN114581681B (zh) * | 2022-05-06 | 2022-07-29 | 南京航空航天大学 | 一种面向飞机整机测量数据的机身轮廓度分析方法 |
CN115512121B (zh) * | 2022-08-01 | 2023-05-16 | 南京林业大学 | 非完全模拟树木水分养分传输的枝干点云骨架提取方法 |
CN116452770B (zh) * | 2023-02-17 | 2023-10-20 | 北京德风新征程科技股份有限公司 | 三维模型重建方法、装置、设备和介质 |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102467753A (zh) * | 2010-11-04 | 2012-05-23 | 中国科学院深圳先进技术研究院 | 基于骨架配准的时变点云重建方法及*** |
CN103065354A (zh) * | 2012-12-24 | 2013-04-24 | 中国科学院深圳先进技术研究院 | 点云优化方法及其装置 |
CN104063898A (zh) * | 2014-06-30 | 2014-09-24 | 厦门大学 | 一种三维点云自动补全方法 |
WO2015149302A1 (zh) * | 2014-04-02 | 2015-10-08 | 中国科学院自动化研究所 | 基于点云与数据驱动的树木模型重建方法 |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8224094B2 (en) * | 2007-06-14 | 2012-07-17 | Siemens Audiologische Technik Gmbh | Method and system for side detection of 3D undetailed ear impressions |
CN101763652B (zh) * | 2009-06-03 | 2012-05-30 | 中国科学院自动化研究所 | 一种基于分叉特征的三维骨架快速提取方法 |
-
2015
- 2015-12-30 WO PCT/CN2015/100005 patent/WO2017113260A1/zh active Application Filing
- 2015-12-30 CN CN201580001258.6A patent/CN107223268B/zh active Active
-
2016
- 2016-12-23 US US15/389,959 patent/US10089781B2/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102467753A (zh) * | 2010-11-04 | 2012-05-23 | 中国科学院深圳先进技术研究院 | 基于骨架配准的时变点云重建方法及*** |
CN103065354A (zh) * | 2012-12-24 | 2013-04-24 | 中国科学院深圳先进技术研究院 | 点云优化方法及其装置 |
WO2015149302A1 (zh) * | 2014-04-02 | 2015-10-08 | 中国科学院自动化研究所 | 基于点云与数据驱动的树木模型重建方法 |
CN104063898A (zh) * | 2014-06-30 | 2014-09-24 | 厦门大学 | 一种三维点云自动补全方法 |
Also Published As
Publication number | Publication date |
---|---|
CN107223268B (zh) | 2020-08-07 |
WO2017113260A8 (zh) | 2017-08-10 |
US10089781B2 (en) | 2018-10-02 |
US20170193692A1 (en) | 2017-07-06 |
CN107223268A (zh) | 2017-09-29 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2017113260A1 (zh) | 一种三维点云模型重建方法及装置 | |
Dinesh et al. | Point cloud denoising via feature graph laplacian regularization | |
Han et al. | Point cloud simplification with preserved edge based on normal vector | |
Li et al. | Optcuts: Joint optimization of surface cuts and parameterization | |
JP7029283B2 (ja) | 画像の補完 | |
Fryazinov et al. | Multi-scale space-variant FRep cellular structures | |
WO2014187046A1 (zh) | 点云骨架提取方法及装置 | |
Ummenhofer et al. | Point-based 3D reconstruction of thin objects | |
Chen et al. | GPU-based polygonization and optimization for implicit surfaces | |
Zwicker et al. | Meshing Point Clouds Using Spherical Parameterization. | |
Dinesh et al. | Fast 3D point cloud denoising via bipartite graph approximation & total variation | |
Dinesh et al. | Local 3D point cloud denoising via bipartite graph approximation & total variation | |
Han et al. | Iterative guidance normal filter for point cloud | |
KR20050063994A (ko) | 경계셀 축소 방식을 이용하여 3차원 측정점들로부터 3차원메쉬를 생성하는 방법 | |
Zeng et al. | Least squares quasi-developable mesh approximation | |
CN112991504A (zh) | 一种改进的基于tof相机三维重建的补空洞方法 | |
Wei et al. | Field‐aligned Quadrangulation for Image Vectorization | |
Wang et al. | 3D incomplete point cloud surfaces reconstruction with segmentation and feature-enhancement | |
Valenzuela et al. | Automatic knot adjustment for b-spline smoothing approximation using improved clustering algorithm | |
Xu et al. | Anisotropic surface remeshing without obtuse angles | |
Gong et al. | Isotropic Gaussian Splatting for Real-Time Radiance Field Rendering | |
Wang et al. | A new interpolation method in mesh reconstruction from 3D point cloud | |
Yang et al. | A new RBF reflection model for shape from shading | |
Akdim et al. | A Study and Comparison of Different 3D Reconstruction Methods Following Quality Criteria. | |
Nie et al. | Bas-relief generation from point clouds based on normal space compression with real-time adjustment on CPU |
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: 15911888 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: 15911888 Country of ref document: EP Kind code of ref document: A1 |
|
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 13.11.2018) |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 15911888 Country of ref document: EP Kind code of ref document: A1 |