CN114972788A - Outlier extraction method and device of three-dimensional point cloud - Google Patents
Outlier extraction method and device of three-dimensional point cloud Download PDFInfo
- Publication number
- CN114972788A CN114972788A CN202210585410.1A CN202210585410A CN114972788A CN 114972788 A CN114972788 A CN 114972788A CN 202210585410 A CN202210585410 A CN 202210585410A CN 114972788 A CN114972788 A CN 114972788A
- Authority
- CN
- China
- Prior art keywords
- point cloud
- feature
- layer
- target
- dimensional
- Prior art date
- Legal status (The legal status 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 status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Length Measuring Devices By Optical Means (AREA)
Abstract
The invention provides a method and a device for extracting outliers of three-dimensional point cloud, wherein the method comprises the following steps: acquiring a target outlier extraction model which is trained completely, wherein the target outlier extraction model comprises a point cloud rotation module, a point cloud feature extraction module and a segmentation module; acquiring a three-dimensional point cloud to be extracted; rotating the three-dimensional point cloud to be extracted based on the point cloud rotating module to obtain a rotating point cloud; performing feature extraction on the rotating point cloud based on the point cloud feature extraction module to obtain target point cloud features; and extracting outliers in the three-dimensional point cloud to be extracted based on the segmentation module and the target point cloud characteristics. The method can improve the efficiency and accuracy of extracting the outliers of the three-dimensional point cloud.
Description
Technical Field
The invention relates to the technical field of three-dimensional point cloud outlier extraction, in particular to an outlier extraction method and device of a three-dimensional point cloud.
Background
The three-dimensional point cloud is a point data set of the product appearance surface obtained by a measuring instrument, and as the three-dimensional point cloud data acquisition technology is mature day by day, high-precision mass point cloud data on the surface of an object can be rapidly acquired in a short time by using advanced measuring equipment.
However, in the actual point cloud data collection process, the point cloud collected by the sensor inevitably contains noise and abnormal values (point cloud with wrong position information, i.e. outliers) under the influence of factors such as inherent errors of the collection equipment, illumination of the scene object surface, reflection properties or workpiece structure. The problems not only affect the precision of the point cloud model, but also have great influence on subsequent curved surface reconstruction, point cloud classification and the like, and for the problems, a global filtering method is adopted in the prior art to extract outliers.
The above solution has the following technical problems: the outliers extracted based on the global filtering method are not suitable for outliers generated due to reflection of the surface of the workpiece, and if the outliers are applied to a scene where the outliers generated due to reflection of the surface of the workpiece are removed, the global filtering method needs to be greatly improved, parameters are adjusted, the amount of workpieces is too large, and therefore the outlier extraction efficiency is low.
Disclosure of Invention
In view of the above, there is a need to provide a method and an apparatus for extracting outliers of a three-dimensional point cloud, so as to solve the technical problem in the prior art that the outliers generated by reflection on the surface of a workpiece are removed in a filtering and denoising manner, and the extraction efficiency is low.
In one aspect, the invention provides an outlier extraction method for three-dimensional point cloud, comprising the following steps:
acquiring a target outlier extraction model which is trained completely, wherein the target outlier extraction model comprises a point cloud rotation module, a point cloud feature extraction module and a segmentation module;
acquiring a three-dimensional point cloud to be extracted;
rotating the three-dimensional point cloud to be extracted based on the point cloud rotating module to obtain a rotating point cloud;
performing feature extraction on the rotating point cloud based on the point cloud feature extraction module to obtain target point cloud features;
and extracting outliers in the three-dimensional point cloud to be extracted based on the segmentation module and the target point cloud characteristics.
In some possible implementations, the point cloud rotation module includes a first upscaling layer, a first convolution layer, an average pooling layer, a first dimension adjustment layer, a first fully-connected layer, a second fully-connected layer, a third fully-connected layer, a second dimension adjustment layer, and a rotated point cloud determination layer;
the dimension raising layer is used for carrying out dimension raising operation on the three-dimensional point cloud to be extracted to obtain a first dimension raising point cloud;
the convolution layer is used for performing convolution operation on the first ascending-dimensional point cloud to obtain a first convolution characteristic;
the average pooling layer is used for carrying out average pooling operation on the first convolution characteristics to obtain first average pooling characteristics;
the first dimension adjusting layer is used for carrying out dimension adjustment on the first average pooling feature to obtain a first dimension adjusting feature;
the first full-connection layer is used for mapping the first dimension adjustment feature to a sample point cloud space to obtain a first mapping feature, and activating and carrying out batch standardization processing on the first mapping feature to obtain a first standard feature;
the second full-connection layer is used for mapping the first standard features to a sample point cloud space to obtain second mapping features, and activating and carrying out batch standardization processing on the second mapping features to obtain second standard features;
the third full-connection layer is used for mapping the second standard feature to a sample point cloud space to obtain a third mapping feature;
the second dimension adjusting layer is used for carrying out dimension adjustment on the third mapping characteristic to obtain a second dimension adjusting characteristic;
the rotating point cloud determining layer is used for obtaining the rotating point cloud according to the second dimension adjusting feature and the three-dimensional point cloud to be extracted.
In some possible implementations, the point cloud feature extraction module includes a preliminary feature extraction unit, a spatial feature alignment unit, and a target feature extraction unit;
the preliminary feature extraction unit is used for performing preliminary feature extraction on the rotating point cloud to obtain a low-dimensional preliminary point cloud feature, a medium-dimensional preliminary point cloud feature and a high-dimensional preliminary point cloud feature;
the spatial feature alignment unit is used for extracting features of the high-dimensional preliminary point cloud features to obtain spatial point cloud features;
the target feature extraction unit is used for obtaining the target point cloud features according to the low-dimensional preliminary point cloud features, the medium-dimensional preliminary point cloud features, the high-dimensional preliminary point cloud features and the space point cloud features.
In some possible implementations, the preliminary feature extraction unit includes a preliminary upscale layer, a first preliminary convolution layer, a second preliminary convolution layer, and a third preliminary convolution layer;
the initial ascending layer is used for carrying out ascending operation on the rotating point cloud to obtain an initial ascending point cloud;
the first preliminary convolution layer is used for performing convolution operation on the preliminary ascending-dimensional point cloud to obtain the low-dimensional preliminary point cloud characteristic;
the second preliminary convolution layer is used for performing convolution operation on the low-dimensional preliminary point cloud feature to obtain the medium-dimensional preliminary point cloud feature;
and the first preliminary convolution layer is used for performing convolution operation on the medium-dimensional preliminary point cloud feature to obtain the high-dimensional preliminary point cloud feature.
In some possible implementations, the preliminary feature extraction unit includes a preliminary upscale layer, a first preliminary convolution layer, a second preliminary convolution layer, and a third preliminary convolution layer;
the initial ascending layer is used for performing ascending operation on the rotating point cloud to obtain an initial ascending point cloud;
the first primary convolution layer is used for performing convolution operation on the primary upscaled point cloud to obtain the low-dimensional primary point cloud characteristic;
the second preliminary convolution layer is used for performing convolution operation on the low-dimensional preliminary point cloud feature to obtain the medium-dimensional preliminary point cloud feature;
and the first preliminary convolution layer is used for performing convolution operation on the medium-dimensional preliminary point cloud feature to obtain the high-dimensional preliminary point cloud feature.
In some possible implementations, the spatial feature alignment unit includes a first spatial convolution layer, a second spatial convolution layer, a spatial maximum pooling layer, a first spatial dimension adjustment layer, a first spatial fully-connected layer, a second spatial fully-connected layer, and a second spatial dimension adjustment layer;
the first space convolution layer is used for performing convolution operation on the high-dimensional preliminary point cloud feature to obtain a first space convolution feature;
the second space convolution layer is used for performing convolution operation on the first space convolution characteristic to obtain a second space convolution characteristic;
the spatial maximum pooling layer is used for performing maximum pooling operation on the second spatial convolution characteristics to obtain spatial maximum pooling characteristics;
the first space dimension adjusting layer is used for carrying out dimension adjustment on the space maximum pooling feature to obtain a first space dimension adjusting feature;
the first space full-connection layer is used for mapping the first space dimension adjustment feature to a sample space to obtain a first space mapping feature, and activating and carrying out batch standardization processing on the first space mapping feature to obtain a first space standard feature;
the second spatial full-link layer is used for mapping the first spatial standard feature to a sample space to obtain a second spatial mapping feature;
the second space dimension adjustment layer is used for carrying out dimension adjustment on the second space mapping characteristic to obtain the space point cloud characteristic.
In some possible implementations, the target feature extraction unit includes a first target dimension adjustment layer, a target processing layer, a first target dimension-increasing layer, a first target convolution layer, a second target convolution layer, a target maximum pooling layer, a leveling layer, a stitching layer, a plurality of convolution layers, and a second target dimension adjustment layer;
the first target dimension adjusting layer is used for adjusting the dimension of the high-dimensional preliminary point cloud feature to obtain a first target dimension adjusting feature;
the target processing layer is used for multiplying the target dimension adjusting layer and the space point cloud characteristics to obtain a first multiplication characteristic;
the first target dimension raising layer is used for carrying out dimension raising processing on the first multiplied feature to obtain a first target dimension raising feature;
the first target convolution layer is used for performing convolution processing on the first target dimension-increasing feature to obtain a first target convolution feature;
the second convolution layer is used for performing convolution processing on the first target convolution characteristic to obtain a second target convolution characteristic;
the target maximum pooling layer is used for performing maximum pooling processing on the second target convolution characteristics to obtain target pooling characteristics;
the flat layer is used for carrying out flat laying operation on the target pooling characteristic to obtain a target flat laying characteristic;
the splicing layer is used for splicing the low-dimensional preliminary point cloud feature, the medium-dimensional preliminary point cloud feature, the high-dimensional preliminary point cloud feature, the first target convolution feature, the second target convolution feature and the target tiling feature to obtain a target splicing feature;
the convolution layers are used for sequentially carrying out convolution processing on the target splicing characteristics to obtain third target convolution characteristics;
the second target dimension adjustment layer is used for carrying out dimension adjustment on the third target convolution characteristics to obtain the target point cloud characteristics.
In some possible implementations, the obtaining a well-trained target outlier extraction model includes:
constructing an initial outlier extraction model;
constructing a point cloud outlier sample set;
and training the initial outlier extraction model according to the point cloud outlier sample collection to obtain a target outlier extraction model with complete training.
In some possible implementations, the constructing a point cloud outlier sample set includes:
acquiring an initial three-dimensional point cloud of a reflective workpiece with a concave structure;
preprocessing the initial three-dimensional point cloud to obtain a three-dimensional point cloud to be marked;
marking outliers and point clouds around holes in the three-dimensional point cloud to be marked to obtain a marked three-dimensional point cloud;
and performing downsampling on the marked three-dimensional point cloud to obtain a point cloud outlier sample set.
In some possible implementation manners, the preprocessing the initial three-dimensional point cloud to obtain a three-dimensional point cloud to be marked includes:
determining a reflective workpiece area and a non-reflective workpiece area in the initial three-dimensional point cloud, and eliminating the non-reflective workpiece area to obtain a workpiece three-dimensional point cloud;
and removing noise point clouds in the three-dimensional point clouds of the workpiece to obtain the three-dimensional point clouds to be marked.
On the other hand, the invention also provides an outlier extracting device of the three-dimensional point cloud, which comprises the following steps:
the model acquisition unit is used for acquiring a target outlier extraction model which is completely trained, and the target outlier extraction model comprises a point cloud rotation module, a point cloud feature extraction module and a segmentation module;
the point cloud obtaining unit is used for obtaining a three-dimensional point cloud to be extracted;
the point cloud rotating unit is used for rotating the three-dimensional point cloud to be extracted based on the point cloud rotating module to obtain a rotating point cloud;
the characteristic extraction unit is used for extracting the characteristics of the rotating point cloud based on the point cloud characteristic extraction module to obtain the target point cloud characteristics;
and the outlier extraction unit is used for extracting outliers in the three-dimensional point cloud to be extracted based on the segmentation module and the target point cloud characteristics.
The beneficial effects of adopting the above embodiment are: according to the method for extracting the outlier of the three-dimensional point cloud, the outlier of the three-dimensional point cloud to be extracted is automatically extracted by acquiring the target outlier extraction model which is trained completely, and the efficiency of extracting the outlier of the three-dimensional point cloud is improved.
Furthermore, the point cloud rotating module is arranged to rotate the three-dimensional point cloud to the pose with more obvious characteristics, so that the accuracy of extracting outliers of the three-dimensional point cloud can be improved.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
FIG. 1 is a schematic flow chart of an embodiment of a method for extracting outliers of a three-dimensional point cloud according to the present invention;
FIG. 2 is a schematic structural diagram of an embodiment of a target outlier extraction model provided in the present invention;
FIG. 3 is a schematic structural diagram of an embodiment of a preliminary feature extraction unit provided in the present invention;
FIG. 4 is a schematic structural diagram of an embodiment of a spatial feature alignment unit provided in the present invention;
FIG. 5 is a schematic structural diagram of an embodiment of a target feature extraction unit provided in the present invention;
FIG. 6 is a schematic flow chart of one embodiment of S101 of FIG. 1;
FIG. 7 is a flowchart illustrating an embodiment of S602 in FIG. 6 according to the present invention;
FIG. 8 is a flowchart illustrating an embodiment of S702 of FIG. 7 according to the present invention;
FIG. 9 is a schematic structural diagram of an embodiment of an initial three-dimensional point cloud provided by the present invention;
FIG. 10 is a schematic structural diagram of an embodiment of a three-dimensional point cloud of a workpiece according to the present invention;
FIG. 11 is a schematic structural diagram of an embodiment of a three-dimensional point cloud to be marked according to the present invention;
FIG. 12 is a schematic structural diagram of an embodiment of a marked three-dimensional point cloud provided by the present invention;
FIG. 13 is a structural diagram of an embodiment of a sample set of point cloud outliers provided in the present invention;
FIG. 14 is a schematic structural diagram illustrating an outlier extracting apparatus for three-dimensional point cloud according to an embodiment of the present invention;
fig. 15 is a schematic structural diagram of an embodiment of an electronic device provided in the present invention.
Detailed Description
The technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. It is to be understood that the described embodiments are merely exemplary of the invention, and not restrictive of the full scope of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be understood that the schematic drawings are not necessarily to scale. The flowcharts used in this invention illustrate operations performed in accordance with some embodiments of the present invention. It should be understood that the operations of the flow diagrams may be performed out of order, and that steps without logical context may be performed in reverse order or concurrently. One skilled in the art, under the direction of this summary, may add one or more other operations to, or remove one or more operations from, the flowchart.
Some of the block diagrams shown in the figures are functional entities and do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in the form of software, or in one or more hardware modules or integrated circuits, or in different networks and/or processor systems and/or microcontroller systems.
Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the invention. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. It is explicitly and implicitly understood by one skilled in the art that the embodiments described herein can be combined with other embodiments.
The embodiment of the invention provides a method and a device for extracting outliers of three-dimensional point cloud, which are respectively explained below.
Before the embodiments are shown, the three-dimensional point cloud in the embodiments of the present invention is explained.
The three-dimensional point cloud in the embodiment of the invention is obtained by scanning the light-reflecting workpiece through the line structured light, and the light-reflecting workpiece comprises a concave surface area.
Fig. 1 is a schematic flowchart of an embodiment of a method for extracting outliers of a three-dimensional point cloud provided by the present invention, and fig. 2 is a schematic structural diagram of an embodiment of a target outlier extraction model provided by the present invention; as shown in fig. 1 and 2, the method for extracting outliers of a three-dimensional point cloud includes:
s101, acquiring a target outlier extraction model with complete training, wherein the target outlier extraction model comprises a point cloud rotation module, a point cloud feature extraction module and a segmentation module;
s102, acquiring three-dimensional point cloud to be extracted;
s103, rotating the three-dimensional point cloud to be extracted based on the point cloud rotating module to obtain a rotating point cloud;
s104, performing feature extraction on the rotating point cloud based on a point cloud feature extraction module to obtain target point cloud features;
and S105, extracting outliers in the three-dimensional point cloud to be extracted based on the segmentation module and the target point cloud characteristics.
Compared with the prior art, the method for extracting the outlier of the three-dimensional point cloud provided by the embodiment of the invention has the advantages that the outlier of the three-dimensional point cloud to be extracted is automatically extracted by acquiring the target outlier extraction model with complete training, so that the efficiency of extracting the outlier of the three-dimensional point cloud is improved.
Furthermore, the point cloud rotating module is arranged to rotate the three-dimensional point cloud to the pose with more obvious features, so that the accuracy of extracting outliers of the three-dimensional point cloud can be improved.
In some embodiments of the invention, as shown in fig. 2, the point cloud rotation module includes a first upscaling layer, a first convolution layer, an average pooling layer, a first dimension adjustment layer, a first fully-connected layer, a second fully-connected layer, a third fully-connected layer, a second dimension adjustment layer, and a rotated point cloud determination layer;
the dimension increasing layer is used for performing dimension increasing operation on the three-dimensional point cloud to be extracted to obtain a first dimension increasing point cloud;
the convolution layer is used for performing convolution operation on the first ascending-dimensional point cloud to obtain a first convolution characteristic;
the average pooling layer is used for carrying out average pooling operation on the first convolution characteristics to obtain first average pooling characteristics;
the first dimension adjusting layer is used for carrying out dimension adjustment on the first average pooling characteristic to obtain a first dimension adjusting characteristic;
the first full-connection layer is used for mapping the first dimension adjustment feature to a sample point cloud space to obtain a first mapping feature, and activating and carrying out batch standardization processing on the first mapping feature to obtain a first standard feature;
the second full-connection layer is used for mapping the first standard features to the sample point cloud space to obtain second mapping features, and activating and carrying out batch standardization processing on the second mapping features to obtain second standard features;
the third full-connection layer is used for mapping the second standard feature to the sample point cloud space to obtain a third mapping feature;
the second dimension adjusting layer is used for carrying out dimension adjustment on the third mapping characteristics to obtain second dimension adjusting characteristics;
and the rotating point cloud determining layer is used for obtaining a rotating point cloud according to the second dimension adjusting characteristic and the three-dimensional point cloud to be extracted.
According to the line-structured light progressive scanning characteristic, the point cloud rotation module comprises the average pooling layer, and the characteristic that the difference of certain channel coordinate values in the outliers is too large is reduced through average pooling, so that the rotation matrix closer to the ordered visual angle is found, the three-dimensional point cloud is rotated to the pose with more obvious characteristics, the rotation point cloud is obtained, and the accuracy of subsequent outlier point extraction is improved.
In a specific embodiment of the present invention, the size of the three-dimensional point cloud to be extracted is (32 × 2048 × 3), and the size of the first raised-dimensional point cloud is (32 × 2048 × 3 × 1); convolution kernel size of convolution layer is (1 x 512), step size is (1 x 1), first convolution feature is (32 x 2048 x 1 x 1024), first average pooling feature size is (32 x 1 x 1024), first dimension adjustment feature size is (32 x 1024); the first standard feature has a size of (32 x 256), the second standard feature has a size of (32 x 64), the third mapped feature has a size of (32 x 9), and the second dimension-adjusted feature has a size of (32 x 3).
In some embodiments of the present invention, as shown in fig. 2, the point cloud feature extraction module includes a preliminary feature extraction unit, a spatial feature alignment unit, and a target feature extraction unit;
the preliminary feature extraction unit is used for performing preliminary feature extraction on the rotating point cloud to obtain a low-dimensional preliminary point cloud feature, a medium-dimensional preliminary point cloud feature and a high-dimensional preliminary point cloud feature;
the spatial feature alignment unit is used for extracting features of the high-dimensional preliminary point cloud features to obtain spatial point cloud features;
the target feature extraction unit is used for obtaining target point cloud features according to the low-dimensional preliminary point cloud features, the medium-dimensional preliminary point cloud features, the high-dimensional preliminary point cloud features and the space point cloud features.
In some embodiments of the present invention, as shown in fig. 3, the preliminary feature extraction unit includes a preliminary dimension-increasing layer, a first preliminary convolution layer, a second preliminary convolution layer, and a third preliminary convolution layer;
the initial ascending layer is used for performing ascending operation on the rotating point cloud to obtain an initial ascending point cloud;
the first preliminary convolution layer is used for performing convolution operation on the preliminary ascending-dimensional point cloud to obtain low-dimensional preliminary point cloud characteristics;
the second preliminary convolution layer is used for performing convolution operation on the low-dimensional preliminary point cloud characteristic to obtain a medium-dimensional preliminary point cloud characteristic;
and the third preliminary convolution layer is used for performing convolution operation on the medium-dimensional preliminary point cloud characteristic to obtain the high-dimensional preliminary point cloud characteristic.
In an embodiment of the invention, the size of the preliminary ascending point cloud is (32 × 2048 × 3 × 1), the size of the low-dimensional preliminary point cloud feature is (32 × 2048 × 1 × 64), the size of the medium-dimensional preliminary point cloud feature is (32 × 2048 × 1 × 128), the size of the high-dimensional preliminary point cloud feature is (32 × 2048 × 1 × 256),
in some embodiments of the present invention, as shown in fig. 4, the spatial feature alignment unit includes a first spatial convolution layer, a second spatial convolution layer, a spatial max-pooling layer, a first spatial dimension adjustment layer, a first spatial fully-connected layer, a second spatial fully-connected layer, and a second spatial dimension adjustment layer;
the first space convolution layer is used for performing convolution operation on the high-dimensional preliminary point cloud characteristic to obtain a first space convolution characteristic;
the second space convolution layer is used for performing convolution operation on the first space convolution characteristic to obtain a second space convolution characteristic;
the spatial maximum pooling layer is used for performing maximum pooling operation on the second spatial convolution characteristics to obtain spatial maximum pooling characteristics;
the first space dimension adjusting layer is used for carrying out dimension adjustment on the space maximum pooling characteristic to obtain a first space dimension adjusting characteristic;
the first space full-connection layer is used for mapping the first space dimension adjustment feature to a sample space to obtain a first space mapping feature, and activating and carrying out batch standardization processing on the first space mapping feature to obtain a first space standard feature;
the second spatial full-link layer is used for mapping the first spatial standard feature to a sample space to obtain a second spatial mapping feature;
the second space dimension adjustment layer is used for carrying out dimension adjustment on the second space mapping characteristics to obtain space point cloud characteristics.
In a specific embodiment of the invention, the first spatial convolution feature has a size of (32 × 2048 × 1 × 512), the second spatial convolution feature has a size of (32 × 2048 × 1 × 1024), the spatially largest pooled feature has a size of (32 × 1 × 1024), the first spatial dimension adjustment feature has a size of (32 × 1024), the first spatial standard feature has a size of (32 × 512), the second spatial mapping feature has a size of (32 × 65536), and the spatial point cloud feature has a size of (32 × 256).
Wherein the convolution kernel size of the first spatial convolution layer is (1 × 512) and the step size is (1 × 1); the convolution kernel size of the second spatial convolution layer is (1 × 1024) and the step size is (1 × 1).
In some embodiments of the present invention, as shown in fig. 5, the target feature extraction unit includes a first target dimension adjustment layer, a target processing layer, a first target dimension-increasing layer, a first target convolutional layer, a second target convolutional layer, a target max-pooling layer, a leveling layer, a stitching layer, a plurality of convolutional layers, and a second target dimension adjustment layer;
the first target dimension adjusting layer is used for adjusting the dimension of the high-dimensional preliminary point cloud feature to obtain a first target dimension adjusting feature;
the target processing layer is used for multiplying the target dimension adjusting layer and the space point cloud characteristics to obtain a first multiplication characteristic;
the first target dimension raising layer is used for carrying out dimension raising processing on the first multiplied feature to obtain a first target dimension raising feature;
the first target convolution layer is used for performing convolution processing on the first target dimension-increasing feature to obtain a first target convolution feature;
the second tag convolution layer is used for carrying out convolution processing on the first target convolution characteristic to obtain a second target convolution characteristic;
the target maximum pooling layer is used for performing maximum pooling processing on the second target convolution characteristics to obtain target pooling characteristics;
the tiling layer is used for carrying out tiling operation on the target pooling characteristics to obtain target tiling characteristics;
the splicing layer is used for splicing the low-dimensional preliminary point cloud feature, the medium-dimensional preliminary point cloud feature, the high-dimensional preliminary point cloud feature, the first target convolution feature, the second target convolution feature and the target tiling feature to obtain a target splicing feature;
the convolution layers are used for sequentially carrying out convolution processing on the target splicing characteristics to obtain third target convolution characteristics;
the second target dimension adjustment layer is used for carrying out dimension adjustment on the third target convolution characteristics to obtain target point cloud characteristics.
In a specific embodiment of the invention, the first target dimension scaling feature is (32 × 2048 × 256), the first multiplied feature is (32 × 2048 × 256), the first target raised dimension feature is (32 × 2048 × 1 × 256), the first target convolved feature is (32 × 2048 × 1 × 512), the second target convolved feature is (32 × 2048 × 1 × 1028), the target pooled feature is (32 × 2041 × 2048), the target tiled feature is (32 × 2048), the target stitched feature is (32 × 2048 × 1 5056), the third target convolved feature is (32 × 2048), and the target stitched feature is (32 × 2048).
In some embodiments of the present invention, as shown in fig. 6, step S101 includes:
s601, constructing an initial outlier extraction model;
s602, constructing a point cloud outlier sample set;
s603, training an initial outlier extraction model according to the point cloud outlier sample collection to obtain a target outlier extraction model which is completely trained.
In an embodiment of the present invention, as shown in fig. 7, step S602 includes:
s701, acquiring an initial three-dimensional point cloud of a reflective workpiece with a concave structure;
s702, preprocessing the initial three-dimensional point cloud to obtain a three-dimensional point cloud to be marked;
s703, marking outliers and point clouds around holes in the three-dimensional point cloud to be marked to obtain a marked three-dimensional point cloud;
s704, down-sampling is conducted on the marked three-dimensional point cloud, and a point cloud outlier sample set is obtained.
According to the embodiment of the invention, the outliers and the point clouds around the cavities in the three-dimensional point cloud to be marked are marked, instead of only marking the outliers, the robustness of the trained target outlier extraction model can be improved, the stability of the target outlier extraction model can be improved, and the occurrence of overfitting is reduced.
Furthermore, by marking the outliers and the points around the holes, the interference of the non-effective outliers can be effectively reduced, meanwhile, the model precision of the target outlier extraction model is improved, the three-dimensional point cloud with obvious outlier characteristics is reserved, the point cloud data scale is reduced, and the extraction efficiency and the accuracy of the subsequent outlier extraction are improved.
Furthermore, the sampling number in the marked three-dimensional point cloud can be reduced by performing downsampling on the marked three-dimensional point cloud, so that the method is suitable for a target outlier extraction model.
The down-Sampling method includes, but is not limited to, Fast Methods for Random Sampling (FMRS), uniform down-Sampling, geometric down-Sampling, lattice down-Sampling, and the like.
In a specific embodiment of the present invention, the down-sampling mode is FMRS.
In some embodiments of the present invention, as shown in fig. 8, step S702 comprises:
s801, determining a reflective workpiece area and a non-reflective workpiece area in the initial three-dimensional point cloud, and eliminating the non-reflective workpiece area to obtain a workpiece three-dimensional point cloud;
s802, removing noise point clouds in the three-dimensional point clouds of the workpiece to obtain the three-dimensional point clouds to be marked.
By eliminating the non-reflective workpiece area, the embodiment of the invention can ensure that the three-dimensional point cloud of the workpiece has no other objects, such as: the object stage and other non-reflective workpieces such as the object stage and the like included in the initial three-dimensional point cloud are removed, and the data scale of the three-dimensional point cloud of the workpiece can be reduced, so that the extraction efficiency and the accuracy of the outlier extraction method are improved.
Furthermore, the embodiment of the invention can further reduce the data scale of the three-dimensional point cloud of the workpiece by removing the noisy point cloud in the three-dimensional point cloud of the workpiece, thereby further improving the extraction efficiency and accuracy of the outlier extraction method. In addition, the detail contour of the denoised three-dimensional point cloud to be marked is clearer, the characteristics are more obvious, and the extraction accuracy can be further improved.
In some embodiments of the present invention, step S802 specifically includes: and removing noise point clouds in the three-dimensional point cloud of the workpiece by using a filtering algorithm. The filtering algorithm includes, but is not limited to, a spherical low-pass filtering algorithm, a neighbor statistical gaussian noise filtering algorithm, a least square filtering, a wiener filtering, a smoothing filtering, and the like.
It should be understood that: the specific filtering algorithm to be used should be adjusted according to the actual application scenario, and is not described in detail herein.
It should be noted that: in order to increase the number of samples in the point cloud outlier sample set and further increase the generalization capability of the target outlier extraction model, in some embodiments of the present invention, dithering and rigid transformation may be added to the three-dimensional point cloud to be marked to simulate the number of samples in the point cloud outlier sample set.
In the embodiment of the present invention, as shown in fig. 9 to 13, the number of point clouds of the initial three-dimensional point cloud, the workpiece three-dimensional point cloud, the three-dimensional point cloud to be marked, the marked three-dimensional point cloud, and the point cloud outlier sample set is gradually smaller, and the details of the point clouds are gradually clear, so that: through the processing process, the training speed of the initial outlier extraction model can be improved.
In order to better implement the method for extracting outliers of three-dimensional point cloud in the embodiment of the present invention, on the basis of the method for extracting outliers of three-dimensional point cloud, correspondingly, the embodiment of the present invention further provides an apparatus for extracting outliers of three-dimensional point cloud, as shown in fig. 14, the apparatus 1400 for extracting outliers of three-dimensional point cloud includes:
a model obtaining unit 1401, configured to obtain a target outlier extraction model with complete training, where the target outlier extraction model includes a point cloud rotation module, a point cloud feature extraction module, and a segmentation module;
a point cloud obtaining unit 1402, configured to obtain a three-dimensional point cloud to be extracted;
a point cloud rotation unit 1403, configured to rotate the three-dimensional point cloud to be extracted based on the point cloud rotation module, so as to obtain a rotated point cloud;
a feature extraction unit 1404, configured to perform feature extraction on the rotation point cloud based on the point cloud feature extraction module, to obtain a target point cloud feature;
and an outlier extracting unit 1405, configured to extract outliers in the three-dimensional point cloud to be extracted based on the segmentation module and the target point cloud feature.
The outlier extracting apparatus 1400 of the three-dimensional point cloud provided in the above-mentioned embodiment may implement the technical solutions described in the above-mentioned embodiments of the outlier extracting method of the three-dimensional point cloud, and the specific implementation principles of the above-mentioned modules or units may refer to the corresponding contents in the above-mentioned embodiments of the outlier extracting method of the three-dimensional point cloud, and are not described herein again.
As shown in fig. 15, the present invention also provides an electronic device 1500 accordingly. The electronic device 1500 includes a processor 1501, a memory 1502, and a display 1503. Fig. 15 shows only some of the components of the electronic device 1500, but it is to be understood that not all of the shown components are required to be implemented, and that more or fewer components can be implemented instead.
The processor 1501, which in some embodiments may be a Central Processing Unit (CPU), microprocessor or other data Processing chip, is configured to run program code stored in the memory 1502 or process data, such as the outlier extraction method of three-dimensional point clouds of the present invention.
In some embodiments, processor 1501 may be a single server or a group of servers. The server groups may be centralized or distributed. In some embodiments, processor 1501 may be local or remote. In some embodiments, processor 1501 may be implemented in a cloud platform. In an embodiment, the cloud platform may include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an intra-site, a multi-cloud, and the like, or any combination thereof.
The storage 1502 may be, in some embodiments, an internal storage unit of the electronic device 1500, such as a hard disk or memory of the electronic device 1500. The memory 1502 may also be an external storage device of the electronic device 1500 in other embodiments, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), etc. provided on the electronic device 1500.
Further, the memory 1502 may also include both internal storage units and external storage devices of the electronic device 1500. The memory 1502 is used for storing application software and various data for installing the electronic apparatus 1500.
The display 1503 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch device, and the like in some embodiments. The display 1503 is used for displaying information at the electronic device 1500 and for displaying a visual user interface. The components 1501-1503 of the electronic device 1500 communicate with each other via a system bus.
In one embodiment, when the processor 1501 executes an outlier extraction procedure for a three-dimensional point cloud in the memory 1502, the following steps may be implemented:
acquiring a target outlier extraction model which is trained completely, wherein the target outlier extraction model comprises a point cloud rotation module, a point cloud feature extraction module and a segmentation module;
acquiring a three-dimensional point cloud to be extracted;
rotating the three-dimensional point cloud to be extracted based on a point cloud rotating module to obtain a rotating point cloud;
performing feature extraction on the rotating point cloud based on a point cloud feature extraction module to obtain target point cloud features;
and extracting outliers in the three-dimensional point cloud to be extracted based on the segmentation module and the target point cloud characteristics.
It should be understood that: the processor 1501, when executing the outlier extraction procedure for the three-dimensional point cloud in the memory 1502, may also implement other functions in addition to the above functions, which may be specifically referred to the description of the corresponding method embodiment above.
Further, the type of the mentioned electronic device 1500 is not specifically limited in the embodiments of the present invention, and the electronic device 1500 may be a portable electronic device such as a mobile phone, a tablet computer, a Personal Digital Assistant (PDA), a wearable device, and a laptop computer (laptop). Exemplary embodiments of portable electronic devices include, but are not limited to, portable electronic devices that carry an IOS, android, microsoft, or other operating system. The portable electronic device may also be other portable electronic devices such as laptop computers (laptop) with touch sensitive surfaces (e.g., touch panels), etc. It should also be understood that in other embodiments of the present invention, electronic device 1500 may not be a portable electronic device, but may be a desktop computer having a touch-sensitive surface (e.g., a touch panel).
Accordingly, the present application further provides a computer-readable storage medium, which is used for storing a computer-readable program or instruction, and when the program or instruction is executed by a processor, the steps or functions of the method for extracting outliers of a three-dimensional point cloud provided in the above method embodiments can be implemented.
Those skilled in the art will appreciate that all or part of the flow of the method implementing the above embodiments may be implemented by instructing relevant hardware (such as a processor, a controller, etc.) by a computer program, and the computer program may be stored in a computer readable storage medium. The computer readable storage medium is a magnetic disk, an optical disk, a read-only memory or a random access memory.
The method and the device for extracting outliers of three-dimensional point cloud provided by the invention are described in detail, and a specific example is applied in the text to explain the principle and the implementation mode of the invention, and the description of the embodiment is only used for helping to understand the method and the core idea of the invention; meanwhile, for those skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed, and in summary, the content of the present specification should not be construed as limiting the present invention.
Claims (10)
1. An outlier extraction method of a three-dimensional point cloud is characterized by comprising the following steps:
acquiring a target outlier extraction model which is trained completely, wherein the target outlier extraction model comprises a point cloud rotation module, a point cloud feature extraction module and a segmentation module;
acquiring a three-dimensional point cloud to be extracted;
rotating the three-dimensional point cloud to be extracted based on the point cloud rotating module to obtain a rotating point cloud;
performing feature extraction on the rotating point cloud based on the point cloud feature extraction module to obtain target point cloud features;
and extracting outliers in the three-dimensional point cloud to be extracted based on the segmentation module and the target point cloud characteristics.
2. The method of claim 1, wherein the point cloud rotation module comprises a first upscaling layer, a first convolution layer, an average pooling layer, a first dimension adjustment layer, a first fully-connected layer, a second fully-connected layer, a third fully-connected layer, a second dimension adjustment layer, and a rotated point cloud determination layer;
the dimension raising layer is used for carrying out dimension raising operation on the three-dimensional point cloud to be extracted to obtain a first dimension raising point cloud;
the convolution layer is used for performing convolution operation on the first ascending-dimensional point cloud to obtain a first convolution characteristic;
the average pooling layer is used for carrying out average pooling operation on the first convolution characteristics to obtain first average pooling characteristics;
the first dimension adjusting layer is used for carrying out dimension adjustment on the first average pooling feature to obtain a first dimension adjusting feature;
the first full-connection layer is used for mapping the first dimension adjustment feature to a sample point cloud space to obtain a first mapping feature, and activating and carrying out batch standardization processing on the first mapping feature to obtain a first standard feature;
the second full-connection layer is used for mapping the first standard features to a sample point cloud space to obtain second mapping features, and activating and carrying out batch standardization processing on the second mapping features to obtain second standard features;
the third full-connection layer is used for mapping the second standard feature to a sample point cloud space to obtain a third mapping feature;
the second dimension adjusting layer is used for carrying out dimension adjustment on the third mapping characteristics to obtain second dimension adjusting characteristics;
the rotating point cloud determining layer is used for obtaining the rotating point cloud according to the second dimension adjusting feature and the three-dimensional point cloud to be extracted.
3. The method of claim 1, wherein the point cloud feature extraction module comprises a preliminary feature extraction unit, a spatial feature alignment unit, and a target feature extraction unit;
the preliminary feature extraction unit is used for performing preliminary feature extraction on the rotating point cloud to obtain a low-dimensional preliminary point cloud feature, a medium-dimensional preliminary point cloud feature and a high-dimensional preliminary point cloud feature;
the spatial feature alignment unit is used for extracting features of the high-dimensional preliminary point cloud features to obtain spatial point cloud features;
the target feature extraction unit is used for obtaining the target point cloud features according to the low-dimensional preliminary point cloud features, the medium-dimensional preliminary point cloud features, the high-dimensional preliminary point cloud features and the space point cloud features.
4. The method of claim 3, wherein the preliminary feature extraction unit comprises a preliminary upscaling layer, a first preliminary convolution layer, a second preliminary convolution layer, and a third preliminary convolution layer;
the initial ascending layer is used for performing ascending operation on the rotating point cloud to obtain an initial ascending point cloud;
the first preliminary convolution layer is used for performing convolution operation on the preliminary ascending-dimensional point cloud to obtain the low-dimensional preliminary point cloud characteristic;
the second preliminary convolution layer is used for performing convolution operation on the low-dimensional preliminary point cloud feature to obtain the medium-dimensional preliminary point cloud feature;
and the third preliminary convolution layer is used for performing convolution operation on the medium-dimensional preliminary point cloud feature to obtain the high-dimensional preliminary point cloud feature.
5. The method of claim 4, wherein the spatial feature alignment unit comprises a first spatial convolution layer, a second spatial convolution layer, a spatial maximum pooling layer, a first spatial dimension adjustment layer, a first spatial full-link layer, a second spatial full-link layer, and a second spatial dimension adjustment layer;
the first space convolution layer is used for performing convolution operation on the high-dimensional preliminary point cloud feature to obtain a first space convolution feature;
the second space convolution layer is used for performing convolution operation on the first space convolution characteristic to obtain a second space convolution characteristic;
the spatial maximum pooling layer is used for performing maximum pooling operation on the second spatial convolution characteristics to obtain spatial maximum pooling characteristics;
the first space dimension adjusting layer is used for carrying out dimension adjustment on the space maximum pooling feature to obtain a first space dimension adjusting feature;
the first space full-connection layer is used for mapping the first space dimension adjustment feature to a sample space to obtain a first space mapping feature, and activating and standardizing the first space mapping feature in batches to obtain a first space standard feature;
the second spatial full-link layer is used for mapping the first spatial standard feature to a sample space to obtain a second spatial mapping feature;
the second space dimension adjustment layer is used for carrying out dimension adjustment on the second space mapping characteristic to obtain the space point cloud characteristic.
6. The method of extracting outliers of a three-dimensional point cloud of claim 5, wherein said target feature extraction unit comprises a first target dimension adjustment layer, a target processing layer, a first target dimension-up layer, a first target convolution layer, a second target convolution layer, a target maximum pooling layer, a leveling layer, a stitching layer, a plurality of convolution layers, and a second target dimension adjustment layer;
the first target dimension adjusting layer is used for adjusting the dimension of the high-dimensional preliminary point cloud feature to obtain a first target dimension adjusting feature;
the target processing layer is used for multiplying the target dimension adjusting layer and the space point cloud characteristics to obtain a first multiplication characteristic;
the first target dimension raising layer is used for carrying out dimension raising processing on the first multiplied feature to obtain a first target dimension raising feature;
the first target convolution layer is used for performing convolution processing on the first target dimension-increasing feature to obtain a first target convolution feature;
the second convolution layer is used for performing convolution processing on the first target convolution characteristic to obtain a second target convolution characteristic;
the target maximum pooling layer is used for performing maximum pooling processing on the second target convolution characteristics to obtain target pooling characteristics;
the flat layer is used for carrying out flat laying operation on the target pooling characteristic to obtain a target flat laying characteristic;
the splicing layer is used for splicing the low-dimensional preliminary point cloud feature, the medium-dimensional preliminary point cloud feature, the high-dimensional preliminary point cloud feature, the first target convolution feature, the second target convolution feature and the target tiling feature to obtain a target splicing feature;
the convolution layers are used for sequentially carrying out convolution processing on the target splicing characteristics to obtain third target convolution characteristics;
the second target dimension adjustment layer is used for carrying out dimension adjustment on the third target convolution characteristics to obtain the target point cloud characteristics.
7. The method of claim 1, wherein the obtaining a well-trained target outlier extraction model comprises:
constructing an initial outlier extraction model;
constructing a point cloud outlier sample set;
and training the initial outlier extraction model according to the point cloud outlier sample collection to obtain a target outlier extraction model with complete training.
8. The method of claim 7, wherein constructing a sample set of point cloud outliers comprises:
acquiring an initial three-dimensional point cloud of a reflective workpiece comprising a concave structure;
preprocessing the initial three-dimensional point cloud to obtain a three-dimensional point cloud to be marked;
marking outliers and point clouds around holes in the three-dimensional point cloud to be marked to obtain a marked three-dimensional point cloud;
and performing downsampling on the marked three-dimensional point cloud to obtain the point cloud outlier sample set.
9. The method of claim 8, wherein the pre-processing the initial three-dimensional point cloud to obtain a three-dimensional point cloud to be labeled comprises:
determining a reflective workpiece area and a non-reflective workpiece area in the initial three-dimensional point cloud, and eliminating the non-reflective workpiece area to obtain a workpiece three-dimensional point cloud;
and removing noise point clouds in the three-dimensional point clouds of the workpiece to obtain the three-dimensional point clouds to be marked.
10. An outlier extracting apparatus for three-dimensional point cloud, comprising:
the model acquisition unit is used for acquiring a target outlier extraction model which is completely trained, and the target outlier extraction model comprises a point cloud rotation module, a point cloud feature extraction module and a segmentation module;
the point cloud obtaining unit is used for obtaining a three-dimensional point cloud to be extracted;
the point cloud rotating unit is used for rotating the three-dimensional point cloud to be extracted based on the point cloud rotating module to obtain a rotating point cloud;
the characteristic extraction unit is used for extracting the characteristics of the rotating point cloud based on the point cloud characteristic extraction module to obtain the target point cloud characteristics;
and the outlier extraction unit is used for extracting outliers in the three-dimensional point cloud to be extracted based on the segmentation module and the target point cloud characteristics.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210585410.1A CN114972788A (en) | 2022-05-25 | 2022-05-25 | Outlier extraction method and device of three-dimensional point cloud |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210585410.1A CN114972788A (en) | 2022-05-25 | 2022-05-25 | Outlier extraction method and device of three-dimensional point cloud |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114972788A true CN114972788A (en) | 2022-08-30 |
Family
ID=82956662
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210585410.1A Pending CN114972788A (en) | 2022-05-25 | 2022-05-25 | Outlier extraction method and device of three-dimensional point cloud |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114972788A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115586506A (en) * | 2022-12-13 | 2023-01-10 | 南京慧尔视智能科技有限公司 | Anti-interference target classification method and device |
-
2022
- 2022-05-25 CN CN202210585410.1A patent/CN114972788A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115586506A (en) * | 2022-12-13 | 2023-01-10 | 南京慧尔视智能科技有限公司 | Anti-interference target classification method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11151780B2 (en) | Lighting estimation using an input image and depth map | |
| CN111178250B (en) | Object identification positioning method and device and terminal equipment | |
| CN110222703B (en) | Image contour recognition method, device, equipment and medium | |
| CN110221722B (en) | Picture processing method, electronic device and storage medium | |
| CN116563493A (en) | Model training method based on three-dimensional reconstruction, three-dimensional reconstruction method and device | |
| CN115053234A (en) | Deep Photometric Learning (DPL) systems, devices, and methods | |
| CN114445904A (en) | Iris segmentation method, apparatus, medium, and device based on full convolution neural network | |
| CN114972788A (en) | Outlier extraction method and device of three-dimensional point cloud | |
| US20220375164A1 (en) | Method and apparatus for three dimensional reconstruction, electronic device and storage medium | |
| Lichy et al. | Fast light-weight near-field photometric stereo | |
| CN110163095B (en) | Loop detection method, loop detection device and terminal equipment | |
| CN113791425A (en) | Radar P display interface generation method and device, computer equipment and storage medium | |
| CN113610958A (en) | 3D image construction method and device based on style migration and terminal | |
| CN115018763A (en) | Pit detection method, device and equipment based on curved surface fitting and storage medium | |
| CN113032071B (en) | Page element positioning method, page testing method, device, equipment and medium | |
| CN115496196A (en) | Fusion processing of successive mathematical operators | |
| CN113610856A (en) | Method and device for training image segmentation model and image segmentation | |
| CN111915615A (en) | Image segmentation method and device, electronic equipment and computer readable storage medium | |
| CN116452655B (en) | Laminating and positioning method, device, equipment and medium applied to MPIS industrial control main board | |
| CN117496161B (en) | Point cloud segmentation method and device | |
| CN112232143B (en) | Face point cloud optimization method and device, machine readable medium and equipment | |
| US11657562B2 (en) | Utilizing hemispherical clamping for importance sampling of image-based light to render a virtual environment | |
| CN114926333A (en) | Image super-resolution reconstruction method and device | |
| CN112541535B (en) | Three-dimensional point cloud classification method based on complementary multi-branch deep learning | |
| CN114926654A (en) | Polar line correction method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |