CN106650916B - A kind of mesh segmentation method based on ant group optimization - Google Patents
A kind of mesh segmentation method based on ant group optimization Download PDFInfo
- Publication number
- CN106650916B CN106650916B CN201611247939.3A CN201611247939A CN106650916B CN 106650916 B CN106650916 B CN 106650916B CN 201611247939 A CN201611247939 A CN 201611247939A CN 106650916 B CN106650916 B CN 106650916B
- Authority
- CN
- China
- Prior art keywords
- grid
- label
- sdf
- mesh
- vertex
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Generation (AREA)
Abstract
The invention belongs to graphics and geometry digital processing technology field, and in particular to a kind of mesh segmentation method based on ant group optimization.A kind of mesh segmentation method based on ant group optimization, comprising the following steps: (1) three-dimensional mesh data to be split and parameter are inputted, subsequently into step (2);(2) grid property is calculated, subsequently into step (3);(3) data prediction, subsequently into step (4);(4) seed point is selected in grid to be split, subsequently into step (5);(5) mesh tab initialization is carried out, subsequently into step (6);(6) ant group optimization iteration is carried out until meeting iteration standard, subsequently into step (7);(7) region merging technique, subsequently into step (8);(8) segmentation result is exported.
Description
Technical field
The invention belongs to graphics and geometry digital processing technology field, and in particular to a kind of grid based on ant group optimization
Dividing method.
Background technique
Over the past decade, the progress of three-dimensional data capture device and its technology has pushed computer vision, medical imaging, base
Develop in depth in technologies such as image modelings, produce the threedimensional model of many complexity so that is based on grid model (especially three
Angle grid model) the relevant technologies such as geometric manipulations become CAD in recent years (CAD) and the important of graphics is ground
Study carefully hot spot.
Initial three-dimensional grid model lacks enough structure features and semantic information, the understanding to initial three-dimensional grid model
As the major issue that many geometric manipulations problems demands solve, mesh segmentation by according to certain segmentation criterion by original three
Dimension module is decomposed into different component or patch, facilitates related geometric manipulations problem (such as surface compression, grid reconstruction, ginseng
Numberization, texture mapping, model index) effective solution, actually exactly come from texture mapping, parametrization, mesh animation, grid
The geometric manipulations problem that the demand of the problems such as deformation keeps mesh segmentation important as one starts to attract people's attention.
In computer vision, model is divided into different piece, facilitates the feature identification for carrying out model, such as by people
Face three-dimensional model identifies cheek, nose, eyes etc. by segmentation.In mesh parameterization and texture mapping, by by model
A series of flat regions are divided into, the torsional deformation of parametrization and texture mapping can be reduced, parametrization is improved and texture reflects
The quality penetrated.
In the prior art, some methods need man-machine interactively, some need method complex, some methods need multiple
Miscellaneous data structure.The method of the present invention is relatively simple, does not need complicated data structure, can be full-automatic, and due to ant
The characteristics of group's algorithm, parallel processing can be carried out.
The Inspiration Sources of ant group optimization (ant colony optimization, ACO) are in the process of Ant Search food.
Ants are found using leaving and detecting this indirect positive feedback mechanism of pheromone (Stigmergy) on the path passed by
Shortest path of the nest to food source.In this way, a common combinatorial optimization problem has just been transformed into a restricted shortest path
Diameter problem.Ant group optimization is initially to study traveling salesman problem (TSP) and be suggested, and is used to many engineering problems later
On.
Summary of the invention
Goal of the invention: the present invention has made improvements in view of the above-mentioned problems of the prior art, i.e., the invention discloses one kind
Mesh segmentation method based on ant group optimization.In the present invention, each grid of grid to be split is considered as an ant, passed through
Ant group optimization iteration constantly updates the label of each grid.When initialization, all grids first assign a background label.So
After seed point is randomly generated, each seed point assigns one and is different from background label and unique label, and for each
Seed point, the mesh point around neighborhood assign the label as seed point.With the iteration of ant group optimization, the mark of seed point
It signs to external diffusion, and tag update is to be carried out under conditions of meeting segmentation standard by the update mechanism of ant group optimization, until
Meet iteration standard.After the completion of ant group optimization, carries out region merging technique and be finally completed by lesser region merging technique at large area
Segmentation.
A kind of technical solution: mesh segmentation method based on ant group optimization, comprising the following steps:
(1) three-dimensional mesh data to be split and parameter are inputted, subsequently into step (2);
(2) grid property is calculated, subsequently into step (3);
(3) data prediction, subsequently into step (4);
(4) seed point is selected in grid to be split, subsequently into step (5);
(5) mesh tab initialization is carried out, subsequently into step (6);
(6) ant group optimization iteration is carried out until meeting iteration standard, subsequently into step (7);
(7) region merging technique, subsequently into step (8);
(8) segmentation result is exported.
Further, the grid data to be split inputted in step (1) includes the geometric coordinate information and topology of three-dimensional grid
Information.
Further, grid property finger-type shape diameter function SDF (Shape Diameter Function) in step (2),
Calculating process it is as follows:
(21), for a vertex on surface mesh, make one using the vertex as conical tip, vertex scheme vector it is inverse
Direction is the cone in center line direction;
(22) ray drawn from the vertex within the scope of several cones meets at surface mesh, by constructing Octree
Structure calculates the ray intersected with grid;
(23) ray identical with vertex normal direction is removed, takes remaining length in all ray length median standard deviations
Within ray be weighted and averaged to get arrive the vertex SDF value;
Wherein: the length of a ray refers to that the vertex of the ray and ray normal vector reverse direction are met between surface mesh
Distance;
Ray length median refers to vertex scheme vector reverse direction for all ray lengths in the cone in center line direction
Median;
It, can be by using a position as in for finding out middle after all observed value height sequences for limited manifold
Number;
Weight is ray to the dihedral of the angle of circular cone center line.
Further, step (3) data prediction includes the following steps
(31) prepare the secondary data structure of each grid;
(32) neighborhood of each grid is calculated;
The grid being directly associated with the vertex is found by the data of search input grid to each grid vertex
Then vertex is put into these grid vertexes as the field on the vertex in secondary data structure;
(33) SDF is normalized
SDF is converted into the numerical value between 0 to 1, reduction formula is as follows:
Wherein,
SDFoldFor conversion before numerical value,
SDFnewFor conversion after numerical value,
SDFminThe minimum value of all grid SDF before converting,
SDFmaxThe maximum value of all grid SDF before converting.
Further, random method choice seed point is taken in step (4).
Further, step (5) mesh tab initialization includes the following steps:
(51) all grids first assign a background label;
(52) a unique label then is assigned for each seed point, the mesh point of surrounding assigns and seed point one
The label of sample.
Further, step (6) includes:
(61) for each iteration, each grid finds optimal label from Γ seed point label, if found most
Good label and last optimal label are different, and the label of grid is updated;
(62) the residual risk concentration of each label in each grid is updated.
Further, step (61) finds optimal label referring to following formula:
M=argmaxu{p(u)},u∈Γ
Wherein, p (u) is the transition probability of a seed point label u, and calculation formula is as follows:
Wherein:
τ (u) indicates that the grid uses the residual risk concentration of a certain candidate label u;
η (u) indicates that the grid uses the inspiration value of a certain candidate label u;
τ (v) indicates that the grid uses the residual risk concentration of a certain candidate label v;
η (v) indicates that the grid uses the inspiration value of a certain candidate label v;
α and β is two factor coefficients for controlling τ (u) and η (u) relative equilibrium;
Further, the calculation formula of η (u) is as follows:
ε (u) is the characteristic value that grid corresponds to u label.
Further, the calculating of grid search-engine value is exactly the mark for calculating grid neighborhood internal net point SDF in step (61)
Quasi- poor, specific formula for calculation is as follows:
Wherein,
SDFiFor in the grid neighborhood correspond to u label i-th of mesh point SDF value,
μSDFFor in the grid neighborhood correspond to u label SDF average value,
N is the mesh point number for corresponding to u label in the grid neighborhood.
Further, in step (62), after each iteration, the residual risk concentration of each label is meeting more in each grid
It is updated under conditions of new, the condition of update is as follows:
(a) its neighborhood contains the non-background label higher than pre-set threshold number;
(b) at least one seed point label meets segmentation condition in its neighborhood;
If the initial value before τ (u) first time iteration is τ0, and be updated at the end of each iteration, more new formula is as follows:
τ(u)←ρ·τ(u)+γ·Δτ(u)
Wherein,
ρ is coefficient of heredity, it controls grid current residual information concentration to the influence degree of next iteration, which inherits
The residual risk concentration of iteration before;
Δ τ (u) is the space constraint component of residual risk concentration increments, and calculation formula is as follows:
Wherein,
B is to assume current grid,
B ' is one of the grid for meeting segmentation condition in neighborhood,
NSFor Size of Neighborhood.
Further, segmentation condition is described as follows:
Ant b to be updated for one, current optimum label are M, it is assumed that its neighborhood contains several non-background marks
Label.For one of candidate label u, its segmentation condition are as follows:
And
Ω (b, M) refers to the grid point set that label is M in mesh point b neighborhood,
Refer to all grid point sets that candidate label is u in mesh point b neighborhood;
STD (SDF [Ω (b, M)]) refers to the standard deviation of the SDF value for the grid point set that label is M in mesh point b neighborhood.
Further, iteration standard is that the ratio that mesh tab updates is less than threshold value in step (61).
Further, in step (7), after the completion of ant colony iteration, adjacent region is carried out in the case where meeting segmentation condition
Merge.
The utility model has the advantages that a kind of mesh segmentation method based on ant group optimization disclosed by the invention has the advantages that
In the prior art, some methods need man-machine interactively, some need method complex, some methods need multiple
Miscellaneous data structure.The method of the present invention is relatively simple, does not need complicated data structure, can be full-automatic, and due to ant
The characteristics of group's algorithm, parallel processing can be carried out.
Detailed description of the invention
Fig. 1 is the schematic diagram of mesh tab initialization;
Fig. 2 is grid neighborhood schematic diagram, and wherein Q is the neighborhood of V3.
Specific embodiment:
Detailed description of specific embodiments of the present invention below.
A kind of mesh segmentation method based on ant group optimization, comprising the following steps:
(1) three-dimensional mesh data to be split and parameter are inputted, subsequently into step (2);
(2) grid property is calculated, subsequently into step (3);
(3) data prediction, subsequently into step (4);
(4) seed point is selected in grid to be split, subsequently into step (5);
(5) as shown in Figure 1, mesh tab initialization is carried out, subsequently into step (6);
(6) ant group optimization iteration is carried out until meeting iteration standard, subsequently into step (7);
(7) region merging technique, subsequently into step (8);
(8) segmentation result is exported.
Further, the grid data to be split inputted in step (1) includes the geometric coordinate information and topology of three-dimensional grid
Information.
Further, grid property finger-type shape diameter function SDF (Shape Diameter Function) in step (2),
Calculating process it is as follows:
(21), for a vertex on surface mesh, make one using the vertex as conical tip, vertex scheme vector it is inverse
Direction is the cone in center line direction;
(22) ray drawn from the vertex within the scope of several cones meets at surface mesh, by constructing Octree
Structure calculates the ray intersected with grid;
(23) ray identical with vertex normal direction is removed, takes remaining length in all ray length median standard deviations
Within ray be weighted and averaged to get arrive the vertex SDF value;
Wherein: the length of a ray refers to that the vertex of the ray and ray normal vector reverse direction are met between surface mesh
Distance;
Ray length median refers to vertex scheme vector reverse direction for all ray lengths in the cone in center line direction
Median;
It, can be by using a position as in for finding out middle after all observed value height sequences for limited manifold
Number;
Weight is ray to the dihedral of the angle of circular cone center line.
Further, step (3) data prediction includes the following steps
(31) prepare the secondary data structure of each grid;
(32) neighborhood of each grid is calculated;
As shown in Fig. 2, by the data of search input grid, being found and the direct phase in the vertex to each grid vertex
Then the grid vertex of connection is put into these grid vertexes as the field on the vertex in secondary data structure;Such as V3 in Fig. 2
Label is neighborhood mesh point 4 of 3, and label is all mesh points 8 of 2.
(33) SDF is normalized
SDF is converted into the numerical value between 0 to 1, reduction formula is as follows:
Wherein,
SDFoldFor conversion before numerical value,
SDFnewFor conversion after numerical value,
SDFminThe minimum value of all grid SDF before converting,
SDFmaxThe maximum value of all grid SDF before converting.
Further, random method choice seed point is taken in step (4).
Further, step (5) mesh tab initialization includes the following steps:
(51) all grids first assign a background label;
(52) a unique label then is assigned for each seed point, the mesh point of surrounding assigns and seed point one
The label of sample.
Further, step (6) includes:
(61) for each iteration, each grid finds optimal label from Γ seed point label, if found most
Good label and last optimal label are different, and the label of grid is updated;
(62) the residual risk concentration of each label in each grid is updated.
Further, step (61) finds optimal label referring to following formula:
M=argmaxu{p(u)},u∈Γ
Wherein, p (u) is the transition probability of a seed point label u, and calculation formula is as follows:
Wherein:
τ (u) indicates that the grid uses the residual risk concentration of a certain candidate label u;
η (u) indicates that the grid uses the inspiration value of a certain candidate label u;
τ (v) indicates that the grid uses the residual risk concentration of a certain candidate label v;
η (v) indicates that the grid uses the inspiration value of a certain candidate label v;
α and β is two factor coefficients for controlling τ (u) and η (u) relative equilibrium;
Further, the calculation formula of η (u) is as follows:
ε (u) is the characteristic value that grid corresponds to u label.
Further, the calculating of grid search-engine value is exactly the mark for calculating grid neighborhood internal net point SDF in step (61)
Quasi- poor, specific formula for calculation is as follows:
Wherein,
SDFiFor in the grid neighborhood correspond to u label i-th of mesh point SDF value,
μSDFFor in the grid neighborhood correspond to u label SDF average value,
N is the mesh point number for corresponding to u label in the grid neighborhood.
Further, in step (62), after each iteration, the residual risk concentration of each label is meeting more in each grid
It is updated under conditions of new, the condition of update is as follows:
(a) its neighborhood contains the non-background label higher than pre-set threshold number;
(b) at least one seed point label meets segmentation condition in its neighborhood;
If the initial value before τ (u) first time iteration is τ0, and be updated at the end of each iteration, more new formula is as follows:
τ(u)←ρ·τ(u)+γ·Δτ(u)
Wherein,
ρ is coefficient of heredity, it controls grid current residual information concentration to the influence degree of next iteration, which inherits
The residual risk concentration of iteration before;
Δ τ (u) is the space constraint component of residual risk concentration increments, and calculation formula is as follows:
Wherein,
B is to assume current grid,
B ' is one of the grid for meeting segmentation condition in neighborhood,
NSFor Size of Neighborhood.
Further, segmentation condition is described as follows:
Ant b to be updated for one, current optimum label are M, it is assumed that its neighborhood contains several non-background marks
Label.For one of candidate label u, its segmentation condition are as follows:
And
Ω (b, M) refers to the grid point set that label is M in mesh point b neighborhood,
Refer to all grid point sets that candidate label is u in mesh point b neighborhood;
STD (SDF [Ω (b, M)]) refers to the standard deviation of the SDF value for the grid point set that label is M in mesh point b neighborhood.
Further, iteration standard is that the ratio that mesh tab updates is less than threshold value in step (61).
Further, in step (7), after the completion of ant colony iteration, adjacent region is carried out in the case where meeting segmentation condition
Merge.
Embodiments of the present invention are elaborated above.But present invention is not limited to the embodiments described above,
Technical field those of ordinary skill within the scope of knowledge, can also do without departing from the purpose of the present invention
Various change out.
Claims (8)
1. a kind of mesh segmentation method based on ant group optimization, which comprises the following steps:
(1) three-dimensional mesh data to be split and parameter are inputted, subsequently into step (2);
(2) grid property is calculated, subsequently into step (3);
(3) data prediction, subsequently into step (4);
(4) seed point is selected in grid to be split, subsequently into step (5);
(5) mesh tab initialization is carried out, subsequently into step (6);
(6) ant group optimization iteration is carried out until meeting iteration standard, subsequently into step (7);
(7) region merging technique, subsequently into step (8);
(8) segmentation result is exported, in which:
The grid data to be split inputted in step (1) includes the geometric coordinate information and topology information of three-dimensional grid;
Grid property finger-type shape diameter function SDF in step (2), calculating process it is as follows:
(21), for a vertex on surface mesh, make one using the vertex as the reverse direction of conical tip, vertex scheme vector
For the cone in center line direction;
(22) ray drawn from the vertex within the scope of several cones meets at surface mesh, by constructing octree structure
To calculate the ray intersected with grid;
(23) ray identical with vertex normal direction is removed, takes remaining length within all ray length median standard deviations
Ray be weighted and averaged to get arrive the vertex SDF value;
Wherein: the length of a ray refer to the vertex of the ray and ray normal vector reverse direction meet between surface mesh away from
From;
Ray length median refers to vertex scheme vector reverse direction for the middle position of all ray lengths in the cone in center line direction
Number;
It, can be by finding out one of middle after all observed values height are sorted as median for limited manifold;
Weight is ray to the dihedral of the angle of circular cone center line.
2. a kind of mesh segmentation method based on ant group optimization according to claim 1, which is characterized in that step (3) number
Data preprocess includes the following steps
(31) prepare the secondary data structure of each grid;
(32) neighborhood of each grid is calculated;
The grid vertex being directly associated with the vertex is found by the data of search input grid to each grid vertex,
Then it is put into these grid vertexes as the field on the vertex in secondary data structure;
(33) SDF is normalized
SDF is converted into the numerical value between 0 to 1, reduction formula is as follows:
Wherein,
SDFoldFor conversion before numerical value,
SDFnewFor conversion after numerical value,
SDFminThe minimum value of all grid SDF before converting,
SDFmaxThe maximum value of all grid SDF before converting.
3. a kind of mesh segmentation method based on ant group optimization according to claim 1, which is characterized in that in step (4)
Take random method choice seed point.
4. a kind of mesh segmentation method based on ant group optimization according to claim 1, which is characterized in that step (5) net
The initialization of case marker label includes the following steps:
(51) all grids first assign a background label;
(52) a unique label then is assigned for each seed point, the mesh point of surrounding assigns as seed point
Label.
5. a kind of mesh segmentation method based on ant group optimization according to claim 1, which is characterized in that step (6) packet
It includes:
(61) for each iteration, each grid finds optimal label from Γ seed point label, if found optimal
Label and last optimal label are different, and the label of grid is updated;
(62) the residual risk concentration of each label in each grid is updated.
6. a kind of mesh segmentation method based on ant group optimization according to claim 5, which is characterized in that step (61) is looked for
To optimal label referring to following formula:
M=argmaxu{p(u)},u∈Γ
Wherein, p (u) is the transition probability of a seed point label u, and calculation formula is as follows:
Wherein:
τ (u) indicates that the grid uses the residual risk concentration of a certain candidate label u;
η (u) indicates that the grid uses the inspiration value of a certain candidate label u;
τ (v) indicates that the grid uses the residual risk concentration of a certain candidate label v;
η (v) indicates that the grid uses the inspiration value of a certain candidate label v;
α and β is two factor coefficients for controlling τ (u) and η (u) relative equilibrium;
The calculation formula of η (u) is as follows:
ε (u) is the characteristic value that grid corresponds to u label.
7. a kind of mesh segmentation method based on ant group optimization according to claim 5, which is characterized in that in step (61)
The calculating of grid search-engine value is exactly to calculate the standard deviation of grid neighborhood internal net point SDF, and specific formula for calculation is as follows:
Wherein,
SDFiFor in the grid neighborhood correspond to u label i-th of mesh point SDF value,
μSDFFor in the grid neighborhood correspond to u label SDF average value,
N is the mesh point number for corresponding to u label in the grid neighborhood.
8. a kind of mesh segmentation method based on ant group optimization according to claim 1, which is characterized in that step (62)
In, after each iteration, the residual risk concentration of each label is updated under conditions of meeting and updating in each grid, update
Condition is as follows:
(a) its neighborhood contains the non-background label higher than pre-set threshold number;
(b) at least one seed point label meets segmentation condition in its neighborhood;
If the initial value before τ (u) first time iteration is τ0, and be updated at the end of each iteration, more new formula is as follows:
τ(u)←ρ·τ(u)+γ·Δτ(u)
Wherein,
ρ is coefficient of heredity, it controls grid current residual information concentration to the influence degree of next iteration, before which inherits
The residual risk concentration of iteration;
Δ τ (u) is the space constraint component of residual risk concentration increments, and calculation formula is as follows:
Wherein,
B is to assume current grid,
B ' is one of the grid for meeting segmentation condition in neighborhood,
NSFor Size of Neighborhood.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611247939.3A CN106650916B (en) | 2016-12-29 | 2016-12-29 | A kind of mesh segmentation method based on ant group optimization |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611247939.3A CN106650916B (en) | 2016-12-29 | 2016-12-29 | A kind of mesh segmentation method based on ant group optimization |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106650916A CN106650916A (en) | 2017-05-10 |
CN106650916B true CN106650916B (en) | 2019-02-01 |
Family
ID=58837183
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201611247939.3A Active CN106650916B (en) | 2016-12-29 | 2016-12-29 | A kind of mesh segmentation method based on ant group optimization |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106650916B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110288606B (en) * | 2019-06-28 | 2024-04-09 | 中北大学 | Three-dimensional grid model segmentation method of extreme learning machine based on ant lion optimization |
CN111696111B (en) * | 2020-06-15 | 2023-04-18 | 重庆大学 | 3D model mesh segmentation method based on SSDF attenuation map clustering |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104732522A (en) * | 2015-02-05 | 2015-06-24 | 江西科技学院 | Image segmentation method based on polymorphic ant colony algorithm |
CN105139409A (en) * | 2015-09-11 | 2015-12-09 | 浙江工商大学 | Two-dimensional image segmentation method based on ant colony algorithm |
-
2016
- 2016-12-29 CN CN201611247939.3A patent/CN106650916B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104732522A (en) * | 2015-02-05 | 2015-06-24 | 江西科技学院 | Image segmentation method based on polymorphic ant colony algorithm |
CN105139409A (en) * | 2015-09-11 | 2015-12-09 | 浙江工商大学 | Two-dimensional image segmentation method based on ant colony algorithm |
Non-Patent Citations (3)
Title |
---|
Ant Colony Optimization Inspired Algorithm for 3D Object Segmentation;Rafael Arnay, et al.;《Advances in Computational Intelligence》;20131231;262-269 |
一种基于蚁群聚类的图像分割方法;汤可宗等;《科技视界》;20131017(第25期);50-51 |
蚁群优化在超声图像运动矢量估计中的应用;张耀楠等;《东北大学学报(自然科学版)》;20120331;第33卷(第3期);327-331 |
Also Published As
Publication number | Publication date |
---|---|
CN106650916A (en) | 2017-05-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Muzahid et al. | CurveNet: Curvature-based multitask learning deep networks for 3D object recognition | |
CN103207879B (en) | The generation method and apparatus of image index | |
CN105389569B (en) | A kind of estimation method of human posture | |
CN109993748B (en) | Three-dimensional grid object segmentation method based on point cloud processing network | |
CN109559320A (en) | Realize that vision SLAM semanteme builds the method and system of figure function based on empty convolution deep neural network | |
CN109934121A (en) | A kind of orchard pedestrian detection method based on YOLOv3 algorithm | |
CN109635843B (en) | Three-dimensional object model classification method based on multi-view images | |
CN112257597B (en) | Semantic segmentation method for point cloud data | |
CN109508675B (en) | Pedestrian detection method for complex scene | |
CN109191455A (en) | A kind of field crop pest and disease disasters detection method based on SSD convolutional network | |
CN105139379B (en) | Based on the progressive extracting method of classified and layered airborne Lidar points cloud building top surface | |
CN110378997A (en) | A kind of dynamic scene based on ORB-SLAM2 builds figure and localization method | |
CN109493344A (en) | A kind of semantic segmentation method of large-scale city three-dimensional scenic | |
CN105354593B (en) | A kind of threedimensional model sorting technique based on NMF | |
CN110827398A (en) | Indoor three-dimensional point cloud automatic semantic segmentation algorithm based on deep neural network | |
CN109558902A (en) | A kind of fast target detection method | |
CN108154104A (en) | A kind of estimation method of human posture based on depth image super-pixel union feature | |
CN103559705A (en) | Computer method for comparing similarity of different plant forms | |
CN105678747A (en) | Tooth mesh model automatic segmentation method based on principal curvature | |
CN110084136A (en) | Context based on super-pixel CRF model optimizes indoor scene semanteme marking method | |
CN101877146A (en) | Method for extending three-dimensional face database | |
CN106650916B (en) | A kind of mesh segmentation method based on ant group optimization | |
CN112396655A (en) | Point cloud data-based ship target 6D pose estimation method | |
CN108830222A (en) | A kind of micro- expression recognition method based on informedness and representative Active Learning | |
CN109766790A (en) | A kind of pedestrian detection method based on self-adaptive features channel |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |