CN107133284A - A kind of view method for searching three-dimension model based on prevalence study - Google Patents
A kind of view method for searching three-dimension model based on prevalence study Download PDFInfo
- Publication number
- CN107133284A CN107133284A CN201710251632.9A CN201710251632A CN107133284A CN 107133284 A CN107133284 A CN 107133284A CN 201710251632 A CN201710251632 A CN 201710251632A CN 107133284 A CN107133284 A CN 107133284A
- Authority
- CN
- China
- Prior art keywords
- model
- view
- mapping function
- characteristic vector
- popular world
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Library & Information Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Databases & Information Systems (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of view method for searching three-dimension model based on prevalence study, including：In training pattern storehouse, carry out data mark and build energy function, using optimum theory, the mapping function that study classic view feature maps to popular world, using mapping function is locally linear embedding into, model parameter is obtained by sample training；Characteristic vector of the test data under popular world is calculated using mapping function, property data base is used as；An object is randomly choosed from view model storehouse as inquiry target, then chooses any object as comparison object；Similarity between model two-by-two is calculated using Euclidean distance；Characteristic vector of all models of the target in the characteristic vector and property data base of popular world in popular world will be inquired about and carry out matching probability descending arrangement, final retrieval result is obtained.Present invention, avoiding redundancy in the retrieval of traditional view, the complexity of similarity measurement between model is reduced, the precision of three-dimensional model search is improved.
Description
Technical field
The present invention relates to view three-dimensional model search field, more particularly to a kind of view threedimensional model based on prevalence study
Search method.
Background technology
In recent years, with the fast development of multimedia technology, the information that people obtain from sound, image and video is
Through demand can not be met.Threedimensional model arises at the historic moment, and its application field also becomes more and more extensive[1].Threedimensional model can be with
More directly perceived, vivid visual experience is provided, more object informations are included than two dimensional image, therefore be widely used in 3D
The fields such as game, virtual reality, industrial manufacture, medical image.Nowadays on internet threedimensional model quantity be on the increase, pattern number
Also constantly increasing according to storehouse, in face of huge three-dimensional modeling data storehouse, how to cause user accurately and rapidly searches out to meet to need
The threedimensional model asked, and then realize the quick using as numerous scholar's research focuses of resource.How three-dimensional mould reasonably to be described
Type is that feature extraction turns into difficulties of the three-dimensional model search firstly the need of solution[2].View three-dimensional model search turns into instantly
The study hotspot in the field.
View three-dimensional model search is based on computer vision, Digital Image Processing, multimedia messages analysis and machine
The technologies such as study, by computer processing technology, are handled, analyzed and are compared to the view of the threedimensional model in database
Process.Currently, three-dimensional model search technology is broadly divided into two classes：Text based retrieval and content-based retrieval.
Wherein, text based retrieval mode is labeled and divided to threedimensional model using the mode of text or coding
Class, serves huge effect in the three-dimensional model search of early stage[3].This method is simple and clear, it is easy to operation and left-hand seat, but
Excessive subjectivity is take part in when being due to mark, with very strong one-sidedness, so can not fully and accurately reflect
Full detail representated by original three-dimensional model.Retrieval result can not presentation user well intention.
Wherein, content-based retrieval mode is then mainly by studying the spatial distribution characteristic of threedimensional model, after pretreatment
By feature extraction function, the correlated characteristic of threedimensional model is extracted, complicated threedimensional model is abstract for being capable of accurate description
Description of original three-dimensional model, then carries out similarity measurement.This mode avoids manual intervention, and inspection is improved well
The degree of accuracy of rope.Two class methods respectively have quality, but content-based retrieval can utilize the two dimensional image for developing more maturation
Treatment technology and be widely used.
The difficult point that is run into the three-dimensional model search based on view is at present：When gathering view, due to each three
Dimension module is made up of multiple views, and redundancy is excessive between view, causes the difficulty increase of Similarity Measure between model.
The content of the invention
The invention provides a kind of multiview three-dimensional model retrieval method based on prevalence study, it is to avoid traditional view
Redundancy in retrieval, reduces the complexity of similarity measurement between model, improves the precision of three-dimensional model search, refers to down
Text description：
It is a kind of based on prevalence study view method for searching three-dimension model, the view method for searching three-dimension model include with
Lower step：
In training pattern storehouse, carry out data mark and build energy function, using optimum theory, learn classic view feature
The mapping function mapped to popular world, using mapping function is locally linear embedding into, model parameter is obtained by sample training；
Characteristic vector of the test data under popular world is calculated using mapping function, property data base is used as；From view
An object is randomly choosed in model library as inquiry target, then chooses any object as comparison object；
Theory analysis carries out Similarity Measure, and the similarity between model two-by-two is calculated using Euclidean distance；Mesh will be inquired about
All models being marked in the characteristic vector of popular world and property data base are matched in the characteristic vector of popular world
Probability descending is arranged, and obtains final retrieval result.
Wherein, the training pattern storehouse is specially：
The view of fractional object in Selection Model database, training pattern is defined as by total view-set of the fractional object
Storehouse.
Wherein, the utilization mapping function calculates characteristic vector of the test data under popular world, is used as characteristic
The step of storehouse is specially：
Calculate K Neighbor Points of each sample of test data characteristic vector；Calculate partial reconstruction weight matrix, definition weight
Structure error function；
The initial characteristicses vector set of inquiry target and comparison object is mapped using mapping function is locally linear embedding into,
Obtain the set of eigenvectors under popular world.
The characteristic value of loss function value is subjected to ascending order arrangement from small to large, taken corresponding to the characteristic value between 2~d+1
Characteristic vector be used as output result.
The beneficial effect for the technical scheme that the present invention is provided is：
1st, the present invention carries out feature extraction, popular world mapping by the view of the threedimensional model to acquisition, obtains and retrieve
Similarity between target and database object, improves the accuracy of various visual angles target retrieval；
2nd, the problem of there is redundancy between view in the three-dimensional model searching algorithm based on view is solved, is reduced
The difficulty of similarity measurement；
3rd, retrieval of the threedimensional model under popular world is realized using being locally linear embedding into mapping function (LLE)；
4th, view three-dimensional model search is applied to popular world, is effectively maintained the popular structure of threedimensional model, than
Simple calculating Euclidean distance, effect is more preferable.
Brief description of the drawings
Fig. 1 is the flow chart of the multiview three-dimensional model retrieval method based on prevalence study；
Fig. 2 is the popular three-dimensional model search block diagram learnt；
Fig. 3 is the schematic diagram that threedimensional model maps in popular world；
Fig. 4 is the schematic diagram of colored views sample and initial views sample.
Embodiment
To make the object, technical solutions and advantages of the present invention clearer, further is made to embodiment of the present invention below
It is described in detail on ground.
In order to solve problem above, it is desirable to be able to comprehensive, automatic, accurate to extract various visual angles clarification of objective, prevalence is carried out
The mapping in space, finally carries out retrieval matching.Research shows：Feature of the test data in popular world is obtained using mapping function
Vector is matched, it is to avoid the problem of the redundancy of traditional views registered presence.
Embodiment 1
The embodiment of the present invention proposes the view method for searching three-dimension model based on prevalence study, referring to Fig. 1 and Fig. 2, in detail
See below description：
101：The various visual angles colored views of each object are gathered, the initial views collection that each object is obtained after mask are extracted, by institute
The collection that always attempts to for having object is defined as model database；
102：The view of fractional object in Selection Model database, total view-set of the fractional object is defined as to train mould
Type storehouse；
103：In training pattern storehouse, carry out data mark and build energy function, using optimum theory, learn classic view
The mapping function that feature maps to popular world, using mapping function is locally linear embedding into, obtains model by sample training and joins
Number；
104：Characteristic vector of the test data under popular world is calculated using mapping function, property data base is used as；From
An object is randomly choosed in view model storehouse as inquiry target, then chooses any object as comparison object, retrieval tasks are
The object similar to inquiry target is found from view model storehouse；
105：Theory analysis carries out Similarity Measure, without loss of generality, using between Euclidean distance calculating two-by-two model
Similarity；Next comparison model in selected characteristic database, repeats the above steps, until all in traversal model library
Model；
106：All models of the target in the characteristic vector and property data base of popular world will be inquired about in popular world
Characteristic vector carry out matching probability descending arrangement, obtain final retrieval result.
Wherein, above-mentioned steps 104 are specially：Inquiry and is compared target using mapping function (LLE) is locally linear embedding into
The initial characteristicses vector set of target is mapped, and obtains the set of eigenvectors under popular world.
In summary, this method avoids redundancy in traditional view retrieval by above-mentioned steps 101- steps 106,
The complexity of similarity measurement between model is reduced, the precision of three-dimensional model search is improved.
Embodiment 2
The scheme in embodiment 1 is further introduced with reference to Fig. 3 Fig. 4 and specific calculation formula, in detail
See below description：
201：The various visual angles colored views of N number of object are gathered first, and the initial of each object can be obtained after extracting mask
Depending on collection Vi, by total view-set V={ V of all objects1,V2,…Vi,…,VsIt is defined as view model storehouse MD (Model
Database), wherein i ∈ { 1,2 ..., S }；
The various visual angles colored views of N number of object are gathered in this method first, process is as follows：By three KINECT cameras
(this camera is known to those skilled in the art, full name for " first generation KINECT " of XBOX 360, model 1414) respectively
It is positioned over the level of rotary table, vertical and three visual angles of oblique direction 45 °.Each object is revolved on the table
Turn to shoot AiColored views (because each model complexity is different, AiParticular number can be set as different numerical value,
Three cameras have extraction 3A altogetheriOpen in colored views, the embodiment of the present invention and be set as Ai=36), schematic diagram is as shown in Figure 3.
Then enter line mask extraction to each view, that is, separate foreground model and background area.Separation foundation is that model is regarded
The region decision that the RGB numerical value of figure meets G-R/2-B/2=0 conditions is model, is otherwise judged as background.Extract after mask, i.e.,
It can obtain the initial views collection of each model objectSchematic diagram is as shown in Figure 3.
Wherein,For a initial views of i-th of object, a ∈ { 1,2 ... 3Ai, 3AiFor the initial of each object
View sum.The initial views collection of N number of object is merged and obtains total initial views collection V={ Vi,Vi,…,Vi,…,VS, by its
It is defined as various visual angles model library MD (Model DataBase), wherein i ∈ { 1,2 ..., S }.
202：Selection is locally linear embedding into mapping function and sets up tranining database；
Wherein, it is locally linear embedding into mapping function (Locally-Linear embedding, LLE) and is reflected according to this
Penetrate known to those skilled in the art, the embodiment of the present invention pair of the step of function sets up tranining database, training pattern parameter
This is not repeated.
203：Characteristic vector pickup is carried out in popular world to test data according to mapping function is locally linear embedding into, as
Property data base；
The step is specially：
1st, K Neighbor Points of each sample of test data characteristic vector are calculated；
Wherein, K value is set according to the need in practical application.
2nd, partial reconstruction weight matrix is calculated, reconstructed error function is defined；
Wherein, XiFor i-th of sample, Xij(j=1,2 ..., k) are XiK Neighbor Points, EijIt is XiWith XijBetween power
Value, and without loss of generality, weights are normalized, that is, meet condition ∑j Eij=1；N is number of samples.
It follows that seeking partial reconstruction weight matrix, it is necessary to obtain local covariance matrix R：
Wherein,Represent the local covariance matrix of j Neighbor Points of i-th of sample；T represents transposed matrix；Xim(m<j)
For XiM-th of Neighbor Points.
By ∑j Eij=1 and local covariance matrixIt is combined, and uses method of Lagrange multipliers, you can obtains part
Rebuild weight matrix：
Wherein, RiThe singular matrix tieed up for i, needs to carry out regularization conversion in follow-up calculate；K be j (j=1,2 ...,
K) maximum occurrences.
All sample points are mapped in lower dimensional space, condition is met：
In above formula, ∈ (Y) is loss function value, YiIt is XiOutput valve, Yij(j=1,2 ... k) it is YiK Neighbor Points,
And meet two conditions, i.e.,：
Wherein, I is m*m unit matrix.Here Eij(j=1,2 ..., can n) be stored in n*n sparse matrix E,
Work as XiIt is XjNeighbor Points when, loss function can be rewritten as：
M is n*n symmetrical matrix, M=(I-E)T(I-E), E is the sparse matrix that n*n is tieed up, and T represents matrix transposition.
Loss function is set to reach minimum, then optimal solution Y*Take the feature corresponding to ∈ (Y) minimum d nonzero eigenvalue
Vector.That is, in processing procedure, ∈ (Y) characteristic value is carried out into ascending order arrangement from small to large, first characteristic value is several
Close to zero, then cast out first characteristic value, the characteristic vector corresponding to the characteristic value between 2~d+1 is generally taken to make
For output result.
204：An object is randomly choosed from view model storehouse as inquiry target, then chooses any object as comparing mesh
Mark, retrieval tasks are that the object similar to inquiry target is found from view model storehouse；
205：Theory analysis carries out Similarity Measure, calculates the similarity between model two-by-two；In selected characteristic database
Next comparison model, repeat the above steps, until traversal model library in all models；
After the threedimensional model maps feature vectors of higher-dimension to low-dimensional, without loss of generality, using Euclidean (Euclideans
Distance) similarity between model is measured.Specific formula for calculation is as follows：
Wherein, wiFor the weight of different characteristic component.Obtained result carries out descending arrangement, then can obtain optimal inspection
Hitch fruit.
206：All models of the target in the characteristic vector and property data base of popular world will be inquired about in popular world
Characteristic vector carry out matching probability descending arrangement, obtain final retrieval result.
In faceform, each angular views regard a data point as, and each pixel is a dimension, then a n*m
Image be exactly a point in nm dimension theorem in Euclid space.The free degree of view is gathered according to concrete model, to determine these
The popular dimension of point distribution.Such as, if the view free degree that model is gathered out is 3, then these points are just distributed across in fact
In three-dimensional prevalence；Therefore, Model Matching is carried out in popular world, the popular world structure in view of model is can be very good,
With this so that matching is more accurate.
In summary, this method avoids redundancy in traditional view retrieval by above-mentioned steps 201- steps 206,
The complexity of similarity measurement between model is reduced, the precision of three-dimensional model search is improved.
Bibliography：
[1] Feng Yi climbs three-dimensional model search technical research [D] Zhejiang Polytechnical University of the based on view, 2012.
[2] threedimensional model Study on Feature Extraction summary [J] Chinese new traffics of Liu Yupeng, Hou Yu the elder brother based on view,
2016 (2016 06):99-99,100.
[3] three-dimensional model search technical research [D] the Central China University of Science and Technology of the stone forest woods based on view, 2014.
[4] three-dimensional model search technology [D] Guangxi Normal University Master's thesis of the Xu Peng victories based on content, 2010.
[5] Zheng Baichuan, Peng Wei, Zhang Yin, wait [J] the CADs of .3D model indexs technology summary and graphics
Report, 2004,16 (7)：873-881.
[6]Kim T K,O,Cipolla R.Boosted manifold principal angles
for image set-based recognition[J].Pattern Recognition,2007,40(9):2475-2484.
[7]Pless R,Souvenir R.A Survey of Manifold Learning for Images[J]
.Ipsj Transactions on Computer Vision &Applications,2009,1(1):83-94.
[8]Tahmasbi A,Saki F,Aghapanah H,et al.A novel breast mass diagnosis
system based on Zernike moments as shape and density descriptors[C]//
Biomedical Engineering(ICBME),2011 18th Iranian Conference of.IEEE,2011:100-
104.
[9]Shih J L,Lee C H,Wang J T.A new 3D model retrieval approach based
on the elevation descriptor[J].Pattern Recognition,2007,40(1):283-295.
[10]Ansary T F,Daoudi M,Vandeborre J P.A bayesian 3-d search engine
using adaptive views clustering[J].Multimedia,IEEE Transactions on,2007,9(1):
78-88.
It will be appreciated by those skilled in the art that accompanying drawing is the schematic diagram of a preferred embodiment, the embodiments of the present invention
Sequence number is for illustration only, and the quality of embodiment is not represented.
The foregoing is only presently preferred embodiments of the present invention, be not intended to limit the invention, it is all the present invention spirit and
Within principle, any modification, equivalent substitution and improvements made etc. should be included in the scope of the protection.
Claims (3)
1. a kind of view method for searching three-dimension model based on prevalence study, it is characterised in that the view three-dimensional model search
Method comprises the following steps：
In training pattern storehouse, carry out data mark and build energy function, using optimum theory, learn classic view feature to stream
The mapping function of row space reflection, using mapping function is locally linear embedding into, model parameter is obtained by sample training；
Characteristic vector of the test data under popular world is calculated using mapping function, property data base is used as；From view model
An object is randomly choosed in storehouse as inquiry target, then chooses any object as comparison object；
Theory analysis carries out Similarity Measure, and the similarity between model two-by-two is calculated using Euclidean distance；Inquiry target is existed
All models in the characteristic vector and property data base of popular world carry out matching probability in the characteristic vector of popular world
Descending is arranged, and obtains final retrieval result.
2. a kind of view method for searching three-dimension model based on prevalence study according to claim 1, it is characterised in that institute
Stating training pattern storehouse is specially：
The view of fractional object in Selection Model database, training pattern storehouse is defined as by total view-set of the fractional object.
3. a kind of view method for searching three-dimension model based on prevalence study according to claim 1, it is characterised in that institute
State and calculate characteristic vector of the test data under popular world using mapping function, be specially the step of as property data base：
Calculate K Neighbor Points of each sample of test data characteristic vector；Partial reconstruction weight matrix is calculated, definition reconstruct is missed
Difference function；
The initial characteristicses vector set of inquiry target and comparison object is mapped using mapping function is locally linear embedding into, obtained
Set of eigenvectors under popular world.
The characteristic value of loss function value is subjected to ascending order arrangement from small to large, the spy corresponding to the characteristic value between 2~d+1 is taken
Vector is levied as output result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710251632.9A CN107133284A (en) | 2017-04-18 | 2017-04-18 | A kind of view method for searching three-dimension model based on prevalence study |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710251632.9A CN107133284A (en) | 2017-04-18 | 2017-04-18 | A kind of view method for searching three-dimension model based on prevalence study |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107133284A true CN107133284A (en) | 2017-09-05 |
Family
ID=59716701
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710251632.9A Pending CN107133284A (en) | 2017-04-18 | 2017-04-18 | A kind of view method for searching three-dimension model based on prevalence study |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107133284A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110059213A (en) * | 2019-03-20 | 2019-07-26 | 杭州电子科技大学 | A kind of threedimensional model classification retrieving method based on Density Estimator |
| CN110543581A (en) * | 2019-09-09 | 2019-12-06 | 山东省计算中心（国家超级计算济南中心） | Multi-view three-dimensional model retrieval method based on non-local graph convolution network |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104090972A (en) * | 2014-07-18 | 2014-10-08 | 北京师范大学 | Image feature extraction and similarity measurement method used for three-dimensional city model retrieval |
| CN105243139A (en) * | 2015-10-10 | 2016-01-13 | 天津大学 | Deep learning based three-dimensional model retrieval method and retrieval device thereof |
-
2017
- 2017-04-18 CN CN201710251632.9A patent/CN107133284A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104090972A (en) * | 2014-07-18 | 2014-10-08 | 北京师范大学 | Image feature extraction and similarity measurement method used for three-dimensional city model retrieval |
| CN105243139A (en) * | 2015-10-10 | 2016-01-13 | 天津大学 | Deep learning based three-dimensional model retrieval method and retrieval device thereof |
Non-Patent Citations (2)
| Title |
|---|
| MEGUMI ENDOH，ET AL.: "Efficient manifold learning for 3D model retrieval by using clustering-based training sample reduction", 《2012 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS，SPEECH AND SIGNAL PROCESSING》 * |
| 王庆刚.: "流形学习算法及若干应用研究", 《中国博士学位论文全文数据库 信息科技辑》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110059213A (en) * | 2019-03-20 | 2019-07-26 | 杭州电子科技大学 | A kind of threedimensional model classification retrieving method based on Density Estimator |
| CN110543581A (en) * | 2019-09-09 | 2019-12-06 | 山东省计算中心（国家超级计算济南中心） | Multi-view three-dimensional model retrieval method based on non-local graph convolution network |
| CN110543581B (en) * | 2019-09-09 | 2023-04-04 | 山东省计算中心（国家超级计算济南中心） | Multi-view three-dimensional model retrieval method based on non-local graph convolution network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhu et al. | Vpfnet: Improving 3d object detection with virtual point based lidar and stereo data fusion | |
| Wang et al. | SaliencyGAN: Deep learning semisupervised salient object detection in the fog of IoT | |
| Guo et al. | Efficient center voting for object detection and 6D pose estimation in 3D point cloud | |
| CN109063139B (en) | Three-dimensional model classification and retrieval method based on panorama and multi-channel CNN | |
| CN108319957A (en) | A kind of large-scale point cloud semantic segmentation method based on overtrick figure | |
| CN103927511B (en) | image identification method based on difference feature description | |
| CN110069656A (en) | A method of threedimensional model is retrieved based on the two-dimension picture for generating confrontation network | |
| CN112907602B (en) | Three-dimensional scene point cloud segmentation method based on improved K-nearest neighbor algorithm | |
| CN111625667A (en) | Three-dimensional model cross-domain retrieval method and system based on complex background image | |
| CN105930382A (en) | Method for searching for 3D model with 2D pictures | |
| CN106844620B (en) | View-based feature matching three-dimensional model retrieval method | |
| CN105868706A (en) | Method for identifying 3D model based on sparse coding | |
| CN105183795B (en) | Remote Sensing Imagery Change Detection information retrieval method based on content | |
| CN108537887A (en) | Sketch based on 3D printing and model library 3-D view matching process | |
| Feng et al. | 3D shape retrieval using a single depth image from low-cost sensors | |
| Zhang et al. | Perception-based shape retrieval for 3D building models | |
| Zhou et al. | Learning transferable and discriminative representations for 2D image-based 3D model retrieval | |
| CN112668662B (en) | Outdoor mountain forest environment target detection method based on improved YOLOv3 network | |
| CN107133284A (en) | A kind of view method for searching three-dimension model based on prevalence study | |
| CN117788810A (en) | Learning system for unsupervised semantic segmentation | |
| Pan et al. | Online human action recognition based on improved dynamic time warping | |
| Wang et al. | Swimmer’s posture recognition and correction method based on embedded depth image skeleton tracking | |
| CN111597367A (en) | Three-dimensional model retrieval method based on view and Hash algorithm | |
| Proenca et al. | SHREC’15 Track: Retrieval of Oobjects captured with kinect one camera | |
| Cao et al. | Stable image matching for 3D reconstruction in outdoor |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170905 |
|
| RJ01 | Rejection of invention patent application after publication |