CN106529394B - A kind of indoor scene object identifies simultaneously and modeling method - Google Patents
A kind of indoor scene object identifies simultaneously and modeling method Download PDFInfo
- Publication number
- CN106529394B CN106529394B CN201610832845.6A CN201610832845A CN106529394B CN 106529394 B CN106529394 B CN 106529394B CN 201610832845 A CN201610832845 A CN 201610832845A CN 106529394 B CN106529394 B CN 106529394B
- Authority
- CN
- China
- Prior art keywords
- feature
- dimensional
- similarity
- identified
- fpfh
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- 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
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/64—Three-dimensional objects
- G06V20/653—Three-dimensional objects by matching three-dimensional models, e.g. conformal mapping of Riemann surfaces
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Geometry (AREA)
- Computer Graphics (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of indoor scene object while identification and modeling methods, comprising steps of input RGB-D image;Object segmentation;Extract SIFT feature and FPFH feature;SIFT feature and FPFH Fusion Features;Object identification;Object modeling.Wherein object modeling is the position orientation relation calculated between object parts and sets a threshold value, if pose variation merges two object parts less than threshold value, and a node is used as in view figure, otherwise, two object parts all remain into view figure, as two nodes.Compared with prior art, object online recognition and modeling may be implemented in the present invention, and proposes improved view drawing method, reduces data redundancy, reduces data storage burden, improve recognition efficiency.
Description
Technical field
The present invention relates to field of information processing more particularly to a kind of indoor scene object while identification and modeling methods.
Background technique
Currently, in robot object identification and scene understanding research field, a large amount of algorithm researches all concentrate on passing through by
Test picture matches with training dataset or achievees the purpose that identification by training classifier, and does not all take into account that object generally
The pose problem of body.As Liang Mingjie et al. proposes the object model representation method based on view figure, and indicated in the object
On the basis of provide the probability observation model of object, identification with modeling is finally attributed to probability inference, and estimate by maximum likelihood
Optimization while meter realizes identification and modeling;Alvaro Collet Romea proposes the indoor object based on constraint frame
Body identification and modeling method, this method can automatically handle entire video data stream, and it was found that object, most important spy
Point is to joined constraint frame, certain constraints in daily life can be dissolved into frame by mathematical way, can be with
According to the different additions of scene, constraint condition is reduced or changed, increases object identification rate and object identification efficiency.
But accurate posture information, such as robot manipulation can be all used in many object identifications application, path rule
It draws, augmented reality etc., and algorithm needs training data, so being unable to reach online recognition effect, has seriously affected machine
People's performance in scene indoors.In terms of modeling, robot walks in environment indoors for a long time, the observation portion of indoor object
Branch is often acquired, so that the model established generates redundant data, the time of robot ambulation is longer, obtains same
The observation data of a object are more, and the object model of foundation is bigger, and redundant data can also increase, this builds not only bad for object
Mould and data storage, also have a adverse impact to later period object identification efficiency.
Summary of the invention
In order to overcome the deficiencies of the prior art, object online recognition and modeling are realized, and solves the problems, such as data redundancy, is reduced
Memory burden, effectively improves recognition efficiency, and the present invention proposes that a kind of indoor scene object identifies and modeling method simultaneously.
The technical scheme of the present invention is realized as follows:
A kind of indoor scene object identification and modeling method, including step S1 simultaneously: input RGB-D image;
S2: object segmentation;
S3: SIFT feature and FPFH feature are extracted;
S4:SIFT feature and FPFH Fusion Features, object identification;
S5: object modeling calculates the position orientation relation between object parts and sets a threshold value, if pose variation is small
In threshold value, then two object parts are merged, and is used as a node, otherwise, two object parts in view figure
It all remains into view figure, as two nodes.
Further, the step S2 includes step
S21: planar point cloud is extracted using RANSAC method;
S22: the point cloud unrelated with object data is rejected;
S23: object point cloud is obtained by Euclidean distance clustering algorithm;
S24: point cloud data is mapped back into RGB-D image
Further: step S3 includes step
S31: object two dimension local feature region and Feature Descriptor are extracted using SIFT feature extracting method;
S32: the two-dimentional local feature region is mapped in three-dimensional space, three-dimensional feature point is obtained;
S33: using FPFH algorithm, generates three-dimensional feature description.
Further, step S4 includes step
S41: the match point of some mistakes is rejected using RANSAC algorithm;
S42: the two dimensional character calculated between object to be identified and object model is registrated distance;
S43: it calculates three-dimensional feature between object to be identified and object model and is registrated distance;
S44: fusion two-dimensional signal and three-dimensional information, and save similarity result;
S45: comparing all similarity results, take out the maximum result of similarity as this object identification most
Terminate fruit.
The beneficial effects of the present invention are, compared with prior art, object online recognition and modeling may be implemented in the present invention,
And improved view drawing method is proposed, data redundancy is reduced, data storage burden is reduced, improves recognition efficiency.
Detailed description of the invention
Fig. 1 is indoor scene object of the present invention while identification and modeling method flow chart.
Fig. 2 is the original point cloud chart of one embodiment of the invention.
Fig. 3 is the filtered point cloud chart of Fig. 2.
Fig. 4 is one embodiment of the invention RANSAC segmentation point cloud chart.
Fig. 5 is cluster segmentation point cloud chart.
Fig. 6 is the object figure before segmentation.
Fig. 7 is object figure of the Fig. 6 after over-segmentation.
Fig. 8 is partial error point registration result schematic diagram.
Fig. 9, which is Fig. 8, rejects registration result schematic diagram after the match points of some mistakes using RANSAC algorithm.
Figure 10 is misregistration result schematic diagram.
Figure 11 is Figure 10 three-dimensional registration result schematic diagram.
The recognizer flow chart of Figure 12 fusion two and three dimensions information.
Figure 13 is view graph model schematic diagram.
Figure 14 is the improved view graph model schematic diagram of Figure 13.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
Referring to Figure 1, a kind of indoor scene object identification and modeling method, including step S1 simultaneously: input RGB-D figure
Picture;
S2: object segmentation;
S3: SIFT feature and FPFH feature are extracted;
S4:SIFT feature and FPFH Fusion Features, object identification;
S5: object modeling calculates the position orientation relation between object parts and sets a threshold value, if pose variation is small
In threshold value, then two object parts are merged, and is used as a node, otherwise, two object parts in view figure
It all remains into view figure, as two nodes.
In step S2, firstly, calculating point cloud data C by RGB image and depth image, as shown in Figure 2.Due to acquisition
Point cloud data is too big, has an impact to operations such as later period segmentation, identifications, then passes through voxelization grid method and realizes down-sampling,
Point cloud data is reduced, as shown in Figure 3.In order to obtain the object in scene, RANSAC method (random sampling consistency is used herein
Algorithm) planar point cloud is extracted, since planar point cloud is generally the point cloud unrelated with object data such as desktop or metope, it is possible to
It is rejected, as shown in Figure 4.Remaining cloud includes object and noise as seen from Figure 4, is clustered finally by Euclidean distance
Algorithm obtains object point cloud { Cm, m=1 ... M, as shown in Figure 5.Point cloud data maps back RGB and schemes the object after available segmentation
Body RGB schemes { Im, m=1 ... M, as shown in Figure 6,7.Then the object after segmentation can be expressed as { Im, Cm, m=1 ... M.
In step S3, a series of object { I can be partitioned into for each frame scene figurem, Cm, we use SIFT (ruler
Spend invariant features transformation) feature extracting method RGB scheme ImMiddle extraction object two dimension local feature region and Feature Descriptor { Xmr,
Smr, two-dimentional local feature region is mapped in three-dimensional space, obtains three-dimensional feature point, and (quick point feature is straight with FPFH
Side's figure) algorithm generation three-dimensional feature description son { Zmr, Fmr, wherein X indicates that two dimensional character point, S indicate two dimensional character description, Z
Indicate that three-dimensional feature point, F indicate that three-dimensional feature description, m indicate that m-th of object, r indicate r-th of characteristic point.
In step S4, we identify object by fusion two and three dimensions information, improve object identification
Rate.Firstly, using the SIFT feature and Feature Descriptor of object to be identified, using KNN (K arest neighbors) algorithmic match object mould
Type, but SIFT feature registration equally can also generate the match point of mistake, as shown in figure 8, producing several couples of wrong spies in figure
Sign point a registration result, if be utilized mistake as a result, if can seriously affect later period object pose calculating accuracy, so
We reject the match point of some mistakes using RANSAC algorithm, as shown in Figure 9.But sometimes due to two-dimensional image surface can recognize
Property it is too weak, textural characteristics are unobvious, and causing SIFT feature registration to generate very big mistake will lead to final knowledge as shown in Figure 10
Not mistake.We have incorporated three-dimensional object structure feature on the basis of two-dimensional, and three-dimensional feature registration result is as shown in figure 11, have
The deficiency for compensating for two dimensional character registration of effect.
Recognizer using fusion two and three dimensions information is as follows:
Step1: calculating two dimensional character between object to be identified and object model and be registrated distance, as shown in formula (1):
C1(s, t)=d (SS,St) (1)
Wherein s indicates that object to be identified, t indicate that object model, d () indicate that Euclidean distance, S indicate SIFT feature description
Son, C1(s, t) indicates the Euclidean distance between object SIFT feature and object model SIFT feature to be identified, i.e., two-dimentional similar
Degree, and be normalized, C is required herein1When (s, t) < 0.5, two articles are similar, execute step2 and otherwise continue to hold
Row step1.
Step2: calculating three-dimensional feature between object to be identified and object model and be registrated distance, as shown in formula (2):
C2(s, t)=1-d (FS,Ft) (2)
Wherein s indicates that object to be identified, t indicate that object model, d () indicate that Euclidean distance, F indicate the description of FPFH feature
Son, C2(s, t) indicates the Euclidean distance between object FPFH feature and object model FPFH feature to be identified, i.e., three-dimensional similar
Degree, and be normalized, C is required herein2When (s, t) > 0.5, two articles are similar, execute step3, otherwise execute
step1。
Step3: fusion two-dimensional signal and three-dimensional information, and similarity result is saved, as shown in formula (3):
T (s, t)=C2(s,t)·exp(-k·C1(s,t)) (3)
Wherein T (s, t) indicates the similarity after fusion two and three dimensions information, by C1Index location is placed on mainly to examine
Two-dimentional SIFT feature has been considered for object identification with more judgment and meaning, the adjustable weight size of k.If matched
Object model is the last one model, then executes step4, otherwise execute step1.
Step4: comparing all similarity results, takes out the maximum result of similarity as this object identification
Final result.Algorithm flow chart is as shown in figure 12.
In step S5, by calculating the position orientation relation between object parts, and a threshold value is set, if pose changes
Less than threshold value, then two object parts are merged, and is used as a node, otherwise, two object portions in view figure
Divide and all remain into view figure, as two nodes in this way, not only remain the function that view figure is easily modified, but also reduce
The redundancy of data, improves operational efficiency.
The data that fusion between similar view retains have: (1) model view relative pose, will not for figure after fusion
New node is generated, but new side can be generated, so needing record cast view relative pose;(2) two view feature points
Position, the characteristic point position of new view, the feature that two views are mutually matched after the characteristic point position of master mould view and transformation
Point position only retains once;(3) Feature Descriptor, including SIFT feature description son and FPFH Feature Descriptor, for view it
Between matching, the Feature Descriptor that two views are mutually matched only retains once;The point cloud data of (4) two views, master mould
The point cloud data of new view after the point cloud data of view and transformation, fused point cloud data is real by voxelization grid method
It now samples, reduces point cloud data.The production process for so improving view figure can be by following algorithmic notation:
Wherein V0Indicate original view node of graph, E0Indicate original view figure side, hiIndicate newly generated i-th of view
Figure, P (hi) indicate hiThe similarity of view and database object model, pminIndicate similarity threshold, Pangle(hi) indicate and count
According to the relative angle between the most similar view in library, PangleminIndicate angle threshold, Pdistance(hi) indicate close with database
View between relative distance, PdisminIndicate distance threshold, i.e., if hiBetween view and the most similar view of database
Relative angle and relative distance are both less than some threshold value, then merge two views, otherwise without fusion, are applied directly to
In view figure, a node is generated.P(hi, hj) indicate that i-th of view is similar between j-th of view to database model
Degree illustrates that there are similarity relations between two views, then by i-th of view and jth if this similarity is greater than threshold value
Transformation relation between a view is added in database object model, becomes a new side.Figure 13 and Figure 14 is view figure
Comparison between model and improved view graph model, wherein dark node indicates that fused node, grayed-out nodes are new add
The node entered, dotted line is when expression is newly added or changes, and improved view graph model does not increase section it can be seen from upper figure
Point increases a line newly, and view graph model has then newly added a node and three sides, so methods herein can theoretically drop
Low data redundancy reaches better object modeling effect.
The above is a preferred embodiment of the present invention, it is noted that for those skilled in the art
For, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also considered as
Protection scope of the present invention.
Claims (4)
1. a kind of indoor scene object identifies simultaneously and modeling method, which is characterized in that including step
S1: input RGB-D image;
S2: object segmentation;
S3: SIFT feature and FPFH feature are extracted;
S4:SIFT feature and FPFH Fusion Features, object identification;
S5: object modeling calculates the position orientation relation between object parts and sets a threshold value, if pose variation is less than threshold
Value, then merge two object parts, and a node is used as in view figure, and otherwise, two object parts are all protected
It is left in view figure, as two nodes;
The step 4 includes step
Step1: calculating two dimensional character between object to be identified and object model and be registrated distance, as shown in formula (1):
C1(s, t)=d (SS,St) (1)
Wherein s indicates that object to be identified, t indicate that object model, d () indicate that Euclidean distance, S indicate SIFT feature description, C1
(s, t) indicates the Euclidean distance between object SIFT feature and object model SIFT feature to be identified, i.e. two dimension similarity, and
It is normalized, works as C1When (s, t) < 0.5, two articles are similar, execute step2 and otherwise continue to execute step1;
Step2: calculating three-dimensional feature between object to be identified and object model and be registrated distance, as shown in formula (2):
C2(s, t)=1-d (FS,Ft) (2)
Wherein s indicates that object to be identified, t indicate that object model, d () indicate that Euclidean distance, F indicate FPFH Feature Descriptor, C2
(s, t) indicates the Euclidean distance between object FPFH feature and object model FPFH feature to be identified, i.e. three-dimensional similarity, and
It is normalized, works as C2When (s, t) > 0.5, two articles are similar, execute step3, otherwise execute step1;
Step3: fusion two-dimensional signal and three-dimensional information, and similarity result is saved, as shown in formula (3):
T (s, t)=C2(s,t)·exp(-k·C1(s,t)) (3)
Wherein T (s, t) indicates the similarity after fusion two and three dimensions information, by C1Index location is placed on mainly to consider
Two-dimentional SIFT feature has more judgment and meaning, the adjustable weight size of k, if matched object for object identification
Body Model is the last one model, then executes step4, otherwise execute step1;
Step4: comparing all similarity results, takes out the maximum result of similarity as the final of this object identification
As a result.
2. indoor scene object as described in claim 1 identifies simultaneously and modeling method, which is characterized in that the step S2 packet
Include step
S21: planar point cloud is extracted using RANSAC method;
S22: the point cloud unrelated with object data is rejected;
S23: object point cloud is obtained by Euclidean distance clustering algorithm;
S24: point cloud data is mapped back into RGB-D image.
3. indoor scene object as described in claim 1 identifies simultaneously and modeling method, step S3 include step
S31: object two dimension local feature region and Feature Descriptor are extracted using SIFT feature extracting method;
S32: the two-dimentional local feature region is mapped in three-dimensional space, three-dimensional feature point is obtained;
S33: using FPFH algorithm, generates three-dimensional feature description.
4. indoor scene object as described in claim 1 identifies simultaneously and modeling method, step S4 include step
S41: the match point of some mistakes is rejected using RANSAC algorithm;
S42: the two dimensional character calculated between object to be identified and object model is registrated distance;
S43: it calculates three-dimensional feature between object to be identified and object model and is registrated distance;
S44: fusion two-dimensional signal and three-dimensional information, and save similarity result;
S45: comparing all similarity results, takes out most termination of the maximum result of similarity as this object identification
Fruit.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610832845.6A CN106529394B (en) | 2016-09-19 | 2016-09-19 | A kind of indoor scene object identifies simultaneously and modeling method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610832845.6A CN106529394B (en) | 2016-09-19 | 2016-09-19 | A kind of indoor scene object identifies simultaneously and modeling method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106529394A CN106529394A (en) | 2017-03-22 |
CN106529394B true CN106529394B (en) | 2019-07-19 |
Family
ID=58344928
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610832845.6A Active CN106529394B (en) | 2016-09-19 | 2016-09-19 | A kind of indoor scene object identifies simultaneously and modeling method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106529394B (en) |
Families Citing this family (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107123027B (en) * | 2017-04-28 | 2021-06-01 | 广东工业大学 | Deep learning-based cosmetic recommendation method and system |
CN108133181B (en) * | 2017-12-12 | 2022-03-18 | 北京小米移动软件有限公司 | Method for acquiring indication information, AR device and storage medium |
CN108537163A (en) * | 2018-04-04 | 2018-09-14 | 天目爱视(北京)科技有限公司 | A kind of biological characteristic 4 D data recognition methods taken pictures based on visible light and system |
CN110553849A (en) * | 2018-06-01 | 2019-12-10 | 上汽通用汽车有限公司 | Driving condition evaluation system and method |
CN109360174B (en) * | 2018-08-29 | 2020-07-07 | 清华大学 | Three-dimensional scene reconstruction method and system based on camera pose |
CN109360234B (en) * | 2018-08-29 | 2020-08-21 | 清华大学 | Three-dimensional scene reconstruction method and system based on total uncertainty |
CN109242959B (en) * | 2018-08-29 | 2020-07-21 | 清华大学 | Three-dimensional scene reconstruction method and system |
CN109344813B (en) * | 2018-11-28 | 2023-11-28 | 北醒(北京)光子科技有限公司 | RGBD-based target identification and scene modeling method |
EP3894788A4 (en) * | 2018-12-13 | 2022-10-05 | Continental Automotive GmbH | Method and system for generating an environment model for positioning |
CN110097598B (en) * | 2019-04-11 | 2021-09-07 | 暨南大学 | Three-dimensional object pose estimation method based on PVFH (geometric spatial gradient frequency) features |
CN111797268B (en) * | 2020-07-17 | 2023-12-26 | 中国海洋大学 | RGB-D image retrieval method |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101271469A (en) * | 2008-05-10 | 2008-09-24 | 深圳先进技术研究院 | Two-dimension image recognition based on three-dimensional model warehouse and object reconstruction method |
CN101877143A (en) * | 2009-12-09 | 2010-11-03 | 中国科学院自动化研究所 | Three-dimensional scene reconstruction method of two-dimensional image group |
CN102609979A (en) * | 2012-01-17 | 2012-07-25 | 北京工业大学 | Fourier-Mellin domain based two-dimensional/three-dimensional image registration method |
CN104715254A (en) * | 2015-03-17 | 2015-06-17 | 东南大学 | Ordinary object recognizing method based on 2D and 3D SIFT feature fusion |
CN105261060A (en) * | 2015-07-23 | 2016-01-20 | 东华大学 | Point cloud compression and inertial navigation based mobile context real-time three-dimensional reconstruction method |
-
2016
- 2016-09-19 CN CN201610832845.6A patent/CN106529394B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101271469A (en) * | 2008-05-10 | 2008-09-24 | 深圳先进技术研究院 | Two-dimension image recognition based on three-dimensional model warehouse and object reconstruction method |
CN101877143A (en) * | 2009-12-09 | 2010-11-03 | 中国科学院自动化研究所 | Three-dimensional scene reconstruction method of two-dimensional image group |
CN102609979A (en) * | 2012-01-17 | 2012-07-25 | 北京工业大学 | Fourier-Mellin domain based two-dimensional/three-dimensional image registration method |
CN104715254A (en) * | 2015-03-17 | 2015-06-17 | 东南大学 | Ordinary object recognizing method based on 2D and 3D SIFT feature fusion |
CN105261060A (en) * | 2015-07-23 | 2016-01-20 | 东华大学 | Point cloud compression and inertial navigation based mobile context real-time three-dimensional reconstruction method |
Non-Patent Citations (3)
Title |
---|
Generic Object Recognition Based on the Fusion of 2D and 3D SIFT Descriptors;Miaomiao Liu 等;《Information Fusion》;20150709;1085-1092 |
HerbDisc: Towards Lifelong Robotic Object Discovery;Alvaro Collet 等;《Robotics Research archive》;20150131;第34卷(第1期);3-25 |
Object Recognition Robust to Imperfect Depth Data;David F. Fouhey 等;《Computer Vision》;20121013;第2卷;83-92 |
Also Published As
Publication number | Publication date |
---|---|
CN106529394A (en) | 2017-03-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106529394B (en) | A kind of indoor scene object identifies simultaneously and modeling method | |
WO2016110005A1 (en) | Gray level and depth information based multi-layer fusion multi-modal face recognition device and method | |
CN107886129B (en) | Mobile robot map closed-loop detection method based on visual word bag | |
US9940577B2 (en) | Finding semantic parts in images | |
CN107424161B (en) | Coarse-to-fine indoor scene image layout estimation method | |
CN106407958B (en) | Face feature detection method based on double-layer cascade | |
JP5394959B2 (en) | Discriminator generating apparatus and method, and program | |
Xiao et al. | Joint affinity propagation for multiple view segmentation | |
CN110866934B (en) | Normative coding-based complex point cloud segmentation method and system | |
CN103530894B (en) | A kind of video object method for tracing based on multiple dimensioned piece of rarefaction representation and system thereof | |
JP2016018538A (en) | Image recognition device and method and program | |
CN106815824B (en) | A kind of image neighbour's optimization method improving extensive three-dimensional reconstruction efficiency | |
CN109492589A (en) | The recognition of face working method and intelligent chip merged by binary features with joint stepped construction | |
CN112562081B (en) | Visual map construction method for visual layered positioning | |
WO2023151237A1 (en) | Face pose estimation method and apparatus, electronic device, and storage medium | |
CN109101864A (en) | The upper half of human body action identification method returned based on key frame and random forest | |
CN107291936A (en) | The hypergraph hashing image retrieval of a kind of view-based access control model feature and sign label realizes that Lung neoplasm sign knows method for distinguishing | |
CN104834894B (en) | A kind of gesture identification method of combination binary coding and class-Hausdorff distances | |
Liu et al. | Regularization based iterative point match weighting for accurate rigid transformation estimation | |
CN104732247B (en) | A kind of human face characteristic positioning method | |
Du | High-precision portrait classification based on mtcnn and its application on similarity judgement | |
CN110942453A (en) | CT image lung lobe identification method based on neural network | |
WO2020232697A1 (en) | Online face clustering method and system | |
CN113887509B (en) | Rapid multi-modal video face recognition method based on image set | |
CN114445649A (en) | Method for detecting RGB-D single image shadow by multi-scale super-pixel fusion |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | 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 |