CN107274483A - A kind of object dimensional model building method - Google Patents
A kind of object dimensional model building method Download PDFInfo
- Publication number
- CN107274483A CN107274483A CN201710447564.3A CN201710447564A CN107274483A CN 107274483 A CN107274483 A CN 107274483A CN 201710447564 A CN201710447564 A CN 201710447564A CN 107274483 A CN107274483 A CN 107274483A
- Authority
- CN
- China
- Prior art keywords
- image
- characteristic point
- pixel
- point
- dimensional model
- 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
- 238000000034 method Methods 0.000 title claims abstract description 53
- 238000010276 construction Methods 0.000 claims abstract description 8
- 238000010586 diagram Methods 0.000 claims description 20
- 238000004364 calculation method Methods 0.000 claims description 7
- 230000004044 response Effects 0.000 claims description 6
- 238000001228 spectrum Methods 0.000 claims description 5
- 238000000205 computational method Methods 0.000 abstract description 2
- 239000011159 matrix material Substances 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 4
- 238000000605 extraction Methods 0.000 description 4
- 238000012216 screening Methods 0.000 description 3
- 238000001514 detection method Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
- G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/50—Depth or shape recovery
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2200/00—Indexing scheme for image data processing or generation, in general
- G06T2200/08—Indexing scheme for image data processing or generation, in general involving all processing steps from image acquisition to 3D model generation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
- G06T2207/10012—Stereo images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20228—Disparity calculation for image-based rendering
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computer Graphics (AREA)
- Geometry (AREA)
- Software Systems (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of object dimensional model building method, including:First camera device and the second camera device are shot simultaneously with different angles to object, are obtained subject image, are respectively depicted as the first image and the second image;Characteristic point is extracted in the picture;It is characterized structure description;In the first image and the second image, the characteristic point matched is searched out according to description of characteristic point, matching relationship is set up and obtains disparity map;Using disparity map, the intrinsic parameter of the first camera device or the second camera device and outer parameter, the locus of characteristic point correspondence spatial point is obtained, to construct object dimensional model.Object dimensional model building method of the present invention, by extracting characteristic point and being characterized structure description, characteristic matching is carried out to image using efficient feature point detecting method and feature point description sub- computational methods, realize construction body three-dimensional models, compared with existing object dimensional model building method, calculate more efficient.
Description
Technical field
The present invention relates to image procossing application field, more particularly to a kind of object dimensional model building method.
Background technology
Technique of binocular stereoscopic vision is a kind of three-dimensional measurement technology of passive type, including camera calibration, image are to matching, three
Tie up the main process such as information reverting.Object dimensional model building method based on binocular stereo vision has been demarcated by two
, the monocular camera put with different angles be tested object image while shooting two width, then set up two images pixel it
Between corresponding relation, obtain measured object three-dimensional point cloud coordinate data, so as to build measured object threedimensional model.
Stereovision technique can obtain dense measured object three-dimensional point cloud coordinate, without to testee projection grating etc.
Auxiliary information, man-machine interaction is friendly, and hardware configuration requires low, and cost is cheap, and automaticity is high, and can realize and adopt in real time
Collection, is currently popular a kind of technology in object dimensional reconstruction field.
Solid matching method is the key problem of technique of binocular stereoscopic vision.However, existing feature based Point matching
In object dimensional model building method, the extraction of characteristic point and computational efficiency are low, the characteristic point of robust such as SIFT feature
Point or SURF characteristic points, it is longer that it calculates the time accordingly, causes computational efficiency low.Therefore working as needs to extract more spy
It is just improper using these characteristic points when levying a little for three-dimensional reconstruction.Because the quantity of characteristic point is more, object reconstruction system
The time spent in the extraction and calculating of characteristic point is longer, and then causes the reduction of object reconstruction overall system performance, it is impossible to
Meet the growing demand to three-dimensional reconstruction performance of people.
The content of the invention
It is an object of the invention to provide a kind of object dimensional model building method, realized based on image procossing and build object three
Dimension module, the extraction and image procossing to image characteristic point is more efficient.
To achieve the above object, the present invention provides following technical scheme:
A kind of object dimensional model building method, including:
First camera device and the second camera device are shot simultaneously with different angles to object, obtain subject image, respectively
It is described as the first image and the second image;
Characteristic point is extracted in the picture, including:For pixel in image, if being preset in the pixel in neighborhood, pixel value is full
The quantity of the pixel of the first preparatory condition of foot meets the second preparatory condition, then is characterized the pixel definition a little;
Description is built for the characteristic point, description of the characteristic point is in each layer convolution direction by the characteristic point
The vector that the value of respective pixel is constituted in figure;
In described first image and second image, the spy matched is searched out according to description of the characteristic point
Levy a little, set up matching relationship and obtain disparity map;
Using the intrinsic parameter and outer parameter of the disparity map, first camera device or second camera device, obtain
The locus of the characteristic point correspondence spatial point is obtained, with construction body three-dimensional models.
Alternatively, the characteristic point of extracting in the picture also includes:
The corresponding response of each key point is calculated using Harris receptance functions, according to response size from big to small
Each key point is sorted, top n key point is taken as final set of keypoints.
Alternatively, it is described to be included for characteristic point structure description attached bag:
Gradient is asked with y directions in the x-direction respectively to the image of shooting, gradient map is obtained;
Each pair gradient map is projected to multiple directions in plane, gradient direction figure is calculated to each pair gradient map;
The convolution algorithm of at least three kinds different Gaussian kernels is carried out to the obtained gradient direction figure, convolution direction is obtained
Figure;
It is that the characteristic point builds description according to the obtained convolution directional diagram.
Alternatively, it is described to project each pair gradient map to multiple directions in plane, gradient is calculated to each pair gradient map
Directional diagram includes:
The gradient map dx and dy obtained based on calculating, according to calculating formula dx cos θ1+dy sinθ2Gradient direction figure is calculated,
WhereinT represents T direction gradient figure of calculating, and T is positive integer, i=0,1 ..., T-1.
Alternatively, to the gradient direction figure in 8 directions in each pair gradient map Calculation Plane;
The convolution algorithm of three kinds of different Gaussian kernels is carried out to the obtained gradient direction figure, 24 convolution directions are obtained
Figure, is expressed asWherein, Gaussian kernel is expressed asJ=0,1,2, Q=3.
Alternatively, the intrinsic parameter using the disparity map, first camera device or second camera device
With outer parameter, obtaining the locus of the characteristic point correspondence spatial point includes:
Using similar triangles property, the depth of the characteristic point correspondence spatial point is calculated by the disparity map;
The characteristic point depth that intrinsic parameter based on the first camera device, outer parameter and calculating are obtained, calculates the feature
The locus of point correspondence spatial point.
Alternatively, it is described in described first image and in second image, found according to the description of characteristic point
Going out the characteristic point matched includes:
For each characteristic point in the first image, matching characteristic point is searched in the second image, calculates in two images and searches for
Euclidean distance between description of the matching characteristic point gone out, when Euclidean distance is less than first threshold, the match is successful.
Alternatively, it is described in described first image and in second image, found according to the description of characteristic point
Going out the characteristic point matched also includes:The characteristic point that the match is successful is screened using constraints, rejecting is unsatisfactory for
The characteristic point of constraints.
Alternatively, the constraints includes:
Unique constraints condition, be specially:Second image is matched by described first image and obtains the first disparity map, by
The second images match described first image obtains error after the second disparity map, same pixel matching in two disparity maps not
More than default error allowed band;
Or/and Ordinal Consistency constraints, it is specially:If pixel (u in described first image0, v0) matching described second
Pixel (u, v) in image, as pixel (u in described first image0+ 1, v0) match second image when, matched position can not go out
Present pixel (u, v) left side;
Or/and disparity continuity constraints, it is specially:Pixel (the u in disparity map0, v0) parallax and its neighborhood in it is each
The difference of pixel parallax is no more than Second Threshold.
Alternatively, the characteristic point that the match is successful is screened using constraints described, rejecting is unsatisfactory for about
Also include after the characteristic point of beam condition:
In default confining spectrum, the characteristic point for being unsatisfactory for constraints is matched again, if being defined in described preset
In the range of can not find match point, then build regarding for its correspondence position in the obtained disparity map using linear interpolation method
Difference.
As shown from the above technical solution, object dimensional model building method provided by the present invention, first, by the first shooting
Device and the second camera device are shot simultaneously with different angles to object, obtain subject image, be respectively depicted as the first image and
Second image, secondly extracts characteristic point in the picture, and is characterized structure description, then in the first image and the second image
In, the characteristic point that matches is searched out according to description of characteristic point, matching relationship is set up and obtains disparity map, further using regarding
The intrinsic parameter and outer parameter of difference figure, the first camera device or the second camera device, obtain the space bit of characteristic point correspondence spatial point
Put, to construct object dimensional model.
Object dimensional model building method of the present invention, by extracting characteristic point and being characterized structure description, utilizes spy
Description levied a little carries out characteristic matching and carrys out construction body three-dimensional models, compared with existing object dimensional model building method, meter
It is more efficient.
Brief description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
There is the accompanying drawing used required in technology description to be briefly described, it should be apparent that, drawings in the following description are only this
Some embodiments of invention, for those of ordinary skill in the art, on the premise of not paying creative work, can be with
Other accompanying drawings are obtained according to these accompanying drawings.
Fig. 1 is a kind of flow chart of object dimensional model building method provided in an embodiment of the present invention;
Fig. 2 be the embodiment of the present invention in image calculate convolution directional diagram computational methods schematic diagram;
Fig. 3 describes the schematic diagram of son to be built in the embodiment of the present invention to image pixel;
Fig. 4 is that the schematic diagram of picture depth is sought by disparity map in the embodiment of the present invention;
Fig. 5 is the method flow diagram that is matched to characteristic point in image in the embodiment of the present invention.
Embodiment
In order that those skilled in the art more fully understand the technical scheme in the present invention, below in conjunction with of the invention real
The accompanying drawing in example is applied, the technical scheme in the embodiment of the present invention is clearly and completely described, it is clear that described implementation
Example only a part of embodiment of the invention, rather than whole embodiments.Based on the embodiment in the present invention, this area is common
The every other embodiment that technical staff is obtained under the premise of creative work is not made, should all belong to protection of the present invention
Scope.
It refer to Fig. 1, a kind of object dimensional model building method provided in an embodiment of the present invention, including step:
S10:First camera device and the second camera device are shot simultaneously with different angles to object, obtain subject image,
It is respectively depicted as the first image and the second image.
Such as, for human face three-dimensional model build, using two completed demarcation camera devices, two camera devices with
Different angles are shot to face simultaneously, and each shoot obtains piece image, correspond to be described as the first image and the second image respectively.
S11:Characteristic point is extracted in the picture, including:For pixel in image, if being preset in the pixel in neighborhood, pixel
The quantity that value meets the pixel of the first preparatory condition meets the second preparatory condition, then is characterized the pixel definition a little.
In this step, whether it is characterized a little for each pixel detection in image, detection method is:For picture in image
Element, if being preset in the pixel in neighborhood, the quantity that pixel value meets the pixel of the first preparatory condition meets the second preparatory condition, then
The pixel is defined as characteristic point.
Wherein, pixel value refers to the gray value of pixel.First preparatory condition is pixel value higher than predetermined threshold value or less than pre-
If threshold value.
Such as, around pixel to be detected in 16 pixel regions, pixel value meets the pixel quantity of the first preparatory condition
Account for the area pixel total quantity 3/4 and more than, then the pixel definition is characterized a little.
Characteristic point is extracted by the above method, situation about too assembling easily occurs in obtained characteristic point, and extracts
Characteristic point is of low quality, in consideration of it, in the another specific embodiment of this method, being extracted in this step by the above method
To after characteristic point, in addition to:
The corresponding response of each characteristic point is calculated using Harris receptance functions, to each characteristic point according to response
Value size sorts from big to small, takes top n characteristic point as final set of characteristic points.
Wherein, Harris receptance functions are expressed as R=Det (M)-kTr (M)2, k represents threshold value, in the specific implementation
0.04, Det (M) is typically set in OpenCV and represents that the product ab, Tr (M) of the oval major and minor axis of Harris represent Harris ellipses
Major and minor axis sum a+b.
S12:Description is built for the characteristic point, description of the characteristic point is in each layer convolution by the characteristic point
The vector that the value of respective pixel is constituted in directional diagram.
Structure description is characterized in the picture, and this step specifically includes procedure below:
S120:Gradient is asked with y directions in the x-direction respectively to the image of shooting, gradient map is obtained.Specifically obtain and original graph
As the consistent gradient image of size, dx and dy are expressed as.
S121:Each pair gradient map is projected to multiple directions in plane, gradient direction figure is calculated to each pair gradient map.
The gradient image dx and dy obtained according to upper step, by dx cos θ1+dy sinθ2Calculating obtains gradient direction figure,
WhereinT represents T direction gradient figure of calculating, and T is positive integer, i=0,1 ..., T-1.
The direction quantity planar selected during calculating gradient direction figure is more, and computational accuracy is higher, but can reduce therewith
Data operation speed, therefore in the specific implementation, the quantity of projecting direction can be according to circumstances selected, to take into account computational accuracy and fortune
Calculate efficiency.
It is preferred that, the gradient side in 8 directions in computational accuracy and operation efficiency, Calculation Plane is considered in the present embodiment
Xiang Tu, accordinglyI=0,1 ..., 7, T=8.
S122:The convolution algorithm of at least three kinds different Gaussian kernels is carried out to the obtained gradient direction figure, convolution is obtained
Directional diagram.
The kind number for carrying out different Gaussian kernel convolution algorithms to image is more, to image characteristics extraction and object dimensional model
The precision of structure is higher, but can reduce PDR, increases amount of calculation.Therefore in the specific implementation, be according to actual feelings
Condition, takes into account consideration computational accuracy and data operation efficiency to select the kind number of Gaussian convolution computing.
It is preferred that, the convolution algorithm of three kinds of different Gaussian kernels is carried out in the present embodiment to image.With to 8 directions in plane
In case of calculating gradient direction figure, it refer to shown in Fig. 2 and Fig. 3, the 8 gradient direction figures obtained to upper step enter respectively
The Gaussian convolution of the different Gaussian kernels of three kinds of row, can obtain 24 convolution directional diagrams, is expressed as Wherein, Gaussian kernel is expressed asJ=0,1,2, Q=3.Wherein, Q=3 represents current picture
Description of element is determined that R represents current pixel in third layer by the convolution directional diagram of the pixel of three layers of fixed range around
Pixel distance.
S123:It is that the characteristic point builds description according to the obtained convolution directional diagram.
In case of carrying out the convolution algorithm of three kinds of different Gaussian kernels in examples detailed above to gradient direction figure, it can obtain
24 convolution gradient direction figures.
The distance of each layer of convolution directional diagram distance feature point pixel is expressed as(i=0,1,2, Q=3).
Wherein, it is r to characteristic point pixel distance0Pixel from Gaussian kernel be ∑0Convolution directional diagramIn take the picture of its correspondence position
Element value, is r to characteristic point pixel distance1Pixel from Gaussian kernel be ∑1Convolution directional diagramIn take the picture of its correspondence position
Element value, is r to characteristic point pixel distance2Pixel from Gaussian kernel be ∑2Convolution directional diagramIn take the picture of its correspondence position
Element value, then the description sublist of image pixel (u, v) is shown as:
S13:In described first image and second image, searched out and matched according to description of the characteristic point
Characteristic point, set up matching relationship and obtain disparity map.
Characteristic matching is carried out in this step, phase is found using description of characteristic point in the first image and the second image
The characteristic point matched somebody with somebody.Specifically, specifically including following steps in this step:
S130:For each characteristic point in the first image, matching characteristic point is searched in the second image, using quickly most
Nearest neighbor algorithm finds matching characteristic point in the second image, then calculates description of the matching characteristic point searched out in two images
Between Euclidean distance, when Euclidean distance be less than first threshold when the match is successful.Matching relationship is then set up, disparity map is obtained.
S14:Utilize the intrinsic parameter and outer ginseng of the disparity map, first camera device or second camera device
Number, obtains the spatial position data of the characteristic point correspondence spatial point, with construction body three-dimensional models.
Procedure below is specifically included in this step:
S140:Using similar triangles property, the depth of the characteristic point correspondence spatial point is calculated by the disparity map.
With reference to shown in Fig. 4, P (X, Y, Z) represents the corresponding spatial point of characteristic point, X1Representation space point P is in the first shooting dress
The X-coordinate value of (i.e. in the first image) picture point, X are set in image planes2Representation space point P on the second camera device imaging surface (i.e.
In second image) X-coordinate value of picture point, f represents camera device focal length, TXThe distance of photocentre, that is, exist between expression two images
The distance of photocentre between two images described in the same space coordinate system.Constituted using P points and its picture point in two images
Triangle, and P points and photocentre O1、O2Similar quality between the triangle of composition, calculates P point depth, and specific formula for calculation is represented
For:Wherein d represents the parallax of matching characteristic point.
S141:The characteristic point depth that intrinsic parameter based on the first camera device, outer parameter and calculating are obtained, calculates described
The locus of characteristic point correspondence spatial point.
In the specific implementation, calculation formula is expressed as:Wherein,Represent feature
Point image plane coordinate,Represent the position of characteristic point correspondence spatial point, ([R1|t1])-1Represent first camera device
The inverse matrix of outer parameter matrix,Represent the inverse matrix of the first camera device Intrinsic Matrix.
It should be noted that in this method other embodiments, also can using the second camera device Intrinsic Matrix, outer
Parameter matrix is calculated.
The three-dimensional point cloud locus obtained according to calculating, constructs the threedimensional model of object, further exportable display.
The present embodiment object model three-dimensional construction method, is counted using efficient feature point detecting method and feature point description
Calculation method carries out characteristic matching to image, calculates more efficient with existing object dimensional model building method.
It refer to Fig. 5, the object dimensional model building method that further embodiment of this invention is provided, in above-described embodiment content
On the basis of, also include in the step S13:
Step S131:The characteristic point that the match is successful is screened using constraints, rejecting is unsatisfactory for constraining bar
The characteristic point of part.
Specifically include and characteristic point is screened according to following several constraintss:
The first screening technique:Judge whether to meet unique constraints condition.Unique constraints condition is specially:By described
The second image obtains the first disparity map described in first images match, and second is obtained by the second images match described first image
Error after disparity map, same pixel matching in two disparity maps is no more than default error allowed band.If not satisfying the constraint
Condition then rejects the matching characteristic point.
Second of screening technique:Judge whether to meet Ordinal Consistency constraints.Ordinal Consistency constraints is specific
For:If pixel (u in described first image0, v0) pixel (u, v) in second image is matched, when pixel in described first image
(u0+ 1, v0) match second image when, matched position can not appear in pixel (u, v) left side.If not satisfying the constraint bar
Part, then reject the matching characteristic point.
The third screening technique, judges whether to meet disparity continuity constraints.Disparity continuity constraints is specific
For:Pixel (the u in disparity map0, v0) parallax and its neighborhood in each pixel parallax difference no more than Second Threshold.If no
The constraints is met, then is rejected the matching characteristic point.
For above-mentioned three kinds restrictive condition judgments, in the method when it is implemented, can use therein any one
Kind, two kinds or three kinds, in a kind of embodiment, above-mentioned three kinds can be carried out successively and restrictive is judged.By to
Restrictive judgement is carried out with characteristic point, wherein Mismatching point is deleted.
Further, step S132 is also included after step S131:In default confining spectrum, to being unsatisfactory for constraining bar
The characteristic point of part is matched again, if can not find match point in the default confining spectrum, is existed using linear interpolation method
The parallax of its correspondence position is built in the obtained disparity map.
The present embodiment method passes through weight matching process, further Optimized Matching result.
Therefore, the three-dimensional construction method of the present embodiment object model, first with the two camera devices collection two demarcated
Width testee image, then to image zooming-out key feature points and calculates description and just matching, then rejected with constraints
Error hiding, and matched again, matching result is obtained, measured object three-dimensional point cloud coordinate is finally reconstructed and shows, obtain precision
Height, the good measured object Three-dimension Reconstruction Model of effect, it is achieved that the thing that a kind of accuracy rate is high, efficiency is fast, easy to operate, inexpensive
Body three-dimensional models construction method.
This method can be applied to face three-dimensional reconstruction, can also be applied to other measured object threedimensional models and builds.
A kind of object dimensional model building method provided by the present invention is described in detail above.It is used herein
Specific case is set forth to the principle and embodiment of the present invention, and the explanation of above example is only intended to help and understands
The method and its core concept of the present invention.It should be pointed out that for those skilled in the art, not departing from this
On the premise of inventive principle, some improvement and modification can also be carried out to the present invention, these are improved and modification also falls into the present invention
In scope of the claims.
Claims (10)
1. a kind of object dimensional model building method, it is characterised in that including:
First camera device and the second camera device are shot simultaneously with different angles to object, are obtained subject image, are described respectively
For the first image and the second image;
Characteristic point is extracted in the picture, including:For pixel in image, if being preset in the pixel in neighborhood, pixel value meets the
The quantity of the pixel of one preparatory condition meets the second preparatory condition, then is characterized the pixel definition a little;
For the characteristic point build description son, the characteristic point description son be by the characteristic point in each layer convolution directional diagram
The vector that the value of respective pixel is constituted;
In described first image and second image, the feature matched is searched out according to description of the characteristic point
Point, sets up matching relationship and obtains disparity map;
Using the intrinsic parameter and outer parameter of the disparity map, first camera device or second camera device, institute is obtained
The locus of characteristic point correspondence spatial point is stated, with construction body three-dimensional models.
2. object dimensional model building method according to claim 1, it is characterised in that described to extract feature in the picture
Point also includes:
The corresponding response of each key point is calculated using Harris receptance functions, according to response size from big to small to each
The key point sequence, takes top n key point as final set of keypoints.
3. object dimensional model building method according to claim 1, it is characterised in that described to be built for the characteristic point
Description attached bag is included:
Gradient is asked with y directions in the x-direction respectively to the image of shooting, gradient map is obtained;
Each pair gradient map is projected to multiple directions in plane, gradient direction figure is calculated to each pair gradient map;
The convolution algorithm of at least three kinds different Gaussian kernels is carried out to the obtained gradient direction figure, convolution directional diagram is obtained;
It is that the characteristic point builds description according to the obtained convolution directional diagram.
4. object dimensional model building method according to claim 3, it is characterised in that it is described by each pair gradient map to flat
Multiple directions are projected in face, and calculating gradient direction figure to each pair gradient map includes:
The gradient map dx and dy obtained based on calculating, according to calculating formula dxcos θ1+dysinθ2Gradient direction figure is calculated, whereinT represents T direction gradient figure of calculating, and T is positive integer, i=0,1 ..., T-1.
5. object dimensional model building method according to claim 4, it is characterised in that to each pair gradient map Calculation Plane
The gradient direction figure in interior 8 directions;
The convolution algorithm of three kinds of different Gaussian kernels is carried out to the obtained gradient direction figure, 24 convolution directional diagrams, table are obtained
It is shown asWherein, Gaussian kernel is expressed asj
=0,1,2, Q=3.
6. object dimensional model building method according to claim 1, it is characterised in that the utilization disparity map,
The intrinsic parameter and outer parameter of first camera device or second camera device, obtain the characteristic point correspondence spatial point
Locus includes:
Using similar triangles property, the depth of the characteristic point correspondence spatial point is calculated by the disparity map;
The characteristic point depth that intrinsic parameter based on the first camera device, outer parameter and calculating are obtained, calculates the characteristic point pair
Answer the locus of spatial point.
7. the object dimensional model building method according to claim any one of 1-6, it is characterised in that described described
In one image and second image, the characteristic point matched is searched out according to description of the characteristic point to be included:
For each characteristic point in the first image, matching characteristic point is searched in the second image, calculates what is searched out in two images
Euclidean distance between description of matching characteristic point, when Euclidean distance is less than first threshold, the match is successful.
8. object dimensional model building method according to claim 7, it is characterised in that it is described in described first image and
In second image, the characteristic point matched is searched out according to description of the characteristic point also to be included:Utilize constraints
The characteristic point that the match is successful is screened, the characteristic point for being unsatisfactory for constraints is rejected.
9. object dimensional model building method according to claim 8, it is characterised in that the constraints includes:
Unique constraints condition, be specially:Second image is matched by described first image and obtains the first disparity map, by described
The error that second images match described first image obtains after the second disparity map, same pixel matching in two disparity maps is no more than
Default error allowed band;
Or/and Ordinal Consistency constraints, it is specially:If pixel (u in described first image0, v0) matching second image
Middle pixel (u, v), as pixel (u in described first image0+ 1, v0) match second image when, matched position can not be appeared in
Pixel (u, v) left side;
Or/and disparity continuity constraints, it is specially:Pixel (the u in disparity map0, v0) parallax and its neighborhood in each pixel
The difference of parallax is no more than Second Threshold.
10. object dimensional model building method according to claim 8, it is characterised in that in the utilization constraints
The characteristic point that the match is successful is screened, rejecting is unsatisfactory for also including after the characteristic point of constraints:
In default confining spectrum, the characteristic point for being unsatisfactory for constraints is matched again, if in the default confining spectrum
It is interior to find match point, then the parallax of its correspondence position is built in the obtained disparity map using linear interpolation method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710447564.3A CN107274483A (en) | 2017-06-14 | 2017-06-14 | A kind of object dimensional model building method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710447564.3A CN107274483A (en) | 2017-06-14 | 2017-06-14 | A kind of object dimensional model building method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107274483A true CN107274483A (en) | 2017-10-20 |
Family
ID=60066720
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710447564.3A Pending CN107274483A (en) | 2017-06-14 | 2017-06-14 | A kind of object dimensional model building method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107274483A (en) |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108725044A (en) * | 2018-05-21 | 2018-11-02 | 贵州民族大学 | A kind of mechano-electronic teaching drafting machine |
CN108764024A (en) * | 2018-04-09 | 2018-11-06 | 平安科技(深圳)有限公司 | Generating means, method and the computer readable storage medium of human face recognition model |
CN109191509A (en) * | 2018-07-25 | 2019-01-11 | 广东工业大学 | A kind of virtual binocular three-dimensional reconstruction method based on structure light |
CN109857895A (en) * | 2019-01-25 | 2019-06-07 | 清华大学 | Stereoscopic vision search method and system based on polycyclic road view convolutional neural networks |
CN110009722A (en) * | 2019-04-16 | 2019-07-12 | 成都四方伟业软件股份有限公司 | Three-dimensional rebuilding method and device |
CN110021065A (en) * | 2019-03-07 | 2019-07-16 | 杨晓春 | A kind of indoor environment method for reconstructing based on monocular camera |
CN110837297A (en) * | 2019-10-31 | 2020-02-25 | 联想(北京)有限公司 | Information processing method and AR equipment |
CN110874818A (en) * | 2018-08-31 | 2020-03-10 | 阿里巴巴集团控股有限公司 | Image processing and virtual space construction method, device, system and storage medium |
CN110942479A (en) * | 2018-09-25 | 2020-03-31 | Oppo广东移动通信有限公司 | Virtual object control method, storage medium, and electronic device |
CN111160232A (en) * | 2019-12-25 | 2020-05-15 | 上海骏聿数码科技有限公司 | Front face reconstruction method, device and system |
WO2020181465A1 (en) | 2019-03-11 | 2020-09-17 | Moqi Technology (beijing) Co., Ltd. | Device and method for contactless fingerprint acquisition |
CN112132960A (en) * | 2020-09-28 | 2020-12-25 | 北京博能科技股份有限公司 | Three-dimensional reconstruction method and device and electronic equipment |
CN112951014A (en) * | 2021-01-29 | 2021-06-11 | 中南大学湘雅三医院 | Intelligent medical education system, method, terminal and medium integrating teaching and practical training |
CN115082563A (en) * | 2021-03-15 | 2022-09-20 | 北京小米移动软件有限公司 | Image processing method and device, electronic equipment and storage medium |
CN117372647A (en) * | 2023-10-26 | 2024-01-09 | 天宫开物(深圳)科技有限公司 | Rapid construction method and system of three-dimensional model for building |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20080002409A (en) * | 2006-06-30 | 2008-01-04 | 연세대학교 산학협력단 | Device and method for transforming 2-d image into 3-d image |
CN102722731A (en) * | 2012-05-28 | 2012-10-10 | 南京航空航天大学 | Efficient image matching method based on improved scale invariant feature transform (SIFT) algorithm |
CN104166995A (en) * | 2014-07-31 | 2014-11-26 | 哈尔滨工程大学 | Harris-SIFT binocular vision positioning method based on horse pace measurement |
CN106485690A (en) * | 2015-08-25 | 2017-03-08 | 南京理工大学 | Cloud data based on a feature and the autoregistration fusion method of optical image |
-
2017
- 2017-06-14 CN CN201710447564.3A patent/CN107274483A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20080002409A (en) * | 2006-06-30 | 2008-01-04 | 연세대학교 산학협력단 | Device and method for transforming 2-d image into 3-d image |
CN102722731A (en) * | 2012-05-28 | 2012-10-10 | 南京航空航天大学 | Efficient image matching method based on improved scale invariant feature transform (SIFT) algorithm |
CN104166995A (en) * | 2014-07-31 | 2014-11-26 | 哈尔滨工程大学 | Harris-SIFT binocular vision positioning method based on horse pace measurement |
CN106485690A (en) * | 2015-08-25 | 2017-03-08 | 南京理工大学 | Cloud data based on a feature and the autoregistration fusion method of optical image |
Non-Patent Citations (2)
Title |
---|
蓝福明: "双目立体视觉的摄像机标定与特征点匹配技术研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
陈济棠: "双目视觉三维测量技术研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2019196308A1 (en) * | 2018-04-09 | 2019-10-17 | 平安科技(深圳)有限公司 | Device and method for generating face recognition model, and computer-readable storage medium |
CN108764024A (en) * | 2018-04-09 | 2018-11-06 | 平安科技(深圳)有限公司 | Generating means, method and the computer readable storage medium of human face recognition model |
CN108725044A (en) * | 2018-05-21 | 2018-11-02 | 贵州民族大学 | A kind of mechano-electronic teaching drafting machine |
CN109191509A (en) * | 2018-07-25 | 2019-01-11 | 广东工业大学 | A kind of virtual binocular three-dimensional reconstruction method based on structure light |
CN110874818B (en) * | 2018-08-31 | 2023-06-23 | 阿里巴巴集团控股有限公司 | Image processing and virtual space construction method, device, system and storage medium |
CN110874818A (en) * | 2018-08-31 | 2020-03-10 | 阿里巴巴集团控股有限公司 | Image processing and virtual space construction method, device, system and storage medium |
CN110942479A (en) * | 2018-09-25 | 2020-03-31 | Oppo广东移动通信有限公司 | Virtual object control method, storage medium, and electronic device |
CN110942479B (en) * | 2018-09-25 | 2023-06-02 | Oppo广东移动通信有限公司 | Virtual object control method, storage medium and electronic device |
CN109857895B (en) * | 2019-01-25 | 2020-10-13 | 清华大学 | Stereo vision retrieval method and system based on multi-loop view convolutional neural network |
CN109857895A (en) * | 2019-01-25 | 2019-06-07 | 清华大学 | Stereoscopic vision search method and system based on polycyclic road view convolutional neural networks |
CN110021065A (en) * | 2019-03-07 | 2019-07-16 | 杨晓春 | A kind of indoor environment method for reconstructing based on monocular camera |
WO2020181465A1 (en) | 2019-03-11 | 2020-09-17 | Moqi Technology (beijing) Co., Ltd. | Device and method for contactless fingerprint acquisition |
US11734948B2 (en) | 2019-03-11 | 2023-08-22 | Moqi Technology (beijing) Co., Ltd. | Device and method for contactless fingerprint acquisition |
CN110009722A (en) * | 2019-04-16 | 2019-07-12 | 成都四方伟业软件股份有限公司 | Three-dimensional rebuilding method and device |
CN110837297A (en) * | 2019-10-31 | 2020-02-25 | 联想(北京)有限公司 | Information processing method and AR equipment |
CN111160232A (en) * | 2019-12-25 | 2020-05-15 | 上海骏聿数码科技有限公司 | Front face reconstruction method, device and system |
CN112132960A (en) * | 2020-09-28 | 2020-12-25 | 北京博能科技股份有限公司 | Three-dimensional reconstruction method and device and electronic equipment |
CN112132960B (en) * | 2020-09-28 | 2024-01-30 | 北京博能科技股份有限公司 | Three-dimensional reconstruction method and device and electronic equipment |
CN112951014A (en) * | 2021-01-29 | 2021-06-11 | 中南大学湘雅三医院 | Intelligent medical education system, method, terminal and medium integrating teaching and practical training |
CN115082563A (en) * | 2021-03-15 | 2022-09-20 | 北京小米移动软件有限公司 | Image processing method and device, electronic equipment and storage medium |
CN117372647A (en) * | 2023-10-26 | 2024-01-09 | 天宫开物(深圳)科技有限公司 | Rapid construction method and system of three-dimensional model for building |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107274483A (en) | A kind of object dimensional model building method | |
CN106355570B (en) | A kind of binocular stereo vision matching method of combination depth characteristic | |
US11521311B1 (en) | Collaborative disparity decomposition | |
Hirschmuller | Stereo processing by semiglobal matching and mutual information | |
US8326025B2 (en) | Method for determining a depth map from images, device for determining a depth map | |
CN106485690A (en) | Cloud data based on a feature and the autoregistration fusion method of optical image | |
CN108288292A (en) | A kind of three-dimensional rebuilding method, device and equipment | |
CN106651942A (en) | Three-dimensional rotation and motion detecting and rotation axis positioning method based on feature points | |
CN107230225A (en) | The method and apparatus of three-dimensional reconstruction | |
Agrawal et al. | Analytical forward projection for axial non-central dioptric and catadioptric cameras | |
CN108596975A (en) | A kind of Stereo Matching Algorithm for weak texture region | |
CN102903101B (en) | Method for carrying out water-surface data acquisition and reconstruction by using multiple cameras | |
CN110567441B (en) | Particle filter-based positioning method, positioning device, mapping and positioning method | |
EP1063614A2 (en) | Apparatus for using a plurality of facial images from different viewpoints to generate a facial image from a new viewpoint, method thereof, application apparatus and storage medium | |
CN111754618B (en) | Object-oriented live-action three-dimensional model multi-level interpretation method and system | |
Kuschk | Large scale urban reconstruction from remote sensing imagery | |
CN113192179A (en) | Three-dimensional reconstruction method based on binocular stereo vision | |
CN107590444A (en) | Detection method, device and the storage medium of static-obstacle thing | |
CN105654547A (en) | Three-dimensional reconstruction method | |
CN113362457A (en) | Stereoscopic vision measurement method and system based on speckle structured light | |
CN110096993A (en) | The object detection apparatus and method of binocular stereo vision | |
CN115035235A (en) | Three-dimensional reconstruction method and device | |
Song et al. | Volumetric stereo and silhouette fusion for image-based modeling | |
CN113989758A (en) | Anchor guide 3D target detection method and device for automatic driving | |
CN114332125A (en) | Point cloud reconstruction method and device, electronic equipment and storage medium |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20171020 |