Nothing Special   »   [go: up one dir, main page]

CN106097436B - A kind of three-dimensional rebuilding method of large scene object - Google Patents

A kind of three-dimensional rebuilding method of large scene object Download PDF

Info

Publication number
CN106097436B
CN106097436B CN201610406617.2A CN201610406617A CN106097436B CN 106097436 B CN106097436 B CN 106097436B CN 201610406617 A CN201610406617 A CN 201610406617A CN 106097436 B CN106097436 B CN 106097436B
Authority
CN
China
Prior art keywords
sequence
photographs
dough sheet
unit
sequence photographs
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
Application number
CN201610406617.2A
Other languages
Chinese (zh)
Other versions
CN106097436A (en
Inventor
林靖宇
郑恩
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangxi University
Original Assignee
Guangxi University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Guangxi University filed Critical Guangxi University
Priority to CN201610406617.2A priority Critical patent/CN106097436B/en
Publication of CN106097436A publication Critical patent/CN106097436A/en
Application granted granted Critical
Publication of CN106097436B publication Critical patent/CN106097436B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T17/00Three dimensional [3D] modelling, e.g. data description of 3D objects

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Computer Graphics (AREA)
  • Geometry (AREA)
  • Software Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Image Processing (AREA)

Abstract

A kind of three-dimensional rebuilding method of large scene object, it is characterised in that: the following steps are included: 1) with camera with different view, position several sequence photographs are acquired to scenario objects;2) sequence photographs, are divided into several unit Ni, 3), feature detection, matching carried out using three sequences piece of the SIFT algorithm to each unit;4) by unit N1, first sequence photographs initialization of (i=1), basis matrix F, the essential matrix E of other two sequence photographs of calculating;5) the camera parameter R of every sequence photographs in other units, is solvedi、TiWith three-dimensional point cloud coordinate Mj);7), the extension of seed dough sheet;8), the filtering of seed dough sheet;The present invention obtains accurate basis matrix using the effective removal Mismatching point of three-view diagram constraint, the precision of reconstruction is high, and coordinate conversion is not needed between unit, computation complexity is reduced, it is denser by carrying out the point cloud that dense reconstruction makes to obtain threedimensional model on the basis of exercise recovery structure.

Description

A kind of three-dimensional rebuilding method of large scene object
Technical field
The present invention relates to a kind of computer vision and field of Computer Graphics, especially a kind of three-dimensional of large scene object Method for reconstructing.
Background technique
The final purpose of three-dimensional reconstruction is to restore the three-dimensional structure of target scene, and the three-dimensional reconstruction based on image sequence is to obtain Take one of the main means of three-dimensional structure.This method can be regarded as the inverse process of photograph, advantage of lower cost, it is only necessary to general Logical video camera, it is easy to operate, easy to carry, be by the Feature Points Matching of image sequence and multiple view stereo reconstruction based on Research foundation and emphasis in calculation machine vision technique.
The present more method of three-dimensional rebuilding method for large scene object is SFM (Structure From Motion), the disadvantage is that feature detection is constrained with when matching using Epipolar geometry, match point can only be determined in corresponding polar curve On, without can determine that match point accurate location, this constraint is weaker.Once Mismatching point deletion is not fallen, basis matrix is solved Mistake, obtained three-dimensional point cloud coordinate are also mistake, and the corresponding cloud of other images match points acquired with this is sat Mark is also mistake, with regard to will appear error accumulation problem.Also one is the three-dimensional reconstructions based on independent three-view diagram, every three width Figure carries out three-dimensional reconstruction as a separate unit, is coordinately transformed between each unit after reconstruction, transformation of scale is all Three-dimensional point cloud be transformed under the same coordinate, this method computation complexity is bigger.In addition both methods have one it is common The problem of be rebuild point cloud all than sparse, be not able to satisfy visualization effect.It also needs to carry out after general sparse reconstruction Dense reconstruction, the most of dense method for reconstructing are three-dimensional reconstruction (the Patched-based Multi-View based on dough sheet Stereo, PMVS).The input of PMVS method be one group of image arrangement set and it is sparse rebuild camera parameter R, T for acquiring, so The point off density cloud of threedimensional model is exported afterwards.
Summary of the invention
In view of the deficiencies of the prior art, the present invention provides a kind of three-dimensional rebuilding method of large scene object.
The technical solution of the present invention is as follows: a kind of three-dimensional rebuilding method of large scene object, it is characterised in that: including following step It is rapid:
1) with camera with different view, position several sequence photographs are acquired to scenario objects, and according to the successive of acquisition Sequence numbers several sequence photographs for Ik, k=1,2,3...N;
2) sequence photographs are divided into several unit N according to the sequencing of numberi, i=1,2,3...n, each unit Including three sequence photographs successively acquired, and by unit Ni, i=1,2,3...n third sequence photographs are as under adjacent One unit Ni+1, i=1,2,3...n-1 first sequence photographs, and using the sequence photographs as the public of two adjacent cells Sequence photographs;
3) by using SIFT algorithm to each unit Ni, i=1,2, the 3...n three sequence photographs progress features for including Detection, matching, and utilize three-view diagram constraint removal Mismatching point;
4) by unit Ni, the camera parameter initialization of first sequence photographs of i=1, and found out using RANSANC algorithm Unit Ni, basis matrix F, the essential matrix E of other two sequence photographs in i=1 obtain every in other two sequence photographs Open the camera parameter R of sequence photographsh、TgWith three-dimensional point cloud coordinate Mj
5) by unit N in adjacent Unit twoi, i=1, the camera parameter R of common sequence photo in 2,3...nh、TgAs it Adjacent next unit Ni+1, i=1, the initial value of the camera parameter of 2,3...n-1 common sequence photo, and solve its it is adjacent under One unit Ni+1, i=1, the camera parameter R of other two sequence photographs in 2,3...n-1h、TgWith three-dimensional point cloud coordinate Mj, with this Calculate unit Ni+1, i=1, the camera parameter R of every sequence photographs in 2,3...n-1 other two sequence photographsh、TgWith three Dimension point cloud coordinate Mj
6) by the camera parameter R of all sequence photographsi、TiWith three-dimensional point cloud coordinate MjAs the input of PMVS algorithm, from And obtain the seed dough sheet of scenario objects;
7) extension of seed dough sheet: by sequence photographs IkIt is divided into several image blocks Co(x, y) utilizes neighboring seeds face There is piece position and the similar characteristic of normal direction to be extended to seed dough sheet, so that the seed dough sheet after extension be made gradually to cover mesh Mark the surface of object;
8) seed dough sheet filters: it is wrong that extension is rejected using local gray level consistency constraint and globally visible consistency constraint Seed dough sheet accidentally.
In above-mentioned technical proposal, when camera surrounds scene objects acquisition sequence photo in step 1), adjacent two sequences are shone Angle between piece is 5 °~15 °, and the sequence photographs quantity of acquisition is odd number, and the sequence photographs of acquisition are at least 3.
In above-mentioned technical proposal, the sequence photographs acquired in step 1) are 45.
In above-mentioned technical proposal, feature detection, matching are carried out to sequence photographs using SIFT algorithm in step 3), to count Calculate the corresponding image points m of three sequence photographs in each unitkj, wherein the corresponding image points of any two sequence photographs is at another Match point on sequence photographs is on the intersection point that any two sequences piece corresponding image points corresponds to polar curve, if match point and intersection point Distance be more than that two pixels are then used as Mismatching point to delete.
Unit N in above-mentioned technical proposal, in step 4) after initializationi, the camera ginseng of first sequence photographs of i=1 Number R1For 3 × 3 unit matrix, T1For 3 × 1 null matrix;
Basis matrix F is solved using RANSANC algorithm, combining camera internal reference K solves essential matrix E, then to essential square Battle array E carries out singular value decomposition and obtains camera parameter Rh、Tg, and obtain three-dimensional point cloud coordinate Mj, then adjusted using boundling The camera parameter R of (Bundle Adjustment) algorithm local optimum sequence photographsh、Tg, three-dimensional point cloud coordinate Mj, pass through boundling Adjustment (bundle adjustment) algorithm makes three-dimensional point cloud coordinate MjRe-projection point in the of the same name of kth sequence photographs Picture point mkiThe quadratic sum of difference is minimum, calculating formula are as follows:
Wherein D is Euclidean distance, and N is sequence photographs number, and M is three-dimensional point cloud number of coordinates.
In above-mentioned technical proposal, the extension in step 7) by seed dough sheet in each image block so that at least reconstruct One seed dough sheet.
In above-mentioned technical proposal, the seed dough sheet of extended error includes external seed dough sheet and internal seeds face in step 8) Piece by the extension of seed dough sheet, is completely covered target object surface by dense seed dough sheet filtration cycle iteration 3 times.
The invention has the benefit that can effectively remove Mismatching point using three-view diagram constraint obtains accurate basis Matrix, the precision of reconstruction is high, and coordinate conversion is not needed between unit, thus greatly reduces computation complexity, by It is denser that the point cloud that dense reconstruction makes to obtain threedimensional model is carried out on the basis of exercise recovery structure.
Detailed description of the invention
Fig. 1 is that three-view diagram of the invention constrains schematic diagram;
Specific embodiment
Specific embodiments of the present invention will be further explained with reference to the accompanying drawing:
As shown in Figure 1, a kind of three-dimensional rebuilding method of large scene object, it is characterised in that: the following steps are included:
1) with camera with different view, position several sequence photographs are acquired to scenario objects, and according to the successive of acquisition Sequence numbers several sequence photographs for Ik, k=1,2,3...N;
2) sequence photographs are divided into several unit N according to the sequencing of numberi, (i=1,2,3...n), Mei Gedan Member includes three sequence photographs successively acquiring, and by unit Ni, the third sequence photographs of (i=1,2,3...n) are as phase Adjacent next unit Ni+1, i=1,2,3...n-1 first sequence photographs, and using the sequence photographs as two adjacent cells Common sequence photo;
3) by using SIFT algorithm to each unit Ni, three sequences piece that (i=1,2,3...n) includes carries out special Sign detection, matching, and utilize three-view diagram constraint removal Mismatching point;
4) by unit Ni, the camera parameter initialization of first sequence photographs of (i=1), and asked using RANSANC algorithm Unit N outi, basis matrix F, the essential matrix E of other two sequence photographs in (i=1) obtain in two sequence photographs every Open the camera parameter R of sequence photographsh、TgWith three-dimensional point cloud coordinate Mj
5) a unit N oni, the camera parameter R of common sequence photo in (i=1,2,3...n)h、TgAs next unit Ni+1, i=1,2,3...n-1 first open the initial value of the camera parameter of sequence photographs, and solve next unit Ni+1, i=1,2, The camera parameter R of two sequence photographs of other in 3...n-1h、TgWith three-dimensional point cloud coordinate Mj, unit N is calculated with thisi+1, i= The camera parameter R of every sequence photographs in 1,2,3...n-1 other two sequence photographsh、TgWith three-dimensional point cloud coordinate Mj
6) by the camera parameter R of all sequence photographsi、TiWith three-dimensional point cloud coordinate MjAs the input of PMVS algorithm, from And obtain the seed dough sheet of scenario objects;
7) extension of seed dough sheet: by sequence photographs IkIt is divided into several image blocks Co(x, y) utilizes neighboring seeds face There is piece position and the similar characteristic of normal direction to be extended to seed dough sheet, so that the seed dough sheet after extension be made gradually to cover mesh Mark the surface of object;
8) seed dough sheet filters: it is wrong that extension is rejected using local gray level consistency constraint and globally visible consistency constraint Seed dough sheet accidentally.
In above-mentioned technical proposal, when camera surrounds scene objects acquisition sequence photo in step 1), adjacent two sequences are shone Angle between piece is 5 °~15 °, and the sequence photographs quantity of acquisition is odd number, and the sequence photographs of acquisition are at least 3.
In above-mentioned technical proposal, the sequence photographs acquired in step 1) are 45.
In above-mentioned technical proposal, feature detection, matching are carried out to sequence photographs using SIFT algorithm in step 3), to count Calculate the corresponding image points m of three sequence photographs in each unitkj, wherein the corresponding image points of any two sequence photographs is at another Match point on sequence photographs is on the intersection point that any two sequences piece corresponding image points corresponds to polar curve, if match point and intersection point Distance be more than that two pixels are then used as Mismatching point to delete, x, x ', x in Fig. 1 " be any unit three open the of the same name of sequence photographs Picture point, wherein the corresponding image points x ' of two sequence photographs, x " match point put two sequence photographs corresponding image points x ', the x " are right The polar curve l answered12、l13Intersection point on, deleted if more than 2 pixels at a distance from match point with this intersection point as Mismatching point It removes.
Unit N in above-mentioned technical proposal, in step 4) after initializationi, the camera of first sequence photographs of (i=1) Parameter R1For 3 × 3 unit matrix, T1For 3 × 1 null matrix;
Basis matrix F is solved using RANSANC algorithm, combining camera internal reference K solves essential matrix E, then to essential square Battle array E carries out singular value decomposition and obtains camera parameter Rh、Tg, and obtain three-dimensional point cloud coordinate Mj, then adjusted using boundling The camera parameter R of (Bundle Adjustment) algorithm local optimum sequence photographsh、Tg, three-dimensional point cloud coordinate Mj, pass through boundling Adjustment (bundle adjustment) algorithm makes three-dimensional point cloud coordinate MjRe-projection point in the of the same name of kth sequence photographs Picture point mkjThe quadratic sum of difference is minimum, calculating formula are as follows:
Wherein D is Euclidean distance, and N is sequence photographs number, and M is three-dimensional point cloud number of coordinates.
In above-mentioned technical proposal, using the characteristic point of Harris and DOG operator detection sequence photo in step 6), find full The potential matching double points of sufficient Epipolar geometry constraint acquire three-dimensional point cloud coordinate using triangle Principle Method to potential matching double points Mi, seed dough sheet center c (p) is three-dimensional point cloud coordinate Mj, sequence photographs collection is combined into Im={ Ik| k=1,2,3 ..., N }, space The unit normal vector n (p) of dough sheet p is the camera photocentre that reference image R (p) is directed toward by space dough sheet center c (p), it may be assumed that
c(p)←{Triangulation from f and f′}
R(p)←Ik
V (p), V are initialized simultaneously*(p), visual picture collection subject to V (p), the unit of the space dough sheet p of V (p) sequence photographs Angle where normal vector n (p) and space dough sheet center c (p) between light is less than 60 °, it may be assumed that
Wherein O (Ik) it is sequence photographs IkThe optical center of corresponding camera,
V*It (p) is visual picture collection, V*(p) the space dough sheet center c (p) of the sequence photographs in projects to reference image R (p) value of the gray consistency function between the sequence photographs is greater than threshold alpha=0.4, it may be assumed that
V*(p)=I | I ∈ V (p), h (p, I, R (p)) > α },
Wherein h (p, I, R (p)) refers to the gray scale uniform metric between sequence photographs I and reference image R (p), is taken as 1 and subtracts Its NCC (normalization intersection correlation) value is gone, then with space dough sheet p in set V*(p) the gray consistency function on
Come advanced optimize space dough sheet center c (p) and The unit normal vector n (p) of space dough sheet, with the unit normal vector n (p) of seed dough sheet center c (p) after optimization and seed dough sheet come Update V (p), the V of dough sheet*(p), if | V*(p) | > γ then generates the success of seed dough sheet, then by every sequence photographs IkIt draws It is divided into the image block C that size is β × β (β=32)o(x, y), arbitrary image block Co(x, y) there are two corresponding set Qi(x, y), Qi *(x, y) is used to store the corresponding real space dough sheet of the image block, wherein x, y are sequence photographs IiIndex;Then life At seed dough sheet project on the sequence photographs of division, record seed dough sheet where image block Co(x, y),
In above-mentioned technical proposal, the extension in step 7) by seed dough sheet p in each image block so that at least rebuild A seed dough sheet out,
In above-mentioned technical proposal, the seed dough sheet p of extended error includes external seed dough sheet and internal seeds in step 8) Dough sheet by the extension of seed dough sheet, is completely covered target object surface by dense seed dough sheet filtration cycle iteration 3 times.
The above embodiments and description only illustrate the principle of the present invention and most preferred embodiment, is not departing from this Under the premise of spirit and range, various changes and improvements may be made to the invention, these changes and improvements both fall within requirement and protect In the scope of the invention of shield.

Claims (5)

1. a kind of three-dimensional rebuilding method of large scene object, it is characterised in that: the following steps are included:
1) by camera with different view, position several sequence photographs are acquired to scenario objects, and according to the successive suitable of acquisition Sequence numbers several sequence photographs for Ik, k=1,2,3...N;
2) sequence photographs are divided into several unit N according to the sequencing of numberi, i=1,2,3...n, each unit packet Include three sequence photographs successively acquired, and by unit NiThird sequence photographs as adjacent next unit Ni+1First Sequence photographs are opened, and using the sequence photographs as the common sequence photo of two adjacent cells;
3) by using SIFT algorithm to each unit Ni, i=1,2,3...n three sequences for including piece progress feature detection, Matching, and Mismatching point is removed using three-view diagram constraint, specifically:
Feature detection, matching are carried out to sequence photographs by SIFT algorithm, to calculate three sequence photographs in each unit Corresponding image points mij, wherein match point of the corresponding image points of any two sequence photographs on another sequence photographs is any at this Two sequences piece corresponding image points correspond on the intersection point of polar curve, the conduct mistake if more than two pixels at a distance from the match point with intersection point Match point deletion;
4) by unit Ni, the camera parameter initialization of first sequence photographs of i=1, and unit is found out using RANSANC algorithm Ni, basis matrix F, the essential matrix E of other two sequence photographs in i=1 obtain every sequence in two sequence photographs and shine The camera parameter R of pieceh、TgWith three-dimensional point cloud coordinate Mj, specifically:
Using RANSANC algorithm solve basis matrix F, combining camera internal reference K solve essential matrix E, then to essential matrix E into Row singular value decomposition obtains camera parameter Rh、Tg, and obtain three-dimensional point cloud coordinate Mj, then locally excellent using boundling adjustment algorithm Change the camera parameter R of sequence photographsh、Tg, three-dimensional point cloud coordinate Mj, three-dimensional point cloud coordinate M is made by boundling adjustment algorithmjWeight Subpoint and the corresponding image points m in kth sequence photographskiThe quadratic sum of difference is minimum, calculating formula are as follows:
Wherein D is Euclidean distance, and N is sequence photographs number, and M is three-dimensional point cloud number of coordinates;
5) by unit N in adjacent Unit twoiThe camera parameter R of middle common sequence photoh、TgAs its adjacent next unit Ni+1's The initial value of the camera parameter of common sequence photo, and solve its adjacent next unit Ni+1In other two sequence photographs phase Machine parameter Rh、TgWith three-dimensional point cloud coordinate Mj, unit N is calculated with thisi+1Every sequence photographs in other two sequence photographs Camera parameter Rh、TgWith three-dimensional point cloud coordinate Mj
6) by the camera parameter R of all sequence photographsh、TgWith three-dimensional point cloud coordinate MjAs the input of PMVS algorithm, thus To the seed dough sheet of scenario objects;
7) extension of seed dough sheet: by sequence photographs IkIt is divided into several image blocks Co(x, y) has using neighboring seeds dough sheet There are position and the similar characteristic of normal direction to be extended seed dough sheet, to make the gradually coverage goal object of the seed dough sheet after extension The surface of body;
8) seed dough sheet filters: rejecting extended error using local gray level consistency constraint and globally visible consistency constraint Seed dough sheet.
2. a kind of three-dimensional rebuilding method of large scene object according to claim 1, it is characterised in that: camera in step 1) When around scene objects acquisition sequence photo, the angle between adjacent two sequence photographs is 5 °~15 °, the sequence of acquisition Number of pictures is odd number, and the sequence photographs of acquisition are at least 3.
3. a kind of three-dimensional rebuilding method of large scene object according to claim 1, it is characterised in that: pass through in step 7) The extension of seed dough sheet in each image block so that at least reconstruct a seed dough sheet.
4. a kind of three-dimensional rebuilding method of large scene object according to claim 1, it is characterised in that: extension in step 8) The seed dough sheet of mistake includes external seed dough sheet and internal seeds dough sheet, and the extension of seed dough sheet, filtering are circuited sequentially iteration 3 It is secondary that target object surface is completely covered by dense seed dough sheet.
5. a kind of three-dimensional rebuilding method of large scene object according to claim 2, it is characterised in that: the sequence of acquisition is shone Piece is 45.
CN201610406617.2A 2016-06-12 2016-06-12 A kind of three-dimensional rebuilding method of large scene object Active CN106097436B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610406617.2A CN106097436B (en) 2016-06-12 2016-06-12 A kind of three-dimensional rebuilding method of large scene object

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610406617.2A CN106097436B (en) 2016-06-12 2016-06-12 A kind of three-dimensional rebuilding method of large scene object

Publications (2)

Publication Number Publication Date
CN106097436A CN106097436A (en) 2016-11-09
CN106097436B true CN106097436B (en) 2019-06-25

Family

ID=57228655

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610406617.2A Active CN106097436B (en) 2016-06-12 2016-06-12 A kind of three-dimensional rebuilding method of large scene object

Country Status (1)

Country Link
CN (1) CN106097436B (en)

Families Citing this family (21)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106910241A (en) * 2017-01-20 2017-06-30 徐迪 The reconstructing system and method for the three-dimensional human head based on cell-phone camera and Cloud Server
CN107330934B (en) * 2017-05-23 2021-12-07 上海交通大学 Low-dimensional cluster adjustment calculation method and system
CN107247834B (en) * 2017-05-31 2019-09-24 华中科技大学 A kind of three dimensional environmental model reconstructing method, equipment and system based on image recognition
CN107590825B (en) * 2017-07-31 2020-10-02 东南大学 Point cloud hole repairing method based on SFM
CN107862742B (en) * 2017-12-21 2020-08-14 华中科技大学 Dense three-dimensional reconstruction method based on multi-hypothesis joint view selection
CN108550166B (en) * 2018-03-26 2022-04-15 北京航空航天大学 Spatial target image matching method
CN108648264B (en) * 2018-04-25 2020-06-23 吉林大学 Underwater scene reconstruction method based on motion recovery and storage medium
CN108592919B (en) * 2018-04-27 2019-09-17 百度在线网络技术(北京)有限公司 Drawing and localization method, device, storage medium and terminal device
CN108961151B (en) * 2018-05-08 2019-06-11 中德(珠海)人工智能研究院有限公司 A method of the three-dimensional large scene that ball curtain camera obtains is changed into sectional view
CN108648270B (en) * 2018-05-12 2022-04-19 西北工业大学 Unmanned aerial vehicle real-time three-dimensional scene reconstruction method capable of realizing real-time synchronous positioning and map construction
CN108986218B (en) * 2018-06-06 2023-03-31 东南大学 PMVS-based rapid building dense point cloud reconstruction method
CN110021065A (en) * 2019-03-07 2019-07-16 杨晓春 A kind of indoor environment method for reconstructing based on monocular camera
CN110060240B (en) * 2019-04-09 2023-08-01 南京链和科技有限公司 Tire contour measurement method based on image pickup
CN110163903B (en) * 2019-05-27 2022-02-25 百度在线网络技术(北京)有限公司 Three-dimensional image acquisition and image positioning method, device, equipment and storage medium
CN111080653B (en) * 2019-11-06 2022-09-20 广西大学 Method for simplifying multi-view point cloud by using region segmentation and grouping random simplification method
CN111047631B (en) * 2019-12-04 2023-04-07 广西大学 Multi-view three-dimensional point cloud registration method based on single Kinect and round box
CN111325780B (en) * 2020-02-17 2021-07-27 天目爱视(北京)科技有限公司 3D model rapid construction method based on image screening
CN112258409B (en) * 2020-10-22 2024-10-18 中国人民武装警察部队工程大学 Method and device for recovering absolute scale of unmanned monocular camera
CN114445550A (en) * 2020-11-06 2022-05-06 中煤航测遥感集团有限公司 Three-dimensional reconstruction method and device, electronic equipment and storage medium
CN112270755B (en) * 2020-11-16 2024-04-05 Oppo广东移动通信有限公司 Three-dimensional scene construction method and device, storage medium and electronic equipment
CN112633293B (en) * 2020-11-24 2022-05-20 北京航空航天大学青岛研究院 Three-dimensional sparse point cloud reconstruction image set classification method based on image segmentation

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102034274A (en) * 2010-12-10 2011-04-27 中国人民解放军国防科学技术大学 Method for calculating corresponding relationship between frame and frame in 3D point cloud sequence
CN103021017A (en) * 2012-12-04 2013-04-03 上海交通大学 Three-dimensional scene rebuilding method based on GPU acceleration
CN105184863A (en) * 2015-07-23 2015-12-23 同济大学 Unmanned aerial vehicle aerial photography sequence image-based slope three-dimension reconstruction method
CN107655459A (en) * 2017-09-07 2018-02-02 南京理工大学 A kind of measurement of field rock texture surface roughness and computational methods

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102034274A (en) * 2010-12-10 2011-04-27 中国人民解放军国防科学技术大学 Method for calculating corresponding relationship between frame and frame in 3D point cloud sequence
CN103021017A (en) * 2012-12-04 2013-04-03 上海交通大学 Three-dimensional scene rebuilding method based on GPU acceleration
CN105184863A (en) * 2015-07-23 2015-12-23 同济大学 Unmanned aerial vehicle aerial photography sequence image-based slope three-dimension reconstruction method
CN107655459A (en) * 2017-09-07 2018-02-02 南京理工大学 A kind of measurement of field rock texture surface roughness and computational methods

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
Accurate Camera Calibration from Multi-View Stereo and Bundle Adjustment;Yasutaka Furukawa 等;《Proceedings of the 2008 IEEE conference on Computer Vision and Pattern Recognition》;20080930;1-8
图像序列三维重建方法研究与实现;李聪;《中国优秀硕士学位论文全文数据库 信息科技辑》;20151231;第2015年卷(第12期);第9-11、36、56、67-68页,第五章
基于立体视觉的三维重建;李竞超;《中国优秀硕士学位论文全文数据库 信息科技辑》;20101031;第2010年卷(第10期);第41-48页

Also Published As

Publication number Publication date
CN106097436A (en) 2016-11-09

Similar Documents

Publication Publication Date Title
CN106097436B (en) A kind of three-dimensional rebuilding method of large scene object
CN111462329B (en) Three-dimensional reconstruction method of unmanned aerial vehicle aerial image based on deep learning
CN105844696B (en) Image position method and device based on ray model three-dimensionalreconstruction
CN110569768B (en) Construction method of face model, face recognition method, device and equipment
CN108961390A (en) Real-time three-dimensional method for reconstructing based on depth map
CN108280814B (en) Light field image angle super-resolution reconstruction method based on perception loss
CN110223383A (en) A kind of plant three-dimensional reconstruction method and system based on depth map repairing
CN101877143B (en) Three-dimensional scene reconstruction method of two-dimensional image group
CN110163953A (en) Three-dimensional facial reconstruction method, device, storage medium and electronic device
CN104539928B (en) A kind of grating stereo printing image combining method
CN105303616B (en) Embossment modeling method based on single photo
CN110211223A (en) A kind of increment type multiview three-dimensional method for reconstructing
CN103021017A (en) Three-dimensional scene rebuilding method based on GPU acceleration
CN105787989B (en) A kind of measurement material geometric properties reconstructing method based on photometric stereo vision
CN101916455B (en) Method and device for reconstructing three-dimensional model of high dynamic range texture
CN107767339B (en) Binocular stereo image splicing method
CN110633628B (en) RGB image scene three-dimensional model reconstruction method based on artificial neural network
CN104240289A (en) Three-dimensional digitalization reconstruction method and system based on single camera
CN108648264A (en) Underwater scene method for reconstructing based on exercise recovery and storage medium
CN117115359B (en) Multi-view power grid three-dimensional space data reconstruction method based on depth map fusion
CN106228603A (en) A kind of three-dimensional model reconfiguration system and method based on Euclidean distance statistics splicing
CN106683163A (en) Imaging method and system used in video monitoring
EP4455875A1 (en) Feature map generation method and apparatus, storage medium, and computer device
CN107610219A (en) The thick densification method of Pixel-level point cloud that geometry clue perceives in a kind of three-dimensional scenic reconstruct
Jancosek et al. Scalable multi-view stereo

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant