CN115063460A - High-precision self-adaptive homonymous pixel interpolation and optimization method - Google Patents
High-precision self-adaptive homonymous pixel interpolation and optimization method Download PDFInfo
- Publication number
- CN115063460A CN115063460A CN202111597172.8A CN202111597172A CN115063460A CN 115063460 A CN115063460 A CN 115063460A CN 202111597172 A CN202111597172 A CN 202111597172A CN 115063460 A CN115063460 A CN 115063460A
- Authority
- CN
- China
- Prior art keywords
- homonymous
- adaptive
- neighborhood
- optimization
- matching
- 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
Images
Classifications
-
- 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
- G06T7/37—Determination of transform parameters for the alignment of images, i.e. image registration using transform domain methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- 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
- G06T2207/10032—Satellite or aerial image; Remote sensing
-
- 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
- G06T2207/20004—Adaptive image processing
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a high-precision self-adaptive homonymous pixel interpolation and optimization method, which comprises the following steps of: considering the self-adaptive mismatching elimination strategy of the characteristic space distribution; obtaining high-precision geometric transformation parameters of each local neighborhood of the image by a complementary feature neighborhood self-adaptive growth and iterative optimization algorithm, and then carrying out initial dense matching of global and local fusion constraints; and automatically detecting the hole matching area and interpolating the pixels with the same name to obtain a reliable dense matching result of the stereo image pair of the complex scene. The method fully utilizes the stereo image inversion technology and fuses the complementary homonymous feature neighborhood information of the global epipolar line geometry and the local space, and constructs pixel-by-pixel dense interpolation and dense matching strategies under tight constraint.
Description
Technical Field
The invention relates to the technical field related to digital image processing in photogrammetry and remote sensing, in particular to a high-precision self-adaptive homonymous pixel interpolation and optimization method.
Background
The homonymous pixel interpolation and optimization is a process of determining homonymous image points between images pixel by pixel or nearly pixel by pixel in a stereo image and obtaining an optimal three-dimensional point cloud in an object space according to the principle of intersection of homonymous ray pairs. The technology obtains the matching result, has the outstanding advantages of high point cloud density, low cost, stable and vivid matching result and the like, and can be widely applied to the fields of topographic map surveying and mapping, change detection, intelligent driving, virtual reality and the like. However, the current three-dimensional reconstruction technology is relatively mature, dense matching contained in a stereo image still depends on human-computer interaction for batch acquisition, a large amount of labor and material cost is consumed, and the measurement accuracy is often limited by the technical level of operators. In a word, the dense homonymous image point high-precision reliable interpolation of the stereo image is always a bottleneck problem in the technical development and practical application, and also becomes one of the important and difficult problems which need to be overcome in the fields of the remote sensing technology and the artificial intelligence at present.
In the conventional image dense interpolation method, the image primitives can be classified into three types, namely dense interpolation based on gray scale, dense interpolation based on features, dense interpolation based on phase and the like according to the adopted image primitives. Firstly, dense interpolation based on gray level can obtain very dense point clouds of the same name, but the precision and the stability of the point clouds are easily influenced by the image geometry and radiation deformation, and the size and the shape of a pixel point constraint window are difficult to reasonably determine in the interpolation process; secondly, the dense interpolation based on the characteristics can better adapt to various distortion conditions of the image, but the cost of characteristic extraction and calculation is too high, and the matching precision is easily influenced by factors such as ground object shielding, repeated texture and the like; again, dense phase-based interpolation is generally less suitable for optical stereoscopic imagery.
According to the adopted optimization theory, the image dense interpolation can be divided into local optimal dense interpolation and global optimal dense interpolation. The former has more mismatching points and lower precision; the latter is only suitable for matching of plane scenes or layered plane scenes, and greatly restricts the reliability of dense matching of complex scenes.
Disclosure of Invention
How to fully fuse the complementary homonymous feature neighborhood information of the global epipolar line geometry and the local space and construct the pixel-by-pixel dense interpolation and dense matching strategies under tight constraint becomes a key problem solved by the invention. Aiming at the problems of dense pixel connection of all typical difficult areas such as image texture shortage, parallax break, shadow and shielding, the invention provides the following technical scheme:
a high-precision self-adaptive homonymous pixel interpolation and optimization method at least comprises the following steps:
(1) considering the self-adaptive mismatching elimination strategy of the characteristic space distribution;
(2) obtaining high-precision geometric transformation parameters of each local neighborhood of the image by a complementary feature neighborhood self-adaptive growth and iterative optimization algorithm, and then carrying out initial dense matching of global and local fusion constraints;
(3) and automatically detecting the hole matching area and interpolating the pixels with the same name to obtain a reliable dense matching result of the stereo image pair of the complex scene.
Preferably, in the step (1), the epipolar constraint is utilized to eliminate the mismatching, and the spatial distribution uniformity of the homonymous features is kept as much as possible, so that the subsequent high-precision homonymous pixel matching interpolation is guided.
Preferably, the epipolar line constraint comprises the steps of:
(1a) increasing the homonymous feature weight of the matching sparse region according to a formula
Respectively calculating the weight P of each pair of homonymous features i When the condition P is satisfied i >1, then copy the current number of homonymous features as P i And added to the homonymous feature set for epipolar geometry estimation, where i e [1, K ∈]K is the number of homonymous features, and INT is rounding operation;
(1b) estimating a basic matrix, and estimating a basic matrix F based on the matching feature set obtained in the previous step;
(1c) and (4) eliminating mismatching, generating the epipolar line geometry of the stereopair based on the F, and then eliminating mismatching under the geometric constraint of homonymy epipolar lines.
Preferably, the step (2) includes a classification hierarchical neighborhood complementary growth and iterative optimization segmentation strategy, and the specific steps are as follows:
(2a) performing neighborhood growth and optimization based on the face feature covariance matrix and the neighborhood main gradient azimuth;
(2b) line feature neighborhood growth and optimization, and line segment neighborhood expansion and optimization are carried out by adopting an independent growth strategy of each side;
(2c) and (3) point feature neighborhood growth and optimization, and affine neighborhood expansion and optimization are carried out aiming at the same-name point features of the image residual region.
Preferably, step (3) comprises:
(3a) performing binarization segmentation on the matched cavity region based on the self-adaptive watershed and the image;
(3b) judging and approximating a possible shielding area by adopting a lattice tower criterion, and performing adjacent homonymy matching interpolation on a non-shielding area by adopting a self-adaptive thin plate spline function model to obtain a reliable matching point of a cavity area;
(3c) and eliminating local possibly repeated pixels with the same name.
The invention considers the self-adaptive error matching elimination strategy of the feature space distribution and can reserve the homonymous features of the low-density area to the maximum extent; a complementary feature neighborhood self-adaptive growth and iterative optimization algorithm, in particular to a classification hierarchical neighborhood complementary growth and iterative optimization segmentation strategy, obtains high-precision geometric transformation parameters of each local neighborhood of an image, and is favorable for initial dense matching of global and local fusion constraints; and automatic detection of the hole matching area and homonymous pixel interpolation guarantee to obtain a dense matching result.
In conclusion, aiming at the dense pixel connection problems of all typical difficult areas such as image texture deficiency, parallax fracture, shadow, occlusion and the like, the invention establishes a high-precision self-adaptive homonymous pixel interpolation and optimization method, which makes full use of the stereo image inversion technology and fusion of global epipolar geometry and local space complementary homonymous feature neighborhood information, constructs pixel-by-pixel dense interpolation and dense matching strategies under tight constraint, provides a feasible thought for dense matching of the difficult areas, and lays a technical foundation for the next step of novel fusion processing and analysis of multi-source data.
Drawings
FIG. 1 is a schematic flow diagram of the present invention;
FIG. 2 is a schematic diagram of neighborhood growth and optimization of homonymous surface features;
FIG. 3 shows three cases of homonym feature neighborhood growth;
FIG. 4 is a homonymous point feature triangle neighborhood affine transformation estimation;
FIG. 5 is a diagram illustrating the results of adaptive neighborhood growing and partition optimization for complementary homonymous features.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In an embodiment of the present invention, a high-precision adaptive homonymous pixel interpolation and optimization method as shown in fig. 1 is disclosed, and the specific implementation manner is divided into the following steps.
(1) And considering the self-adaptive mismatching elimination strategy of the characteristic space distribution.
Firstly, point-line-surface homonymous feature sets are integrated, a Random Sample Consensus (RANSAC) algorithm is adopted to estimate a basic matrix for a stereo image pair, a epipolar geometry (also called epipolar geometry in the field of computer vision) self-adaptive constraint strategy considering feature density is adopted, and homonymous features with low density are kept as much as possible while wrong matching is eliminated. Both theory and experiment show that the global epipolar geometry is still the most robust and efficient method for detecting the large-inclination angle stereo matching gross error at present. Based on the method, the error matching is eliminated by utilizing the epipolar constraint, and the spatial distribution uniformity of the homonymous features is kept as much as possible, so that the subsequent high-precision homonymous pixel matching interpolation is guided. The specific strategy is as follows:
s1, increasing homonymous feature weight of matching sparse region
Push type
Respectively calculating the weight P of each pair of homonymous features i (i is the same name feature number, and i is belonged to [1, K ]]K is the number of homonymous features) when the condition P is satisfied i >1, copying the current number of homonymous features to be P i And adding the weighted result into a homonymous feature set for epipolar geometric estimation so as to increase the homonymous feature weight of a matching sparse region and further improve the probability of random extraction of local sparse matching.
S2, estimating a basic matrix
And (4) estimating a basic matrix F by using a RANSAC algorithm based on the matching feature set obtained in the previous step.
S3, eliminating mismatching
Generating epipolar line geometry of the stereopair based on F, and then eliminating mismatching under geometric constraint of homonymy epipolar lines, wherein the maximum allowable gross error of homonymy characteristic epipolar line distance adopts an adaptive threshold value e T ＝(2*P i +1)/P i (unit: pixel) will help preserve the homonymous features of the sparse region.
(2) And (3) performing self-adaptive growth and iterative optimization algorithm on the complementary feature neighborhood.
After the mismatching is eliminated through the epipolar constraint, reliable complementary homonymous characteristics are obtained, and high-precision geometric transformation parameters of each local neighborhood of the image can be obtained by adopting a classification hierarchical neighborhood growing and iteration optimization strategy.
S4, surface feature neighborhood growth and optimization
Firstly, based on a surface feature covariance matrix and a neighborhood main gradient azimuth, obtaining an affine transformation relation of each homonymous feature, and then based on Normalized Cross Correlation (NCC) measurement, calculating an NCC coefficient of each homonymous feature; then, a pair of surface features with the largest NCC coefficient value is selected for neighborhood growing and optimization, as shown in FIG. 2, wherein x and x' represent barycentric coordinates of the same-name surface features, and A 1 Initial affine transformation matrix, omega, representing homonymous features F And Ω G Respective generation surface feature basis neighborhood and growthA neighborhood. The specific neighborhood growing and optimizing steps are expressed as follows:
firstly, on the basis of NCC calculation, Least Square image Matching (LSM) is carried out based on the current correlation window to obtain the value of NCC coefficient as LNCC, and a local affine transformation matrix A is updated 1 ；
Calculating window self-adaptive growth index
R LNCC ＝LNCC f o rmer /LNCC later
Wherein the LNCC former And LNCC later Respectively LNCC values before and after each increase of the correlation window. Such as R LNCC Greater than a threshold value
(the value is 0.99), continuing to execute the growth operation; otherwise, stopping growing and storing the current neighborhood window omega G ；
Executing neighborhood growing operation of delta (value is 4) pixel in step length towards one direction, and calculating R after neighborhood growing each time LNCC A value; if R is LNCC If the threshold value is not met, stopping the growth of the direction; for neighborhood omega F Sequentially carrying out the growth operation in four directions;
fourthly, repeatedly executing the first step to the third step until the R step LNCC And if the threshold condition is not met, exiting the iteration. Similarly, neighborhood growing and optimization of all the face features is completed according to the steps (note that growing is not repeatedly performed on the same-name features which completely fall into the growing region).
And S5, line feature neighborhood growth and optimization.
Considering that most line segment features are located at the junction of a multi-plane scene, the adjacent regions on two sides of the multi-plane scene have significant parallax mutation, and the description is difficult to be carried out by utilizing a uniform local affine transformation relation. Therefore, the segment neighborhood expansion and optimization are performed by adopting the independent growth strategy of each side so as to effectively avoid parallax fracture, and three conditions of automatic growth of the homonymy line neighborhood are shown in fig. 3. The method comprises the following specific steps:
selecting one side of line segment to proceed neighborhood growth, searching all the basic neighborhood range of the one sideAnd estimating an initial affine transformation matrix A by adopting a least square adjustment method 2 Then utilize A 2 Obtaining an NCC coefficient from the single-side basic neighborhood, if the NCC coefficient is larger than a given threshold (the value is 0.85), performing linear feature single-side neighborhood growth by using a reference surface feature neighborhood growth strategy, and if not, giving up the side neighborhood growth;
and secondly, independently increasing and optimizing the other side of the line segment based on the strategy of the first step. And in the same way, completing the automatic growth and optimization of all homonymous line feature neighborhoods.
And S6, point feature neighborhood growth and optimization.
The method completes the growth and optimization of a large part of image areas, and the section mainly carries out affine neighborhood expansion and optimization aiming at the homonymy point characteristics of the rest image areas, and the specific steps comprise:
firstly, constructing a Delaunay triangulation network for the homonymous points of the left image, recording indexes of vertexes of each triangle, and constructing a homonymous triangulation network corresponding to the indexes according to the homonymous points on the right image;
and secondly, finishing corresponding neighborhood affine transformation estimation on any homonymous triangle. As shown in FIG. 4, based on three pairs of coordinates of feature points with the same name, e and e ', f and f ', g and g ', an affine transformation matrix A of the corresponding triangular region can be estimated 3 Obtaining barycentric coordinates x and x' of the same-name triangle, and taking a neighborhood window W with the size of l multiplied by l by taking x as a center, wherein l can be an average value of three side lengths of the triangle; and further utilize A 3 Obtaining a homonymous neighborhood window W ' with x ' as the center, then calculating the NCC coefficient of W and W ', and obtaining the optimal local affine transformation matrix A based on LSM iteration if the NCC coefficient is greater than a given threshold (the value is 0.75) 3 And transmitting the current homonymous triangle neighborhood, otherwise, giving up the current homonymous triangle processing;
and thirdly, traversing all homonymous triangles in sequence by utilizing the strategy II, and then finishing the growth and optimization of the point feature neighborhood. Fig. 5 illustrates the complementary homonymous feature neighborhood adaptive growth and segmentation optimization results generated for a stereopair pair oriented to a complex scene with a large inclination angle based on the classification and grading strategy, wherein homonymous neighborhoods in the stereopair pair are represented by the same gray scale, different gray scale regions represent the difference of the satisfied local affine transformation relationship, and the grown neighborhoods should have good spatial distribution complementarity. Therefore, pixel-by-pixel matching can be respectively carried out under the fusion constraint of global epipolar geometry and local affine transformation, and an initial dense matching result of the stereo image pair is obtained.
(3) Matching hole region detection and homonymous pixel interpolation
There may still be a few matching hole regions in the initial dense result due to the influence of single texture, lighting shadow or occlusion, etc. For initial dense matching, firstly, based on self-adaptive Watershed (Watershed) and image binarization, dividing a matching cavity region; then, under the constraints of edge lines, epipolar line relations, parallax consistency and the like, judging and approximating a possible shielding area by adopting a lattice tower criterion (Gestalt Laws), and carrying out adjacent homonymy matching interpolation on a non-shielding area by adopting an adaptive thin plate spline function model; and finally, eliminating local possibly repeated homonymous pixels, and finally obtaining a reliable dense matching result of the stereo image pair of the complex scene. Finally, it should be noted that: although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that changes may be made in the embodiments and/or equivalents thereof without departing from the spirit and scope of the invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (5)
1. A high-precision self-adaptive homonymous pixel interpolation and optimization method is characterized by at least comprising the following steps:
(1) executing a self-adaptive mismatching elimination strategy considering the characteristic space distribution;
(2) obtaining high-precision geometric transformation parameters of each local neighborhood of the image by utilizing a complementary feature neighborhood self-adaptive growth and iterative optimization algorithm, and then carrying out initial dense matching of global and local fusion constraints;
(3) and carrying out automatic detection and homonymous pixel interpolation on the hole matching area to obtain a reliable dense matching result of the stereo image pair with the complex scene.
2. A high precision adaptive homonymous pixel interpolation and optimization method according to claim 1, wherein the step (1) utilizes epipolar line constraints to eliminate mismatching.
3. The high-precision adaptive homonymous pixel interpolation and optimization method according to claim 2, wherein the epipolar constraint in the step (1) specifically comprises the steps of:
(1a) increasing the homonymous feature weight of the matching sparse region according to a formula
Respectively calculating the weight P of each pair of homonymous features i When the condition P is satisfied i If more than 1, copying the current homonymous features to P i And added to the homonymous feature set for epipolar geometry estimation, where i e [1, K ∈]K is the number of homonymous features, and INT is rounding operation;
(1b) estimating a basic matrix, and estimating a basic matrix F based on the matching feature set obtained in the previous step;
(1c) and (4) eliminating mismatching, generating the epipolar line geometry of the stereopair based on the F, and then eliminating mismatching under the geometric constraint of the homonymy epipolar line.
4. The high-precision self-adaptive homonymous pixel interpolation and optimization method according to claim 1, wherein the step (2) comprises a classification hierarchical neighborhood complementary growing and iterative optimization segmentation strategy, and comprises the following specific steps:
(2a) performing neighborhood growth and optimization based on the face feature covariance matrix and the neighborhood main gradient azimuth;
(2b) line feature neighborhood growth and optimization, and line segment neighborhood expansion and optimization are carried out by adopting an independent growth strategy of each side;
(2c) and (3) point feature neighborhood growth and optimization, and affine neighborhood expansion and optimization are carried out aiming at the same-name point features of the image residual region.
5. A high precision adaptive homonymous pixel interpolation and optimization method according to claim 1, wherein the step (3) comprises:
(3a) performing binarization segmentation on the matched cavity region based on the self-adaptive watershed and the image;
(3b) judging and approximating a possible shielding area by adopting a lattice tower criterion, and performing adjacent homonymy matching interpolation on a non-shielding area by adopting a self-adaptive thin plate spline function model to obtain a reliable matching point of a cavity area;
(3c) and eliminating local possibly repeated pixels with the same name.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111597172.8A CN115063460A (en) | 2021-12-24 | 2021-12-24 | High-precision self-adaptive homonymous pixel interpolation and optimization method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111597172.8A CN115063460A (en) | 2021-12-24 | 2021-12-24 | High-precision self-adaptive homonymous pixel interpolation and optimization method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115063460A true CN115063460A (en) | 2022-09-16 |
Family
ID=83197207
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111597172.8A Pending CN115063460A (en) | 2021-12-24 | 2021-12-24 | High-precision self-adaptive homonymous pixel interpolation and optimization method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115063460A (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2004114063A2 (en) * | 2003-06-13 | 2004-12-29 | Georgia Tech Research Corporation | Data reconstruction using directional interpolation techniques |
| CN106558024A (en) * | 2015-09-28 | 2017-04-05 | 北京大学 | Carry out the autoregression interpolation method and device of self-adapting window expansion |
| CN110285805A (en) * | 2019-06-28 | 2019-09-27 | 北京航空航天大学 | A kind of adaptive-interpolation/division processing method of data void holes |
| CN110909303A (en) * | 2019-11-19 | 2020-03-24 | 湖南大学 | Adaptive space-time heterogeneity inverse distance interpolation method |
| CN112233019A (en) * | 2020-10-14 | 2021-01-15 | 长沙行深智能科技有限公司 | ISP color interpolation method and device based on self-adaptive Gaussian kernel |
| CN112380306A (en) * | 2020-11-11 | 2021-02-19 | 郑州大学 | Adaptive correction method for Kergin spatial interpolation considering distribution balance |
-
2021
- 2021-12-24 CN CN202111597172.8A patent/CN115063460A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2004114063A2 (en) * | 2003-06-13 | 2004-12-29 | Georgia Tech Research Corporation | Data reconstruction using directional interpolation techniques |
| CN106558024A (en) * | 2015-09-28 | 2017-04-05 | 北京大学 | Carry out the autoregression interpolation method and device of self-adapting window expansion |
| CN110285805A (en) * | 2019-06-28 | 2019-09-27 | 北京航空航天大学 | A kind of adaptive-interpolation/division processing method of data void holes |
| CN110909303A (en) * | 2019-11-19 | 2020-03-24 | 湖南大学 | Adaptive space-time heterogeneity inverse distance interpolation method |
| CN112233019A (en) * | 2020-10-14 | 2021-01-15 | 长沙行深智能科技有限公司 | ISP color interpolation method and device based on self-adaptive Gaussian kernel |
| CN112380306A (en) * | 2020-11-11 | 2021-02-19 | 郑州大学 | Adaptive correction method for Kergin spatial interpolation considering distribution balance |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104820991B (en) | A kind of multiple soft-constraint solid matching method based on cost matrix | |
| CN113034568B (en) | Machine vision depth estimation method, device and system | |
| Zhao et al. | Geometric-constrained multi-view image matching method based on semi-global optimization | |
| US20210201570A1 (en) | Method and apparatus for generating digital surface model using satellite imagery | |
| CN110060283B (en) | Multi-measure semi-global dense matching method | |
| CN112163622B (en) | Global and local fusion constrained aviation wide-baseline stereopair line segment matching method | |
| CN107103610B (en) | automatic detection method for suspicious region matched with satellite images in stereo mapping | |
| CN111457930B (en) | High-precision mapping positioning method by combining vehicle-mounted Lidar and unmanned aerial vehicle | |
| CN111881985B (en) | Stereo matching method, device, terminal and storage medium | |
| CN115953535A (en) | Three-dimensional reconstruction method and device, computing equipment and storage medium | |
| CN112435267B (en) | Disparity map calculation method for high-resolution urban satellite stereo image | |
| CN113393577B (en) | Oblique photography terrain reconstruction method | |
| CN113077552A (en) | DSM (digital communication system) generation method and device based on unmanned aerial vehicle image | |
| Patil et al. | A comparative evaluation of SGM variants (including a new variant, tMGM) for dense stereo matching | |
| Zhu et al. | Propagation strategies for stereo image matching based on the dynamic triangle constraint | |
| CN115511759A (en) | Point cloud image depth completion method based on cascade feature interaction | |
| CN114663298A (en) | Disparity map repairing method and system based on semi-supervised deep learning | |
| Zhu et al. | Triangulation of well-defined points as a constraint for reliable image matching | |
| CN117726747A (en) | Three-dimensional reconstruction method, device, storage medium and equipment for complementing weak texture scene | |
| CN115063460A (en) | High-precision self-adaptive homonymous pixel interpolation and optimization method | |
| CN115937465A (en) | Depth map fusion algorithm considering pixel region prediction | |
| CN116704112A (en) | 3D scanning system for object reconstruction | |
| CN115345897A (en) | Three-dimensional reconstruction depth map optimization method and device | |
| CN112991525B (en) | Digital surface model generation method for image space and object space mixed matching primitive | |
| CN116045920A (en) | Method for extracting DEM based on GF7 stereopair |
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 |