CN116403014A - Image feature matching method and system based on rough matching set - Google Patents
Image feature matching method and system based on rough matching set Download PDFInfo
- Publication number
- CN116403014A CN116403014A CN202310366384.8A CN202310366384A CN116403014A CN 116403014 A CN116403014 A CN 116403014A CN 202310366384 A CN202310366384 A CN 202310366384A CN 116403014 A CN116403014 A CN 116403014A
- Authority
- CN
- China
- Prior art keywords
- matching
- image
- algorithm
- feature
- measure
- 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
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 90
- 230000007246 mechanism Effects 0.000 claims abstract description 28
- 230000008569 process Effects 0.000 claims description 20
- 230000008859 change Effects 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 4
- 230000000007 visual effect Effects 0.000 claims description 3
- 238000010586 diagram Methods 0.000 description 7
- 238000001914 filtration Methods 0.000 description 7
- 238000004458 analytical method Methods 0.000 description 5
- 230000008901 benefit Effects 0.000 description 4
- 238000004590 computer program Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 2
- 239000004973 liquid crystal related substance Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 239000011449 brick Substances 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000008094 contradictory effect Effects 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 230000001360 synchronised effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/757—Matching configurations of points or features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/771—Feature selection, e.g. selecting representative features from a multi-dimensional feature space
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/776—Validation; Performance evaluation
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an image feature matching method and system based on a coarse matching set, which relate to the technical field of image matching and comprise the following steps: constructing a picture data set, and obtaining a rough matching set between two frames of images by obtaining a feature descriptor of the picture data set and performing violent matching; taking the measure attribute of each matching pair of the coarse matching set as a weight, constructing a first scoring mechanism, improving the GMS algorithm, and generating an improved GMS algorithm; based on the picture data set, extracting the measure attribute of the feature descriptor of the target image, and carrying out feature matching according to an improved GMS algorithm; the invention establishes a scoring mechanism based on matching measure as weight on the basis of GMS algorithm, and the mechanism can effectively inhibit the error matching score and increase the correct matching score, thereby improving the accuracy and recall rate; the improved algorithm operation efficiency is improved by about 90%, the operation time is reduced, and the application scene of the algorithm is enlarged.
Description
Technical Field
The invention relates to the technical field of image matching, in particular to an image feature matching method and system based on a rough matching set.
Background
The establishment of a correct matching relationship between image feature points is an important work in the field of computer vision, and feature matching plays a role in the vision tasks of synchronous positioning and image construction, stereo matching, image stitching and the like.
The establishment of a good feature matching set can be done in two stages. The first stage firstly utilizes brightness information around feature points to build local descriptors; and performing violent matching on the descriptors to obtain a coarse matching set between two frames of images. However, the related algorithm has high time complexity, by using the FLANN tool, the matching speed can be improved but the matching quality can be reduced, and in the matching set acquired by any method, there is a false matching condition: false positive matches, which results in the need to use a closer match measure as a screening threshold for the purpose of limiting false matches. In order to solve the problems in the first stage, the center of gravity of the current feature matching algorithm is shifted to the number of eliminating false positive matches as much as possible, so that the matching result is optimized in the second stage by selecting various matching filtering algorithms, and the existing method can be roughly divided into a geometric constraint method and a feature similarity constraint method, and a method combining the geometric constraint method and the feature similarity constraint method, and error matching in the rough matching set is filtered.
The GMS algorithm constrained by the feature similarity is excellent in real scene, and the method has the characteristic of similarity to the features according to the surrounding matching, can ensure the rejection of most of error matching under the condition of less calculation resources, and is applied to SLAM scheme. However, the existing GMS have the problems: there are a number of situations where the correct matching pair is mis-screened out. In the existing research, the score of a correct matching point is improved by carrying out selective score assignment on the characteristic point positioned at the edge of the grid so as to reduce the occurrence of false negative matching, but the acquisition of the distance from the characteristic point to the edge of the grid also needs additional operation. As can be seen from the above analysis, there are many research results about feature filtering, but there are limitations in terms of filtering accuracy or efficiency, and it is difficult to combine the two, in which, for a coarse matching set with relatively sparse numbers, the coarse matching set lacks a sufficient number of supports around the correct matching, which results in erroneous filtering, and the situation is more obvious with the decrease of the number of matching pairs in the coarse matching set, the filtering accuracy is lower, and in addition, by combining the feature filtering algorithm with the geometric model estimation algorithm, the real-time performance of program operation is greatly reduced, resulting in limited application scenarios of the algorithm; therefore, an image feature matching method and system based on a coarse matching set are urgently needed, and the method and system are used for improving the image matching speed and the feature filtering precision.
Disclosure of Invention
In order to solve the problems, the invention aims to provide an image feature matching method and system based on a coarse matching set, which utilize the advantage of the existence of near-far matching measures in the feature matching stage of a GMS algorithm, establish a scoring mechanism based on the matching measures as weights according to the measure attribute of each matching pair in the coarse matching set, further avoid the operation of a core algorithm to be executed for a plurality of times, reduce the running time and improve the running efficiency.
In order to achieve the technical purpose, the application provides an image feature matching method based on a rough matching set, which comprises the following steps:
constructing a picture data set, and obtaining a rough matching set between two frames of images by obtaining a feature descriptor of the picture data set and performing violent matching;
taking the measure attribute of each matching pair of the coarse matching set as a weight, constructing a first scoring mechanism, improving the GMS algorithm, and generating an improved GMS algorithm;
based on the picture data set, feature matching is performed according to an improved GMS algorithm by extracting the measure attribute of the feature descriptor of the target image.
Preferably, in constructing the picture data set, the picture data set includes a first image and one or more of a blur change image, a viewpoint change image, a zoom-plus-rotate image, a brightness change image, and a JPEG compressed image of the first image as a second image;
and taking the first image or the second image as a reference image of other images, and establishing a homography matrix corresponding to the reference image so as to generate a picture data set.
Preferably, in the process of improving the GMS algorithm, the process of image matching by the GMS algorithm is as follows:
dividing the first image and the second image into m×n non-overlapping grid cells;
acquiring indexes of the grids of the first image in the maximum corresponding number of grids of the second image, rotating eight surrounding grids according to the rotation condition of the images, and then taking the eight surrounding grids into a second scoring mechanism together as a whole to consider for image matching;
and judging whether the matched pair is correct or not according to the score and the threshold value.
Preferably, in the image matching process by the GMS algorithm, the image matching process is:
in the method, in the process of the invention,
representing a matching grid (a) k ，b k ) K represents the grid where a certain feature point is located and the neighborhood grid.
Preferably, in the process of judging whether the matching pair is correct according to the score and the threshold, the judging process is as follows:
wherein S is ij Representing the final score of each matching pair affected by surrounding matches, T representing a true match, F representing a false match, alpha representing a coefficient factor, n i Represents the average of the number of matches.
Preferably, in generating the modified GMS algorithm, the modified GMS algorithm is generated by replacing the second scoring mechanism of the GMS algorithm by the first scoring system, wherein the first scoring mechanism for the matched pair is generated by generating weights using the matching measure according to the relation of the matching measure and the matching accuracy.
Preferably, in the process of obtaining the first scoring system, the neighborhood score calculation method through the first scoring mechanism is as follows:
where α represents a coefficient factor, c represents a threshold at which a high probability of correct match occurs, d is a match measure for each match, m represents the number of surrounding grids, and n represents the number of matching pairs of the current grid.
Preferably, in the process of obtaining the matching measure of the target image, the matching measure of the target image is obtained through a floating point type description vector algorithm or a binary type description vector algorithm.
Preferably, the image feature matching method is applied to visual SLAM and stereoscopic matching, and is used for improving matching accuracy.
The invention provides an image feature matching system based on a rough matching set, which comprises the following steps:
the data acquisition module is used for generating a picture data set;
the rough matching set generation module is used for obtaining a rough matching set between two frames of images by obtaining a feature descriptor of the picture data set and performing violent matching;
the feature matching module is used for carrying out feature matching according to an improved GMS algorithm by extracting the measure attribute of the feature descriptor of the target image based on the picture data set, wherein the measure attribute of each matching pair of the rough matching set is taken as a weight, a scoring mechanism of the GMS algorithm is improved, and the improved GMS algorithm is generated.
The invention discloses the following technical effects:
the invention establishes a scoring mechanism based on matching measure as weight on the basis of GMS algorithm, and the scoring mechanism can effectively inhibit the error matching score and increase the correct matching score, thereby improving the accuracy and recall rate.
The improved algorithm operation efficiency is improved by about 90%, the operation time is reduced, and the application scene of the algorithm is enlarged.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are needed in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a diagram of mesh-based motion smoothness constraints according to the present invention;
FIG. 2 is a schematic diagram of a neighborhood score distribution of correct matching and incorrect matching according to the present invention;
FIG. 3 is a schematic diagram of the relationship between matching accuracy and matching measure (Hamming distance) according to the present invention;
fig. 4 is a schematic flow chart of the method according to the invention.
Detailed Description
For the purposes of making the objects, technical solutions and advantages of the embodiments of the present application more clear, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is apparent that the described embodiments are only some embodiments of the present application, but not all embodiments. The components of the embodiments of the present application, which are generally described and illustrated in the figures herein, may be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the present application, as provided in the accompanying drawings, is not intended to limit the scope of the application, as claimed, but is merely representative of selected embodiments of the application. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present application without making any inventive effort, are intended to be within the scope of the present application.
As shown in fig. 1-4, the present invention takes advantage of the near-far nature of the matching measure in the feature matching stage. Establishing a scoring mechanism based on the matching measure as a weight by utilizing the measure attribute of each matching pair in the rough matching set; in order to obtain a numerical value serving as a high-accuracy threshold, the invention performs statistics on the relation between the matching measure and the accuracy. In addition, the invention judges the matching of which the score does not exceed the threshold value but is higher than the edge threshold value, so that only a small part of matching pairs are required to be reevaluated, the operation that the core algorithm is executed for a plurality of times is avoided, the running time is shortened, and the running efficiency is improved.
1. Analysis of GMS algorithm:
the motion of the camera sensor in the actual scene is continuous, so that pixels in the neighborhood of the feature points in the image have the same motion state, and therefore, the feature points at the two ends which are correctly matched are in the respective tiny neighborhood, and a certain number of feature points have the same matching condition. As shown in fig. 1, a pair of images { I to be matched exist a ，I b The method comprises the steps of obtaining a bundle of matching sets from the pair of images through a violent matching method, wherein each pair of images respectively has { P, Q } feature points
Where χ represents the number of matching pairs. The matching of each characteristic point is independently carried out, and a matching pair x can be obtained i Neighborhood support matching number S i The probability distribution of (support) approximates to the binomial distribution as shown in expression (1).
Wherein n is the number of neighborhood matching pairs, P t When the representative matching pair is correct, matching to the probability of occurrence of the event of the corresponding region, P f When the representative matching pair is an error, the probability of occurrence of the event in the corresponding region is matched.
Equation (1) shows that both correct matching and incorrect matching have matching support in their corresponding neighborhoods, but the support degree S i As shown in fig. 2, the solid curve indicates that the feature point score is represented as a double peak as a whole, and the distinction between correct matching and incorrect matching can be achieved by selecting an appropriate threshold.
The image is rasterized for faster algorithm execution and the neighborhood of single matches is expanded to all matches of the surrounding grid area. First, image I is processed a And I b Divided into m×n non-overlapping grid cells as shown in fig. 1, and then image I is acquired a Is at I b The index of the maximum corresponding number of grids in the image is considered as a whole by rotating the surrounding eight grids according to the rotation condition of the image and then taking the rotating grids into a scoring mechanism, as shown in the formula (2).
Wherein, the liquid crystal display device comprises a liquid crystal display device,
representing a matching grid (a) k ，b k ) And finally judging whether the matching pair is correct or not according to the score and the threshold value, wherein k represents the grid where a certain feature point is located and the neighborhood grid, and the matching pair is shown in a formula (3).
Wherein S is ij Representing the final score of each matching pair affected by surrounding matches, T representing true matches, F representing false matches, alpha representing coefficientsFactor n i Represents the average of the number of matches.
2. Scoring mechanism based on matching measure:
the feature matching algorithm realizes judgment by comparing the similarity degree of feature descriptors, and the closer the matching measure between two binary descriptors is, the larger the similarity between the two binary descriptors is, and the stronger the reliability that two feature points are the same space point is. In order to establish a reasonable scoring mechanism, 5 scenes in the VGG data set are selected, and 12.5 ten thousand feature matching participation statistics are taken as shown in figure 3.
In fig. 3, each 8 units is a group, and the matching accuracy is obtained by comparing the correct matching number with the rough matching number. The matching images in the graffiti scene change too strongly, the coarse matching set does not get a match in the interval 0 to 8, but starting from 16, the matching accuracy of each scene shows a tendency to decrease with increasing hamming distance.
Through the relation between the matching measure and the precision in the figure 3, the invention utilizes the matching measure to generate the weight and establishes a scoring mechanism of the matching pair so as to reduce the occurrence of error rejection. The neighborhood score calculation mode is shown in the formula (4).
Where α represents a coefficient factor, the number of matches may be changed, default is 1, no addition or deletion is performed, c represents a threshold value at which a high probability of correct matching occurs, and the accuracy of matching may be changed, for example, it may be determined through experiments that the threshold value may be 45, d is a matching measure of each match, m represents the number of surrounding grids, and n represents the number of matching pairs of the current grid. Thus, a new support S can be obtained i As shown in equation (5).
A new probability model is built using equation (5) to increase the score of the correct match while decreasing the score of the incorrect match, as shown by the dashed line in fig. 3, to increase the mean match measure between the correct match and the incorrect match, so that whether the match is correct or not can still be distinguished in fewer coarse match sets.
3. Experiment setting:
3.1 experimental environment and data:
the experimental hardware environment of the invention: the processor is AMD Ryzen TM 75800X Desktop Processors, the memory is 32GB frequency 3600MHz (DDR 4), and GPU acceleration is not started; software environment: the system is a Windows11 professional version, the compiling component is GCC (12.2.0, minGW64 version), and the ORB and other algorithm dependency libraries are OpenCV-4.6.0.
The image data adopts VGG data set, the data set is rich in content, and contains 49 images of 8 scenes, and 2 groups of fuzzy change images (bikes and trees), viewpoint changes (graffiti and bricks), scaling and rotation (trees and boot), brightness changes (cars) and JPEG compression (ubc), wherein the first image in each scene serves as a reference of other images, and contains Homography matrix H (Homography) corresponding to the first image. Therefore, the universality and the robustness of the algorithm under different conditions can be well reflected by experiments performed on the data set.
3.2 evaluation index of experimental result:
the evaluation index includes an accuracy (P), as shown in equation 6,
recall (R), as shown in formula (7),
meanwhile, under the condition that P and R indexes are contradictory, the F value is used for comprehensive consideration, as shown in the formula 8,
wherein |χ RM |，
And->
The number of the coarse matching sets, the number of correct matching in the coarse matching sets and the number of correct matching in the sets screened by the algorithm are respectively represented.
Regarding the correct judgment of the matching, firstly, the homography matrix H is utilized to calculate the observed pixel coordinate y 'of the feature point' i As shown in the formula (9),
where ρ represents the scale factor, defaulting to 1, H is provided by the dataset, then the error L is exploited 2 The paradigm determines whether the match is correct as shown in equation (9).
e i ＝||x i -y′ i || 2 (10)
Wherein x is i Is the real pixel coordinate of the feature point, x' i Representation and x i Matched pixel coordinates, when e i A correct match is considered below a threshold, which typically takes 3 pixel units.
4. Results and analysis:
the coarse matching set is acquired by using a 0RB algorithm, and 10000, 6000 and 2000 characteristic points are respectively selected for quantitative comparison of the algorithm under the condition that 1000 characteristic points are extracted by using a characteristic matching algorithm for convenience of display. In order to verify the improvement of the algorithm in performance, the experimental result is comprehensively evaluated in terms of precision, recall rate, F value and calculation time.
Table 1 shows that the improved algorithm of the present invention has significant improvement in precision (P), recall (R) and F values over the original GMS algorithm in most scenarios. Because the matching frames in the 'body' scene are obviously scaled, the phenomenon that the feature points extracted from the image by the ORB algorithm are locally dense appears, and the grids established on the basis of the phenomenon that the number of the feature points in some grids is large and the feature points in most grids are sparse or even not appears. In addition, the core program of the original GMS algorithm can shift the grid right, up and right up, and the threshold value and the score value of the matched pair are more accurate by increasing the number of the characteristic points; however, in the algorithm, in order to reduce the running time, grid offset is abandoned, so that the matching pair only depends on the characteristic points in a certain grid scale, but the characteristic points of the reduced image are concentrated densely, and the problem that the accuracy is reduced in a high-zoom scene is finally caused because the neighborhood of the matching pair is not different from the neighborhood of the matching pair.
TABLE 1
The number of correct matches obtained by the improved algorithm in the table 2 is higher than that of the original algorithm, and the more matches are positioned at the sparse positions of the surrounding characteristic points, which is the starting point and the target of the improved algorithm. The scene of 'graffiti' is severely changed, and is trapped in that the ORB algorithm has too low scene accuracy and contains a large number of mismatching, so that the accuracy of the algorithm is not obviously improved, but the improved algorithm can be obtained through the number of the correct matching in the table 2, and the correct matching is basically extracted. Table 3 shows that the average execution time of the modified algorithm is only fifteen times longer than that of the original GMS algorithm, and the real-time performance of the system is improved.
TABLE 2
TABLE 3 Table 3
As can be seen from the result analysis in the graph, when the number of screening results of the basic coarse matching algorithm is large, compared with the original algorithm, the improved algorithm has less calculation time and no remarkable improvement, but when the number of coarse matching sets is sparse, the improved algorithm can recall more correct matching on the premise of ensuring the precision. By combining the feature matching mode and a large number of statistical result analyses, when the coarse matching set is sparse, the situation that the matching measure is closer is more likely to be reserved, and the probability that the measure is correct matching is higher, so that the invention gives a high score to the matching of the measure, and ensures that the correct matching can be reserved more easily when the surrounding lacks sufficient feature point support. The original matching algorithm greatly consumes the running time of the algorithm by shifting the neighborhood positions of the matching to improve the number of the feature points around each matching, and the original matching algorithm still has enough matching numbers around the correct matching to support the correct matching, so that the matching accuracy can be well improved while the running speed is ensured by using a scoring mechanism established by taking the matching measure as a weight.
The improved algorithm provided by the invention establishes a weight system based on the matching measure, and redesigns the scoring mechanism of the original GMS algorithm so as to solve the problem that the correct matching is incorrectly screened out due to lack of matching support in the neighborhood of the correct matching. By using the test data set to carry out test comparison, the improved algorithm improves the matching precision under the condition of reducing the running time, and obviously improves the recall rate of correct matching. Therefore, the algorithm can be well applied to the field of visual SLAM, stereo matching and other image vision, wherein for the matching measure attribute, the feature descriptor is required to be generated by using an algorithm which can generate the matching measure attribute, such as a floating point type description vector or a binary type description vector.
In short, the matching score mechanism is established by utilizing the matching measure attribute of the matching pair, the score mechanism of the original GMS algorithm is redesigned, the correct matching score is increased, and meanwhile, the error matching score is reduced, so that the algorithm keeps correct matching as much as possible in a sparse coarse matching set; meanwhile, grid offset is not needed when the improved algorithm calculates the matching score, so that the instantaneity of the algorithm is greatly improved. Compared with the original algorithm, the improved algorithm has the advantages that the matching precision and recall rate are improved, the running time is reduced by 90% compared with the original algorithm, and the improved algorithm is suitable for high-real-time scenes.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
In the description of the present invention, it should be understood that the terms "first," "second," and the like are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more such feature. In the description of the present invention, the meaning of "a plurality" is two or more, unless explicitly defined otherwise.
It will be apparent to those skilled in the art that various modifications and variations can be made to the present invention without departing from the spirit or scope of the invention. Thus, it is intended that the present invention also include such modifications and alterations insofar as they come within the scope of the appended claims or the equivalents thereof.
Claims (10)
1. The image feature matching method based on the rough matching set is characterized by comprising the following steps of:
constructing a picture data set, and obtaining a rough matching set between two frames of images by obtaining a feature descriptor of the picture data set and performing violent matching;
taking the measure attribute of each matching pair of the coarse matching set as a weight, constructing a first scoring mechanism, improving the GMS algorithm, and generating an improved GMS algorithm;
and based on the picture data set, extracting the measure attribute of the feature descriptor of the target image, and performing feature matching according to the improved GMS algorithm.
2. The image feature matching method based on the rough matching set as claimed in claim 1, wherein:
in the process of constructing a picture data set, the picture data set comprises a first image and one or more of a blur change image, a viewpoint change image, a zoom and rotation image, a brightness change image and a JPEG compressed image of the first image as a second image;
and taking the first image or the second image as a reference image of other images, and establishing a homography matrix corresponding to the reference image so as to generate the picture data set, wherein the other images are used for representing images except the first image or the second image.
3. The image feature matching method based on the rough matching set as claimed in claim 2, wherein:
in the process of improving the GMS algorithm, the process of performing image matching through the GMS algorithm is as follows:
dividing the first image and the second image into m×n non-overlapping grid cells;
acquiring indexes of the grids of the first image in the maximum corresponding number of grids of the second image, rotating eight surrounding grids according to the rotation condition of the images, and then taking the eight surrounding grids into a second scoring mechanism together as a whole to consider, and carrying out image matching;
and judging whether the matched pair is correct or not according to the score and the threshold value.
4. A method of matching image features based on a coarse matching set as claimed in claim 3, wherein:
in the process of image matching by the GMS algorithm, the image matching process is as follows:
in the method, in the process of the invention,
representing a matching grid (a) k ，b k ) K represents the grid where a certain feature point is located and the neighborhood grid.
5. The image feature matching method based on the coarse matching set as claimed in claim 4, wherein:
in the process of judging whether the matched pair is correct according to the score and the threshold value, the judging process is as follows:
wherein S is ij Representing the final score of each matching pair affected by surrounding matches, T representing a true match, F representing a false match, alpha representing a coefficient factor, n i Represents the average of the number of matches.
6. The image feature matching method based on the coarse matching set according to claim 5, wherein:
in the process of generating an improved GMS algorithm, replacing the second scoring mechanism of the GMS algorithm by the first scoring mechanism to generate the improved GMS algorithm, wherein the first scoring mechanism for the matching pair is generated by utilizing the matching measure to generate a weight according to the relation between the matching measure and the matching precision.
7. The image feature matching method based on the coarse matching set as claimed in claim 6, wherein:
in the process of acquiring the first scoring system, the neighborhood score calculation mode through the first scoring mechanism is as follows:
where α represents a coefficient factor, c represents a threshold at which a high probability of correct match occurs, d is a match measure for each match, m represents the number of surrounding grids, and n represents the number of matching pairs of the current grid.
8. The image feature matching method based on the coarse matching set as claimed in claim 7, wherein:
in the process of acquiring the matching measure of the target image, the matching measure of the target image is acquired through a floating point type description vector algorithm or a binary type description vector algorithm.
9. The image feature matching method based on the coarse matching set as claimed in claim 8, wherein:
the image feature matching method is applied to visual SLAM, and is used for improving matching accuracy.
10. An image feature matching system based on a coarse matching set, comprising:
the data acquisition module is used for generating a picture data set;
the rough matching set generation module is used for obtaining a rough matching set between two frames of images by obtaining the feature descriptors of the picture data set and performing violent matching;
a feature matching module for performing feature matching according to an improved GMS algorithm by extracting the measure attribute of the feature descriptor of the target image based on the picture dataset, wherein,
and improving a scoring mechanism of the GMS algorithm by taking the measure attribute of each matching pair of the coarse matching set as a weight, and generating the improved GMS algorithm.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310366384.8A CN116403014A (en) | 2023-04-07 | 2023-04-07 | Image feature matching method and system based on rough matching set |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310366384.8A CN116403014A (en) | 2023-04-07 | 2023-04-07 | Image feature matching method and system based on rough matching set |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116403014A true CN116403014A (en) | 2023-07-07 |
Family
ID=87013841
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310366384.8A Pending CN116403014A (en) | 2023-04-07 | 2023-04-07 | Image feature matching method and system based on rough matching set |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116403014A (en) |
-
2023
- 2023-04-07 CN CN202310366384.8A patent/CN116403014A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109859296B (en) | Training method of SMPL parameter prediction model, server and storage medium | |
| CN110490919B (en) | Monocular vision depth estimation method based on deep neural network | |
| CN107220997B (en) | Stereo matching method and system | |
| CN111369623B (en) | Lung CT image identification method based on deep learning 3D target detection | |
| US20230169677A1 (en) | Pose Estimation Method and Apparatus | |
| CN114897779A (en) | Cervical cytology image abnormal area positioning method and device based on fusion attention | |
| WO2022077863A1 (en) | Visual positioning method, and method for training related model, related apparatus, and device | |
| CN112446891A (en) | Medical image segmentation method based on U-Net network brain glioma | |
| CN110033514B (en) | Reconstruction method based on point-line characteristic rapid fusion | |
| CN110738241A (en) | binocular stereo vision matching method based on neural network and operation frame thereof | |
| CN111209811A (en) | Method and system for detecting eyeball attention position in real time | |
| CN115631112B (en) | Building contour correction method and device based on deep learning | |
| CN115601574A (en) | Unmanned aerial vehicle image matching method for improving AKAZE characteristics | |
| CN116091490A (en) | Lung nodule detection method based on YOLOv4-CA-CBAM-K-means++ -SIOU | |
| CN110992310A (en) | Method and device for determining partition where mediastinal lymph node is located | |
| CN110728692A (en) | Image edge detection method based on Scharr operator improvement | |
| CN113496260A (en) | Grain depot worker non-standard operation detection method based on improved YOLOv3 algorithm | |
| CN113344110A (en) | Fuzzy image classification method based on super-resolution reconstruction | |
| CN116403014A (en) | Image feature matching method and system based on rough matching set | |
| CN111192302A (en) | Feature matching method based on motion smoothness and RANSAC algorithm | |
| CN102496022B (en) | Effective feature point description I-BRIEF method | |
| CN111967579A (en) | Method and apparatus for performing convolution calculation on image using convolution neural network | |
| CN115063715A (en) | ORB-SLAM3 loop detection acceleration method based on gray level histogram | |
| CN110070110B (en) | Adaptive threshold image matching method | |
| CN114331919A (en) | Depth recovery method, electronic device, 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 |