WO2021017699A1 - 异常点对的检测方法、图像拼接方法、装置及设备 - Google Patents

异常点对的检测方法、图像拼接方法、装置及设备 Download PDF

Info

Publication number
WO2021017699A1
WO2021017699A1 PCT/CN2020/098110 CN2020098110W WO2021017699A1 WO 2021017699 A1 WO2021017699 A1 WO 2021017699A1 CN 2020098110 W CN2020098110 W CN 2020098110W WO 2021017699 A1 WO2021017699 A1 WO 2021017699A1
Authority
WO
WIPO (PCT)
Prior art keywords
matrix
point pair
point
pair set
images
Prior art date
Application number
PCT/CN2020/098110
Other languages
English (en)
French (fr)
Inventor
陈伟
朱飞
吴腾
杜凌霄
Original Assignee
广州市百果园信息技术有限公司
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 广州市百果园信息技术有限公司 filed Critical 广州市百果园信息技术有限公司
Publication of WO2021017699A1 publication Critical patent/WO2021017699A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
    • G06T3/4038—Image mosaicing, e.g. composing plane images from plane sub-images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
    • G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • 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

Definitions

  • This application relates to the field of image processing technology, such as detection methods, image stitching methods, devices and equipment for abnormal point pairs.
  • the methods of removing matching abnormal feature points in the related art mostly eliminate the matching abnormal feature points by forming a correspondence between the points in one image and the line in the projection image corresponding to another perspective, but this method still cannot effectively eliminate the abnormal feature points. All abnormal feature point pairs will still provide wrong feature point matching information for subsequent operations in image processing, which will affect the result of image processing.
  • the embodiments of the present application provide an abnormal point pair detection method, image stitching method, device, and equipment, which accurately and effectively realizes the detection of an abnormal point pair among the matched feature point pairs between images.
  • the embodiment of the application provides a method for detecting abnormal point pairs, including:
  • the feature point pair set between the images to be detected is filtered according to the first matrix to obtain the normal point pair set between the images to be detected, and the point pairs in the first matrix and the normal point pair set satisfy the first constraint condition;
  • the normal point pair set is screened according to the second matrix to obtain the abnormal point pair set between the images to be detected, and the second matrix and the normal point pair set except for the abnormal point pair set
  • the second constraint condition is satisfied, and the constraint force of the second constraint condition is stronger than the constraint force of the first constraint condition.
  • the embodiment of the present application provides an image stitching method, including:
  • the embodiment of the present application provides an abnormal point pair detection device, including:
  • the first set determining module is configured to filter the feature point pair set between the images to be detected according to the first matrix to obtain the normal point pair set between the images to be detected, the first matrix and the normal point pair set The point pair in satisfies the first constraint condition;
  • the second set determining module is configured to filter the normal point pair set according to a second matrix to obtain an abnormal point pair set between the images to be detected, and divide the second matrix and the normal point pair set by Point pairs outside the set of abnormal point pairs satisfy a second constraint condition, and the constraint force of the second constraint condition is stronger than that of the first constraint condition.
  • An embodiment of the application provides an image splicing device, including:
  • the abnormal set determination module is provided with an abnormal point pair detection device as provided in the embodiment of the present application, and is set to obtain the abnormal point pair set between the images to be spliced according to the feature point pair set between the images to be spliced;
  • the target image determination module is configured to splice the image to be spliced according to the feature point pair set and the abnormal point pair set between the images to be spliced to obtain the target image.
  • An embodiment of the present application provides a computer device, including:
  • One or more processors are One or more processors;
  • Storage device set to store one or more programs
  • the one or more programs are executed by the one or more processors, so that the one or more processors implement the method provided in any embodiment of the present application.
  • the embodiment of the present application provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, the method provided in any embodiment of the present application is implemented.
  • FIG. 1 is a schematic flowchart of an abnormal point pair detection method provided in Embodiment 1 of this application;
  • FIG. 2 is a schematic flowchart of an abnormal point pair detection method provided by Embodiment 2 of the application;
  • Fig. 3 shows an example flowchart of determining the first candidate matrix in the second embodiment of the present application
  • FIG. 4 shows an example flow chart of determining the second candidate matrix in Embodiment 2 of the present application
  • FIG. 5 is a schematic flowchart of another abnormal point pair detection method provided in Embodiment 2 of the application.
  • FIG. 6 is a schematic flowchart of an image stitching method provided in Embodiment 3 of this application.
  • FIG. 7 shows an original example diagram before image stitching is performed in Embodiment 3 of the present application.
  • FIG. 8 shows an example diagram of the effect of detecting abnormal point pairs in the third embodiment of the present application.
  • FIG. 9 shows an effect display diagram after image stitching processing in the third embodiment of the present application.
  • FIG. 10 is a structural block diagram of an abnormal point pair detection device provided in the fourth embodiment of the application.
  • FIG. 11 is a structural block diagram of an image stitching device provided in Embodiment 5 of the application.
  • FIG. 12 is a schematic diagram of the hardware structure of a computer device according to Embodiment 6 of this application.
  • the application scenarios of the embodiments of the present application are in a variety of image processing fields related to feature point matching processing, and are suitable for detecting abnormal point pairs on a point pair set after feature point matching. Based on the method provided in the embodiments of the present application, It can effectively detect the abnormal point pairs in the feature point pair set.
  • the method can be implemented by an abnormal point pair detection device, which can be implemented by software and/or hardware, and generally can be integrated in a computer device, which can be an image processing server with image processing functions.
  • FIG. 1 is a schematic flowchart of an abnormal point pair detection method provided in Embodiment 1 of the application. As shown in FIG. 1, the method includes the following operations:
  • S110 Filter a set of feature point pairs between the images to be detected according to the first matrix to obtain a set of normal point pairs between the images to be detected.
  • the image to be detected can be understood as the processing object in image processing.
  • the feature points and the two waiting images in the two images to be detected can be obtained through feature point recognition and feature point matching algorithms.
  • a set of matching feature point pairs between the detected images. Therefore, the feature point pair set in this step can be regarded as a point pair set of matching point pairs obtained after feature point matching is performed on the feature points of the two images to be detected.
  • any point pair in the feature point pair set is composed of one feature point in two images to be detected, and the feature point in the point pair has corresponding feature point information.
  • the first matrix can be understood as a matrix for verifying whether a point pair in a feature point pair set belongs to a normally matched point pair.
  • the point pairs in the feature point pair set and the first matrix meet the first constraint condition, the point pairs can be determined as the normally matched point pairs, that is, the normal point pairs, thereby forming the normal point pair set, which can be It is considered that the first matrix and the point pairs in the normal point pair set satisfy the first constraint condition, and the first matrix can be determined by the characteristic point pair set combined with the first matrix parameter model and random sampling consistency sampling strategy.
  • the above S110 can obtain the normal point pair set that meets the first constraint condition with the first matrix.
  • the normal point pair set obtained after the above S110 processing can already be used as the matching feature points in the two images to be detected.
  • the normal point pair set obtained may include abnormal point pairs that do not actually match. This embodiment This step is used to perform matching judgment on the point pairs in the normal point pair set, and then obtain the abnormal point pair set that does not actually match from the normal point pair set.
  • the second matrix can be understood as a matrix for verifying whether a point pair in a normal point pair set belongs to a truly matched point pair.
  • the normal point pair can be determined as a true matching point pair, and the normal point pair can be deleted from the normal point pair set .
  • This step can operate on all point pairs in the normal point pair set, so that after all normal point pairs are operated, the undeleted point pairs can be determined as abnormal point pairs.
  • the second matrix and the normal point pair set except for the abnormal point pair set satisfy the second constraint condition, and the binding force of the second constraint condition is stronger than the first constraint
  • the constraints of the conditions, and in this embodiment, the second matrix can also be determined by the normal point pair set combined with the second matrix parameter model and the sampling strategy of random sampling consistency.
  • the first constraint condition can be understood as an epipolar constraint.
  • the pair of points (X 1 , X 2 ) that are normally matched in the two images to be detected it is equivalent to satisfying the first
  • the feature point X 1 has a corresponding epipolar line in the second to-be-detected image
  • P 1 is the eigenvector of the first feature point X 1
  • P 2 is the second feature point X
  • the eigenvector of 2 can be considered that the number of rows and the number of columns of the first matrix are both 3 and the rank is 2.
  • the second constraint condition can be understood as a point constraint.
  • the first matrix is a basic matrix
  • the second matrix is a homography matrix.
  • the feature point pair set between the images to be detected is first screened according to the first matrix to obtain the normal point pair set between the images to be detected, the first matrix The normal point pair set satisfies the first constraint condition, and then the normal point pair set is filtered according to the second matrix to obtain the abnormal point pair set between the images to be detected.
  • the second matrix and the normal point The point pairs in the set other than the set of abnormal point pairs satisfy the second constraint condition, and the constraint force of the second constraint condition is stronger than that of the first constraint condition.
  • this method uses multiple constraint conditions to filter the feature point set, and the constraint of different constraint conditions is gradually increased, which can effectively detect the image to be detected.
  • the set of abnormal point pairs between the images to ensure the accuracy of the matching feature point pairs between the images to be detected, thereby reducing the negative impact of abnormal points on the entire image processing, and improving the robustness of the entire image processing application.
  • Fig. 2 is a schematic flow chart of a method for detecting abnormal point pairs provided in the second embodiment of the application.
  • the second embodiment of the present application is described based on the above-mentioned first embodiment.
  • the to-be-detected The feature point pair set between the images to obtain the normal point pair set between the images to be detected may include: determining at least one first candidate matrix based on the point pairs in the feature point pair set between the images to be detected; For each first candidate matrix, determine the number of point pairs in the feature point pair set that meets the first constraint condition with each first candidate matrix; compare the corresponding points of the at least one first candidate matrix The first candidate matrix corresponding to the maximum number of point pairs is determined as the first matrix; a set of normal point pairs between the images to be detected is formed based on matching point pairs corresponding to the first matrix.
  • filtering the normal point pair set according to the second matrix to obtain the abnormal point pair set between the images to be detected includes: determining at least one second point pair set based on the point pairs in the normal point pair set Candidate matrix; for each second candidate matrix, determine the number of point pairs in the normal point pair set and each second candidate matrix that meets the second constraint condition; compare the at least one first The number of point pairs corresponding to the two candidate matrices, the second candidate matrix corresponding to the largest number of point pairs is determined as the second matrix; the matching point pairs corresponding to the second matrix are deleted from the set of normal point pairs to obtain the A collection of abnormal point pairs between test images.
  • the process of performing the first matrix screening to obtain the set of normal point pairs and the process of obtaining the set of abnormal point pairs through the second matrix screening are divided into two stages.
  • the acquisition of the normal point pair set mainly includes the first matrix
  • the determination of and corresponding to the first matrix determines the set of normal point pairs;
  • the acquisition of the set of abnormal point pairs mainly includes the determination of the second matrix and the determination of the set of abnormal point pairs corresponding to the second matrix.
  • a schematic flow chart of a method for detecting abnormal point pairs provided in the second embodiment of the present application mainly includes the following operations:
  • S210 to S240 in the following steps of this embodiment are operations for determining the first matrix and its corresponding normal point pair set;
  • S250 to S280 are operations for determining the second matrix and its corresponding abnormal point pair set.
  • S210 Determine at least one first candidate matrix based on the point pairs in the feature point pair set between the images to be detected.
  • a number of first candidate matrices are determined by combining a set of feature point pairs with a random sampling consistency method, and then a first matrix that satisfies the selection condition is selected from the first candidate matrices.
  • the first matrix is a matrix with 3 rows and columns, and a rank of 2.
  • the first matrix is restricted by the first constraint condition.
  • a first candidate matrix model is formed.
  • the unknown parameter values in the first candidate matrix model can be obtained, and a first candidate matrix can be obtained.
  • the first candidate matrix can be understood as a candidate matrix that satisfies the characteristics of the first matrix described above.
  • this step eight known point pairs used to determine the first candidate matrix can be randomly selected from the set of feature point pairs, and then the parameter values of multiple unknown parameters corresponding to the currently selected eight known point pairs can be determined.
  • a first candidate matrix corresponding to the current random selection operation is formed.
  • the first candidate matrix determined based on a random point pair selection can be regarded as the first matrix.
  • this step will perform the first candidate Cyclic determination of the matrix, that is, each time randomly select 8 point pairs that are not exactly the same as the previous selection, and re-determine the first candidate matrix, thereby performing a set number of cycles to obtain at least one determined first candidate matrix .
  • Fig. 3 shows an example flow chart of determining the first candidate matrix in the second embodiment of the present application.
  • the at least one first candidate is determined based on the point pairs in the set of feature point pairs between the images to be detected.
  • the matrix includes the following steps:
  • the value of the current cycle times can be initialized and set to 0.
  • the determined current cycle times of 0 can be obtained.
  • the first threshold may be a threshold set based on historical experience to limit the number of cycles and terminate the cycle determination operation of the first candidate matrix. In this step, when the current cycle number is less than or equal to the first threshold, the first candidate matrix determination operation can be performed, and the first candidate matrix determination operation can be ended when the current cycle number is greater than the first threshold value.
  • This step is equivalent to one of the steps of determining the first candidate matrix.
  • a first preset number of first sample point pairs can be randomly selected from the set of obtained feature point pairs.
  • the first preset number is greater than or equal to 8.
  • S2140 Based on the feature point information of the feature points in the first preset number of first sample point pairs, determine parameter values of multiple pending parameters in the first candidate matrix model to form a current first candidate matrix.
  • the first candidate matrix model is a parameter matrix model in which the number of rows and the number of columns is 3, and the rank is 2, and the parameter matrix model includes a plurality of undetermined parameters for determining the first candidate matrix.
  • the feature point information of the selected first preset number of first sample point pairs can be combined to form a set of equations containing 8 independent parameter equations.
  • Each independent parameter equation can be based on a feature point pair and the first sample point pair.
  • a candidate matrix model is formed, and finally by solving the equations, the parameter value of each undetermined parameter in the first candidate matrix model can be obtained, and the determined parameter value is substituted into the first candidate matrix model to obtain the first in the current cycle.
  • Candidate matrix is a parameter matrix model in which the number of rows and the number of columns is 3, and the rank is 2, and the parameter matrix model includes a plurality of undetermined parameters for determining the first candidate matrix.
  • the current loop count is increased by 1, and a new round of determining the first candidate matrix is returned.
  • S220 For each first candidate matrix, determine the number of point pairs in the feature point pair set that meets the first constraint condition with each first candidate matrix.
  • this embodiment can pass the points of each first candidate matrix and feature point pair set that satisfy the first constraint condition A first candidate matrix is selected as the first matrix based on the number. This step is used to determine the number of matching point pairs corresponding to each first candidate matrix.
  • the matching point pairs can be understood as the set of feature point pairs and The point pair of the first candidate matrix that meets the first constraint condition.
  • this step can be based on the mathematical expression of the first constraint condition to find the matching point pair that satisfies the mathematical expression with the first candidate matrix from the set of feature point pairs, that is, matching
  • the product of the transpose of the eigenvector of the first feature point in the point pair and the first candidate matrix is multiplied by the eigenvector of the second feature point to form a zero matrix; then this step can count the matching point pairs that satisfy the above expression
  • the number of matching point pairs corresponding to each first candidate matrix is obtained.
  • determining the number of point pairs in the feature point pair set that matches the first candidate matrix that meets the first constraint condition may include: for each feature point pair set For the first point pair, the product of the transpose of the eigenvector of the first feature point in each first point pair and the first candidate matrix is used as the first product matrix; When the product of the feature vector of the second feature point in each first point pair is a zero matrix, use each first point pair as a matching point pair corresponding to each first candidate matrix; The number of matching point pairs corresponding to each first candidate matrix in the feature point pair set.
  • the multiplication with the first candidate matrix is mainly the feature vector formed by the first feature point and the second feature point in any point pair of the feature point pair set according to the corresponding feature point information.
  • S230 Compare the number of point pairs corresponding to the at least one first candidate matrix, and determine the first candidate matrix corresponding to the largest number of point pairs as the first matrix.
  • this step can start from the above
  • the first candidate matrix corresponding to the largest number of point pairs is determined from the at least one first candidate matrix, and this is used as the first matrix.
  • S240 Form a set of normal point pairs between the images to be detected based on matching point pairs corresponding to the first matrix.
  • the matching point pairs corresponding to the first matrix can be determined.
  • the matching point pairs corresponding to the first matrix can be considered as normal matching point pairs in the feature point pair set. Therefore, the matching point pairs corresponding to the first matrix can be filtered out from the feature point pair set to form a normal point pair set.
  • the normal point pair set determined based on the above steps may include actual unmatched point pairs.
  • the abnormal point pair detection can be performed again from the normal point pair set based on the following steps.
  • S250 Determine at least one second candidate matrix based on the point pairs in the normal point pair set.
  • a random sampling consistency method is also adopted in combination with a normal point pair set to first determine a plurality of second candidate matrices, and then a second matrix that meets the selection condition is selected from the second candidate matrices.
  • the second matrix is a matrix with the number of rows and columns both 3 and the element value of the third row and third column is 1, and the second matrix is constrained by the second Conditional constraints, which can form a second candidate matrix model based on the above characteristics.
  • the unknown parameter value in the second candidate matrix model can be finally obtained, and a second candidate matrix can be obtained.
  • the second candidate matrix can be understood as a candidate matrix that satisfies the characteristics of the second matrix.
  • the four known point pairs determined as the second candidate matrix can be randomly selected from the normal point pair set, and then the parameter values corresponding to the multiple unknown parameters of the currently selected four known point pairs can be determined.
  • the second candidate matrix determined based on a random point pair selection can be regarded as the second matrix, but in practical applications, considering the lack of globality of the 4 randomly selected point pairs, this step will perform the second candidate Cyclic determination of the matrix, that is, randomly selecting 4 point pairs that are not exactly the same as the previous selection, and re-determining the second candidate matrix, thereby performing a set number of cycles to obtain at least one determined second candidate matrix .
  • Fig. 4 shows an example flow chart of determining the second candidate matrix in the second embodiment of the present application.
  • the at least one first candidate is determined based on the point pairs in the set of feature point pairs between the images to be detected
  • the matrix can include the following steps:
  • the value of the current execution times can be initialized and set as 0.
  • the current execution times with the determined value of 0 can be obtained.
  • the second threshold may be a threshold set based on historical experience for limiting the number of executions and ending the loop determination operation of the second candidate matrix. In this step, when the current execution times are less than or equal to the second threshold, the second candidate matrix determination operation can be performed, and when the current execution times are greater than the second threshold, the second candidate matrix determination operation can be ended.
  • This step is equivalent to one of the steps of determining the second candidate matrix.
  • a second preset number of second sample point pairs can be randomly selected from the obtained normal point pair set.
  • the second preset number is greater than or equal to 4.
  • S2540 Determine parameter values of multiple pending parameters in the second candidate matrix model based on the feature point information of the feature points in the at least a second preset number of second sample point pairs to form a current second candidate matrix.
  • the second candidate matrix model is a parameter matrix model in which the number of rows and the number of columns are both 3 and the element value of the third row and third column is 1, and the parameter matrix model includes the second candidate
  • this step can combine the feature point information of the selected second preset number of second sample point pairs to form an equation composed of at least 4 pairs of matching points and the second candidate matrix model
  • the parameter value of each undetermined parameter in the second candidate matrix model can be obtained, and the determined parameter value is substituted into the second candidate matrix model to obtain the second candidate matrix in the current cycle.
  • the current execution times may be increased by 1, and then a new round of determining the second candidate matrix may be returned.
  • S260 For each second candidate matrix, determine the number of point pairs in the normal point pair set that meets the second constraint condition with each second candidate matrix.
  • this embodiment can pass the points of each second candidate matrix and the normal point pair set that satisfy the second constraint condition.
  • a second candidate matrix is selected as the second matrix based on the number. This step is used to determine the number of matching point pairs corresponding to each second candidate matrix.
  • the matching point pairs can be understood as the normal point pair set and The second candidate matrix satisfies the second constraint condition.
  • a matching point pair that satisfies the mathematical expression with the second candidate matrix can be found from the set of normal point pairs, that is, matching The product of the eigenvector of the third feature point in the point pair and the second candidate matrix is equal to the eigenvector of the fourth feature point; in this step, the number of matching point pairs that satisfy the above expression can be counted, thereby obtaining multiple first The number of point pairs corresponding to the two candidate matrices.
  • the determining the number of point pairs in the normal point pair set that meets the second constraint condition with each second candidate matrix may include: for the normal point pair set For each second point pair in each second point pair, the product of the feature vector of the third feature point in each second candidate matrix and each second point pair is equal to the fourth point in each second point pair In the case of the feature vector of the feature point, each second point pair is used as the matching point pair corresponding to each second candidate matrix; the statistics of the normal point pair set corresponding to each second candidate matrix The number of matching point pairs.
  • the multiplication with the second candidate matrix is mainly the feature vector formed by the third feature point and the fourth feature point in any point pair of the normal point pair set according to the corresponding feature point information.
  • S270 Compare the number of point pairs corresponding to the at least one second candidate matrix, and determine the second candidate matrix corresponding to the largest number of point pairs as the second matrix.
  • this step can start from the above at least one first Determine the second candidate matrix corresponding to the largest number of point pairs from the two candidate matrices, and use this as the second matrix.
  • S280 Delete matching point pairs corresponding to the second matrix from the normal point pair set, and obtain an abnormal point pair set between the images to be tested.
  • the matching point pair corresponding to the second matrix can be determined.
  • the matching point pair corresponding to the second matrix can be considered as the true matching point pair in the normal point pair set. This can delete the matching point pairs corresponding to the second matrix from the normal point pair set, and the remaining pairs after the deletion can be considered as the abnormally matched point pairs of feature points.
  • the normal point pair set will be deleted from the matching point pairs corresponding to the second matrix. The latter point pair is recorded as an abnormal point pair set.
  • the second embodiment of the present application provides an abnormal point pair detection method, which explains the screening operation of the first matrix screening feature point pair collection and the second matrix screening the normal point pair collection screening operation, through the feature point pair collection based on
  • the layer-by-layer screening of multiple constraint conditions finally detects the abnormal point pairs that are abnormally matched, so as to obtain the matching point pairs that truly match between the images to be detected.
  • this method adopts multiple constraint conditions to achieve the screening of feature point sets, and the constraint strength of different constraint conditions is gradually increased, which can effectively detect the abnormal point pair sets between the images to be detected. Ensure the accuracy of the matching feature point pairs between the images to be detected, thereby reducing the negative impact of abnormal points on the entire image processing, and improving the robustness of the entire image processing application.
  • filtering the normal point pair set according to the second matrix to obtain the abnormal point pair set between the images to be detected includes: determining where the normal point pair set is The total number of point pairs containing point pairs; in the case that the total number of point pairs is greater than the set reference value, a second matrix is determined according to the normal point pair set, and the normal point pair set is filtered based on the determined second matrix to obtain Correspond to the set of intermediate point pairs of the second matrix, and use the set of intermediate point pairs as a new set of normal point pairs, and return to continue to determine the total number of point pairs included in the normal point pair set; When the total number of pairs is less than or equal to the set reference value, the point pairs included in the normal point pair set are determined to be abnormal point pairs to form an abnormal point pair set including the abnormal point pairs; the intermediate point The pair set includes the point pairs in the normal point pair set that do not satisfy the second constraint condition with the second matrix.
  • This second embodiment provides two implementations of abnormal point pair detection.
  • the one implementation of abnormal point pairs given in S210 to S280 above is equivalent to obtaining the results by one screening of the first matrix and one screening of the second matrix.
  • Set of abnormal point pairs in order to further ensure the accuracy of detection of abnormal point pairs, this embodiment 2 proposes another implementation method for detecting abnormal point pairs based on the above-mentioned embodiment, which can be regarded as a pair of feature points.
  • the set and the first matrix are screened once to obtain a normal point pair set, and then a new second matrix is determined based on the normal point pair set cycle for multiple times, and abnormal point pair detection is performed based on the new second matrix multiple times.
  • this implementation manner further enhances the detection accuracy of abnormal point pairs.
  • FIG. 5 is a schematic flowchart of another abnormal point pair detection method provided in the second embodiment of the application. As shown in FIG. 5, the abnormal point pair detection method includes the following steps:
  • This step can be implemented based on the description of S210 to S240 in the first embodiment or the second embodiment above, and will not be repeated here.
  • the following S320 to S350 give another implementation of detecting abnormal point pairs from a set of normal point pairs.
  • S320 Determine the total number of point pairs included in the normal point pair set.
  • the total number of point pairs included in the normal point pair set can be counted in this step, which is recorded as the total number of point pairs.
  • S330 Determine whether the total number of point pairs is greater than the set reference value, and execute S340 in response to the determination result that the total number of point pairs is greater than the set reference value; and respond to the determination that the total number of point pairs is less than or equal to the set reference value As a result, S350 is executed.
  • the normal point pair set still includes enough points when the total number of point pairs is greater than the set reference value. There are many feature matching points, so the detection operation of abnormal point pairs can be continued through S340.
  • the set reference value can be regarded as a lower limit threshold set based on historical experience as an end condition for looping to filter abnormal point pairs.
  • S340 Determine a second matrix using the normal point pair set, and filter the normal point pair set based on the second matrix to obtain an intermediate point pair set corresponding to the second matrix, and use the intermediate point pair set as For the new normal point pair set, return to execute S320.
  • the second matrix used When screening the normal point pair set based on the second matrix in this step, the second matrix used also needs to be determined.
  • the method of S250-S270 can be used to determine the corresponding first point based on the normal point pair set determined after each cycle.
  • Second matrix after that, the matching point pair corresponding to the second matrix in the normal point pair set can be used as the true matching point pair, and the set of point pairs remaining after the true matching point pair is deleted from the normal point pair set can be used as the intermediate point Pair collection.
  • the determined intermediate point pair set can be returned to S320 as a new normal point pair set to perform a new round of cyclic determination to determine whether to determine the second matrix and the corresponding matching point pair again.
  • S350 Determine the point pairs included in the normal point pair set as abnormal point pairs, and form an abnormal point pair set including the abnormal point pairs.
  • the method for detecting abnormal point pairs provided in the above optional embodiment adopts a method of cyclically filtering the set of normal point pairs based on the second matrix, thereby improving the strength of the constraint on matching point pairs when detecting abnormal point pairs , So as to better improve the detection accuracy of abnormal point detection, effectively ensure the robustness of image processing applications, and reduce the negative impact of abnormal points on image processing.
  • Fig. 6 is a schematic flow chart of an image stitching method provided in the third embodiment of the application.
  • the method is suitable for the case where two images are stitched together.
  • the method can be implemented by an image stitching device, which can be software and/or Hardware implementation, and generally can be integrated in a computer device, which can be a server or terminal for image processing.
  • the third embodiment proposes an application scenario of image processing.
  • the abnormal point pair detection method provided in the above-mentioned embodiment of this application can be used to detect abnormal point pairs in the image, which is similar to the method in the related art.
  • this method enhances the processing accuracy and robustness of image processing in this application scenario.
  • the image stitching method includes:
  • this step provides the implementation of abnormal point pair detection based on the set of feature point pairs between two images to be stitched.
  • this step please refer to the above-mentioned embodiment 1 and/or embodiment 2
  • the detection method of abnormal point pairs is realized, and it will not be repeated here.
  • S420 Splicing the image to be spliced according to the feature point pair set and the abnormal point pair set between the images to be spliced to obtain a target image.
  • the normal point pair set can be filtered from the feature point pair set based on step S410, and the abnormal point pair set can be detected from the normal point pair set. Therefore, it can be considered that the normal point pair set is removed from the normal point pair set.
  • the remaining point pairs after the two abnormal point pairs are equivalent to the true matching feature point pairs in the two to be spliced images.
  • the true matching feature points can be found from the two to be spliced images, and then based on the true The matched multiple feature points are stitched together for the two images to be stitched, and finally the target image is obtained.
  • FIG. 7 shows the original example image before image splicing in the third embodiment of the present application.
  • FIG. 7 mainly includes a first image 31 to be spliced and a second image 32 to be spliced.
  • Correspondingly shows multiple feature points in the feature point pair set obtained after feature point matching.
  • the figure does not show which points in the two images constitute the feature point pair, it can still be seen that the two images to be stitched in Figure 7
  • the feature points shown in has the problem of abnormal matching.
  • FIG. 8 shows an example diagram of the effect of abnormal point pair detection in the third embodiment of the present application. Compared with FIG.
  • Figure 9 shows the effect of the image stitching process in the third embodiment of the application. As shown in Figure 9, the figure forms a target image after merging the first image to be stitched and the second image to be stitched. Figure.
  • the third embodiment of the present application provides an image stitching method.
  • the feature point pairs between the images to be stitched are collected and the abnormal point pair detection method provided in the embodiments of the present application is executed to obtain the abnormal points between the images to be stitched. Pair set; then according to the feature point pair set and the abnormal point pair set between the images to be spliced splicing the image to be spliced to obtain the target image.
  • the image splicing method uses the method of abnormal point pair detection provided in the above-mentioned embodiment of the application to detect abnormal point pairs, thereby improving the detection accuracy of abnormal point pairs, and then performing image splicing based on the detected feature point pairs, thereby improving The stitching accuracy of the image stitching process.
  • the image stitching method before the collection of the feature point pairs between the images to be stitched and the method for detecting abnormal point pairs further includes: Extracting feature points in the images to be spliced; matching the feature points between the images to be spliced to obtain a set of feature point pairs between the images to be spliced.
  • a method based on edges, gray levels, or a template may be used to perform feature point extraction, or a method based on optical flow tracking or feature information matching may be used to perform feature point matching.
  • FIG. 10 is a structural block diagram of an abnormal point pair detection device provided in the fourth embodiment of the application.
  • the abnormal point pair detection device includes: a first set determination module 41 and a second set determination module 42 .
  • the first set determining module 41 is configured to filter the feature point pair sets between the images to be detected according to the first matrix to obtain the normal point pair sets between the images to be detected, and the first matrix and the normal point pair sets The point pair in satisfies the first constraint condition;
  • the second set determining module 42 is configured to filter the normal point pair set according to a second matrix to obtain an abnormal point pair set between the images to be detected, and divide the second matrix and the normal point pair set by Point pairs outside the set of abnormal point pairs satisfy a second constraint condition, and the constraint force of the second constraint condition is stronger than that of the first constraint condition.
  • the detection device for abnormal point pairs provided in the fourth embodiment of the present application, compared with the abnormal point pair detection device in the related art, adopts multiple constraint conditions to filter the feature point pair set, and the constraint strength of different constraint conditions Gradually increase, which can effectively detect the set of abnormal point pairs between the images to be detected, thereby ensuring the accuracy of the matching feature point pairs between the images to be detected, and thereby reducing the negative impact of abnormal points on the entire image processing. Improve the robustness of the entire image processing application.
  • the second set determining module 42 is configured to: determine the total number of point pairs included in the normal point pair set; when the total number of point pairs is greater than a set reference value, according to the normal The point pair set determines a second matrix, the normal point pair set is filtered based on the determined second matrix, the intermediate point pair set corresponding to the second matrix is obtained, and the intermediate point pair set is taken as a new normal point pair set , Return to continue to determine the total number of point pairs included in the normal point pair set; in the case that the total number of point pairs is less than or equal to the set reference value, the point pairs included in the normal point pair set are It is determined as an abnormal point pair to form an abnormal point pair set including the abnormal point pair; the intermediate point pair set includes a point pair in the normal point pair set that does not satisfy the second constraint condition with the second matrix.
  • the number of rows and columns of the first matrix are both 3 and the rank is 2; the number of rows and columns of the second matrix are both 3, and the element value in the third row and third column is 1.
  • the first set determining module 41 may include: a first candidate determining unit configured to determine at least one first candidate matrix based on a point pair in a set of feature point pairs between images to be detected; a first number determining unit , Set to determine, for each first candidate matrix, the number of point pairs in the feature point pair set that meets the first constraint condition with each first candidate matrix; a first matrix determination unit, Set to compare the number of point pairs corresponding to the at least one first candidate matrix, and determine the first candidate matrix corresponding to the largest number of point pairs as the first matrix; the first set determining unit is set to be based on the number of points corresponding to the first matrix The matched point pairs form a set of normal point pairs between the images to be detected.
  • the first candidate determining unit is configured to: compare the determined current cycle number with a first threshold; if the current cycle number is less than or equal to the first threshold, start from the waiting A first preset number of first sample point pairs are randomly selected from the set of feature point pairs between the detected images; based on the feature point information of the feature points in the first preset number of first sample point pairs, the first sample point pair is determined
  • the parameter values of multiple undetermined parameters in a candidate matrix model form the current first candidate matrix; add 1 to the current cycle number as the new current cycle number, and return to continue the comparison between the current cycle number and the first threshold Operation; in the case that the current number of cycles is greater than the first threshold, obtain the determined first candidate matrix.
  • the first number determining unit is configured to: transpose the feature vector of the first feature point in each first point pair in the feature point pair set to the value of each first candidate matrix.
  • the product is taken as the first product matrix; in the case where the product of the first product matrix and the eigenvector of the second feature point in each first point pair is a zero matrix, each first point pair is taken as Corresponding to the matching point pairs of each first candidate matrix; and counting the number of matching point pairs corresponding to each of the first candidate matrixes in the feature point pair set.
  • the second set determining module may include: a second candidate determining unit configured to determine at least one second candidate matrix based on the point pairs in the normal point pair set; the second number determining unit is configured to For each second candidate matrix, determine the number of point pairs in the normal point pair set that meets the second constraint condition with each second candidate matrix; the second matrix determining unit is set to compare The number of point pairs corresponding to the at least one second candidate matrix, the second candidate matrix corresponding to the largest number of point pairs is determined as the second matrix; the second set determining unit is configured to delete the corresponding The matching point pairs of the second matrix obtain the set of abnormal point pairs between the images to be tested.
  • the second candidate determination unit is configured to: compare the determined current execution times with a second threshold; in the case that the current execution times are less than or equal to the second threshold, start from the waiting At least a second preset number of second sample point pairs are randomly selected from the set of feature point pairs between the detected images; based on the feature point information of the feature points in the at least second preset number of second sample point pairs, the second The parameter values of multiple pending parameters in the candidate matrix model form the current second candidate matrix; add 1 to the current execution times as the new current execution times, and return to continue the comparison operation between the current execution times and the second threshold ; In the case that the current execution times is greater than the second threshold, obtain the determined second candidate matrix.
  • the second quantity determining unit is configured to: the product of the eigenvectors of the third feature points in each second candidate matrix and each second point pair in the normal point pair set is equal to all In the case of the feature vector of the fourth feature point in each second point pair, each second point pair is used as a matching point pair corresponding to each second candidate matrix; the normal point pairs are counted The number of matching point pairs corresponding to each second candidate matrix in the set.
  • FIG. 11 is a structural block diagram of an image splicing device provided by Embodiment 5 of the application.
  • the image splicing device is suitable for the case of performing splicing processing on two images to be spliced.
  • the device includes: an abnormal set determination module 51 and a target image determination Module 52.
  • the abnormal set determination module 51 is provided with the abnormal point pair detection device as described in the fourth embodiment of the present application, and is configured to obtain the abnormal point pair set between the images to be spliced according to the feature point pair set between the images to be spliced;
  • the target image determining module 52 is configured to splice the image to be spliced according to the feature point pair set and the abnormal point pair set between the images to be spliced to obtain the target image.
  • the image stitching device provided in the second embodiment of the present application adopts the abnormal point pair detection method provided in the above-mentioned embodiment of the present application to perform abnormal point pair detection, which improves the detection accuracy of abnormal point pairs, and then performs detection based on the detected feature point pairs Image stitching, thereby improving the stitching accuracy during image stitching processing.
  • the device may further include: a feature point extraction module configured to extract feature points in the images to be spliced separately; a feature point matching module configured to match feature points between the images to be spliced , To obtain a set of feature point pairs between the images to be spliced.
  • a feature point extraction module configured to extract feature points in the images to be spliced separately
  • a feature point matching module configured to match feature points between the images to be spliced , To obtain a set of feature point pairs between the images to be spliced.
  • FIG. 12 is a schematic diagram of the hardware structure of a computer device according to Embodiment 6 of this application.
  • the computer equipment may include: a processor 60, a storage device 61, a display screen 62, an input device 63, an output device 64, and a communication device 65.
  • the number of processors 60 in the computer device may be one or more, and one processor 60 is taken as an example in FIG. 12.
  • the number of storage devices 61 in the computer device may be one or more, and one storage device 61 is taken as an example in FIG. 12.
  • the processor 60, the storage device 61, the display screen 62, the input device 63, the output device 64, and the communication device 65 of the computer equipment may be connected by a bus or other methods. In FIG. 12, a bus connection is taken as an example.
  • the storage device 61 can be configured to store software programs, computer-executable programs, and modules, such as the program corresponding to the abnormal point pair detection method and/or the image splicing method described in any embodiment of this application Instructions/modules (for example, the first set determination module 41 and the second set determination module 42 in the abnormal point pair detection device, and so on, another example, the abnormal set determination module 51 and the target image determination module 52 in the image splicing device).
  • the storage device 61 may mainly include a storage program area and a storage data area.
  • the storage program area may store an operating device and an application program required for at least one function; the storage data area may store data created according to the use of computer equipment.
  • the storage device 61 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage devices.
  • the storage device 61 may further include a memory provided remotely with respect to the processor 60, and these remote memories may be connected to the computer device through a network. Examples of the aforementioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.
  • the display screen 62 is a display screen 62 with a touch function, which may be a capacitive screen, an electromagnetic screen or an infrared screen.
  • the display screen 62 is configured to display data according to instructions of the processor 60, and is also configured to receive touch operations on the display screen 62 and send corresponding signals to the processor 60 or other devices.
  • the display screen 62 is an infrared screen
  • the display screen 62 further includes an infrared touch frame, which is arranged around the display screen 62, and the display screen 62 may also be set to receive infrared signals and transmit the infrared signals. Send to the processor 60 or other computer equipment.
  • the input device 63 may be configured to receive input digital or character information and generate key signal inputs related to user settings and function control of the computer equipment.
  • the input device 63 may include a camera configured to obtain images and a pickup for obtaining audio in video data. Tone computer equipment.
  • the output device 64 may include video computer equipment such as a display screen and audio computer equipment such as a speaker. The composition of the input device 63 and the output device 64 can be set according to actual conditions.
  • the communication device 65 may be configured to establish a communication connection with other computer equipment, and the communication device 65 may be a wired communication device and/or a wireless communication device.
  • the processor 60 is configured to execute various functional applications and data processing of the computer equipment by running the software programs, instructions, and modules stored in the storage device 61, that is, to implement the above-mentioned abnormal point pair detection method and/or image stitching method .
  • the processor 60 when the processor 60 executes one or more programs stored in the storage device 61, it implements the following operations: filter the set of feature point pairs between the images to be detected according to the first matrix to obtain The normal point pair set of the first matrix and the normal point pair set satisfy the first constraint condition; the normal point pair set is filtered according to the second matrix to obtain the abnormality between the images to be detected A point pair set, where the second matrix and the normal point pair set except for the abnormal point pair set satisfy a second constraint condition, and the binding force of the second constraint condition is stronger than the first constraint Binding of conditions.
  • the processor 60 executes one or more programs stored in the storage device 61, the following operations are also implemented: the feature point pairs between the images to be spliced are collected to execute the abnormal point pair detection method provided by any embodiment of the present application, Obtain an abnormal point pair set between the images to be spliced; splicing the image to be spliced according to the feature point pair set and the abnormal point pair set between the images to be spliced to obtain a target image.
  • the embodiment of the present application also provides a computer-readable storage medium.
  • the computer device can execute the abnormal point pair detection method and/or as described in the above method embodiment. Or image stitching method.
  • the method for detecting abnormal point pairs includes: screening a set of characteristic point pairs between images to be detected according to a first matrix to obtain a set of normal point pairs between the images to be detected, and the first matrix The point pairs in the normal point pair set satisfy the first constraint condition; the normal point pair set is filtered according to the second matrix to obtain the abnormal point pair set between the images to be detected, and the second matrix and the normal The point pairs in the point pair set except for the abnormal point pair set satisfy a second constraint condition, and the constraint force of the second constraint condition is stronger than that of the first constraint condition.
  • the image splicing method includes: performing the abnormal point pair detection method provided by any embodiment of the present application on the feature point pair set between the images to be spliced to obtain the abnormal point pair set between the images to be spliced;
  • the feature point pair set and the abnormal point pair set between the images to be spliced are spliced on the image to be spliced to obtain the target image.
  • this application can be implemented by software and general hardware, or can be implemented by hardware.
  • the technical solution of this application can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), Random Access Memory (RAM), flash memory (FLASH), hard disk or CD-ROM, etc., including multiple instructions to make a computer device (which can be a robot, personal computer, server, or network device, etc.) execute this Apply for the method for detecting abnormal point pairs and/or the method for image stitching described in any of the embodiments.
  • the multiple units and modules included in the above-mentioned abnormal point pair detection device are only divided according to the functional logic, but are not limited to the above-mentioned division, as long as the corresponding functions can be realized; in addition, The names of multiple functional units are only for the convenience of distinguishing each other, and are not used to limit the protection scope of this application.
  • multiple parts of this application can be implemented by hardware, software, firmware, or a combination thereof.
  • multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution device.
  • a logic gate circuit for implementing logic functions on data signals Discrete logic circuits, application specific integrated circuits with suitable combinational logic gate circuits, Programmable Gate Array (PGA), Field Programmable Gate Array (FPGA), etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Databases & Information Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)

Abstract

本申请公开了异常点对的检测方法、图像拼接方法、装置及设备。该检测方法包括：根据第一矩阵筛选待检测图像之间的特征点对集合，得到待检测图像之间的正常点对集合，第一矩阵与正常点对集合中的点对满足第一约束条件；根据第二矩阵筛选正常点对集合，得到待检测图像之间的异常点对集合，第二矩阵与正常点对集合中除异常点对集合外的点对满足第二约束条件，第二约束条件的约束力强于第一约束条件的约束力。

Description

异常点对的检测方法、图像拼接方法、装置及设备
本申请要求在2019年07月30日提交中国专利局、申请号为201910696884.1的中国专利申请的优先权，该申请的全部内容通过引用结合在本申请中。
技术领域
本申请涉及图像处理技术领域，例如涉及异常点对的检测方法、图像拼接方法、装置及设备。
背景技术
在图像处理技术领域的常见应用中，如图像的拼接处理、即时定位与地图构建(Simultaneous Localization And Mapping，SLAM)以及三维重建等不同的处理场景，其技术实现一个关键环节均在于图像间特征点的匹配。
然而，相关技术中实现图像间特征点匹配的方法如基于特征点描述子匹配的方式以及基于光流追踪的方法等，都会不可避免地引入匹配异常的特征点，即特征点匹配的结果中存在错误匹配，这些异常匹配的特征点会影响整个图像处理的结果，为图像处理带来负面影响，降低图像处理应用的鲁棒性。
相关技术中去除匹配异常特征点的方法，大多通过一个图像中的点与另一视角所对应投影图像中的线形成对应关系的方式来排除匹配异常特征点，但是该种方式仍然无法有效地排除所有异常的特征点对，仍然会为图像处理中的后续操作提供错误的特征点匹配信息，影响图像处理结果。
发明内容
本申请实施例提供了异常点对的检测方法、图像拼接方法、装置及设备，准确有效地实现了图像间所匹配特征点对中异常点对的检测。
本申请实施例提供了一种异常点对的检测方法，包括：
根据第一矩阵筛选待检测图像之间的特征点对集合，得到所述待检测图像之间的正常点对集合，所述第一矩阵与所述正常点对集合中的点对满足第一约束条件；
根据第二矩阵筛选所述正常点对集合，得到所述待检测图像之间的异常点对集合，所述第二矩阵与所述正常点对集合中除所述异常点对集合外的点对满足第二约束条件，所述第二约束条件的约束力强于所述第一约束条件的约束力。
本申请实施例提供了一种图像拼接方法，包括：
将待拼接图像之间的特征点对集合执行本申请实施例提供的异常点对的检测方法，得到所述待拼接图像之间的异常点对集合；
根据所述待拼接图像之间的特征点对集合及异常点对集合拼接所述待拼接图像，获得目标图像。
本申请实施例提供了一种异常点对的检测装置，包括：
第一集合确定模块，设置为根据第一矩阵筛选待检测图像之间的特征点对集合，得到所述待检测图像之间的正常点对集合，所述第一矩阵与所述正常点对集合中的点对满足第一约束条件；
第二集合确定模块，设置为根据第二矩阵筛选所述正常点对集合，得到所述待检测图像之间的异常点对集合，所述第二矩阵与所述正常点对集合中除所述异常点对集合外的点对满足第二约束条件，所述第二约束条件的约束力强于所述第一约束条件的约束力。
本申请实施例提供一种图像拼接装置，包括：
异常集合确定模块，设置有如本申请实施例提供的异常点对的检测装置，设置为根据待拼接图像之间的特征点对集合得到所述待拼接图像之间的异常点对集合；
目标图像确定模块，设置为根据所述待拼接图像之间的特征点对集合及异常点对集合拼接所述待拼接图像，获得目标图像。
本申请实施例提供了一种计算机设备，包括：
一个或多个处理器；
存储装置，设置为存储一个或多个程序；
所述一个或多个程序被所述一个或多个处理器执行，使得所述一个或多个处理器实现本申请任意实施例提供的方法。
本申请实施例提供了一种计算机可读存储介质，其上存储有计算机程序，该程序被处理器执行时实现本申请任意实施例提供的方法。
附图说明
图1为本申请实施例一提供的一种异常点对的检测方法的流程示意图；
图2为本申请实施例二提供的一种异常点对的检测方法的流程示意图；
图3给出了本申请实施例二中确定第一候选矩阵的示例流程图；
图4给出了本申请实施例二中确定第二候选矩阵的示例流程图；
图5为本申请实施例二提供的另一种异常点对的检测方法的流程示意图；
图6为本申请实施例三提供的一种图像拼接方法的流程示意图；
图7给出了本申请实施例三进行图像拼接前的原始示例图；
图8给出了本申请实施例三进行异常点对检测后的效果示例图；
图9给出了本申请实施例三进行图像拼接处理后的效果展示图；
图10为本申请实施例四提供的一种异常点对的检测装置的结构框图；
图11为本申请实施例五提供的一种图像拼接装置的结构框图；
图12为本申请实施例六提供的一种计算机设备的硬件结构示意图。
具体实施方式
下面结合附图和实施例对本申请进行说明。此处所描述的实施例仅仅用于解释本申请，而非对本申请的限定。为了便于描述，附图中仅示出了与本申请相关的部分而非全部结构。
本申请实施例的应用场景为多种涉及到特征点匹配处理的图像处理领域中，适用于对特征点匹配后的点对集合进行异常点对检测的情况，基于本申请实施例提供的方法，能够有效地对特征点对集合中的异常点对进行准确地检测。该方法可以由异常点对的检测装置实现，该装置可以由软件和/或硬件实现，并一般可集成在计算机设备中，该计算机设备可以是具备图像处理功能的图像处理服务器。
图1为本申请实施例一提供的一种异常点对的检测方法的流程示意图，如图1所示，该方法包括如下操作：
S110、根据第一矩阵筛选待检测图像之间的特征点对集合，得到所述待检测图像之间的正常点对集合。
在本实施例中，所述待检测图像可理解为图像处理中的处理对象，在执行本步骤之前，可以通过特征点识别以及特征点匹配算法分别获得两待检测图像中的特征点以及两待检测图像之间相匹配的特征点对集合。因此，本步骤中所述特征点对集合可看作对两待检测图像的特征点进行特征点匹配后获得的相匹配点对的点对集合。示例性地，可认为所述特征点对集合中的任一点对分别由两待检测图像中的一个特征点构成，且点对中特征点具备相应的特征点信息。
在本实施例中，所述第一矩阵可理解为验证特征点对集合中的点对是否属于正常匹配的点对的矩阵。本步骤可以在特征点对集合中的点对与第一矩阵满 足第一约束条件时，将点对确定为正常匹配的点对，即正常点对，由此构成正常点对集合，由此可认为，上述第一矩阵与正常点对集合中的点对满足第一约束条件，该第一矩阵可以通过特征点对集合结合第一矩阵参数模型及随机采样一致性的采样策略确定。
S120、根据第二矩阵筛选所述正常点对集合，得到所述待检测图像之间的异常点对集合。
上述S110可以获得与第一矩阵满足第一约束条件的正常点对集合，在常规的图像处理中，已经可以将经上述S110处理后获得的正常点对集合作为两待检测图像中相匹配特征点的点对集合，考虑到仅通过上述第一矩阵及对应的第一约束条件进行筛选的约束力度相对较弱，其获得正常点对集合中可能包括实际并不匹配的异常点对，本实施例采用本步骤对正常点对集合中的点对进行匹配判定，进而从正常点对集合中获得实际并不匹配的异常点对集合。
在本实施例中，所述第二矩阵可理解为验证正常点对集合中的点对是否属于真正匹配的点对的矩阵。本步骤可以在正常点对集合中正常点对与第二矩阵满足第二约束条件时，将该正常点对确定为真正匹配的点对，并可以将该正常点对从正常点对集合中删除。本步骤可以对正常点对集合中的所有点对进行操作，由此可在对所有正常点对都进行操作后，将未删除的点对确定为异常点对。
由此可认为，所述第二矩阵与所述正常点对集合中除所述异常点对集合外的点对满足第二约束条件，该第二约束条件的约束力强于所述第一约束条件的约束力，且本实施例中也可以通过正常点对集合结合第二矩阵参数模型及随机采样一致性的采样策略来确定第二矩阵。
一实施例中，所述第一约束条件可以为：点对中第一特征点的特征向量P 1和第二特征点的特征向量P 2与所述第一矩阵H 1满足P 1 TH 1P 2＝0；所述第二约束条件可以为：点对中第三特征点的特征向量P 3和第四特征点的特征向量P 4与所述第二矩阵H 2满足P 4＝H 2P 3。
在本实施例中，所述第一约束条件可理解为对极约束，对于两待检测图像中正常匹配的点对(X 1,X 2)，相当于满足第一待检测图像中的第一特征点X 1在第二待检测图像中存在一条相对应的极线，且第二检测图像中的第二特征点X 2处于该极线上，用数学式就可表示为：P 1 TH 1P 2＝0，即P 1的转置与第一矩阵的乘积再与P 2相乘的值为0，P 1为第一特征点X 1的特征向量，P 2为第二特征点X 2的特征向量，可认为上述第一矩阵的行数和列数均为3且秩为2。
在本实施例中，所述第二约束条件可理解为对点约束，对于两待检测图像中真正匹配的点对(X 3,X 4)，相当于满足第一待检测图像中的第三特征点X 3 在第二待检测图像中存在一个映射点，且该映射点为第二待检测图像中的第四特征点X 4，用数学式就可表示为：P 4＝H 2P 3，即第二矩阵与P 3的乘积等于P 4，P 3为第三特征点X 3的特征向量，P 4为第四特征点X 4的特征向量，该第二矩阵行数和列数均为3且第3行第3列的元素值为1。可选地，第一矩阵为一个基础矩阵，第二矩阵为一个单应矩阵。
本申请实施例一提供的一种异常点对的检测方法，首先根据第一矩阵筛选待检测图像之间的特征点对集合，得到所述待检测图像之间的正常点对集合，第一矩阵与该正常点对集合中的点对满足第一约束条件，然后根据第二矩阵筛选所述正常点对集合，得到所述待检测图像之间的异常点对集合，第二矩阵与该正常点对集合中除所述异常点对集合外的点对满足第二约束条件，第二约束条件的约束力强于第一约束条件的约束力。与相关技术中的异常点对检测方案相比，该方法采用了多个约束条件对特征点对集合进行筛选，且不同约束条件的约束力度逐渐加大，由此能够有效检测出待检测图像之间的异常点对集合，从而保证待检测图像之间所匹配特征点对的准确性，进而降低异常点对对整个图像处理的负面影响，提高整个图像处理应用的鲁棒性。
实施例二
图2为本申请实施例二提供的一种异常点对的检测方法的流程示意图，本申请实施例二以上述实施例一为基础进行说明，在本实施例中，根据第一矩阵筛选待检测图像之间的特征点对集合，得到所述待检测图像之间的正常点对集合，可包括：基于待检测图像之间特征点对集合中的点对，确定至少一个第一候选矩阵；针对每个第一候选矩阵，确定所述特征点对集合中与所述每个第一候选矩阵满足所述第一约束条件的匹配点对的点对数量；比较所述至少一个第一候选矩阵对应的点对数量，将最大点对数量对应的第一候选矩阵确定为第一矩阵；基于对应所述第一矩阵的匹配点对形成所述待检测图像之间的正常点对集合。
本实施例中，根据第二矩阵筛选所述正常点对集合，得到所述待检测图像之间的异常点对集合，包括：基于所述正常点对集合中的点对，确定至少一个第二候选矩阵；针对每个第二候选矩阵，确定所述正常点对集合中与所述每个第二候选矩阵满足所述第二约束条件的匹配点对的点对数量；比较所述至少一个第二候选矩阵对应的点对数量，将最大点对数量对应的第二候选矩阵确定为第二矩阵；从所述正常点对集合中删除对应所述第二矩阵的匹配点对，获得所述待测图像之间的异常点对集合。
本实施例中进行第一矩阵筛选获得正常点对集合的过程以及第二矩阵筛选 获得异常点对集合的过程中均分为两个阶段，对于正常点对集合的获得，主要包括对第一矩阵的确定以及对应第一矩阵确定出正常点对集合；对于异常点对集合的获得，主要包括对第二矩阵的确定以及对应第二矩阵确定异常点对集合。
如图2所示，本申请实施例二提供的一种异常点对的检测方法的流程示意图，主要包括如下操作：
可认为本实施例下述步骤中的S210至S240为第一矩阵及其相应正常点对集合的确定操作；S250至S280为第二矩阵及其相应异常点对集合的确定操作。
S210、基于待检测图像之间特征点对集合中的点对，确定至少一个第一候选矩阵。
在本实施例中，主要采取随机采样一致性的方式结合特征点对集合确定出多个第一候选矩阵，然后再从第一候选矩阵中选出一个满足选定条件的第一矩阵。
基于上述实施例的描述，分析第一矩阵的特性，可知第一矩阵为一个行数和列数均为3、秩为2的矩阵，同时第一矩阵受第一约束条件的约束，由此可根据上述特性形成一个第一候选矩阵模型，该模型中存在9个未知参数值，为确定出每个未知参数值，需要选取至少8个点对作为已知信息，结合第一约束条件的约束，最终可以获得第一候选矩阵模型中的未知参数值，获得一个第一候选矩阵。该第一候选矩阵可理解为一个满足上述第一矩阵特性的候选矩阵。
本步骤中可以从特征点对集合中随机选取用于确定第一候选矩阵的8个已知点对，然后确定出对应当前所选取的8个已知点对的多个未知参数的参数值，由此形成对应当前随机选取操作的一个第一候选矩阵。理论上，基于一次随机的点对选取确定的第一候选矩阵就可以看作第一矩阵，但是在实际应用中，考虑到随机选取的8个点对缺少全局性，本步骤将进行第一候选矩阵的循环确定，即每次随机选取与之前所选取不完全相同的8个点对，重新进行第一候选矩阵的确定，由此循环执行设定次数，获得至少一个已确定的第一候选矩阵。
图3给出了本申请实施例二中确定第一候选矩阵的示例流程图，如图3所示，所述基于待检测图像之间特征点对集合中的点对，确定至少一个第一候选矩阵，包括如下步骤：
S2110、获取已确定的当前循环次数。
在本实施例中，可以初始化设定当前循环次数的值为0，本步骤初次执行时，可以获取到已确定的值为0的当前循环次数。
S2120、比较当前循环次数是否小于或等于第一阈值，响应于当前循环次数小于或等于第一阈值的比较结果，执行S2130；响应于当前循环次数大于第一阈 值的比较结果，执行S2160。
在本实施例中，所述第一阈值可以是一个基于历史经验设定的用来限定循环次数以及结束第一候选矩阵的循环确定操作的阈值。本步骤可以在当前循环次数小于或等于第一阈值时，进行第一候选矩阵的确定操作，并在当前循环次数大于第一阈值时结束第一候选矩阵的确定操作。
S2130、从所述待检测图像之间特征点对集合中随机选取第一预设数量的第一样本点对。
本步骤相当于进行第一候选矩阵确定的其中一步，可以从已获得的特征点对集合中随机选取第一预设数量的第一样本点对。可选地，所述第一预设数量大于或等于8。
S2140、基于所述第一预设数量的第一样本点对中特征点的特征点信息，确定第一候选矩阵模型中多个待定参数的参数值，形成当前的第一候选矩阵。
本实施例可认为该第一候选矩阵模型为行数和列数均为3，秩为2的参数矩阵模型，该参数矩阵模型中包含了进行第一候选矩阵确定的多个待定参数。本步骤可以结合所选定第一预设数量的第一样本点对的特征点信息，形成一组包含8个独立参数方程的方程组，每个独立参数方程可以基于一个特征点对与第一候选矩阵模型形成，最终通过求解方程组，可以获得第一候选矩阵模型中的每个待定参数的参数值，将确定的参数值代入第一候选矩阵模型，便获得了当前循环下的第一候选矩阵。
S2150、将所述当前循环次数加1作为新的当前循环次数，返回执行S2110。
在本实施例中，为保证循环的正常进行，可以在确定当前循环下的第一候选矩阵后，将当前循环次数加1，并返回进行新一轮第一候选矩阵的确定操作。
S2160、获取已确定的第一候选矩阵。
在本实施例中，在当前循环次数大于第一阈值时，可认为达到了循环结束条件，此时可获得循环过程中确定出的至少一个第一候选矩阵。
S220、针对每个第一候选矩阵，确定所述特征点对集合中与所述每个第一候选矩阵满足所述第一约束条件的匹配点对的点对数量。
在本实施例中，确定出用于选取第一矩阵的至少一个第一候选矩阵后，本实施例可以通过每个第一候选矩阵与特征点对集合中满足第一约束条件的点对的点对数量来选取一个第一候选矩阵作为第一矩阵，本步骤用于确定每个第一候选矩阵所对应的匹配点对的点对数量，所述匹配点对可理解为特征点对集合中与第一候选矩阵满足第一约束条件的点对。
针对每一个第一候选矩阵而言，本步骤可以基于第一约束条件的数学表达式，从特征点对集合中找出与该第一候选矩阵满足该数学表达式的匹配点对，即，匹配点对中第一特征点的特征向量的转置与第一候选矩阵的乘积再与第二特征点的特征向量相乘后为一个零矩阵；之后本步骤可以统计满足上述表达式的匹配点对的点对数量，由此获得每个第一候选矩阵对应的匹配点对的点对数量。
一实施例中，确定所述特征点对集合中与所述第一候选矩阵满足所述第一约束条件的匹配点对的点对数量，可以包括：针对所述特征点对集合中的每个第一点对，将所述每个第一点对中第一特征点的特征向量的转置与所述第一候选矩阵的乘积作为第一乘积矩阵；在所述第一乘积矩阵与所述每个第一点对中第二特征点的特征向量的乘积为零矩阵的情况下，将所述每个第一点对作为对应所述每个第一候选矩阵的匹配点对；统计所述特征点对集合中对应所述每个第一候选矩阵的匹配点对的点对数量。
本实施例中与第一候选矩阵进行相乘的主要为由特征点对集合的任一点对中的第一特征点以及第二特征点根据相应的特征点信息形成的特征向量。
S230、比较所述至少一个第一候选矩阵对应的点对数量，将最大点对数量对应的第一候选矩阵确定为第一矩阵。
在本实施例中，当一个第一候选矩阵对应的点对数量最大时，可认为该第一候选矩阵最能描述特征点对集合中多个点对的匹配关系，因此，本步骤可以从上述至少一个第一候选矩阵中确定出最大点对数量对应的第一候选矩阵，并将此作为第一矩阵。
S240、基于对应所述第一矩阵的匹配点对形成所述待检测图像之间的正常点对集合。
在本实施例中，进行第一矩阵确定时，可以确定出第一矩阵对应了哪些匹配点对，本步骤可认为第一矩阵对应的匹配点对为特征点对集合中正常匹配的点对，由此可从特征点对集合中筛选出第一矩阵对应的匹配点对构成正常点对集合。
基于上述步骤确定出的正常点对集合中可能包括有实际不匹配的点对，本实施例可以基于下述步骤再次从正常点对集合中进行异常点对的检测。
S250、基于所述正常点对集合中的点对，确定至少一个第二候选矩阵。
在本实施例中，同样采取随机采样一致性的方式结合正常点对集合首先来确定出多个第二候选矩阵，然后再从第二候选矩阵中选出一个满足选定条件的第二矩阵。
基于上述实施例的描述，分析第二矩阵的特性，可知第二矩阵为一个行数和列数均为3且第三行第三列的元素值1的矩阵，同时第二矩阵受第二约束条件的约束，由此可根据上述特性形成一个第二候选矩阵模型，该模型中存在8个未知参数值，为确定出每个未知参数值，需要选取至少需要4个点对作为已知信息，结合第二约束条件的约束，最终可以获得第二候选矩阵模型中的未知参数值，获得一个第二候选矩阵。所述第二候选矩阵可理解为一个满足上述第二矩阵特性的候选矩阵。
本步骤中可以从正常点对集合中随机选取作为第二候选矩阵确定的4个已知点对，然后确定出对应当前所选取的4个已知点对的多个未知参数的参数值，由此形成对应当前随机选取操作的一个第二候选矩阵。理论上，基于一次随机的点对选取确定的第二候选矩阵就可以看作第二矩阵，但是在实际应用中，考虑到随机选取的4个点对缺少全局性，本步骤将进行第二候选矩阵的循环确定，即每次随机选取与之前所选取不完全相同的4个点对，重新进行第二候选矩阵的确定，由此循环执行设定次数，获得至少一个已确定的第二候选矩阵。
图4给出了本申请实施例二中确定第二候选矩阵的示例流程图，如图4所示，所述基于待检测图像之间特征点对集合中的点对，确定至少一个第一候选矩阵，可以包括如下步骤：
S2510、获取已确定的当前执行次数。
在本实施例中，可以初始化设定当前执行次数的值为0，本步骤初次执行时，可以获取到已确定的值为0的当前执行次数。
S2520、比较所述当前执行次数是否小于或等于第二阈值，响应于所述当前执行次数小于或等于第二阈值的比较结果，执行S2530；响应于所述当前执行次数大于第二阈值的比较结果，执行S2540。
在本实施例中，所述第二阈值可以是一个基于历史经验设定的用来限定执行次数以及结束第二候选矩阵的循环确定操作的阈值。本步骤可以在当前执行次数小于或等于第二阈值时，进行第二候选矩阵的确定操作，并在当前执行次数大于第二阈值时结束第二候选矩阵的确定操作。
S2530、从所述待检测图像之间特征点对集合中随机选取至少第二预设数量的第二样本点对。
本步骤相当于进行第二候选矩阵确定的其中一步，可以从已获得的正常点对集合中随机选取第二预设数量的第二样本点对。可选地，所述第二预设数量大于或等于4。
S2540、基于所述至少第二预设数量的第二样本点对中特征点的特征点信 息，确定第二候选矩阵模型中多个待定参数的参数值，形成当前的第二候选矩阵。
本实施例可认为所述第二候选矩阵模型为一个行数和列数均为3且第三行第三列的元素值为1的参数矩阵模型，该参数矩阵模型中包含了进行第二候选矩阵确定所需的多个待定参数，本步骤可以结合所选定第二预设数量的第二样本点对的特征点信息，形成一个由至少4对匹配点与第二候选矩阵模型构成的方程组，最终通过求解该方程组，可以获得第二候选矩阵模型中的每个待定参数的参数值，将确定的参数值代入第二候选矩阵模型，便获得了当前循环下的第二候选矩阵。
S2550、将所述当前执行次数加1作为新的当前执行次数，返回S2510。
在本实施例中，为保证循环的正常进行，可以在确定当前循环下的第二候选矩阵后，将当前执行次数加1，并返回进行新一轮第二候选矩阵的确定操作。
S2560、获取已确定的第二候选矩阵。
在本实施例中，在当前执行次数大于第二阈值时，可认为达到了循环结束条件，此时可获得循环过程中确定出的至少一个第二候选矩阵。
S260、针对每个第二候选矩阵，确定所述正常点对集合中与所述每个第二候选矩阵满足所述第二约束条件的匹配点对的点对数量。
在本实施例中，确定出用于选取第二矩阵的至少一个第二候选矩阵后，本实施例可以通过每个第二候选矩阵与正常点对集合中满足第二约束条件的点对的点对数量来选取一个第二候选矩阵作为第二矩阵，本步骤用于确定每个第二候选矩阵所对应的匹配点对的点对数量，所述匹配点对可理解为正常点对集合中与第二候选矩阵满足第二约束条件的点对。
针对每一个第二候选矩阵而言，本步骤可以基于第二约束条件的数学表达式，从正常点对集合中找出与该第二候选矩阵满足该数学表达式的匹配点对，即，匹配点对中第三特征点的特征向量与第二候选矩阵的乘积等于第四特征点的特征向量；之后本步骤可以统计满足上述表达式的匹配点对的点对数量，由此获得多个第二候选矩阵对应的点对数量。
一实施例中，所述确定所述正常点对集合中与所述每个第二候选矩阵满足所述第二约束条件的匹配点对的点对数量，可包括：针对所述正常点对集合中的每个第二点对，在所述每个第二候选矩阵与所述每个第二点对中第三特征点的特征向量的乘积等于所述每个第二点对中的第四特征点的特征向量的情况下，将所述每个第二点对作为对应所述每个第二候选矩阵的匹配点对；统计所述正常点对集合中对应所述每个第二候选矩阵的匹配点对的点对数量。
本实施例中与第二候选矩阵进行相乘的主要为由正常点对集合的任一点对中的第三特征点以及第四特征点根据相应的特征点信息形成的特征向量。
S270、比较所述至少一个第二候选矩阵对应的点对数量，将最大点对数量对应的第二候选矩阵确定为第二矩阵。
在本实施例中，当一个第二候选矩阵对应的点对数量最大时，可认为该第二候选矩阵最能够描述正常点对集合中点对的匹配关系，因此本步骤可以从上述至少一个第二候选矩阵中确定出最大点对数量对应的第二候选矩阵，并将此作为第二矩阵。
S280、从所述正常点对集合中删除对应所述第二矩阵的匹配点对，获得所述待测图像之间的异常点对集合。
在本实施例中，进行第二矩阵确定时，可以确定出第二矩阵对应的匹配点对，本步骤可认为第二矩阵对应的匹配点对为正常点对集合中真正匹配的点对，由此可从正常点对集合中删除第二矩阵对应的匹配点对，删除后余下的则可认为是特征点异常匹配的点对，本步骤将正常点对集合删除第二矩阵对应的匹配点对后的点对记为异常点对集合。
本申请实施例二提供的一种异常点对的检测方法，对第一矩阵筛选特征点对集合的筛选操作以及第二矩阵筛选正常点对集合的筛选操作进行说明，通过对特征点对集合基于多个约束条件的层层筛选，最终检测出了异常匹配的异常点对，从而获得了待检测图像之间真正匹配的匹配点对。与相关技术相比，该方法采用了多个约束条件实现特征点集合的筛选，且不同约束条件的约束力度逐渐加大，由此能够有效检测出待检测图像之间的异常点对集合，从而保证待检测图像之间所匹配特征点对的准确性，进而降低异常点对对整个图像处理的负面影响，提高整个图像处理应用的鲁棒性。
在本申请实施例二的一个可选实施例中，根据第二矩阵筛选所述正常点对集合，得到所述待检测图像之间的异常点对集合，包括：确定所述正常点对集合所包含点对的点对总数；在所述点对总数大于设定基准值的情况下，根据所述正常点对集合确定第二矩阵，基于确定的第二矩阵筛选所述正常点对集合，获得对应所述第二矩阵的中间点对集合，并将所述中间点对集合作为新的正常点对集合，返回继续确定所述正常点对集合所包含点对的点对总数；在所述点对总数小于或等于所述设定基准值的情况下，将所述正常点对集合中包含的点对确定为异常点对，形成包含所述异常点对的异常点对集合；所述中间点对集合包含了所述正常点对集合中与所述第二矩阵不满足第二约束条件的点对。
本实施例二提供了异常点对检测的两种实现方式，上述S210至S280给出 的异常点对的一种实现方式，相当于通过第一矩阵的一次筛选以及第二矩阵的一次筛选获得了异常点对集合，为了进一步保证异常点对检测的精准度，本实施例二在上述实施例的基础上，提出了另一种异常点对检测的实现方式，该实现方式可看作对特征点对集合以及第一矩阵进行一次筛选获得一个正常点对集合，后续则采用了多次基于正常点对集合循环确定新的第二矩阵，并多次基于新的第二矩阵进行异常点对检测的方案，该种实现方式，相比于本实施例二提供的前一种实现方式，进一步增强了异常点对的检测精度。
图5为本申请实施例二提供的另一种异常点对的检测方法的流程示意图，如图5所示，该异常点对的检测方法包括如下步骤：
S310、根据第一矩阵筛选待检测图像之间的特征点对集合，得到所述待检测图像之间的正常点对集合。
本步骤可以基于上述实施例一或实施例二中S210至S240的描述来实现，这里不再赘述。下述S320至S350给出了从正常点对集合检测异常点对的另一种实现。
S320、确定所述正常点对集合所包含点对的点对总数。
在确定出正常点对集合后，本步骤可以统计出正常点对集合中包含的点对的总数量，记为点对总数。
S330、判断所述点对总数是否大于设定基准值，响应于所述点对总数大于设定基准值的判断结果，执行S340；响应于所述点对总数小于或等于设定基准值的判断结果，执行S350。
本实施例考虑根据正常点对集合中点对的点对总数来确定是否继续基于第二矩阵进行点对筛选，可以在点对总数大于设定基准值时，认为正常点对集合中还包括足够多的特征匹配点，由此可以继续通过S340执行异常点对的检测操作，在点对总数小于或等于设定基准值时，可认为正常点对集合中的点对少于一定数量，不适合再次筛选异常点对，可以跳转至S350，所述设定基准值可看做基于历史经验设定的下限阈值，以作为循环筛选异常点对的结束条件。
S340、采用所述正常点对集合确定第二矩阵，并基于所述第二矩阵筛选所述正常点对集合，获得对应所述第二矩阵的中间点对集合，将所述中间点对集合作为新的正常点对集合，返回执行S320。
本步骤基于第二矩阵进行正常点对集合的筛选时，同样需要确定所采用的第二矩阵，这里可以采用上述S250-S270的方法基于每次循环后确定的正常点对集合来确定对应的第二矩阵，之后，可以将正常点对集合中第二矩阵对应的匹配点对作为真正匹配的点对，并可将正常点对集合中删除真正匹配点对后余 下的点对的集合作为中间点对集合。
之后可将确定的中间点对集合作为新的正常点对集合返回S320执行新一轮循环判定，以确定是否再次进行第二矩阵的确定及相应匹配点对的确定。
S350、将所述正常点对集合中包含的点对确定为异常点对，形成包含所述异常点对的异常点对集合。
在本实施例中，当正常点对集合中的点对总数小于设定基准值时，可认为当前的正常点对集合中包含的点对就是异常点对。
本实施例二上述可选实施例提供的异常点对的检测方法，采用了循环基于第二矩阵筛选正常点对集合的方式，由此提高了异常点对检测时的对匹配点对的约束力度，从而更好地提高了异常点对检测的检测精度，有效保证了图像处理应用的鲁棒性，降低了异常点对对图像处理的负面影响。
实施例三
图6为本申请实施例三提供的一种图像拼接方法的流程示意图，该方法适用于对两张图像进行拼接处理的情况，该方法可以由图像拼接装置实现，该装置可以由软件和/或硬件实现，并一般可集成在计算机设备中，该计算机设备可以是用于图像处理的服务器或终端。
本实施例三提出了一种图像处理的应用场景，该应用场景下可以先采用本申请上述实施例提供的异常点对检测方法对图像中的异常点对进行检测，与相关技术中的方式相比，该方法增强了该应用场景下图像处理的处理精度及鲁棒性。
如图6所示，该图像拼接方法包括：
S410、将待拼接图像之间的特征点对集合执行本申请任意实施例提供的异常点对的检测方法，得到所述待拼接图像之间的异常点对集合。
在本实施例中，本步骤给出了基于两个待拼接图像之间的特征点对集合，进行异常点对检测的实现操作，本步骤可以参考上述实施例一和/或实施例二提供的异常点对的检测方法来实现，这里不再赘述。
S420、根据所述待拼接图像之间的特征点对集合及异常点对集合拼接所述待拼接图像，获得目标图像。
在本实施例中，可以基于步骤S410从特征点对集合中筛选出正常点对集合，并从正常点对集合中检测出异常点对集合，由此，可认为从正常点对集合中除去多个异常点对后余下的点对相当于两待拼接图像中真正匹配的特征点对，本 步骤可以从两待拼接图像中找出真正匹配的多个特征点，然后基于两待拼接图像中真正匹配的多个特征点进行两待拼接图像的拼接处理，最终获得目标图像。
示例性地，图7给出了本申请实施例三进行图像拼接前的原始示例图，如图7所示，主要包括了第一待拼接图像31和第二待拼接图像32，两待拼接图像中相应展示了进行特征点匹配后所获得的特征点对集合中的多个特征点，图中尽管未示出两图像中哪些点构成特征点对，但还是可以看出图7两待拼接图像中所展示的特征点存在匹配异常的问题。图8给出了本申请实施例三进行异常点对检测后的效果示例图，与图7相比，可以看出图8中第一待拼接图像31和第二待拼接图像32之间分别展示的特征点呈现出真正匹配的效果。图9给出了本申请实施例三进行图像拼接处理后的效果展示图，如图9所示，图中形成了一张将第一待拼接图像和第二待拼接图像合并后的目标图像效果图。
本申请实施例三提供的一种图像拼接方法，首先将待拼接图像之间的特征点对集合执行本申请实施例提供的异常点对的检测方法，得到所述待拼接图像之间的异常点对集合；然后根据所述待拼接图像之间的特征点对集合及异常点对集合拼接所述待拼接图像，获得目标图像。该图像拼接方法采用了本申请上述实施例提供的异常点对检测的方法进行异常点对检测，提高了异常点对的检测精度，进而基于检测后的特征点对进行图像拼接，由此提高了图像拼接处理时的拼接精度。
在本申请实施例三的一个可选实施例中，该图像拼接方法在将待拼接图像之间的特征点对集合执行本申请任意实施例提供的异常点对的检测方法之前，还包括：分别提取所述待拼接图像中的特征点；将所述待拼接图像之间的特征点进行匹配，得到所述待拼接图像之间的特征点对集合。
在本实施例中，图像拼接之前，需要对两待拼接图像进行特征点的提取处理，以及特征点的匹配处理，由此获得初始的特征点对集合。示例性地，可以采用基于边缘、灰度或者模板的方式来进行特征点提取，也可以基于光流追踪或特征信息匹配的方式来进行特征点匹配。
对于方法实施例，为了简单描述，将其都表述为一系列的动作组合，但是本领域技术人员应该知悉，本申请实施例并不受所描述的动作顺序的限制，因为依据本申请实施例，某些步骤可以采用其他顺序或者同时进行。
实施例四
图10为本申请实施例四提供的一种异常点对的检测装置的结构框图，如图10所示，该异常点对的检测装置包括：第一集合确定模块41和第二集合确定模 块42。
第一集合确定模块41设置为根据第一矩阵筛选待检测图像之间的特征点对集合，得到所述待检测图像之间的正常点对集合，所述第一矩阵与所述正常点对集合中的点对满足第一约束条件；
第二集合确定模块42设置为根据第二矩阵筛选所述正常点对集合，得到所述待检测图像之间的异常点对集合，所述第二矩阵与所述正常点对集合中除所述异常点对集合外的点对满足第二约束条件，所述第二约束条件的约束力强于所述第一约束条件的约束力。
本申请实施例四提供的一种异常点对的检测装置，与相关技术中的异常点对检测装置相比，采用了多个约束条件对特征点对集合进行筛选，且不同约束条件的约束力度逐渐加大，由此能够有效检测出待检测图像之间的异常点对集合，从而保证待检测图像之间所匹配特征点对的准确性，进而降低异常点对对整个图像处理的负面影响，提高整个图像处理应用的鲁棒性。
一实施例中，第二集合确定模块42是设置为：确定所述正常点对集合所包含点对的点对总数；在所述点对总数大于设定基准值的情况下，根据所述正常点对集合确定第二矩阵，基于确定的第二矩阵筛选所述正常点对集合，获得对应所述第二矩阵的中间点对集合，并将所述中间点对集合作为新的正常点对集合，返回继续确定所述正常点对集合所包含点对的点对总数；在所述点对总数小于或等于所述设定基准值的情况下，将所述正常点对集合中包含的点对确定为异常点对，形成包含所述异常点对的异常点对集合；所述中间点对集合包含了所述正常点对集合中与所述第二矩阵不满足第二约束条件的点对。
一实施例中，所述第一约束条件可以为：点对中第一特征点的特征向量P 1和第二特征点的特征向量P 2与所述第一矩阵H 1满足P 1 TH 1P 2＝0；所述第二约束条件可以为：点对中第三特征点的特征向量P 3和第四特征点的特征向量P 4与所述第二矩阵H 2满足P 4＝H 2P 3。
一实施例中，所述第一矩阵的行数和列数均为3且秩为2；所述第二矩阵行数和列数均为3且第3行第3列的元素值为1。
一实施例中，第一集合确定模块41可以包括：第一候选确定单元，设置为基于待检测图像之间特征点对集合中的点对，确定至少一个第一候选矩阵；第一数量确定单元，设置为针对每个第一候选矩阵，确定所述特征点对集合中与所述每个第一候选矩阵满足所述第一约束条件的匹配点对的点对数量；第一矩阵确定单元，设置为比较所述至少一个第一候选矩阵对应的点对数量，将最大点对数量对应的第一候选矩阵确定为第一矩阵；第一集合确定单元，设置为基 于对应所述第一矩阵的匹配点对形成所述待检测图像之间的正常点对集合。
一实施例中，第一候选确定单元是设置为：将已确定的当前循环次数与第一阈值进行比较；在所述当前循环次数小于或等于所述第一阈值的情况下，从所述待检测图像之间的特征点对集合中随机选取第一预设数量的第一样本点对；基于所述第一预设数量的第一样本点对中特征点的特征点信息，确定第一候选矩阵模型中多个待定参数的参数值，形成当前的第一候选矩阵；将所述当前循环次数加1作为新的当前循环次数，返回继续进行当前循环次数与所述第一阈值的比较操作；在所述当前循环次数大于所述第一阈值的情况下，获取已确定的第一候选矩阵。
一实施例中，第一数量确定单元是设置为：将所述特征点对集合中的每个第一点对中第一特征点的特征向量的转置与所述每个第一候选矩阵的乘积作为第一乘积矩阵；在所述第一乘积矩阵与所述每个第一点对中第二特征点的特征向量的乘积为零矩阵的情况下，将所述每个第一点对作为对应所述每个第一候选矩阵的匹配点对；统计所述特征点对集合中对应所述每个第一候选矩阵的匹配点对的点对数量。
一实施例中，第二集合确定模块可以包括：第二候选确定单元，设置为基于所述正常点对集合中的点对，确定至少一个第二候选矩阵；第二数量确定单元，设置为针对每个第二候选矩阵，确定所述正常点对集合中与所述每个第二候选矩阵满足所述第二约束条件的匹配点对的点对数量；第二矩阵确定单元，设置为比较所述至少一个第二候选矩阵对应的点对数量，将最大点对数量对应的第二候选矩阵确定为第二矩阵；第二集合确定单元，设置为从所述正常点对集合中删除对应所述第二矩阵的匹配点对，获得所述待测图像之间的异常点对集合。
一实施例中，第二候选确定单元是设置为：将已确定的当前执行次数与第二阈值进行比较；在所述当前执行次数小于或等于所述第二阈值的情况下，从所述待检测图像之间特征点对集合中随机选取至少第二预设数量的第二样本点对；基于所述至少第二预设数量的第二样本点对中特征点的特征点信息，确定第二候选矩阵模型中多个待定参数的参数值，形成当前的第二候选矩阵；将所述当前执行次数加1作为新的当前执行次数，返回继续进行当前执行次数与所述第二阈值的比较操作；在所述当前执行次数大于所述第二阈值的情况下，获取已确定的第二候选矩阵。
一实施例中，第二数量确定单元是设置为：在所述每个第二候选矩阵与所述正常点对集合中的每个第二点对中第三特征点的特征向量的乘积等于所述每个第二点对中的第四特征点的特征向量的情况下，将所述每个第二点对作为对 应所述每个第二候选矩阵的匹配点对；统计所述正常点对集合中对应所述每个第二候选矩阵的匹配点对的点对数量。
实施例五
图11为本申请实施例五提供的一种图像拼接装置的结构框图，该图像拼接装置适用于对两个待拼接图像进行拼接处理的情况，该装置包括：异常集合确定模块51和目标图像确定模块52。
异常集合确定模块51设置有如本申请实施例四所述的异常点对的检测装置，设置为根据待拼接图像之间的特征点对集合得到所述待拼接图像之间的异常点对集合；
目标图像确定模块52设置为根据所述待拼接图像之间的特征点对集合及异常点对集合拼接所述待拼接图像，获得目标图像。
本申请实施例二提供的图像拼接装置，采用了本申请上述实施例提供的异常点对检测的方法进行异常点对检测，提高了异常点对的检测精度，进而基于检测后的特征点对进行图像拼接，由此提高了图像拼接处理时的拼接精度。
一实施例中，该装置还可包括：特征点提取模块，设置为分别提取所述待拼接图像中的特征点；特征点匹配模块，设置为将所述待拼接图像之间的特征点进行匹配，得到所述待拼接图像之间的特征点对集合。
实施例六
图12为本申请实施例六提供的一种计算机设备的硬件结构示意图。如图12所示，该计算机设备可以包括：处理器60、存储装置61、显示屏62、输入装置63、输出装置64以及通信装置65。该计算机设备中处理器60的数量可以是一个或者多个，图12中以一个处理器60为例。该计算机设备中存储装置61的数量可以是一个或者多个，图12中以一个存储装置61为例。该计算机设备的处理器60、存储装置61、显示屏62、输入装置63、输出装置64以及通信装置65可以通过总线或者其他方式连接，图12中以通过总线连接为例。
存储装置61作为一种计算机可读存储介质，可设置为存储软件程序、计算机可执行程序以及模块，如本申请任意实施例所述的异常点对的检测方法和/或图像拼接方法对应的程序指令/模块(例如，异常点对的检测装置中的第一集合确定模块41和第二集合确定模块42等，又如，图像拼接装置中的异常集合确定模块51和目标图像确定模块52)。存储装置61可主要包括存储程序区和存 储数据区，存储程序区可存储操作装置、至少一个功能所需的应用程序；存储数据区可存储根据计算机设备的使用所创建的数据等。此外，存储装置61可以包括高速随机存取存储器，还可以包括非易失性存储器，例如至少一个磁盘存储器件、闪存器件、或其他非易失性固态存储器件。在一些实例中，存储装置61可进一步包括相对于处理器60远程设置的存储器，这些远程存储器可以通过网络连接至计算机设备。上述网络的实例包括但不限于互联网、企业内部网、局域网、移动通信网及其组合。
显示屏62为具有触摸功能的显示屏62，其可以是电容屏、电磁屏或者红外屏。一般而言，显示屏62设置为根据处理器60的指示显示数据，还设置为接收作用于显示屏62的触摸操作，并将相应的信号发送至处理器60或其他装置。可选的，当显示屏62为红外屏时，显示屏62还包括红外触摸框，该红外触摸框设置在显示屏62的四周，显示屏62还可以设置为接收红外信号，并将该红外信号发送至处理器60或者其他计算机设备。
输入装置63可设置为接收输入的数字或者字符信息，以及产生与计算机设备的用户设置以及功能控制有关的键信号输入，输入装置63可以包括设置为获取图像的摄像头以及获取视频数据中音频的拾音计算机设备。输出装置64可以包括显示屏等视频计算机设备以及扬声器等音频计算机设备。输入装置63和输出装置64的组成可以根据实际情况设定。
通信装置65可设置为与其他计算机设备建立通信连接，通信装置65可以是有线通信装置和/或无线通信装置。
处理器60设置为通过运行存储在存储装置61中的软件程序、指令以及模块，从而执行计算机设备的多种功能应用以及数据处理，即实现上述的异常点对的检测方法和/或图像拼接方法。
本实施例中，处理器60执行存储装置61中存储的一个或多个程序时，实现如下操作：根据第一矩阵筛选待检测图像之间的特征点对集合，得到所述待检测图像之间的正常点对集合，所述第一矩阵与所述正常点对集合中的点对满足第一约束条件；根据第二矩阵筛选所述正常点对集合，得到所述待检测图像之间的异常点对集合，所述第二矩阵与所述正常点对集合中除所述异常点对集合外的点对满足第二约束条件，所述第二约束条件的约束力强于所述第一约束条件的约束力。
此外，处理器60执行存储装置61中存储的一个或多个程序时，也实现如下操作：将待拼接图像之间的特征点对集合执行本申请任意实施例提供的异常点对的检测方法，得到所述待拼接图像之间的异常点对集合；根据所述待拼接图像之间的特征点对集合及异常点对集合拼接所述待拼接图像，获得目标图像。
本申请实施例还提供一种计算机可读存储介质，该存储介质中的程序由计算机设备的处理器执行时，使得计算机设备能够执行如上述方法实施例所述的异常点对的检测方法和/或图像拼接方法。示例性的，该异常点对的检测方法包括：根据第一矩阵筛选待检测图像之间的特征点对集合，得到所述待检测图像之间的正常点对集合，所述第一矩阵与所述正常点对集合中的点对满足第一约束条件；根据第二矩阵筛选所述正常点对集合，得到所述待检测图像之间的异常点对集合，所述第二矩阵与所述正常点对集合中除所述异常点对集合外的点对满足第二约束条件，所述第二约束条件的约束力强于所述第一约束条件的约束力。或者，该图像拼接方法包括：将待拼接图像之间的特征点对集合执行本申请任意实施例提供的异常点对的检测方法，得到所述待拼接图像之间的异常点对集合；根据所述待拼接图像之间的特征点对集合及异常点对集合拼接所述待拼接图像，获得目标图像。
对于装置、计算机设备、存储介质实施例而言，由于其与方法实施例基本相似，所以描述的比较简单，相关之处参见方法实施例的部分说明即可。
通过以上关于实施方式的描述，所属领域的技术人员可以了解到，本申请可借助软件及通用硬件来实现，也可以通过硬件实现。基于这样的理解，本申请的技术方案可以以软件产品的形式体现出来，该计算机软件产品可以存储在计算机可读存储介质中，如计算机的软盘、只读存储器(Read-Only Memory,ROM)、随机存取存储器(Random Access Memory,RAM)、闪存(FLASH)、硬盘或光盘等，包括多个指令用以使得一台计算机设备(可以是机器人，个人计算机，服务器，或者网络设备等)执行本申请任意实施例所述的异常点对的检测方法和/或图像拼接方法。
值得注意的是，上述异常点对的检测装置中，所包括的多个单元和模块只是按照功能逻辑进行划分的，但并不局限于上述的划分，只要能够实现相应的功能即可；另外，多个功能单元的名称也只是为了便于相互区分，并不用于限制本申请的保护范围。
应当理解，本申请的多个部分可以用硬件、软件、固件或它们的组合来实现。在上述实施方式中，多个步骤或方法可以用存储在存储器中且由合适的指令执行装置执行的软件或固件来实现。例如，如果用硬件来实现，和在另一实施方式中一样，可用本领域公知的下列技术中的任一项或他们的组合来实现：具有用于对数据信号实现逻辑功能的逻辑门电路的离散逻辑电路，具有合适的组合逻辑门电路的专用集成电路，可编程门阵列(Programmable Gate Array，PGA)，现场可编程门阵列(Field Programmable Gate Array，FPGA)等。

Claims (16)

  1. 一种异常点对的检测方法，包括：
    根据第一矩阵筛选待检测图像之间的特征点对集合，得到所述待检测图像之间的正常点对集合，所述第一矩阵与所述正常点对集合中的点对满足第一约束条件；
    根据第二矩阵筛选所述正常点对集合，得到所述待检测图像之间的异常点对集合，所述第二矩阵与所述正常点对集合中除所述异常点对集合外的点对满足第二约束条件，所述第二约束条件的约束力强于所述第一约束条件的约束力。
  2. 根据权利要求1所述的方法，其中，所述根据第二矩阵筛选所述正常点对集合，得到所述待检测图像之间的异常点对集合，包括：
    确定所述正常点对集合所包含点对的点对总数；
    在所述点对总数大于设定基准值的情况下，根据所述正常点对集合确定第二矩阵，基于确定的第二矩阵筛选所述正常点对集合，获得对应所述第二矩阵的中间点对集合，并将所述中间点对集合作为新的正常点对集合，返回继续确定所述正常点对集合所包含点对的点对总数；
    在所述点对总数小于或等于所述设定基准值的情况下，将所述正常点对集合中包含的点对确定为异常点对，形成包含所述异常点对的异常点对集合；
    其中，所述中间点对集合包含了所述正常点对集合中与所述第二矩阵不满足第二约束条件的点对。
  3. 根据权利要求1所述的方法，其中，所述第一约束条件为：点对中第一特征点的特征向量P 1和第二特征点的特征向量P 2与所述第一矩阵H 1满足P 1 TH 1P 2＝0；
    所述第二约束条件为：点对中第三特征点的特征向量P 3和第四特征点的特征向量P 4与所述第二矩阵H 2满足P 4＝H 2P 3。
  4. 根据权利要求1所述的方法，其中，所述第一矩阵的行数和列数均为3且秩为2；所述第二矩阵行数和列数均为3且第3行第3列的元素值为1。
  5. 根据权利要求1至4任一项所述的方法，其中，所述根据第一矩阵筛选待检测图像之间的特征点对集合，得到所述待检测图像之间的正常点对集合，包括：
    基于待检测图像之间特征点对集合中的点对，确定至少一个第一候选矩阵；
    针对每个第一候选矩阵，确定所述特征点对集合中与所述每个第一候选矩阵满足所述第一约束条件的匹配点对的点对数量；
    比较所述至少一个第一候选矩阵对应的点对数量，将最大点对数量对应的第一候选矩阵确定为第一矩阵；
    基于对应所述第一矩阵的匹配点对形成所述待检测图像之间的正常点对集合。
  6. 根据权利要求5所述的方法，其中，所述基于待检测图像之间特征点对集合中的点对，确定至少一个第一候选矩阵，包括：
    将已确定的当前循环次数与第一阈值进行比较；
    在所述当前循环次数小于或等于所述第一阈值的情况下，从所述待检测图像之间的特征点对集合中随机选取第一预设数量的第一样本点对；
    基于所述第一预设数量的第一样本点对中特征点的特征点信息，确定第一候选矩阵模型中多个待定参数的参数值，形成当前的第一候选矩阵；
    将所述当前循环次数加1作为新的当前循环次数，返回继续进行当前循环次数与所述第一阈值的比较操作；
    在所述当前循环次数大于所述第一阈值的情况下，获取已确定的第一候选矩阵。
  7. 根据权利要求5所述的方法，其中，所述确定所述特征点对集合中与所述每个第一候选矩阵满足所述第一约束条件的匹配点对的点对数量，包括：
    将所述特征点对集合中的每个第一点对中第一特征点的特征向量的转置与所述每个第一候选矩阵的乘积作为第一乘积矩阵；
    在所述第一乘积矩阵与所述每个第一点对中第二特征点的特征向量的乘积为零矩阵的情况下，将所述每个第一点对作为对应所述每个第一候选矩阵的匹配点对；
    统计所述特征点对集合中对应所述每个第一候选矩阵的匹配点对的点对数量。
  8. 根据权利要求1至4任一项所述的方法，其中，所述根据第二矩阵筛选所述正常点对集合，得到所述待检测图像之间的异常点对集合，包括：
    基于所述正常点对集合中的点对，确定至少一个第二候选矩阵；
    针对每个第二候选矩阵，确定所述正常点对集合中与所述每个第二候选矩阵满足所述第二约束条件的匹配点对的点对数量；
    比较所述至少一个第二候选矩阵对应的点对数量，将最大点对数量对应的第二候选矩阵确定为第二矩阵；
    从所述正常点对集合中删除对应所述第二矩阵的匹配点对，获得所述待测图像之间的异常点对集合。
  9. 根据权利要求8所述的方法，其中，所述基于所述正常点对集合中的点对，确定至少一个第二候选矩阵，包括：
    将已确定的当前执行次数与第二阈值进行比较；
    在所述当前执行次数小于或等于所述第二阈值的情况下，从所述待检测图像之间特征点对集合中随机选取至少第二预设数量的第二样本点对；
    基于所述至少第二预设数量的第二样本点对中特征点的特征点信息，确定第二候选矩阵模型中多个待定参数的参数值，形成当前的第二候选矩阵；
    将所述当前执行次数加1作为新的当前执行次数，返回继续进行当前执行次数与所述第二阈值的比较操作；
    在所述当前执行次数大于所述第二阈值的情况下，获取已确定的第二候选矩阵。
  10. 根据权利要求8所述的方法，其中，所述确定所述正常点对集合中与所述每个第二候选矩阵满足所述第二约束条件的匹配点对的点对数量，包括：
    在所述每个第二候选矩阵与所述正常点对集合中的每个第二点对中第三特征点的特征向量的乘积等于所述每个第二点对中的第四特征点的特征向量的情况下，将所述每个第二点对作为对应所述每个第二候选矩阵的匹配点对；
    统计所述正常点对集合中对应所述每个第二候选矩阵的匹配点对的点对数量。
  11. 一种图像拼接方法，包括：
    将待拼接图像之间的特征点对集合执行如权利要求1-10任一项所述的异常点对的检测方法，得到所述待拼接图像之间的异常点对集合；
    根据所述待拼接图像之间的特征点对集合及异常点对集合拼接所述待拼接图像，获得目标图像。
  12. 根据权利要求11所述的方法，在得到所述待拼接图像之间的异常点对集合之前，还包括：
    分别提取所述待拼接图像中的特征点；
    将所述待拼接图像之间的特征点进行匹配，得到所述待拼接图像之间的特征点对集合。
  13. 一种异常点对的检测装置，包括：
    第一集合确定模块，设置为根据第一矩阵筛选待检测图像之间的特征点对集合，得到所述待检测图像之间的正常点对集合，所述第一矩阵与所述正常点对集合中的点对满足第一约束条件；
    第二集合确定模块，设置为根据第二矩阵筛选所述正常点对集合，得到所述待检测图像之间的异常点对集合，所述第二矩阵与所述正常点对集合中除所述异常点对集合外的点对满足第二约束条件，所述第二约束条件的约束力强于所述第一约束条件的约束力。
  14. 一种图像拼接装置，包括：
    异常集合确定模块，其中设置有如权利要求13所述的异常点对的检测装置，设置为根据待拼接图像之间的特征点对集合得到所述待拼接图像之间的异常点对集合；
    目标图像确定模块，设置为根据所述待拼接图像之间的特征点对集合及异常点对集合拼接所述待拼接图像，获得目标图像。
  15. 一种计算机设备，包括：
    至少一个处理器；
    存储装置，设置为存储至少一个程序；
    所述至少一个程序被所述至少一个处理器执行，使得所述至少一个处理器实现如权利要求1-12任一项所述的方法。
  16. 一种计算机可读存储介质，存储有计算机程序，所述计算机程序被处理器执行时实现如权利要求1-12任一项所述的方法。
PCT/CN2020/098110 2019-07-30 2020-06-24 异常点对的检测方法、图像拼接方法、装置及设备 WO2021017699A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910696884.1 2019-07-30
CN201910696884.1A CN110458875B (zh) 2019-07-30 2019-07-30 异常点对的检测方法、图像拼接方法、相应装置及设备

Publications (1)

Publication Number Publication Date
WO2021017699A1 true WO2021017699A1 (zh) 2021-02-04

Family

ID=68484158

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/098110 WO2021017699A1 (zh) 2019-07-30 2020-06-24 异常点对的检测方法、图像拼接方法、装置及设备

Country Status (2)

Country Link
CN (1) CN110458875B (zh)
WO (1) WO2021017699A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112907540A (zh) * 2021-02-22 2021-06-04 浙江大华技术股份有限公司 一种拼接异常检测方法、装置、设备及介质
CN113689555A (zh) * 2021-09-09 2021-11-23 武汉惟景三维科技有限公司 一种双目图像特征匹配方法及系统

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110458875B (zh) * 2019-07-30 2021-06-15 广州市百果园信息技术有限公司 异常点对的检测方法、图像拼接方法、相应装置及设备
CN112365521B (zh) * 2020-12-08 2021-08-27 萱闱(北京)生物科技有限公司 终端设备的速度监测方法、装置、介质和计算设备

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104361626A (zh) * 2014-09-29 2015-02-18 北京理工大学 基于混合匹配策略的皮下静脉三维重建方法
CN104574421A (zh) * 2015-01-29 2015-04-29 北方工业大学 一种大幅面小重合区域高精度多光谱图像配准方法及装置
WO2017132766A1 (en) * 2016-02-03 2017-08-10 Sportlogiq Inc. Systems and methods for automated camera calibration
CN107993193A (zh) * 2017-09-21 2018-05-04 沈阳工业大学 基于光照均衡化和改进surf算法的隧道衬砌图像拼接方法
CN110070564A (zh) * 2019-05-08 2019-07-30 广州市百果园信息技术有限公司 一种特征点匹配方法、装置、设备及存储介质
CN110458875A (zh) * 2019-07-30 2019-11-15 广州市百果园信息技术有限公司 异常点对的检测方法、图像拼接方法、相应装置及设备

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101819680B (zh) * 2010-05-12 2011-08-31 上海交通大学 图像匹配点对的检测方法
CN105761206A (zh) * 2014-12-16 2016-07-13 富泰华工业（深圳）有限公司 点云拼接方法及系统
CN107507226B (zh) * 2017-09-26 2021-04-06 中国科学院长春光学精密机械与物理研究所 一种图像匹配的方法及装置
CN108335319A (zh) * 2018-02-06 2018-07-27 中南林业科技大学 一种基于自适应阈值及ransac的图像角点匹配方法
CN108470324B (zh) * 2018-03-21 2022-02-25 深圳市未来媒体技术研究院 一种鲁棒的双目立体图像拼接方法
CN109741245B (zh) * 2018-12-28 2023-03-17 杭州睿琪软件有限公司 平面信息的插入方法及装置

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104361626A (zh) * 2014-09-29 2015-02-18 北京理工大学 基于混合匹配策略的皮下静脉三维重建方法
CN104574421A (zh) * 2015-01-29 2015-04-29 北方工业大学 一种大幅面小重合区域高精度多光谱图像配准方法及装置
WO2017132766A1 (en) * 2016-02-03 2017-08-10 Sportlogiq Inc. Systems and methods for automated camera calibration
CN107993193A (zh) * 2017-09-21 2018-05-04 沈阳工业大学 基于光照均衡化和改进surf算法的隧道衬砌图像拼接方法
CN110070564A (zh) * 2019-05-08 2019-07-30 广州市百果园信息技术有限公司 一种特征点匹配方法、装置、设备及存储介质
CN110458875A (zh) * 2019-07-30 2019-11-15 广州市百果园信息技术有限公司 异常点对的检测方法、图像拼接方法、相应装置及设备

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112907540A (zh) * 2021-02-22 2021-06-04 浙江大华技术股份有限公司 一种拼接异常检测方法、装置、设备及介质
CN112907540B (zh) * 2021-02-22 2024-05-14 浙江大华技术股份有限公司 一种拼接异常检测方法、装置、设备及介质
CN113689555A (zh) * 2021-09-09 2021-11-23 武汉惟景三维科技有限公司 一种双目图像特征匹配方法及系统
CN113689555B (zh) * 2021-09-09 2023-08-22 武汉惟景三维科技有限公司 一种双目图像特征匹配方法及系统

Also Published As

Publication number Publication date
CN110458875B (zh) 2021-06-15
CN110458875A (zh) 2019-11-15

Similar Documents

Publication Publication Date Title
WO2021017699A1 (zh) 异常点对的检测方法、图像拼接方法、装置及设备
US11055516B2 (en) Behavior prediction method, behavior prediction system, and non-transitory recording medium
KR20210144625A (ko) 영상 데이터 처리 방법, 장치 및 판독 가능 저장 매체
KR101303877B1 (ko) 얼굴 검출과 피부 영역 검출을 적용하여 피부의 선호색변환을 수행하는 방법 및 장치
TWI734375B (zh) 圖像處理方法、提名評估方法及相關裝置
WO2021000830A1 (zh) 人脸识别的方法、装置、计算机设备和存储介质
CN107909569B (zh) 一种花屏检测方法、花屏检测装置及电子设备
CN109598250B (zh) 特征提取方法、装置、电子设备和计算机可读介质
CN108875481A (zh) 用于行人检测的方法、装置、系统及存储介质
CN109408672A (zh) 一种文章生成方法、装置、服务器及存储介质
CN110210558A (zh) 评估神经网络性能的方法及装置
CN109657711A (zh) 一种图像分类方法、装置、设备及可读存储介质
CN104933145A (zh) 照片处理方法及装置与移动终端
CN110472499B (zh) 一种行人重识别的方法及装置
CN112562727B (zh) 应用于音频监控的音频场景分类方法、装置以及设备
WO2017113735A1 (zh) 一种视频格式区分方法及系统
CN116070816B (zh) 一种基于物联网的飞行模拟训练管理方法及系统
CN113010736A (zh) 一种视频分类方法、装置、电子设备及存储介质
CN108665455B (zh) 图像显著性预测结果的评价方法和装置
CN111309948A (zh) 图片筛选方法、图片筛选装置以及电子设备
TW201319956A (zh) 人臉影像辨識方法及裝置
CN112801070B (zh) 一种目标检测方法、装置、设备及存储介质
CN113783795B (zh) 加密流量分类方法及相关设备
CN110738638B (zh) 视觉显著性检测算法适用性预测与性能盲评价方法
WO2021135863A1 (zh) 一种图像识别方法及装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20848668

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20848668

Country of ref document: EP

Kind code of ref document: A1