CN114067421B - Personnel duplicate removal identification method, storage medium and computer equipment - Google Patents
Personnel duplicate removal identification method, storage medium and computer equipment Download PDFInfo
- Publication number
- CN114067421B CN114067421B CN202210046435.4A CN202210046435A CN114067421B CN 114067421 B CN114067421 B CN 114067421B CN 202210046435 A CN202210046435 A CN 202210046435A CN 114067421 B CN114067421 B CN 114067421B
- Authority
- CN
- China
- Prior art keywords
- face
- person
- information
- similarity
- face information
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G07—CHECKING-DEVICES
- G07C—TIME OR ATTENDANCE REGISTERS; REGISTERING OR INDICATING THE WORKING OF MACHINES; GENERATING RANDOM NUMBERS; VOTING OR LOTTERY APPARATUS; ARRANGEMENTS, SYSTEMS OR APPARATUS FOR CHECKING NOT PROVIDED FOR ELSEWHERE
- G07C9/00—Individual registration on entry or exit
- G07C9/10—Movable barriers with registering means
- G07C9/15—Movable barriers with registering means with arrangements to prevent the passage of more than one individual at a time
-
- G—PHYSICS
- G07—CHECKING-DEVICES
- G07C—TIME OR ATTENDANCE REGISTERS; REGISTERING OR INDICATING THE WORKING OF MACHINES; GENERATING RANDOM NUMBERS; VOTING OR LOTTERY APPARATUS; ARRANGEMENTS, SYSTEMS OR APPARATUS FOR CHECKING NOT PROVIDED FOR ELSEWHERE
- G07C9/00—Individual registration on entry or exit
- G07C9/30—Individual registration on entry or exit not involving the use of a pass
- G07C9/32—Individual registration on entry or exit not involving the use of a pass in combination with an identity check
- G07C9/37—Individual registration on entry or exit not involving the use of a pass in combination with an identity check using biometric data, e.g. fingerprints, iris scans or voice recognition
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a person duplicate removal identification method, a storage medium and computer equipment, and relates to the technical field of image processing, wherein the person duplicate removal identification method can effectively perform person distinguishing duplicate removal identification aiming at persons in a list or strangers of persons in a non-list by maintaining a cache queue T; by adopting the face frame position overlapping degree and matching with the similarity double threshold value of the face algorithm, on one hand, the problem of person tracking loss caused by video decoding frame dropping or low detection rate of the face algorithm and the like of the person can be solved, and extra wrong passing records generated when the person passes in the list are effectively reduced; on the other hand, the similarity double thresholds of the face algorithm are adopted to realize effective duplication removal on both static targets and non-static targets; no matter the overlapping degree of the positions of the face frames or the similarity calculation of the face algorithm, the calculation process has low requirements on calculation force and can run smoothly on most equipment.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to a person duplicate removal identification method, a storage medium and computer equipment.
Background
With the maturity and development of face recognition technology, the application of face recognition in intelligent access control systems is more and more extensive, and a plurality of corresponding face recognition access control systems appear in the market. However, when people walk in front of the camera, people often take a snapshot of two face photos, so that the face recognition intelligent access control system can generate face image records of the same person but with a small difference. For this purpose, it is necessary to minimize such recordings by means of a deduplication algorithm. At present, the main face image deduplication algorithms mainly include a deduplication method based on traditional image features (such as a color histogram and an LBP histogram), a fast deduplication method based on motion matching, and a deduplication algorithm based on tracking.
The method for removing the duplicate based on the traditional image characteristics can remove 50% of repeated human faces, and the false duplicate removal ratio is low. However, due to the fact that light differences of different video monitoring scenes are too large, all application scenes cannot be compatible by using the same threshold value, and the method is not universal; the fast duplicate removal method based on motion matching needs to set two parameter values of a pixel difference threshold and a difference ratio, mainly removes static repeated targets, and cannot process duplicate removal of non-static targets; the deduplication algorithm based on tracking requires three parameters to be set: tracking frequency, deduplication frequency and retention frequency, and the deduplication effect depends on the detection algorithm and the tracking algorithm. If the detection frequency is low or the detection rate of the detection algorithm is poor, the person can be lost; and because of having the minimum detection frequency requirement, have higher requirement to machine performance.
Meanwhile, in the above several duplicate removal algorithms, no consideration is given to the classification of people (i.e. people in a list or strangers) in the actual access control application, and no special targeted treatment is performed; particularly, when a person in the list just appears on a camera screen, the captured face may not reach the recognition similarity threshold, and the person is determined as a stranger passing record; when the user walks to the middle area of the video monitoring, the face captured by the capturing device can reach the threshold value of the recognition similarity, and then the user can be judged as a passing record of the people in the list; when people in the list leave the camera screen soon, the face captured by the camera may not reach the recognition similarity threshold, and the person is judged to be a stranger passing record; when people in the list pass through the camera, two different types of traffic records occur; these are additional false passage records that are generated without consideration of the classification of people that may be present in the application.
Disclosure of Invention
The present invention is directed to a person duplicate removal recognition method, a storage medium, and a computer device, so as to solve the problems in the background art.
In order to solve the technical problems, the invention provides the following technical scheme: a person duplicate removal identification method comprises the following steps:
s1, creating an empty buffer queueBuffer queueThe queue element data comprises face imagesPosition of human face frameFace pose scoreFace feature codeClass bitAnd second-level time stamp of element addition time;
S2, circularly taking out the decoded picture from the network camera; suppose that one frame of decoded pictures isAnd obtaining the second-level time stamp of the current time(ii) a Checking buffer queuesThe second-level time stamps of all queue elements in the queue, if the second-level time stamp of the current time isSubtract a queue elementTime stamp of second orderThe obtained differenceFurther determining the queue elementIs classified into a plurality of groupsThe value of (a) is,the time required for a person to normally pass in front of the camera is accurate to seconds; if it isThen push the queue elementAnd from the cache queueDelete the queue element(ii) a If it isThen directly from the buffer queueDelete the teamColumn element;
S3, displaying the pictureFace detection is carried out to obtain picturesExtracting face feature codes of face images of all the face information to obtain face feature code information corresponding to the face images;
s4, aligning the pictures respectivelyPerforming face feature code first similarity on face feature codes extracted from all face information and face feature codes of people in the list one by oneCalculating and sorting the similarity values in a descending order; if the maximum similarity value is larger than or equal to the similarity threshold value of the same person judged by the algorithmIf not, the person corresponding to the face information is regarded as a stranger in the non-list;
s5, aligning the pictures respectivelyAll face information and buffer queue in the systemThe overlapping degree of the positions of the face frames of all the internal elements is carried out one by oneSecond similarity with face feature codeCalculating;
s6, if one element existsWith certain face informationMeet the overlapping degree of the face frame positionThreshold valueOr a second similarity of the face feature codesThreshold valueThen the face information is consideredThe corresponding personnel are already present, and classification judgment is further carried out according to the personnel types;
s7, if one element does not existWith certain face informationMeet the overlapping degree of the face frame positionThreshold valueOr a second similarity of the face feature codesThreshold value(ii) a An element is newly createdTo convert the face informationAnd time stampSupplement to elements(ii) a Further carrying out classification judgment according to personnel types and setting new elementsCorresponding classification zone bit(ii) a Will new elementsStore in buffer queueIn step (2), the addition of the new cache record is completed.
Further, in step S2,is a positive integer; in step S5, the face frame position overlap degreeCalculating the second similarity of face feature code for duplication elimination of the first featureA second feature calculation is performed for deduplication.
Further, the degree of overlap of the positions of the face frames in step S5The calculation method of (2) is as follows:
assume a rectangleThe vertex coordinates of the upper left corner are respectively、Having a height and a width ofAndrectangular shapeThe vertex coordinates of the upper left corner are respectively、Having a height and a width ofAnd;
duplication removing first characteristic face frame position overlapping degreeThe calculation formula of (a) is as follows:
duplication removing first characteristic face frame position overlapping degreeHas a numerical value range of。
Further, overlapping two rectangular edges is not considered overlapping.
Further, in step S6:
s61, determining face informationIs a person in the list, and elementInner pushing zone bitIf the value of (2) is 0, the face information is immediately pushedAnd modifying the elementsInner pushing zone bitHas a value of 1;
s62, if the face informationIs a person in the list, and elementInner pushing zone bitIf the value of (1) is less than the predetermined threshold, the face information does not need to be pushed;
S63, determining face informationIf the person is not in the list, the judgment element is not neededInner pushing zone bitA value of (d);
face pose scoreAccording to the face pose anglePerforming calculations, respectively, of、Andthree angles, all of which have values in the rangeWithin the range of degrees, the closer any angle value is to zero, the better the face pose angle is, and the corresponding face pose scoreThe higher;
wherein the face pose scoreHas a numerical value range of,To calculate the absolute value function of a real number.
Further, in step S64:
s641, if the face informationFace pose score ofHigher than elementInner face pose scoreUsing the face informationFace image ofFace pose scorePosition of human face frameFace feature codeReplacement elementsInformation of the corresponding item in (1);
s642, if the face informationFace pose score ofNot higher than elementInner face pose scoreThen only the face information is usedFace frame position ofFace feature codeReplacement elementsInformation of the corresponding item in (1);
Further, in step S7, if there is no one of the elementsAnd the face informationMeet the overlapping degree of the face frame positionThreshold valueOr a second similarity of the face feature codesThreshold valueThen the face information is consideredCorresponding personnel do not appear, and the face information needs to be addedRecording for new buffer; newly building an elementTo convert the face informationFace frame position ofHuman face pose angleHuman face imageFace feature codeTime stamp of the order of secondsSupplement to elements(ii) a Further carrying out classification judgment according to the personnel types, if the face informationIf the corresponding person is a person in the list, the face information is immediately pushedAnd new elements are combinedClassification markSign positionIs set to 1; if the face informationIf the corresponding person is a stranger of persons in the non-list, the new element is addedPush flag bitIs set to 0; and new elements are addedStore in buffer queueIn step (2), the addition of the new cache record is completed.
The invention also provides a storage medium for person deduplication identification, on which computer instructions are stored, which, when executed by a processor, implement the steps of a person deduplication identification method as described in any one of the above.
The invention also provides computer equipment for person duplicate removal identification, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein when the processor runs the computer program, the person duplicate removal identification method is realized.
Compared with the prior art, the invention has the following beneficial effects:
1. the invention carries out the operation on all the faces recognized by each frame and maintains a buffer queueCan effectively distinguish and duplicate-removing personnel aiming at the personnel in the list or strangers of the personnel in the non-list(ii) a Because other traditional characteristics are not adopted, the required duplication removing characteristics are face pose scores and face similarity values obtained by the position of a face frame and the face pose angle; the characteristics are all information possessed by a general face algorithm, are irrelevant to a camera scene and have universality; by adopting the face frame position overlapping degree and the similarity double threshold value of the face algorithm, on one hand, the problem of person tracking loss caused by frame dropping of video decoding or low detection rate of the face algorithm and the like of the person can be solved, and the extra wrong passing record generated when the person passes in the list is effectively reduced by matching and tracking the second similarity threshold value of the face algorithm again; on the other hand, by adopting the similarity double thresholds of the face algorithm, even if the personnel are still for a long time, the personnel can be effectively compared, and the static target and the non-static target can be effectively removed; no matter the overlapping degree of the positions of the face frames or the similarity calculation of the face algorithm, the calculation process has low requirements on calculation force and can run smoothly on most equipment;
2. when the person in the list just appears in the camera screen, the captured face may not reach the recognition similarity threshold and can be judged as a stranger passing record, but due to the adoption of the delay methodSecond push, so push will not be done immediately at this time, and only buffer queue will existInternal; when the user walks to the middle area of video monitoring, the captured face possibly reaches the threshold value of the recognition similarity, the user can be judged to be the passing record of the people in the list, at the moment, according to the judgment logic of the invention, the information of the people in the list can be immediately pushed, and the information is cached in a cache queueMatching the cache records, updating the data of the records, and updating the classification mark of the records to be 1; when people in the list are about to leave the camera picture, the face captured by the camera may not reach the threshold of the recognition similarity, and the face can be judgedMaking a pass record of strangers, and then according to the cache queueThe internal cache records can be normally matched with the cache records meeting the conditions, the classification mark is 1, and the system cannot push the stranger information; therefore, the invention can ensure that only one correct pass record is generated when a single person passes by the system to the maximum extent.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention and not to limit the invention. In the drawings:
FIG. 1 is a schematic illustration of the present invention when the rectangles are not overlapping;
FIG. 2 is a schematic illustration of the invention when rectangles overlap;
FIG. 3 is a schematic diagram of the face pose angles of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
A person deduplication identification method as shown in fig. 1-3, comprises the following steps:
s1, creating an empty buffer queueBuffer queueThe queue element data comprises face imagesPosition of human face frameFace pose scoreFace feature codeClass bitAnd second-level time stamp of element addition time;
S2, circularly taking out the decoded picture from the network camera; suppose that one frame of decoded pictures isAnd obtaining the second-level time stamp of the current time(ii) a Checking buffer queuesThe second-level time stamps of all queue elements in the queue, if the second-level time stamp of the current time isSubtract a queue elementTime stamp of second orderThe obtained differenceFurther determining the queue elementIs classified into a plurality of groupsThe value of (a) is,the time required for a person to normally pass in front of the camera is accurate to seconds; if it isThen push the queue elementAnd from the cache queueDelete the queue element(ii) a If it isThen directly from the buffer queueDelete the queue elementIn step S2, the process proceeds,is a positive integer; in step S5, the face frame position overlap degreeCalculating the second similarity of face feature code for duplication elimination of the first featureComputing for the de-duplication second features;
s3, displaying the pictureFace detection is carried out to obtain picturesExtracting face feature codes of face images of all the face information to obtain face feature code information corresponding to the face images;
s4, aligning the pictures respectivelyPerforming face feature code first similarity on face feature codes extracted from all face information and face feature codes of people in the list one by oneCalculating and sorting the similarity values in a descending order; if the maximum similarity value is larger than or equal to the similarity threshold value of the same person judged by the algorithmIf not, the person corresponding to the face information is regarded as a stranger in the non-list;
s5, aligning the pictures respectivelyAll face information and buffer queue in the systemThe overlapping degree of the positions of the face frames of all the internal elements is carried out one by oneSecond similarity with face feature codeCalculating;
s6, if one element existsWith certain face informationMeet the overlapping degree of the face frame positionThreshold valueOr a second similarity of the face feature codesThreshold valueThen the face information is consideredThe corresponding personnel are already present, and classification judgment is further carried out according to the personnel types;
s61, determining face informationIs a person in the list, and elementInner pushing zone bitIf the value of (2) is 0, the face information is immediately pushedAnd modifying the elementsInner pushing zone bitHas a value of 1;
s62, if the face informationIs a person in the list, and elementInner pushing zone bitIf the value of (1) is less than the predetermined threshold, the face information does not need to be pushed;
S63, determining face informationIf the person is not in the list, the judgment element is not neededInner pushing zone bitA value of (d);
s641, if the face informationFace pose score ofNumber ofHigher than elementInner face pose scoreUsing the face informationFace image ofFace pose scorePosition of human face frameFace feature codeReplacement elementsInformation of the corresponding item in (1);
s642, if the face informationFace pose score ofNot higher than elementInner face pose scoreThen only the face message is usedInformation processing deviceFace frame position ofFace feature codeReplacement elementsInformation of the corresponding item in (1);
S7, if one element does not existAnd the face informationMeet the overlapping degree of the face frame positionThreshold valueOr a second similarity of the face feature codesThreshold valueThen the face information is consideredCorresponding personnel do not appear, and the face information needs to be addedRecording for new buffer; newly building an elementTo convert the face informationFace frame position ofHuman face pose angleHuman face imageFace feature codeTime stamp of the order of secondsSupplement to elements(ii) a Further carrying out classification judgment according to the personnel types, if the face informationIf the corresponding person is a person in the list, the face information is immediately pushedAnd new elements are combinedClassification zone bitIs set to 1; if the face informationIf the corresponding person is a stranger of persons in the non-list, the new element is addedPush flag bitIs set to 0; and new elements are addedStore in buffer queueIn step (2), the addition of the new cache record is completed.
assume a rectangleThe vertex coordinates of the upper left corner are respectively、Having a height and a width ofAndrectangular shapeThe vertex coordinates of the upper left corner are respectively、Having a height and a width ofAnd;
duplication removing first characteristic face frame position overlapping degreeThe calculation formula of (a) is as follows:
duplication removing first characteristic face frame position overlapping degreeHas a numerical value range ofThe overlap of two rectangular edges is not considered to be an overlap.
face pose scoreAccording to the face pose anglePerforming calculations, respectively, of、Andthree angles, all of which have values in the rangeWithin the range of degrees, the closer any angle value is to zero, the better the face pose angle is, and the corresponding face pose scoreThe higher;
wherein the face pose scoreHas a numerical value range of,To calculate the absolute value function of a real number.
The invention also provides a storage medium for person deduplication identification, on which computer instructions are stored, which, when executed by a processor, implement the steps of a person deduplication identification method as described in any one of the above.
The invention also provides computer equipment for person duplicate removal identification, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein when the processor runs the computer program, the person duplicate removal identification method is realized.
In the face recognition intelligent access control system, when people in a name list are recognized, the recognition records of the related people in the name list are immediately pushed, and the door opening operation is carried out; for stranger records of people not on the list, delay can be adoptedQuasi-real-time pushing is carried out in a second mode, namely a stranger leaves the range of the camera picturePushing the record after the second; the delay pushing of strangers does not affect the main functions of the intelligent face access control system; whereinGenerally equal to the time it takes a person to normally pass in front of the camera (i.e. the time that occurs within the frame of the camera, accurate to seconds); because the focal lengths of the lenses of the cameras are different and the installation heights and the inclination angles of the cameras are different,the value range of (1) can be any positive integer, because the time of each person appearing in the range of the camera is not fixed, the person possibly lingers for a long time; herein, theEmphasis is placed on the time consumed by a person in normal passage;
suppose the picture in the network camera is an image(ii) a Creating an empty buffer queueBuffer queueThe queue element data comprises face imagesPosition of human face frameFace pose scoreFace feature codeClass bitAnd second-level time stamp of element addition time;
Face imageIs not larger than the size of the imageSize of face frame positionIs not larger than the imagePixel range of (2), classification flag bitThe value of (A) can be 0, 1; when in useWhen the queue element is not the push cache record, the queue element is described as the push cache record; when in useWhen the queue element is pushed, the queue element is indicated to be a pushed cache record;
the invention starts to connect the network camera first, and cyclically takes out the decoded picture from the network camera for duplication removal and identification; suppose that one frame of decoded pictures isAnd obtaining the second-level time of the current moment(ii) a Checking buffer queuesThe second-level time stamps of all queue elements in the queue, if the second-level time stamp of the current time isSubtract a queue elementTime stamp of second orderThe obtained difference(A positive integer), the queue element is further determinedIs classified into a plurality of groupsA value of (d); if it isThen push the queue elementAnd from the cache queueDelete the queue element(ii) a If it isThen directly from the buffer queueDelete the queue element;
For the pictureFace detection is carried out to obtain picturesExtracting face feature codes of face images of all face information (including face frame positions and face posture scores) to obtain face feature code information corresponding to the face images;
assuming one of face informationComprises the following steps: face frame positionFace pose scoreHuman face imageFace feature code(ii) a Carrying out first similarity of the face feature codes on the face feature codes and the face feature codes of the persons in the list one by oneCalculating and sorting the similarity values in a descending order; if the maximum similarity value is larger than or equal to the similarity threshold value of the same person judged by the algorithmIf not, the person corresponding to the face information is regarded as a stranger in the non-list;
the face information of the face is processedRespectively associated with the buffer queueThe overlapping degree of the positions of the face frames of all the internal elements is carried out one by one(duplication-removing first feature) and second similarity of face feature code(de-duplication second feature) calculation;
the detailed process of calculating the face frame position overlapping degree (duplication removing first characteristic) is as follows:
because the face frame is generally rectangular, and the sides of the rectangle are generally parallel to the coordinate axes of the pixel points; therefore, the problem of the overlapping degree of the face frames is calculated, namely the problem of the overlapping degree of the two rectangles is calculated; assume a rectangleThe vertex coordinates of the upper left corner are respectively、Having a height and a width ofAndrectangular shapeThe vertex coordinates of the upper left corner are respectively、Having a height and a width ofAnd(ii) a To calculate the overlapping degree problem of two rectangles, firstly, whether the two rectangles are overlapped (the edge overlapping is not considered as overlapping) is judged; because the overlapping conditions of the two rectangles are too many, if the two rectangles are listed one by one according to each condition, the two rectangles are too complicated; therefore, the condition that the two rectangles are not overlapped can be judged in a reverse thinking mode, and then the judgment of the overlapping condition is obtained by taking the inverse;
if the center-most rectangle is a rectangle, as shown in FIG. 1The surrounding rectangle is a rectangleNot in line with rectangleIn various cases where the rectangular shape is superimposed, when any one of the following conditions is satisfied, the rectangular shape is shown in the figureAnd a rectangleNon-overlapping:
the inference based on the above conditions is based on the method of negation, and the rectangle is knownAnd a rectangleOverlapping, the following four conditions need to be satisfied simultaneously:
if rectangularAnd a rectangleIf the two points are overlapped, two intersection points are necessary; as shown in FIG. 3, two points of intersection (where the solid circles are located) are assumed to be points respectivelyAnd pointThe formula is as follows:
combining the above equations and conditions, rectanglesAnd a rectangleOverlap, the following set of conditions holds:
when it is rectangularAnd a rectangleWhen overlapping, the overlapping area is rectangularArea of (2)Comprises the following steps:
as shown in fig. 2, rectangularAnd a rectangleTotal area of formed regionEqual to rectangleArea of (2)And a rectangleArea of (2)And then subtracting the rectangleArea of (2)Namely, the following steps are provided:
rectangleAnd a rectangleDegree of overlapIs rectangularArea of (2)And a rectangleAnd a rectangleTotal area of formed regionThe area ratio of (A) to (B) is as follows:
combining the above equations, the following equations can be derived:
rectangleAnd a rectangleDegree of overlapHas a numerical value range of(ii) a In the invention, the position overlapping degree of the face frame with the first characteristic is removedThe optimal threshold value of (2) is 0.1;
if one of the elements is presentAnd the face informationMeet the overlapping degree of the face frame positionThreshold valueOr a second similarity of the face feature codesThreshold valueThen the face information is consideredThe corresponding personnel are already present, and classification judgment is further carried out according to the personnel types; if the face informationThe corresponding person is a person in the list, and the elementsInner pushing zone bitIf the value of (2) is 0, the face information is immediately pushedAnd modifying the elementsInner pushing zone bitIs 1, then face pose score is performedCalculating; if the face informationIs a person in the list, and elementInner pushing zone bitIf the value of (1) is less than the threshold value, the face pose score is directly performedComputing(ii) a If the face informationIf the corresponding person is not in the list, the judgment element is not neededInner pushing zone bitBy directly performing face pose scoringCalculating;
face pose scoreAccording to the face pose anglePerforming calculations, respectively, of、Andthree angles (as shown in fig. 3), all of which have values in the rangeWithin the range of degrees, the closer any angle value is to zero, the better the face pose angle is, and the corresponding face pose scoreThe higher; in the invention, the face pose scoreThe formula for the calculation is as follows:
face pose scoreHas a numerical value range of,To calculate the absolute value function of the real number;
if the face informationFace pose score ofHigher than elementInner face pose scoreUsing the face informationFace image ofFace pose scorePosition of human face frameFace feature codeReplacement elementsAnd updating the elementTime stamp ofAs a second-order time stamp;
If the face informationFace pose score ofNot higher than elementInner face pose scoreThen only the face information is usedFace frame position ofFace feature codeReplacement elementsAnd updating the elementTime stamp ofAs a second-order time stamp;
If one of the elements is not presentAnd the face informationMeet the overlapping degree of the face frame positionThreshold valueOr a second similarity of the face feature codesThreshold valueThen the face information is consideredCorresponding personnel do not appear, and the face information needs to be addedRecording for new buffer; newly building an elementTo convert the face informationFace frame position ofHuman face pose angleHuman face imageFace feature codeTime stamp of the order of secondsSupplement to elements(ii) a Further carrying out classification judgment according to the personnel types, if the face informationIf the corresponding person is a person in the list, the face information is immediately pushedAnd new elements are combinedClassification zone bitIs set to 1; if the face informationIf the corresponding person is a stranger of persons in the non-list, the new element is addedPush flag bitIs set to 0; and new elements are addedStore in buffer queueIn the middle, the addition of a new cache record is completed;
whereinThe reason is that a general face recognition algorithm has a similarity recommendation threshold value which is used for judging whether the similarity of the face feature codes extracted by two face images can be judged as the same person; however, a single threshold value appears, and the similarity values of the face feature codes extracted from two face images of the same person do not meet the requirement of the threshold value, so that the person can be judged as two persons; for this purpose, the invention sets a second similarity thresholdOr is appropriately lower than the similarity recommendation threshold of the same person determined by the algorithm;
All the faces recognized by each frame are subjected to the operation, and a buffer queue is maintainedThe person distinguishing duplicate removal identification can be effectively carried out on the persons in the list or strangers of the persons in the non-list; because other traditional characteristics are not adopted, the required duplication removing characteristics are face pose scores and face similarity values obtained by the position of a face frame and the face pose angle; all of these featuresThe face recognition method is information which is possessed by a general face algorithm, is irrelevant to a camera scene, and has universality; by adopting the face frame position overlapping degree and the similarity double threshold value of the face algorithm, on one hand, the problem of person tracking loss caused by frame dropping of video decoding or low detection rate of the face algorithm and the like of the person can be solved, and the extra wrong passing record generated when the person passes in the list is effectively reduced by matching and tracking the second similarity threshold value of the face algorithm again; on the other hand, by adopting the similarity double thresholds of the face algorithm, even if the personnel are still for a long time, the personnel can be effectively compared, and the static target and the non-static target can be effectively removed; no matter the overlapping degree of the positions of the face frames or the similarity calculation of the face algorithm, the calculation process has low requirements on calculation force and can run smoothly on most equipment;
by combining the advantages, when people in the list just appear on the camera screen, the captured face may not reach the recognition similarity threshold, and the person is judged to be a stranger passing record due to the adoption of the delaySecond push, so push will not be done immediately at this time, and only buffer queue will existInternal; when the user walks to the middle area of video monitoring, the captured face possibly reaches the threshold value of the recognition similarity, the user can be judged to be the passing record of the people in the list, at the moment, according to the judgment logic of the invention, the information of the people in the list can be immediately pushed, and the information is cached in a cache queueMatching the cache records, updating the data of the records, and updating the classification mark of the records to be 1; when people in the list leave the camera screen soon, the face captured by the camera may not reach the recognition similarity threshold, and the person is judged to be a stranger passing record, and at the moment, the person is judged to be a stranger passing record according to the cache queueThe internal cache records can be normally matched with the cache records meeting the conditions, the classification mark is 1, and the system cannot push the stranger information; therefore, the invention can ensure that only one correct pass record is generated when a single person passes by the system to the maximum extent.
It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus.
Finally, it should be noted that: although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that changes may be made in the embodiments and/or equivalents thereof without departing from the spirit and scope of the invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. A person duplicate removal identification method is characterized by comprising the following steps: the method comprises the following steps:
s1, creating an empty buffer queueBuffer queueThe queue element data comprises face imagesPosition of human face frameFace pose scoreFace feature codeClass bitAnd second-level time stamp of element addition time;
S2, circularly taking out the decoded picture from the network camera; suppose that one frame of decoded pictures isAnd obtaining the second-level time stamp of the current time(ii) a Checking buffer queuesThe second-level time stamps of all queue elements in the queue, if the second-level time stamp of the current time isSubtract a queue elementTime stamp of second orderThe obtained differenceFurther determining the queue elementIs classified into a plurality of groupsThe value of (a) is,the time required for a person to normally pass in front of the camera is accurate to seconds; if it isThen push the queue elementAnd from the cache queueDelete the queue element(ii) a If it isThen directly from the buffer queueDelete the queue element;
S3, displaying the pictureFace detection is carried out to obtain picturesExtracting face feature codes of face images of all the face information to obtain face feature code information corresponding to the face images;
s4, aligning the pictures respectivelyPerforming face feature code first similarity on face feature codes extracted from all face information and face feature codes of people in the list one by oneCalculating and sorting the similarity values in a descending order; if the maximum similarity value is larger than or equal to the similarity threshold value of the same person judged by the algorithmIf not, the person corresponding to the face information is regarded as a stranger in the non-list;
s5, aligning the pictures respectivelyAll face information and buffer queue in the systemThe overlapping degree of the positions of the face frames of all the internal elements is carried out one by oneSecond similarity with face feature codeCalculating;
s6, if one element existsWith certain face informationMeet the overlapping degree of the face frame positionThreshold valueOr a second similarity of the face feature codesThreshold valueThen the face information is consideredThe corresponding personnel are already present, and classification judgment is further carried out according to the personnel types;
s7, if one element does not existWith certain face informationMeet the overlapping degree of the face frame positionThreshold valueOr face feature code secondDegree of similarityThreshold value(ii) a An element is newly createdTo convert the face informationAnd time stampSupplement to elements(ii) a Further carrying out classification judgment according to personnel types and setting new elementsCorresponding classification zone bit(ii) a Will new elementsStore in buffer queueIn step (2), the addition of the new cache record is completed.
2. The person deduplication identification method of claim 1, wherein: in the step S2, in step S2,is a positive integer; in thatIn step S5, the overlap of the face frame positionsCalculating the second similarity of face feature code for duplication elimination of the first featureA second feature calculation is performed for deduplication.
3. The person deduplication identification method of claim 1, wherein: face frame position overlap degree in step S5The calculation method of (2) is as follows:
assume a rectangleThe vertex coordinates of the upper left corner are respectively、Having a height and a width ofAndrectangular shapeThe vertex coordinates of the upper left corner are respectively、Having a height and a width ofAnd;
duplication removing first characteristic face frame position overlapping degreeThe calculation formula of (a) is as follows:
4. A person deduplication identification method according to claim 3, wherein: overlapping two rectangular edges is not considered overlapping.
5. The person deduplication identification method of claim 1, wherein: in step S6:
s61, determining face informationIs a person in the list, and elementInner pushing zone bitIf the value of (2) is 0, the face information is immediately pushedAnd modifying the elementsInner pushing zone bitHas a value of 1;
s62, if the face informationIs a person in the list, and elementInner pushing zone bitIf the value of (1) is less than the predetermined threshold, the face information does not need to be pushed;
S63, determining face informationIf the person is not in the list, the judgment element is not neededInner pushing zone bitA value of (d);
6. The person deduplication identification method of claim 5, wherein: in step S64, the face pose scoreThe calculation method is as follows:
face pose scoreAccording to the face pose anglePerforming calculations, respectively, of、Andthree angles, all of which have values in the rangeWithin the range of degrees, the closer any angle value is to zero, the better the face pose angle is, and the corresponding face pose scoreThe higher;
7. The person deduplication identification method of claim 5, wherein: in step S64:
s641, if the face informationFace pose score ofHigher than elementInner face pose scoreUsing the face informationFace image ofFace pose scorePosition of human face frameFace feature codeReplacement elementsInformation of the corresponding item in (1);
s642, if the face informationFace pose score ofNot higher than elementInner face pose scoreThen only the face information is usedFace frame position ofFace feature codeReplacement elementsInformation of the corresponding item in (1);
8. The person deduplication identification method of claim 1, wherein: in step S7, if there is no one of the elementsAnd the face informationMeet the overlapping degree of the face frame positionThreshold valueOr a second similarity of the face feature codesThreshold valueThen the face information is consideredCorresponding personnel do not appear, and the face information needs to be addedRecording for new buffer; newly building an elementTo convert the face informationFace frame position ofHuman face pose angleHuman face imageFace feature codeTime stamp of the order of secondsSupplement to elements(ii) a Further carrying out classification judgment according to the personnel types, if the face informationIf the corresponding person is a person in the list, the face information is immediately pushedAnd new elements are combinedClassification zone bitIs set to 1; if the face informationIf the corresponding person is a stranger of persons in the non-list, the new element is addedPush flag bitIs set to 0; and new elements are addedStore in buffer queueIn step (2), the addition of the new cache record is completed.
9. A storage medium for person deduplication identification, having computer instructions stored thereon, wherein the computer instructions, when executed by a processor, implement the steps of a person deduplication identification method of any one of claims 1-8.
10. Computer device for person deduplication identification, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor, when executing the computer program, implements a person deduplication identification method as claimed in any of claims 1-8.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210046435.4A CN114067421B (en) | 2022-01-17 | 2022-01-17 | Personnel duplicate removal identification method, storage medium and computer equipment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210046435.4A CN114067421B (en) | 2022-01-17 | 2022-01-17 | Personnel duplicate removal identification method, storage medium and computer equipment |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114067421A CN114067421A (en) | 2022-02-18 |
CN114067421B true CN114067421B (en) | 2022-04-19 |
Family
ID=80231019
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210046435.4A Active CN114067421B (en) | 2022-01-17 | 2022-01-17 | Personnel duplicate removal identification method, storage medium and computer equipment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114067421B (en) |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110084103A (en) * | 2019-03-15 | 2019-08-02 | 深圳英飞拓科技股份有限公司 | A kind of same pedestrian's analysis method and system based on face recognition technology |
CN110222627A (en) * | 2019-05-31 | 2019-09-10 | 深圳算子科技有限公司 | A kind of face amended record method |
WO2020094091A1 (en) * | 2018-11-07 | 2020-05-14 | 杭州海康威视数字技术股份有限公司 | Image capturing method, monitoring camera, and monitoring system |
CN113255621A (en) * | 2021-07-13 | 2021-08-13 | 浙江大华技术股份有限公司 | Face image filtering method, electronic device and computer-readable storage medium |
CN113269127A (en) * | 2021-06-10 | 2021-08-17 | 北京睿芯高通量科技有限公司 | Face recognition and pedestrian re-recognition monitoring method and system for real-time automatic database building |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10691925B2 (en) * | 2017-10-28 | 2020-06-23 | Altumview Systems Inc. | Enhanced face-detection and face-tracking for resource-limited embedded vision systems |
EP3706035A1 (en) * | 2019-03-07 | 2020-09-09 | Koninklijke Philips N.V. | Device, system and method for tracking and/or de-identification of faces in video data |
US11334746B2 (en) * | 2019-05-01 | 2022-05-17 | EMC IP Holding Company LLC | Facial recognition for multi-stream video using high probability group |
CN112241687A (en) * | 2020-09-16 | 2021-01-19 | 安徽超视野智能科技有限公司 | Face recognition method and system with strange face library function |
CN113537107A (en) * | 2021-07-23 | 2021-10-22 | 山东浪潮通软信息科技有限公司 | Face recognition and tracking method, device and equipment based on deep learning |
-
2022
- 2022-01-17 CN CN202210046435.4A patent/CN114067421B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020094091A1 (en) * | 2018-11-07 | 2020-05-14 | 杭州海康威视数字技术股份有限公司 | Image capturing method, monitoring camera, and monitoring system |
CN110084103A (en) * | 2019-03-15 | 2019-08-02 | 深圳英飞拓科技股份有限公司 | A kind of same pedestrian's analysis method and system based on face recognition technology |
CN110222627A (en) * | 2019-05-31 | 2019-09-10 | 深圳算子科技有限公司 | A kind of face amended record method |
CN113269127A (en) * | 2021-06-10 | 2021-08-17 | 北京睿芯高通量科技有限公司 | Face recognition and pedestrian re-recognition monitoring method and system for real-time automatic database building |
CN113255621A (en) * | 2021-07-13 | 2021-08-13 | 浙江大华技术股份有限公司 | Face image filtering method, electronic device and computer-readable storage medium |
Non-Patent Citations (2)
Title |
---|
Face Identification With Top-Push Constrained Generalized Low-Rank Approximation of Matrices;YUANJIAN CHEN .et.al;《IEEE ACCESS》;20191114;第7卷;160998-161007 * |
多路实时监控视频数据处理与分析***的设计与实现;杜爽;《全国优秀硕士学位论文全文数据库 信息科技辑》;20220115(第1期);I136-1957 * |
Also Published As
Publication number | Publication date |
---|---|
CN114067421A (en) | 2022-02-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2019218824A1 (en) | Method for acquiring motion track and device thereof, storage medium, and terminal | |
Bäuml et al. | Multi-pose face recognition for person retrieval in camera networks | |
Kart et al. | How to make an rgbd tracker? | |
US11676390B2 (en) | Machine-learning model, methods and systems for removal of unwanted people from photographs | |
WO2021139167A1 (en) | Method and apparatus for facial recognition, electronic device, and computer readable storage medium | |
Avrithis et al. | Broadcast news parsing using visual cues: A robust face detection approach | |
Yen et al. | Facial feature extraction using genetic algorithm | |
CN113537107A (en) | Face recognition and tracking method, device and equipment based on deep learning | |
Tsai et al. | Robust in-plane and out-of-plane face detection algorithm using frontal face detector and symmetry extension | |
CN111445442A (en) | Crowd counting method and device based on neural network, server and storage medium | |
CN109146913B (en) | Face tracking method and device | |
CN114067421B (en) | Personnel duplicate removal identification method, storage medium and computer equipment | |
e Souza et al. | Survey on visual rhythms: A spatio-temporal representation for video sequences | |
Vural et al. | Multi-view fast object detection by using extended haar filters in uncontrolled environments | |
CN108334811B (en) | Face image processing method and device | |
Wei et al. | Omni-face detection for video/image content description | |
Nakashima et al. | Privacy protection for social video via background estimation and CRF-based videographer's intention modeling | |
CN112258575B (en) | Method for quickly identifying object in synchronous positioning and map construction | |
CN115115976A (en) | Video processing method and device, electronic equipment and storage medium | |
US10147199B2 (en) | Method and apparatus for determining an orientation of a video | |
Duanmu et al. | A multi-view pedestrian tracking framework based on graph matching | |
Nigam et al. | EgoTracker: Pedestrian tracking with re-identification in egocentric videos | |
Schmidt Feris et al. | Detection and tracking of facial features in video sequences | |
Shen et al. | Towards intelligent photo composition-automatic detection of unintentional dissection lines in environmental portrait photos | |
CN117935171B (en) | Target tracking method and system based on gesture key points |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |