Face recognition method and device

Abstract

This application discloses a face recognition method, comprising: identifying a captured video image to obtain multiple tracking sequences of an object to be identified, each tracking sequence comprising multiple video frames; obtaining temporary identity information of the object to be identified based on the tracking sequences; obtaining a face recognition sequence and human body information corresponding to each tracking sequence, each face recognition sequence comprising multiple face recognition results; correcting the face recognition sequence of each tracking sequence using the human body information of the tracking sequence; and weightedly fusing all corrected face recognition sequences of the object to be identified based on the temporary identity information to determine the identity information of the object to be identified. This solution can improve the accuracy of face recognition even when the face is obscured.

Description

Technical Field

[0001] The present application relates to the field of face recognition technology, and in particular to a face recognition method and device. Background Art

[0002] Currently, facial recognition technology primarily relies on collecting data from multiple key facial features (such as the nose, chin, lips, eyes, and jaw) and then using algorithms to achieve recognition. However, if the face is obscured, such as when wearing a mask, the collected facial data is incomplete, reducing the accuracy of facial recognition results and even resulting in failure to recognize the face. Summary of the Invention

[0003] The purpose of this application is to provide a face recognition method and device to at least solve the above technical problems existing in the prior art.

[0004] On one hand, the present application provides a face recognition method, comprising:

[0005] Identify the captured video image to obtain multiple tracking sequences of the object to be identified, wherein the tracking sequences include multiple video frames;

[0006] Obtaining temporary identity information of the object to be identified based on the tracking sequence; and, for each tracking sequence, obtaining a face recognition sequence and body information corresponding to the tracking sequence, wherein the face recognition sequence includes multiple face recognition results;

[0007] For each tracking sequence, the face recognition sequence of the tracking sequence is corrected using the human body information of the tracking sequence;

[0008] According to the temporary identity information, all corrected face recognition sequences of the object to be identified are weightedly fused to determine the identity information of the object to be identified.

[0009] The face recognition result includes identity information and a corresponding first face recognition score, and the face recognition results are sorted from high to low according to the first face recognition score in the face recognition sequence.

[0010] For each tracking sequence, the face recognition sequence of the tracking sequence is corrected using the human body information of the tracking sequence, including:

[0011] Normalizing the first face recognition score in the face recognition sequence to obtain a second face recognition score, wherein the normalized face recognition result includes the identity information and the second face recognition score;

[0012] Each of the second face recognition scores in the face recognition sequence of the tracking sequence is corrected using the human body information of the tracking sequence.

[0013] The step of correcting each second face recognition score in the face recognition sequence of the tracking sequence using the human body information of the tracking sequence includes:

[0014] For each normalized face recognition result, determining a first probability that the object to be recognized is the identity information in the face recognition result;

[0015] Determining, based on the human body information, a posterior probability that the object to be identified is the identity information in the face recognition result;

[0016] The second face recognition score in the face recognition result is corrected according to the first probability and the posterior probability to obtain a third face recognition score.

[0017] Wherein, the human body information includes a plurality of eigenvalues, each eigenvalue having a corresponding dimension;

[0018] Determining, based on the human body information, a posterior probability that the object to be identified is the identity information in the face recognition result, including:

[0019] Obtain the historical human body information corresponding to the identity information in the face recognition result;

[0020] For each dimension in the human body information, obtain all historical feature values ​​corresponding to the dimension from the historical human body information, and determine a first number of all historical feature values;

[0021] determining a second number of historical feature values ​​that are identical to the feature value of the dimension in the human body information;

[0022] Taking the ratio of the second quantity to the first data amount as the probability of the dimension;

[0023] The probabilities of each dimension in the human body information are multiplied to obtain the posterior probability that the object to be identified is the identity information in the face recognition result.

[0024] According to the temporary identity information, all corrected face recognition sequences of the object to be identified are weightedly fused to determine the identity information of the object to be identified, including:

[0025] Merging all corrected face recognition sequences of the object to be identified according to the temporary identity information;

[0026] Performing weighted fusion processing on each third face recognition score in the merged face recognition sequence to obtain a corresponding fourth face recognition score;

[0027] A face recognition result to which a fourth face recognition score that meets the threshold is assigned is selected, and identity information included in the face recognition result is determined as the identity information of the object to be identified.

[0028] Another aspect of the present application provides a face recognition device, comprising:

[0029] A tracking module is used to identify the collected video images and obtain multiple tracking sequences of the object to be identified, wherein the tracking sequences include multiple video frames;

[0030] A first identification module, configured to obtain temporary identity information of the object to be identified according to the tracking sequence;

[0031] A second recognition module is configured to obtain, for each tracking sequence, a face recognition sequence corresponding to the tracking sequence, wherein the face recognition sequence includes a plurality of face recognition results;

[0032] A third identification module is used to obtain human body information corresponding to each tracking sequence;

[0033] A data analysis module is used to correct the face recognition sequence of each tracking sequence based on the human body information of the tracking sequence;

[0034] The fusion module is used to perform weighted fusion on all corrected face recognition sequences of the object to be identified based on the temporary identity information to determine the identity information of the object to be identified.

[0035] The face recognition result includes identity information and a corresponding first face recognition score, and the face recognition results are sorted from high to low according to the first face recognition score in the face recognition sequence;

[0036] The data analysis module is further used to normalize the first face recognition score in the face recognition sequence for each tracking sequence to obtain a second face recognition score, where the normalized face recognition result includes the identity information and the second face recognition score; and is further used to correct each of the second face recognition scores in the face recognition sequence of the tracking sequence using the human body information of the tracking sequence.

[0037] Among them, the second face recognition score is corrected, and the data analysis module is used to determine, for each face recognition result after normalization, a first probability that the object to be identified is the identity information in the face recognition result; and is also used to determine, based on the human body information, a posterior probability that the object to be identified is the identity information in the face recognition result; and based on the first probability and the posterior probability, the second face recognition score in the face recognition result is corrected to obtain a third face recognition score.

[0038] Among them, the identity information of the object to be identified is determined, and the fusion module is used to merge all the corrected face recognition sequences of the object to be identified according to the temporary identity information; perform weighted fusion processing on each third face recognition score in the merged face recognition sequence to obtain a corresponding fourth face recognition score; select the face recognition result to which the fourth face recognition score that meets the threshold belongs, and determine the identity information contained in the face recognition result as the identity information of the object to be identified.

[0039] In the above scheme, the face recognition result of the object to be identified is corrected by the body information of the object to be identified, and the identity information of the object to be identified is determined based on all the corrected face recognition results. In this way, when the face is blocked, although multiple key point data used for face recognition are lost, the face recognition result is constrained by the body information, which can effectively improve the recall rate of the face recognition result; at the same time, all face recognition results of the same object are associated with each other according to the temporary identity information for weighted fusion processing, which can effectively reduce the false recognition rate of the face, thereby improving the accuracy of face recognition as a whole. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A schematic diagram of a face recognition method flow chart provided by an example of the present disclosure is shown;

[0041] Figure 2 A schematic diagram of a process for correcting face recognition results provided by an example of the present disclosure is shown;

[0042] Figure 3 A schematic diagram of a process for weighted fusion processing of face recognition results provided by an example of the present disclosure is shown;

[0043] Figure 4 A schematic diagram of the correction and weighted fusion process of a face recognition sequence provided by an example of the present disclosure is shown;

[0044] Figure 5 A schematic diagram of the structure of a face recognition device provided by an example of the present disclosure is shown. DETAILED DESCRIPTION

[0045] In order to make the purpose, features, and advantages of this application more obvious and easy to understand, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts shall fall within the scope of protection of this application.

[0046] In order to improve the recognition accuracy of occluded faces, Figure 1 As shown, the present application provides a face recognition method, comprising:

[0047] Step 101: Identify the captured video image to obtain multiple tracking sequences of the object to be identified, wherein the tracking sequences include multiple video frames.

[0048] When performing face recognition in the present disclosure, it is necessary to capture a video of the subject to be recognized and analyze the video to obtain multiple video frames. A human tracking algorithm can be used to determine multiple video frames belonging to the same subject in the video. These multiple video frames can then be used to form multiple tracking sequences for the subject. For example, a tracking sequence can be formed by combining consecutive video frames or video frames captured by different video capture devices, although this disclosure is not limited to these.

[0049] Step 102: Acquire temporary identity information of the object to be identified according to the tracking sequence.

[0050] By using the person re-identification algorithm to identify each tracking sequence of the same object to be identified, a unique humanID (i.e., temporary identity information) is generated for the object to be identified. In this way, multiple tracking sequences of the object to be identified can be associated through the humanID.

[0051] Step 103: For each tracking sequence, obtain a face recognition sequence and body information corresponding to the tracking sequence, wherein the face recognition sequence includes multiple face recognition results.

[0052] For each tracking sequence, a face recognition result can be obtained by using a face recognition algorithm, and human body information can be obtained by using a human body recognition algorithm.

[0053] Multiple video frames in a tracking sequence are identified using a face recognition algorithm. Recognizing the facial image in each frame yields a corresponding face recognition result, thereby obtaining multiple face recognition results corresponding to the tracking sequence. In the present disclosure, the multiple face recognition results corresponding to a tracking sequence can be combined into a face recognition sequence. Thus, each tracking sequence can correspond to a face recognition sequence.

[0054] By using a human body recognition algorithm to identify multiple video frames in a tracking sequence, human body information corresponding to the tracking sequence can be obtained.

[0055] Step 104: For each tracking sequence, modify the face recognition sequence of the tracking sequence using the human body information of the tracking sequence.

[0056] In the present disclosure, for each tracking sequence, it is necessary to correct each face recognition result of the tracking sequence using the human body information of the tracking sequence. The correction process will be explained through subsequent examples and will not be repeated here.

[0057] Step 105: Based on the temporary identity information, weighted fusion is performed on all corrected face recognition sequences of the object to be identified to determine the identity information of the object to be identified.

[0058] Since all tracking sequences of the object to be identified can be associated through temporary identity information, after the face recognition sequence corresponding to each tracking sequence is corrected, all corrected face recognition sequences can be weighted fused to determine the identity information of the object to be identified.

[0059] In the above scheme, the face recognition result of the object to be identified is corrected by the body information of the object to be identified, and the identity information of the object to be identified is determined based on all the corrected face recognition results. In this way, when the face is blocked, although multiple key point data used for face recognition are lost, the face recognition result is constrained by the body information, which can effectively improve the recall rate of the face recognition result; at the same time, all face recognition results of the same object are associated with each other according to the temporary identity information for weighted fusion processing, which can effectively reduce the false recognition rate of the face, thereby improving the accuracy of face recognition as a whole.

[0060] It should be noted that there is no fixed execution order between the above steps 102 and 103. Based on a tracking sequence, the pedestrian re-identification algorithm, the face recognition algorithm and the body recognition algorithm can be executed simultaneously without interfering with each other.

[0061] In one example, after a tracking sequence is identified using a face recognition algorithm, the resulting face recognition sequence includes multiple face recognition results, each of which includes identity information and a face recognition score. The face recognition score indicates the confidence level in identifying the face as that identity, with a higher face recognition score indicating a greater likelihood that the face is that identity. A face recognition sequence may include multiple face recognition results with the same identity information. The face recognition scores of these face recognition results with the same identity information can then be merged (e.g., by taking the average of the multiple face recognition scores, taking the sum of the multiple face recognition scores, or performing a weighted summation, etc., which is not limited in this disclosure) to obtain a single face recognition score. This face recognition score and the corresponding identity information form a new face recognition result. After the aforementioned merging process, the k face recognition results with the highest face recognition scores are selected to form a new face recognition sequence. In this new face recognition sequence, the k face recognition results can be sorted from high to low according to their face recognition scores, and the identity information in each of the k face recognition results is different. Unless otherwise specified, the face recognition sequences described herein are all the new face recognition sequences. For the convenience of subsequent description, the face recognition score included in each face recognition result in the face recognition sequence is recorded as the first face recognition score.

[0062] Based on this, in one example, multiple face recognition sequences can be used as values ​​and humanID as keys and stored locally for use in subsequent calculations.

[0063] For the convenience of description, a face recognition sequence of the object to be recognized is recorded as: {x1, x2, ..., x i ..., x k-1 , x k}, the face recognition sequence contains k face recognition results arranged from high to low according to the face recognition score, where x i ={fid i :conf i}, x i The i-th face recognition result, including: fid i is x i Identity information and conf i is x i The first face recognition score.

[0064] For example, the face recognition sequence is: {(A: 0.8), (B: 0.78), (C: 0.75)}, which means: the score of the identity of the object to be identified is 0.8 for A, 0.78 for B, and 0.75 for C, so the possibility that the identity of the object is A is the greatest.

[0065] Based on the above face recognition sequence, in one example, for each tracking sequence described in step 104, the face recognition sequence of the tracking sequence is corrected using the human body information of the tracking sequence, such as Figure 2 As shown, including:

[0066] Step 201 : normalize the first face recognition score in the face recognition sequence to obtain a second face recognition score. The normalized face recognition result includes the identity information and the second face recognition score.

[0067] In this example, the following formula (1) is used to identify the face sequence {x1, x2, ..., x i ..., x k-1 , x k}Perform normalization:

[0068]

[0069] Among them, conf i ^ represents the face recognition score of the normalized face recognition result of the i-th face in the face recognition sequence, recorded as the second face recognition score, conf i The first face recognition score in the i-th face recognition result in the face recognition sequence.

[0070] In one example, the normalized face recognition sequence is recorded as in, represents the normalized i-th face recognition result in the face recognition sequence, Represents the second face recognition score in the i-th face recognition result.

[0071] Step 202: Correct each of the second face recognition scores in the face recognition sequence of the tracking sequence using the human body information of the tracking sequence.

[0072] In the present disclosure, the human body information of the object to be identified includes m human body feature values, which can be recorded as a human body information sequence {y1, y2, ..., y j ,...,y m-1 ,y m}, where y j Represents the jth human feature value. For example, the human information sequence of the object to be identified is: {30, H, 1}. According to the preset human information dimensions: age, hat color, and backpack presence, the age feature value 30 indicates 30 years old, the hat feature value H indicates yellow, and the backpack feature value 0 indicates no backpack.

[0073] Based on this, the correction process described in this example includes:

[0074] First, for each normalized face recognition result, determine the first probability that the object to be recognized is the identity information in the face recognition result, and determine the posterior probability that the object to be recognized is the identity information in the face recognition result based on the body information;

[0075] Then, the second face recognition score in the face recognition result is corrected according to the first probability and the posterior probability to obtain a third face recognition score.

[0076] The implementation process of determining the posterior probability that the object to be identified is the identity information in the face recognition result based on the human body information includes:

[0077] Obtain the historical human body information corresponding to the identity information in the face recognition result;

[0078] For each dimension in the human body information, obtain all historical feature values ​​corresponding to the dimension from the historical human body information, and determine a first number of all historical feature values;

[0079] determining a second number of historical feature values ​​that are identical to the feature value of the dimension in the human body information;

[0080] Taking the ratio of the second quantity to the first data amount as the probability of the dimension;

[0081] The probabilities of each dimension in the human body information are multiplied to obtain the posterior probability that the object to be identified is the identity information in the face recognition result.

[0082] The above correction process can be achieved through the following formula:

[0083]

[0084]

[0085] in:

[0086] is the face recognition score in the corrected i-th face recognition result, recorded as the third face recognition score.

[0087] The object to be identified is The probability of identity information in , that is, the first probability mentioned above, is calculated as follows: The ratio of the second face recognition score in the face recognition sequence to the sum of all the second face recognition scores in the face recognition sequence. For example, if the face recognition sequence is {(A: 0.8), (B: 0.78), (C: 0.75)}, then the probability that the object to be recognized is A is: 0.8 / (0.8+0.78+0.75)=34%;

[0088] The object to be identified is determined based on the human body information of the object to be identified. The posterior probability of the identity information in is calculated as follows: The historical human information corresponding to the identity information in y j The dimension of the dimension is obtained from the historical human body information, and the historical feature value of the dimension is calculated as y j The ratio of the number of times (the first number) to the number of all historical feature values ​​of the dimension (the second number) is used to obtain the probability corresponding to the dimension. The probability of each dimension is multiplied to obtain the object to be identified. The posterior probability of the identity information in the above human body information sequence. For example, the second eigenvalue in the above human body information sequence is H, indicating that the hat color is yellow. Then, combined with the face recognition sequence {(A: 0.8), (B: 0.78), (C: 0.75)}, the historical human body information of the identity information A is obtained for the first face recognition result. Among them, the eigenvalue dimension "hat color" has a total of 10 eigenvalues ​​(the second number), of which 3 are yellow (the first number) and 7 are blue. Then, the probability that the "hat color" is yellow is 30%. And so on, the corresponding probabilities of other dimensions are calculated, and the probabilities of each dimension are multiplied to obtain the posterior probability that the human body information of the object to be identified belongs to A.

[0089] In one example, the corrected face recognition sequence is recorded as is the corrected face recognition result of the i-th person,

[0090] Misidentification of faces in open environments is mostly caused by poor lighting conditions, large face angles, motion blur, etc. Therefore, correcting the face recognition results based on human body information can effectively improve the recall rate of face recognition, but it may also introduce misidentification. In order to reduce misidentification, this disclosure performs weighted fusion on all corrected face recognition sequences of the object to be identified, and then determines the identity information of the object to be identified.

[0091] When performing weighted fusion, a corresponding weight is set for each face recognition sequence of the object to be identified. The weight can be set based on empirical values ​​or according to different face recognition devices. For example, if identification is performed through multiple face recognition devices, the weight of each face recognition sequence can be set according to the corresponding face recognition device. This disclosure does not limit this.

[0092] In one example, the above step 105 performs weighted fusion on all the corrected face recognition sequences of the object to be identified based on the temporary identity information to determine the identity information of the object to be identified, such as Figure 3 As shown, including:

[0093] Step 301: Merge all corrected face recognition sequences of the object to be identified according to the temporary identity information.

[0094] First, all the face recognition sequences of the objects to be identified after correction are merged. The merged face recognition sequence is recorded as: Contains K corrected face recognition results, is the oth modified face recognition result, K is the sum of the number of face recognition results contained in all face recognition sequences. In the merged face recognition sequence, the corrected third face recognition score can be used. Sort the K face recognition results from high to low.

[0095] Step 302: Perform weighted fusion processing on each third face recognition score in the combined face recognition sequence to obtain a corresponding fourth face recognition score.

[0096] Based on the merged face recognition sequence, each face recognition score is weighted and fused again using the following formula, that is, it is corrected again:

[0097]

[0098] Among them, conf′ o is the face recognition score of the oth face recognition result after weighted fusion processing, recorded as the fourth face recognition score, p is the number of face recognition sequences of the object to be recognized (that is, the number of tracking sequences of the object to be recognized), C i is the weight of the i-th face recognition sequence in the p-person face recognition sequence.

[0099] Step 303 : Select the face recognition result to which the fourth face recognition score that meets the threshold belongs, and determine the identity information included in the face recognition result as the identity information of the object to be recognized.

[0100] If the fourth face recognition score conf′ o If the threshold is met, the conf′ o The identity information contained in the face recognition result is used as the identity information of the object to be identified, that is,

[0101] x′ o ={fid o :conf′o fid in} o The identity information of the object to be identified. If there are multiple conf' o If both meet the threshold, the conf′ with the maximum value is selected o The identity information contained in the face recognition result is used as the identity information of the object to be identified.

[0102] The following is an example to illustrate the above correction process. Figure 4 As shown, including:

[0103] 1. Get a face recognition sequence {x1, x2, ..., x i ..., x k-1 , x k}, where x i ={fid i :conf i For each first face recognition score conf in the face recognition sequence i Perform normalization to obtain the corresponding second face recognition score

[0104] 2. The normalized face recognition sequence is recorded as in,

[0105] 3. Obtain the human body information {y1, y2, ..., y j ,...,y m-1 ,y m}.

[0106] 4. By {y1, y2, ..., y j ,...,y m-1 ,y m}right Each second face recognition score in conf i ^ Correction is performed to obtain the third person face recognition score

[0107] 5. The corrected face recognition sequence is recorded as in,

[0108] 6. The p-person face recognition sequence {x1, x2, ..., x i ..., x k-1 , x k}, for each face recognition sequence {x1, x2, ..., x i ..., xk-1 , x k}Execute steps 1-5 above to get p face recognition sequences For each Set the corresponding weight, then p face recognition sequences The corresponding weights are C1~C p .

[0109] 7. P face recognition sequence Merge to get the face recognition sequence For each third face recognition score Perform correction again, that is, weighted fusion processing, to obtain the fourth face recognition score conf′ o , select conf′ that meets the threshold o , then x′ o ={fid o :conf′ o fid in} o This is the identity information of the object to be identified.

[0110] In order to implement the above method, Figure 5 As shown, the present disclosure also provides a face recognition device, comprising:

[0111] The tracking module 10 is used to identify the collected video images and obtain multiple tracking sequences of the object to be identified. The tracking sequences include multiple video frames.

[0112] The first recognition module 20 is configured to obtain temporary identity information of the object to be recognized based on the tracking sequence. In one example, the first recognition module 20 may use a person re-identification algorithm to obtain the temporary identity information of the object to be recognized.

[0113] The second recognition module 30 is configured to obtain, for each tracking sequence, a face recognition sequence corresponding to the tracking sequence, wherein the face recognition sequence includes multiple face recognition results. In one example, the second recognition module 30 may use a face recognition algorithm to obtain the face recognition sequence corresponding to the tracking sequence.

[0114] The third recognition module 40 is configured to obtain, for each tracking sequence, human body information corresponding to the tracking sequence. In one example, the third recognition module 40 may use a human body recognition algorithm to obtain human body information corresponding to the tracking sequence.

[0115] The data analysis module 50 is configured to modify the face recognition sequence of each tracking sequence using the human body information of the tracking sequence.

[0116] The fusion module 60 is configured to perform weighted fusion on all corrected face recognition sequences of the object to be identified based on the temporary identity information, so as to determine the identity information of the object to be identified.

[0117] The face recognition result includes identity information and a corresponding first face recognition score, and the face recognition results are sorted from high to low according to the first face recognition score in the face recognition sequence;

[0118] The data analysis module 50 is further used to normalize the first face recognition score in the face recognition sequence for each tracking sequence to obtain a second face recognition score, where the normalized face recognition result includes the identity information and the second face recognition score; and is further used to correct each of the second face recognition scores in the face recognition sequence of the tracking sequence using the human body information of the tracking sequence.

[0119] In one example, when correcting the second face recognition score, the data analysis module 50 is used to determine, for each face recognition result after normalization, a first probability that the object to be identified is the identity information in the face recognition result; and is also used to determine, based on the human body information, a posterior probability that the object to be identified is the identity information in the face recognition result; and based on the first probability and the posterior probability, correct the second face recognition score in the face recognition result to obtain a third face recognition score.

[0120] In one example, when determining the identity information of the object to be identified, the fusion module 60 is used to merge all the corrected face recognition sequences of the object to be identified based on the temporary identity information; perform weighted fusion processing on each third face recognition score in the merged face recognition sequence to obtain a corresponding fourth face recognition score; select the face recognition result to which the fourth face recognition score that meets the threshold belongs, and determine the identity information contained in the face recognition result as the identity information of the object to be identified.

[0121] In one example, the present disclosure also provides an electronic device including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; the memory is used to store computer programs; and the processor is used to implement the method steps of face recognition when executing the program stored in the memory.

[0122] In one example, the present disclosure further provides a computer storage medium, characterized in that the storage medium stores computer-executable instructions, and when the instructions are executed, the method for performing face recognition is performed.

[0123] The computer-readable storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can, for example, include but is not limited to a system, device or component of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination thereof. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0124] In addition to the above-mentioned methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the steps of the method according to various embodiments of the present application described in the above-mentioned "Exemplary Method" section of this specification.

[0125] The computer program product may be written in any combination of one or more programming languages ​​to implement the program code for performing the operations of the embodiments of the present application, including object-oriented programming languages ​​such as Java, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0126] In the description of this specification, the reference terms "one embodiment," "some embodiments," "example," "specific example," or "some examples" mean that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials, or characteristics described may be combined in any appropriate manner in any one or more embodiments or examples. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification, as well as features of different embodiments or examples, unless they are mutually inconsistent.

[0127] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. Throughout the description of this application, "plurality" means two or more, unless otherwise specifically defined.

[0128] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A face recognition method, comprising: Identify the captured video image to obtain multiple tracking sequences of the object to be identified, wherein the tracking sequences include multiple video frames, and the multiple tracking sequences are from different video capture devices; Acquiring temporary identity information of the object to be identified according to the tracking sequence; And, for each tracking sequence, obtaining a face recognition sequence and human body information corresponding to the tracking sequence, wherein the face recognition sequence includes a plurality of face recognition results; For each tracking sequence, the face recognition sequence of the tracking sequence is corrected using the human body information of the tracking sequence; According to the temporary identity information, all corrected face recognition sequences of the object to be identified are weightedly fused to determine the identity information of the object to be identified.

2. The method according to claim 1, wherein The face recognition results include identity information and a corresponding first face recognition score, and the face recognition results are sorted from high to low according to the first face recognition score in the face recognition sequence.

3. The method according to claim 2, wherein: For each tracking sequence, the face recognition sequence of the tracking sequence is corrected using the human body information of the tracking sequence, including: Normalizing the first face recognition score in the face recognition sequence to obtain a second face recognition score, wherein the normalized face recognition result includes the identity information and the second face recognition score; Each of the second face recognition scores in the face recognition sequence of the tracking sequence is corrected using the human body information of the tracking sequence.

4. The method according to claim 3, wherein: The correcting each second face recognition score in the face recognition sequence of the tracking sequence using the human body information of the tracking sequence includes: For each normalized face recognition result, determining a first probability that the object to be recognized is the identity information in the face recognition result; Determining, based on the human body information, a posterior probability that the object to be identified is the identity information in the face recognition result; The second face recognition score in the face recognition result is corrected according to the first probability and the posterior probability to obtain a third face recognition score.

5. The method according to claim 4, wherein The human body information includes a plurality of eigenvalues, each eigenvalue having a corresponding dimension; Determining, based on the human body information, a posterior probability that the object to be identified is the identity information in the face recognition result, including: Obtain the historical human body information corresponding to the identity information in the face recognition result; For each dimension in the human body information, obtain all historical feature values ​​corresponding to the dimension from the historical human body information, and determine a first number of all historical feature values; determining a second number of historical feature values ​​that are identical to the feature value of the dimension in the human body information; taking the ratio of the second number to the first number as the probability of the dimension; The probabilities of each dimension in the human body information are multiplied to obtain the posterior probability that the object to be identified is the identity information in the face recognition result.

6. The method according to claim 4, wherein, based on the temporary identity information, weighted fusion is performed on all corrected face recognition sequences of the subject to be identified to determine the identity information of the subject to be identified, comprising: Merging all corrected face recognition sequences of the object to be identified according to the temporary identity information; Performing weighted fusion processing on each third face recognition score in the merged face recognition sequence to obtain a corresponding fourth face recognition score; A face recognition result to which a fourth face recognition score that meets the threshold is assigned is selected, and identity information included in the face recognition result is determined as the identity information of the object to be identified.

7. A face recognition device comprising: A tracking module is used to identify the collected video image and obtain multiple tracking sequences of the object to be identified, wherein the tracking sequences include multiple video frames, and the multiple tracking sequences are from different video acquisition devices; A first identification module, configured to obtain temporary identity information of the object to be identified according to the tracking sequence; A second recognition module is configured to obtain, for each tracking sequence, a face recognition sequence corresponding to the tracking sequence, wherein the face recognition sequence includes a plurality of face recognition results; A third identification module is used to obtain human body information corresponding to each tracking sequence; A data analysis module is used to correct the face recognition sequence of each tracking sequence based on the human body information of the tracking sequence; The fusion module is used to perform weighted fusion on all corrected face recognition sequences of the object to be identified based on the temporary identity information to determine the identity information of the object to be identified.

8. The apparatus according to claim 7, wherein the face recognition result comprises identity information and a corresponding first face recognition score, and the face recognition results are sorted in the face recognition sequence from high to low according to the first face recognition score; The data analysis module is further configured to normalize the first face recognition score in the face recognition sequence for each tracking sequence to obtain a second face recognition score, wherein the normalized face recognition result includes the identity information and the second face recognition score; It is also used to correct each of the second face recognition scores in the face recognition sequence of the tracking sequence using the human body information of the tracking sequence.

9. The device according to claim 8 corrects the second face recognition score, and the data analysis module is used to determine, for each face recognition result after normalization, a first probability that the object to be identified is the identity information in the face recognition result; and is also used to determine, based on the human body information, a posterior probability that the object to be identified is the identity information in the face recognition result; and based on the first probability and the posterior probability, correct the second face recognition score in the face recognition result to obtain a third face recognition score.

10. The device according to claim 8, wherein Determining the identity information of the object to be identified, the fusion module is used to merge all the corrected face recognition sequences of the object to be identified based on the temporary identity information; performing weighted fusion processing on each third face recognition score in the merged face recognition sequence to obtain a corresponding fourth face recognition score; A face recognition result to which a fourth face recognition score that meets the threshold is assigned is selected, and identity information included in the face recognition result is determined as the identity information of the object to be identified.

Images