A method, apparatus, device, and medium for identifying a personal file

Abstract

An embodiment of the present application provides a method, device, equipment and medium for identifying a person file. In this method, a first target step length of a target person and a first target type identifier corresponding to the gait in the image collected by an image acquisition device are obtained, and a candidate person file matching the first target step length and the first target type identifier is searched in a database. If there is a standard image in the candidate person file whose similarity to the image is greater than a set similarity threshold, the person file to which the standard image with a similarity greater than the set similarity threshold belongs is output. In the embodiment of the present application, the walking habit of each person is used as a screening condition for person file recognition, narrowing the comparison range of images, thereby improving the efficiency of person file recognition.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular, to a method, device, equipment and medium for identifying a person's file. Background Art

[0002] With the rapid development of face recognition technology, the application scenarios of face recognition technology have been greatly broadened. Whether at home or abroad, relevant departments have built a security business system centered on face recognition by deploying a large number of image acquisition devices during work. Face recognition plays a crucial role in maintaining social security. However, with the continuous operation of the security business system, the captured images are continuously accumulated, accumulating billions or even tens of billions of images. These images are stored in the database and become "sleeping" data, resulting in low application value of the data.

[0003] In order to improve the application value of the accumulated images, the images of the same person are archived to form a person's file with a "file-image" hierarchical ownership relationship, and data retrieval and application expansion are carried out based on the person's file, which can improve the application value of the images. The person's file contains the identity information of the person corresponding to the person's file and the image set of the person being captured.

[0004] In the prior art, usually, a 1:N comparison is made between the image to be recognized and the already archived person's files. For each image to be recognized, a face feature comparison is made with the standard images stored in the existing person's files respectively to determine the person's file to which the image belongs, so as to realize the recognition of the person's file.

[0005] However, since the number of images that need to be compared for face features is very large, a large amount of computing resources are required for face recognition, resulting in a long time required for face recognition and greatly reducing the efficiency of face file recognition. Summary of the Invention

[0006] The embodiments of the present application provide a method, device, equipment and medium for identifying a person's file, which are used to solve the problem of low efficiency in identifying a face file in the prior art.

[0007] In a first aspect, the present application provides a method for identifying a person's file, and the method includes:

[0008] Receiving an image collected by an image acquisition device, inputting the image into a pre-trained classification model, and determining a first target step length and a first target type identifier corresponding to the gait of the target person in the image;

[0009] According to the step lengths corresponding to each person's file recorded in the database and the type identifiers corresponding to the gait, search in the database for candidate person files that match the first target step length and / or the first target type identifier;

[0010] If there is a standard image in the candidate person file whose similarity to the image is greater than a set similarity threshold, output the person file to which the standard image with a similarity greater than the set similarity threshold belongs.

[0011] In a second aspect, the present application further provides a person file recognition device, the device includes:

[0012] A determination module, configured to receive an image collected by an image acquisition device, input the image into a pre-trained classification model, and determine the first target step length of the target person in the image and the first target type identifier corresponding to the gait;

[0013] A search module, configured to search in the database for candidate person files that match the first target step length and / or the first target type identifier according to the step lengths corresponding to each person's file recorded in the database and the type identifiers corresponding to the gait;

[0014] An identification module, configured to output the person file to which the standard image with a similarity greater than the set similarity threshold belongs if there is a standard image in the candidate person file whose similarity to the image is greater than the set similarity threshold.

[0015] In a third aspect, the present application further provides an electronic device, the electronic device at least includes a processor and a memory, and the processor is configured to implement the steps of any one of the above-mentioned person file recognition methods when executing a computer program stored in the memory.

[0016] In a fourth aspect, the present application further provides a computer-readable storage medium, which stores a computer program, and the computer program is configured to implement the steps of any one of the above-mentioned person file recognition methods when executed by a processor.

[0017] The embodiments of the present application provide a method, apparatus, device and medium for identifying a personal file. In this method, an image collected by an image acquisition device is received, and the image is input into a pre-trained classification model to determine the first target step length of the target person in the image and the first target type identifier corresponding to the gait. According to the step length and type identifier corresponding to each personal file recorded in the database, a candidate personal file matching the first target step length and the first target type identifier is searched in the database. If there is a standard image in the candidate personal file whose similarity to the said image is greater than a set similarity threshold, the personal file to which the standard image with a similarity greater than the set similarity threshold belongs is output. Since in the embodiments of the present application, the first target step length of the target person in the image collected by the image acquisition device and the first target type identifier corresponding to the gait are obtained, and a candidate personal file matching the first target step length and the first target type identifier is searched in the database. In the embodiments of the present application, the walking habit of each person is used as a screening condition for personal portrait file identification, narrowing the comparison range of the image, thereby improving the efficiency of personal portrait file identification. Description of the Drawings

[0018] In order to more clearly illustrate the technical solutions of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0019] Figure 1 Schematic diagram of the personal file identification process provided by the embodiments of the present application;

[0020] Figure 2 Schematic diagram of the prompt for the personal file identification process provided by the embodiments of the present application;

[0021] Figure 3 Schematic diagram of the training process of the generative adversarial model including the classification model provided by the embodiments of the present application;

[0022] Figure 4 Schematic diagram of the structure of the personal file identification device provided by the embodiments of the present application;

[0023] Figure 5 Schematic diagram of the structure of an electronic device provided by the embodiments of the present application. Detailed Embodiments

[0024] In order to make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in this application fall within the scope of protection of this application.

[0025] The embodiments of this application provide a method, device, equipment, and medium for identifying a person's file. In this method, an image collected by an image acquisition device is received, and the image is input into a pre-trained classification model to determine the first target step length of the target person in the image and the first target type identifier corresponding to the gait. According to the step length and type identifier corresponding to each person's file recorded in the database, a candidate person's file that matches the first target step length and the first target type identifier is searched for in the database. If there is a standard image in the candidate person's file whose similarity to the said image is greater than the set similarity threshold, the person's file to which the standard image with a similarity greater than the set similarity threshold belongs is output. Since in the embodiments of this application, the first target step length of the target person in the image collected by the image acquisition device and the first target type identifier corresponding to the gait are obtained, and a candidate person's file that matches the first target step length and the first target type identifier is searched for in the database. In the embodiments of this application, the walking habit of each person is used as a screening condition for identifying the portrait file, narrowing the comparison range of the image, thereby improving the efficiency of portrait file identification.

[0026] Embodiment 1:

[0027] Figure 1 It is a schematic diagram of the process for identifying a person's file provided by the embodiments of this application. This process specifically includes the following steps:

[0028] S101: Receive the image collected by the image acquisition device, and input the said image into a pre-trained classification model to determine the first target step length of the target person in the image and the first target type identifier corresponding to the gait.

[0029] The process for identifying a person's file provided by the embodiments of this application is applied to an electronic device, and this electronic device can be a device such as a server or a PC.

[0030] When receiving the image collected by the image acquisition device, in order to be able to determine the person's file to which the target person in the image belongs, a pre-trained classification model can be used to identify the target person in the image. This classification model is pre-trained and can identify the step length and gait of the person in the image, and can also identify information such as the orientation and actions of the person in the image. Input the said image into a pre-trained classification model to determine the first target step length of the target person in the image and the first target type identifier corresponding to the gait.

[0031] S102: According to the step length corresponding to each person's file recorded in the database and the type identifier corresponding to the gait, search in the database for candidate person files that match the first target step length and / or the first target type identifier.

[0032] In order to determine the person file to which the target person in the image belongs, after determining the first target step length of the target person in the image and the first target type corresponding to the gait based on a pre-trained classification model, a candidate person file that matches the first target step length and / or the first target type identifier of the target person can be searched for in the database.

[0033] The database stores the person files of each person. The person file contains the identity information of the person and the image set captured of the person, and can also save other information of the person, specifically including the step length of the person and the type identifier corresponding to the gait. This other information can also include information such as the walking speed of the person and the identifier corresponding to the portrait pose. For the convenience of description, these other information can be referred to as implicit attribute information.

[0034] According to the step length corresponding to each person's file recorded in the database and the type identifier corresponding to the gait, a candidate person file that matches the first target step length and / or the first target type identifier can be searched for in the database.

[0035] If during the search process, the first target step length is the same as the step length corresponding to a certain person file recorded in the database, or the difference between the first target step length and the step length corresponding to a certain person file is less than a preset step length threshold, it can be considered that the step length corresponding to this person file matches the first target step length, and this person file can be determined as a candidate person file of the target person.

[0036] If during the search process, the first target type identifier is the same as the type identifier corresponding to the gait included in the other information of a certain person file recorded in the database, it can be considered that the type identifier corresponding to the gait included in the other information of this person file matches the first target type identifier, and this person file can be determined as a candidate person file of the target person.

[0037] In order to improve the efficiency and accuracy of person file recognition, preferably, if both the first target step length and the first target type match a certain person file recorded in the database, this person file is determined as the candidate file of the target person.

[0038] S103: If there is a standard image in the candidate person file whose similarity to the image is greater than a set similarity threshold, output the person file to which the standard image with a similarity greater than the set similarity threshold belongs.

[0039] After determining the candidate person files corresponding to the target person, in order to accurately identify the person file of the target person, for each determined candidate person file, the standard image stored in the candidate person file can be obtained, and based on the portrait recognition technology, it can be determined whether the similarity between the standard image and the image collected by the image acquisition device is greater than the set similarity threshold. If the similarity between the standard image and the image collected by the image acquisition device is greater than the set similarity threshold, it can be considered that the person corresponding to the person file to which the standard image belongs is the target person, that is, the person file to which the standard image belongs is the person file of the target person, and this person file can be output.

[0040] Since in the embodiment of the present application, the first target step length and the first target type identifier corresponding to the gait of the target person in the image collected by the image acquisition device are obtained, and the candidate person files matching the first target step length and the first target type identifier are searched in the database. In the embodiment of the present application, the walking habit of each person is used as the screening condition for portrait file recognition, narrowing the comparison range of the images, thereby improving the efficiency of portrait file recognition.

[0041] Embodiment 2:

[0042] In order to improve the efficiency of person file recognition, on the basis of the above embodiment, in the embodiment of the present application, before outputting the person file to which the comparison image with a similarity greater than the set similarity threshold belongs if there is a comparison image in the candidate person file with a similarity greater than the set similarity threshold to the image, the method includes:

[0043] Obtain the time and location when the image was captured;

[0044] Input the captured location into the pre-trained spatio-temporal graph modeling model to obtain the target location identifier output by the spatio-temporal graph modeling model;

[0045] Obtain the time identifier corresponding to the preset time range corresponding to the target location identifier, and determine the target time identifier corresponding to the time range to which the captured time belongs according to the captured time and the time identifier;

[0046] According to the time identifier and location identifier corresponding to each person file recorded in the database, search in the database for candidate person files matching the target time identifier and the target location identifier.

[0047] Since the implicit attribute information of the same target person at the same time and place is usually similar or consistent. For example, on the way to work, the facing direction, step length, walking speed, and gait of a certain target person are usually relatively consistent. Therefore, in order to improve the efficiency of portrait file recognition, the scope of portrait file recognition can be further narrowed before similarity comparison.

[0048] In the embodiment of the present application, the captured time and location of the image collected by the image acquisition device are obtained. The captured time obtained can be a specific time or an identifier corresponding to the captured time. The captured location can be the longitude and latitude information of the installation location of the image acquisition device or the road name of the installation location of the image acquisition device.

[0049] After obtaining the captured time and location of the image, the captured location can be input into a pre-trained spatio-temporal graph modeling (GraphWavenet) model, and the spatio-temporal graph modeling model is used to determine the target location identifier corresponding to the input captured location.

[0050] In the embodiment of the present application, the spatio-temporal graph modeling model is a pre-trained unsupervised model. This model can accurately capture the hidden spatial dependence relationships in the input data. When the captured location of the image is input into the pre-trained spatio-temporal graph modeling model, the spatio-temporal graph modeling model can output the target location identifier of the captured location of the image.

[0051] When the captured location is input into the spatio-temporal graph modeling model, the spatio-temporal graph modeling model can extract the association relationship between the captured location and each other location according to the pre-trained association relationship between locations. The location identifier corresponding to the location where the association relationship is greater than the preset distance threshold is determined as the target location identifier of the captured location. The association relationship can be the straight-line distance between portrait acquisition devices or the walking distance of the road between portrait acquisition devices. The specific training process of the spatio-temporal graph modeling model is prior art and will not be elaborated here.

[0052] After determining the target location identifier corresponding to the captured location, the scope of person file recognition can be further determined according to the captured time. In the embodiment of the present application, the time division rule corresponding to the target location can be obtained according to the target location identifier. The time division rule prescribes the time identifiers corresponding to different time ranges in advance. According to the captured time and the time identifiers corresponding to different time ranges corresponding to the target location identifier, the target time identifier corresponding to the time range to which the captured time belongs can be determined.

[0053] In the embodiment of the present application, a time range is preset. Since the time of a day is fixed, a day can be divided according to the preset time range, and the time periods corresponding to each time range are identified. When dividing a day according to the preset time range, it can be divided according to equal time lengths, or according to the number of people in different time periods, it can be divided according to unequal time lengths. It can also be divided according to different road sections according to the number of people in different time periods on that road section according to unequal time lengths. According to the time when the image is captured and the time identifier corresponding to the preset time range, the target time identifier corresponding to the time range to which the captured time belongs can be determined.

[0054] Preferably, in the embodiment of the present application, the preset time range is adjusted according to the tidal law of human activities, which can be understood as presetting multiple time ranges with unequal time lengths. For example, during the morning and evening rush hours, the number of people is large, and congestion is extremely likely to occur on some road sections. The number of people passing through per unit time is less than that in other time periods. Then, for the congested road sections during the morning and evening rush hours, the time range can be set to α 1 , and during non-morning and evening rush hours, the time range can be set to α 2 , where α 1 >α 2 . For example, for a 24-hour day on Road Section 1, the number of people is small in the time period from 00:00:00 to 06:59:59. The time range within this time period can be set to 30 minutes. Then, the time identifier corresponding to 00:00:00 to 00:29:59 is 1, the time identifier corresponding to 00:30:00 to 00:59:59 is 2, and the time identifier corresponding to 01:00:00 to 01:29:59 is 3. The time period from 07:00:00 to 08:59:59 is the morning rush hour with a large number of people and is prone to congestion. The time range within this time period can be set to 40 minutes. The setting of the time range and the setting of the time identifier can be preset according to the actual needs of portrait file recognition and will not be elaborated here.

[0055] In addition, for example, the time when the image is captured may be 07:00:00. The time identifier corresponding to the time range to which 07:00:00 belongs is 5, and the time identifier corresponding to the time range to which 06:59:00 belongs is 4. Only determining the target time identifier corresponding to the captured time based on the time range to which 07:00:00 belongs may reduce the accuracy of portrait file recognition. Therefore, in order to improve the accuracy of portrait file recognition, in the embodiment of the present application, a time threshold of β minutes can be preset, and according to the time ranges to which β minutes before and after the captured time belong, the target time identifier corresponding to the captured time is determined.

[0056] After determining the target location identifier corresponding to the captured location and determining the target time identifier corresponding to the captured time according to the time division rule corresponding to the target location, the target location identifier and the target time identifier can be determined as the spatio-temporal domain identifier of the captured image. For example, if the target location identifier is 55 and the target time identifier is 100, the spatio-temporal domain identifier of the captured image can be expressed as (55, 100), where the first number "55" represents the target location identifier corresponding to the captured location, and the second number "100" represents the target time identifier corresponding to the captured time.

[0057] After determining the spatio-temporal domain identifier of the captured image, the range of person profile comparison can be further narrowed down according to the target time identifier and the target location identifier included in the spatio-temporal domain identifier.

[0058] Each person profile in the database not only contains the identity information, the set of portrait captured images and the implicit attribute information of the person corresponding to the person profile, but also includes the spatio-temporal domain identifier of the captured image. For the convenience of description, the spatio-temporal domain identifier of the captured image can be called spatio-temporal domain information. Since the same person may be captured multiple times, the time and location of each captured image are different. Therefore, there may be multiple sets of time identifiers and location identifiers in the spatio-temporal domain information of the person profile, that is, there are multiple spatio-temporal domain identifiers.

[0059] According to the obtained target time identifier and target location identifier, candidate person profiles that match the target time identifier and the target location identifier can be searched in the database. Since different target location identifiers correspond to different time division rules, when matching the target time identifier and the target location identifier, only when the target time identifier and the target location identifier are exactly the same as a set of time identifier and location identifier, can it be considered a successful match. For example, the spatio-temporal domain information of a certain person profile in the database contains multiple spatio-temporal domain identifiers, which are (51, 66), (51, 88), (2, 88) respectively. According to the target time identifier and target location identifier of the captured image, the spatio-temporal domain identifier is determined to be (51, 88), and this person profile can be determined as a candidate person profile.

[0060] To improve the accuracy of person profile recognition, based on the above embodiments, in the embodiments of the present application, before outputting the person profile to which the comparison image with a similarity greater than the set similarity threshold in the candidate person profile belongs, the method includes:

[0061] Obtaining the target identifier corresponding to the target orientation and posture of the target person output by the pre-trained classification model;

[0062] If it is determined that the target identifier corresponding to the target orientation and posture of the target person is consistent with the identifier corresponding to the orientation and posture included in other information of a certain person file recorded in the database, then this person file can be determined as the candidate person file of the target person.

[0063] Since other information of the person file not only includes the type identifier corresponding to the step length and gait of the person corresponding to this person file, but also includes information such as the walking speed of this person and the identifier corresponding to the portrait posture, etc., therefore, in the embodiment of the present application, based on the pre-trained classification model, not only can the type identifier corresponding to the step length and gait be determined, but also information such as the target identifier corresponding to the target orientation and posture of the target person in the image can be determined. After determining the target identifier corresponding to the target orientation and posture of the target person, the candidate person file that matches the target orientation and the target identifier can be searched for in the database according to the target identifier corresponding to the target orientation and posture.

[0064] Specifically, during the process of searching for the candidate person file in the database, if it is determined that the target identifier corresponding to the target orientation and posture of the target person is consistent with the identifier corresponding to the orientation and posture included in other information of a certain person file recorded in the database, then this person file can be determined as the candidate person file of the target person.

[0065] Next, a specific embodiment is combined to illustrate the person file recognition process provided by the embodiment of the present application. Figure 2 It is a schematic diagram for prompting the person file recognition process provided by the embodiment of the present application, and this process specifically includes the following steps:

[0066] S201: Receive the image collected by the image acquisition device, and respectively execute S202 and S203.

[0067] S202: Obtain the time and location when the received image was captured, determine the target time identifier corresponding to the time range to which the captured time belongs and the target location identifier corresponding to the captured location, search for the candidate person file that matches the target time identifier and the target location identifier in the database, and execute S204.

[0068] S203: Input the received image into the pre-trained classification model, determine the first target type identifier corresponding to the first target step length and gait of the target person in this image, search for the candidate person file that matches the first target step length and the first target type identifier in the database, and execute S204.

[0069] S204: If there is a comparison image in the candidate person file whose similarity to the image is greater than the set similarity threshold, then output the person file to which the comparison image with a similarity greater than the set similarity threshold belongs.

[0070] Embodiment 3:

[0071] In order to further improve the efficiency of identifying personal files, based on the above embodiments, in the embodiments of the present application, after determining the candidate personal files corresponding to the target person and before outputting the personal files to which the comparison images with a similarity greater than the set similarity threshold belong, the method includes:

[0072] For each of the candidate personal files, determine the total number of successful matches of the candidate personal file according to the number of information matches between the candidate personal file and each of the information in the first target step length, the first target type identifier, the target time identifier, and the target location identifier;

[0073] Sort each of the candidate personal files according to the total number of successful matches of each of the candidate personal files, and perform subsequent operations on the sorted candidate personal files.

[0074] Although the candidate personal files corresponding to the target person are determined according to the first target step length, the first target type identifier, the target time identifier, and the target location identifier, which narrows the scope of personal file identification to a certain extent, due to the accumulated data information in the database, there are a large number of captured information in each personal file, and the candidate personal files screened and determined according to the first target step length, the first target type identifier, the target time identifier, and the target location identifier are still relatively numerous. Therefore, in order to further improve the efficiency of personal file identification, each candidate personal file can be sorted, and the image is used to match the standard images in each of the candidate personal files in the sorting result in turn.

[0075] Specifically, in the embodiments of the present application, determine the total number of successful matches of the candidate personal file according to the number of information matches between the candidate personal file and the information such as the first target step length, the first target type identifier, the target time identifier, and the target location identifier. The more the total number of successful matches, the greater the possibility that the person corresponding to the candidate personal file is the target person. Therefore, each candidate personal file can be sorted according to the number of successful matches of each candidate personal file, and the candidate personal file with a large number of successful matches is used as the candidate personal file for priority comparison. For the sorted candidate personal files, perform the comparison operation of image similarity in the sorting order.

[0076] For example, three candidate person files of the target person are found in the database. For the convenience of description, these three candidate person files can be called: candidate person file 1, candidate person file 2, and candidate person file 3. Among them, only candidate person file 1 matches successfully with the first target step length, so the total number of successful matches of candidate person file 1 is 1; candidate person file 2 matches successfully with the first target step length, the first target type, the first target time identifier, and the first target location identifier, so the total number of successful matches of candidate person file 2 is 3; candidate person file 3 matches successfully with the first target step length, the first target time identifier, and the first target location identifier, so the total number of successful matches of candidate person file 3 is 2. Sort each candidate person file according to the total number of successful matches of each candidate person file. The order of the sorted candidate person files is candidate person file 2, candidate person file 3, candidate person file 1. Among them, because the number of successful matches of candidate person file 2 is the largest, it means that the person corresponding to candidate person file 2 is more likely to be the target person. In the subsequent person file recognition process, the collected images are sequentially matched with the standard images in each sorted candidate person file.

[0077] Embodiment 4:

[0078] In order to further improve the efficiency of person file recognition, based on the above embodiments, in the embodiment of the present application, after determining the candidate person file corresponding to the target person and before outputting the person file to which the comparison image whose similarity is greater than the set similarity threshold belongs, the method includes:

[0079] Based on the pre-trained classification model, determine the second target step length and the second target type identifier corresponding to the gait of the people in the same row in the image;

[0080] According to the step length and type identifier corresponding to each person file recorded in the database, search in the database for the candidate person file of the people in the same row that matches the second target step length and the second target type identifier of the people in the same row;

[0081] Obtain the candidate person files that are the same in the candidate person files corresponding to the target person and the candidate person files of the people in the same row corresponding to the people in the same row, and perform subsequent operations on the same candidate person files.

[0082] Since the image acquisition device usually performs global shooting when acquiring images, and the acquired images contain multiple people. In the embodiment of the present application, after determining the candidate person file of the target person and before performing the similarity comparison operation, the people in the same row of the target person included in the acquired images can be determined. The process of determining the people in the same row is a prior art and will not be elaborated here.

[0083] Based on the pre-trained classification model, determine the second target step length and the corresponding second target type identifier of the gait of the accompanying person in the image, and search for the candidate person profile of the accompanying person that matches the second target step length and the second target type identifier in the database. Determining the type identifier corresponding to the step length and the gait, and based on the type identifier corresponding to the step length and the gait, determining the candidate person profile have been described in the above other embodiments and will not be elaborated here.

[0084] Since the accompanying person and the target person generally have the same trajectory, the candidate person profile of the accompanying person determined according to the captured image generally has the same person profile as that in the candidate person profile of the accompanying person. Therefore, after determining the candidate person profile of the accompanying person, the candidate person profile that is the same as that in the candidate person profile of the accompanying person in the candidate person profile corresponding to the target person can be obtained. The possibility that the person profile in the same candidate person profile is the person profile of the target person is greater, and the subsequent operation of image similarity comparison can be first performed on the same candidate person profile.

[0085] Embodiment 5:

[0086] In order to further improve the efficiency of person profile recognition, on the basis of the above embodiments, in the embodiment of the present application, the process of training the generative adversarial model including the classification model includes:

[0087] For each sample image in the sample set, where each sample image corresponds to a target label, and the target label is used to identify the identity information of the target person included in the sample image; input the sample image into the classification model in the original generative adversarial model, and obtain the type identifier corresponding to the step length and the gait output by the classification model. Other sub-models of the original generative adversarial model determine the recognition label of the sample image according to the type identifier corresponding to the step length and the gait.

[0088] According to the recognition label of the sample image and the target label of the sample image, determine the loss value corresponding to the sample image, and adjust the parameters of the original classification model and other sub-models of the original generative adversarial model according to the loss value.

[0089] In order to implement the training of the generative adversarial model including the classification model, in the embodiment of the present application, a sample set is pre-configured. The sample set includes a plurality of sample images. The sample images in the sample set cover images at different times, different locations, different angles, different heights, and different light conditions. The sample images include people with different numbers, different genders, different heights, and different ages.

[0090] To facilitate the training of a generative adversarial model containing a classification model, the target label of each sample image is also saved in the sample set. Specifically, the target label can be used to identify the identity information of the target person in the sample image, such as an ID number, etc.

[0091] Specifically, the generative adversarial model includes a classification model and other sub-models. When training the generative adversarial model, for each sample image in the sample set, the sample image can be input into the original classification model in the original generative adversarial model. Through the processing of the sample image, the original classification model can output the type identifier corresponding to the stride and gait of the target person in the sample image. After the original classification model outputs the type identifier corresponding to the stride and gait of the target person in the sample image, other sub-models of the original generative adversarial model can determine and output the recognition label of the sample image according to the identifier corresponding to the stride and gait of the target person in the sample image.

[0092] In the embodiment of the present application, in order to complete the training of the generative adversarial model containing a classification model, after obtaining the recognition label determined by the original generative adversarial model, the loss value corresponding to the sample image can be determined according to the recognition label and the target label of the sample image, and the parameters of the original classification model and other sub-models of the original generative adversarial model can be adjusted according to the loss value.

[0093] Since the target label corresponding to the sample image input into the original classification model is known, the loss value corresponding to the sample image can be determined according to the target label corresponding to the sample image and the recognition label determined by other sub-models of the original generative adversarial model, and the parameters of the original classification model and other sub-models of the original generative adversarial model can be adjusted according to the determined loss value corresponding to the sample image.

[0094] In the embodiment of the present application, a convergence condition is preset in advance. The convergence condition can be that the number of times the recognition label of the determined sample image is consistent with the target label corresponding to the sample image is greater than a set number; it can also be that the number of iterations of training the original generative adversarial model reaches the set maximum number of iterations, etc. Specifically, the embodiment of the present application does not limit this.

[0095] Specifically, for each sample image in the sample set, the sample image is input into the classification model in the original generative adversarial model. The classification model in the original generative adversarial model first inputs the sample image into the encoder to determine the feature vector corresponding to the sample image. The encoder determines the feature vector corresponding to the sample image based on a convolutional neural network. A commonly used encoder based on a convolutional neural network includes 4 convolutional layers. Each layer uses convolutional kernels of different sizes and different numbers of filters to extract local and global feature vectors in the sample image. After obtaining the feature vectors output by each layer, activation functions such as the Rectified Linear Unit (RELU) can be used to integrate the feature vectors output by different convolutional layers, thereby determining the first feature vector corresponding to the sample image.

[0096] After the encoder determines the first feature vector corresponding to the sample image, the feature vector is input into the view transformer to determine the second feature vector corresponding to the sample image after transformation.

[0097] In order to be able to extract the implicit attribute information of the target person in the sample image, the theory of manifold learning can be used to map the first feature vector of the sample image output by the encoder from a low-dimensional manifold to a high-dimensional space. It can be assumed that the input feature vector lies on a low-dimensional manifold, and the sample image moving along this manifold can achieve a view transformation. Specifically, the transformation process from view a to view b can be represented by the following formula: where, z a represents the feature vector of the portrait in view a, z b represents the feature vector of the portrait in view b, and wi represents the transformation vector from view i - 1 to view i. The specific transformation process can be completed by a fully connected layer without bias to reduce the error accumulated due to the reconstructed view required for view transformation. The weights of the fully connected layer can be represented as W = [w 1 , w 2 , ……, w nb , where nb is the identifier of the angle, and each angle corresponds to a weight. w 1 is the weight of view 1, w 2 is the weight of view 2, and w nb is the weight of view nb. The transformation process from view a to view b represented by a vector is e ab = [e ab 1 , e ab 2 , ……, e ab nb , and e ab iThe range is in {0, 1}. From the above expression, the expression of the perspective converter conversion process can be expressed as: z b = z a + We ab .

[0098] After the perspective converter determines the second feature vector corresponding to the sample image after conversion, it inputs the second feature vector corresponding to the conversion into the pre-trained generator to determine the predicted image feature vector corresponding to the sample image. The role of the generator is to generate a fake image that is indistinguishable from the real image and consists of several transposed convolutional layers. The generator determines the predicted image feature vector corresponding to the sample image, which is the prior art and will not be elaborated here.

[0099] After the generator determines the predicted image feature vector, it inputs the predicted image feature vector and the second feature vector determined by the perspective converter into the pre-trained discriminator. The discriminator mainly includes several convolutional layers. The discriminator can judge the authenticity of the predicted image feature vector generated by the generator and whether the domain of the predicted image generated by the generator is the specified angular domain and state domain. If the discriminator determines that the predicted image feature vector is the feature vector determined by the perspective converter, it inputs the second feature vector into the classifier of the original generative adversarial model, and the classifier determines the type identifier corresponding to the step length and gait of the sample image.

[0100] The trained generative adversarial model is sufficient to distinguish between the sample image and the predicted image. During the discrimination process, different weights are usually assigned to each pixel point of the captured image, so as to reflect the implicit attributes of the sample image.

[0101] Next, a specific embodiment is used to illustrate the training process of the generative adversarial model including the classification model provided by the embodiments of the present application. Figure 3 FIG. is a schematic diagram of the training process of the generative adversarial model including the classification model provided by the embodiments of the present application. This process specifically includes the following steps:

[0102] S301: For each sample image in the sample set, input the sample image into the encoder of the original classification model. The classification model is a part of the generative adversarial model.

[0103] S302: The encoder of the original classification model determines the first feature vector corresponding to the sample image.

[0104] S303: The perspective converter of the original classification model determines the second feature vector corresponding to the sample image after conversion according to the first feature vector corresponding to the sample image and a preset algorithm.

[0105] S304: The generator of the original classification model determines the predicted image feature vector corresponding to the sample image according to the second feature vector corresponding to the sample image after conversion.

[0106] S305: The discriminator of the original classification model determines whether the predicted image feature vector is the second feature vector according to the predicted image feature vector and the second feature vector corresponding to the sample image after perspective transformation. If so, execute S306; if not, execute S304.

[0107] S306: The classifier of the original classification model outputs the type identifier corresponding to the step length and gait of the target person in the sample image according to the second feature vector corresponding to the sample image after perspective transformation.

[0108] S307: Other sub-models of the original generative adversarial model determine the recognition label of the sample image according to the type identifier corresponding to the step length and gait.

[0109] S308: Determine the loss value corresponding to the sample image according to the recognition label of the sample image and the target label of the sample image, and adjust the parameters of the original classification model and other sub-models of the original generative adversarial model according to the loss value.

[0110] Embodiment 6:

[0111] Figure 4 It is a schematic structural diagram of the portrait filing device provided by the embodiment of the present application. As Figure 4 shown, the device includes:

[0112] A determination module 401, configured to receive an image collected by an image acquisition device, input the image into a pre-trained classification model, and determine a first target type identifier corresponding to a first target step length and gait of a target person in the image;

[0113] A search module 402, configured to search in the database for a candidate person file that matches the first target step length and / or the first target type identifier according to the type identifier corresponding to the step length and gait recorded in each person file in the database;

[0114] An identification module 403, configured to output the person file to which the standard image with a similarity greater than a set similarity threshold in the candidate person file belongs if there is a standard image in the candidate person file whose similarity to the image is greater than the set similarity threshold.

[0115] In a possible implementation manner, the determining module 401 is further configured to obtain the time and location when the image is captured; input the captured location into a pre-trained spatio-temporal graph modeling model to obtain a target location identifier output by the spatio-temporal graph modeling model; and determine a target time identifier corresponding to the time range to which the captured time belongs according to the captured time and the label information of each time range corresponding to the target location identifier.

[0116] The searching module 402 is further configured to search in the database for a candidate person file that matches the target time identifier and the target location identifier according to the time identifier and location identifier corresponding to each person file recorded in the database.

[0117] In a possible implementation manner, the determining module 401 is further configured to obtain a target identifier corresponding to the target orientation and posture of the target person output by the pre-trained classification model.

[0118] The searching module 402 is configured to, if it is determined that the target identifier corresponding to the target orientation and posture of the target person is consistent with the identifier corresponding to the orientation and posture included in other information of a certain person file recorded in the database, determine the person file as a candidate person file of the target person.

[0119] In a possible implementation manner, the recognition module 403 is further configured to, for each candidate person file, determine the total number of successful matches of the candidate person file according to the number of information matches between the candidate person file and each of the first target step length, the first target type identifier, the target time identifier, and the target location identifier; sort each candidate person file according to the total number of successful matches of each candidate person file, and perform subsequent operations on the sorted candidate person files.

[0120] In a possible implementation manner, the device further includes:

[0121] A training module 404 is configured to, for each sample image in a sample set, where each sample image corresponds to a target label for identifying the identity information of the target person included in the sample image; input the sample image into the classification model in an original generative adversarial model to obtain a type identifier corresponding to the step length and gait output by the classification model, and other sub-models of the original generative adversarial model determine an identification label of the sample image according to the type identifier corresponding to the step length and gait; determine a loss value corresponding to the sample image according to the identification label of the sample image and the target label of the sample image, and adjust the parameters of the original classification model and other sub-models of the original generative adversarial model according to the loss value.

[0122] Example 7:

[0123] Figure 5 A schematic structural diagram of an electronic device provided for this application. On the basis of the above embodiments, this application also provides an electronic device, as Figure 5 shown, including: a processor 501, a communication interface 502, a memory 503, and a communication bus 504, where the processor 501, the communication interface 502, and the memory 503 complete communication with each other through the communication bus 504;

[0124] A computer program is stored in the memory 503. When the program is executed by the processor 501, the processor 501 is caused to execute the following steps:

[0125] Receive an image collected by an image acquisition device, input the image into a pre-trained classification model, and determine a first target step length and a first target type identifier corresponding to the gait of the target person in the image;

[0126] According to the step length and the type identifier corresponding to the gait recorded in the database for each person's file, search in the database for candidate person files that match the first target step length and / or the first target type identifier;

[0127] If there is a standard image in the candidate person files whose similarity to the image is greater than a set similarity threshold, output the person file to which the standard image with a similarity greater than the set similarity threshold belongs.

[0128] In a possible implementation manner, before the step of if there is a comparison image in the candidate person files whose similarity to the image is greater than a set similarity threshold, output the person file to which the comparison image with a similarity greater than the set similarity threshold belongs, the method includes:

[0129] Obtain the time and location when the image was captured;

[0130] Input the captured location into a pre-trained spatio-temporal graph modeling model, and obtain a target location identifier output by the spatio-temporal graph modeling model;

[0131] According to the captured time and the standard information of each time range corresponding to the target location identifier, determine a target time identifier corresponding to the time range to which the captured time belongs;

[0132] According to the time identifier and location identifier corresponding to each person's file recorded in the database, search in the database for candidate person files that match the target time identifier and the target location identifier.

[0133] In a possible implementation, before outputting the personal profile to which the comparison image with a similarity greater than the set similarity threshold in the candidate personal profile belongs, if there is a comparison image in the candidate personal profile with a similarity greater than the set similarity threshold to the image, the method includes:

[0134] Obtain the target identifier corresponding to the target orientation and posture of the target person output by the pre-trained classification model;

[0135] If it is determined that the target identifier corresponding to the target orientation and posture of the target person is consistent with the identifier corresponding to the orientation and posture included in other information of a certain personal profile recorded in the database, then this personal profile can be determined as the candidate personal profile of the target person.

[0136] In a possible implementation, after determining the candidate personal profile corresponding to the target person and before outputting the personal profile to which the comparison image with a similarity greater than the set similarity threshold belongs, the method includes:

[0137] For each candidate personal profile, determine the total number of successful matches of this candidate personal profile according to the number of information matches between this candidate personal profile and each piece of information in the first target step length, the first target type identifier, the target time identifier, and the target location identifier;

[0138] Sort each candidate personal profile according to the total number of successful matches of each candidate personal profile, and perform subsequent operations on the sorted candidate personal profiles.

[0139] In a possible implementation, after determining the candidate personal profile corresponding to the target person and before outputting the personal profile to which the comparison image with a similarity greater than the set similarity threshold belongs, the method includes:

[0140] Based on the pre-trained classification model, determine the second target step length and the second target type identifier corresponding to the gait of the person walking with the target person in the image;

[0141] According to the step length and type identifier corresponding to each personal profile recorded in the database, search in the database for the candidate personal profile of the person walking with the target person that matches the second target step length and the second target type identifier of the person walking with the target person;

[0142] Obtain the candidate personal profiles that are the same in the candidate personal profile corresponding to the target person and the candidate personal profile of the person walking with the target person corresponding to the person walking with the target person, and perform subsequent operations on the same candidate personal profiles.

[0143] In a possible implementation, the process of training the generative adversarial model including the classification model includes:

[0144] For each sample image in the sample set, where each sample image corresponds to a target label for identifying the identity information of the target person included in the sample image, input the sample image into the classification model in the original generative adversarial model to obtain the type identifier corresponding to the stride and gait output by the classification model, and other sub-models of the original generative adversarial model determine the recognition label of the sample image according to the type identifier corresponding to the stride and gait.

[0145] Determine the loss value corresponding to the sample image according to the recognition label of the sample image and the target label of the sample image, and adjust the parameters of the original classification model and other sub-models of the original generative adversarial model according to the loss value.

[0146] Since the principle of the above electronic device for solving problems is similar to the method for identifying personal files, the implementation of the above electronic device can refer to the above embodiments, and the repeated parts will not be elaborated.

[0147] The communication bus mentioned in the above electronic device may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of representation, only a thick line is used in the figure, but it does not mean that there is only one bus or one type of bus. The communication interface 502 is used for communication between the above electronic device and other devices. The memory may include a Random Access Memory (RAM), and may also include a Non-Volatile Memory (NVM), such as at least one disk memory. Optionally, the memory may also be at least one storage device located far from the aforementioned processor. The above processor may be a general-purpose processor, including a central processor, a Network Processor (NP), etc.; it may also be a Digital Signal Processing (DSP), an application-specific integrated circuit, a field-programmable gate array, or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0148] Embodiment 8:

[0149] Based on the above embodiments, the present application further provides a computer-readable storage medium, in which a computer program executable by a processor is stored. When the program runs on the processor, the processor is caused to execute the following steps:

[0150] Receive the image collected by the image acquisition device, input the image into a pre-trained classification model, and determine the first target step length of the target person in the image and the first target type identifier corresponding to the gait.

[0151] According to the step lengths and the type identifiers corresponding to the gaits recorded in the database for each person's file, search in the database for candidate person files that match the first target step length and / or the first target type identifier.

[0152] If there is a standard image in the candidate person file whose similarity to the image is greater than a set similarity threshold, output the person file to which the standard image with a similarity greater than the set similarity threshold belongs.

[0153] In a possible implementation manner, before the step of if there is a comparison image in the candidate person file whose similarity to the image is greater than a set similarity threshold, output the person file to which the comparison image with a similarity greater than the set similarity threshold belongs, the method includes:

[0154] Obtain the time and location when the image was captured.

[0155] Input the captured location into a pre-trained spatio-temporal graph modeling model, and obtain the target location identifier output by the spatio-temporal graph modeling model.

[0156] According to the captured time and the standard information of each time range corresponding to the target location identifier, determine the target time identifier corresponding to the time range to which the captured time belongs.

[0157] According to the time identifiers and location identifiers corresponding to each person's file recorded in the database, search in the database for candidate person files that match the target time identifier and the target location identifier.

[0158] In a possible implementation manner, before the step of if there is a comparison image in the candidate person file whose similarity to the image is greater than a set similarity threshold, output the person file to which the comparison image with a similarity greater than the set similarity threshold belongs, the method includes:

[0159] Obtain the target identifier corresponding to the target orientation and posture of the target person output by the pre-trained classification model.

[0160] If it is determined that the target identifier corresponding to the target orientation and posture of the target person is consistent with the identifier corresponding to the orientation and posture included in other information of a certain person file in the database, then this person file can be determined as the candidate person file of the target person.

[0161] In a possible implementation, after determining the candidate person profile corresponding to the target person and before outputting the person profile to which the comparison image with a similarity greater than a set similarity threshold belongs, the method includes:

[0162] For each of the candidate person profiles, determine the total number of successful matches of the candidate person profile according to the number of information matches between the candidate person profile and each of the first target step length, the first target type identifier, the target time identifier, and the target location identifier;

[0163] Sort each of the candidate person profiles according to the total number of successful matches of each candidate person profile, and perform subsequent operations on the sorted candidate person profiles.

[0164] In a possible implementation, after determining the candidate person profile corresponding to the target person and before outputting the person profile to which the comparison image with a similarity greater than a set similarity threshold belongs, the method includes:

[0165] Based on a pre-trained classification model, determine the second target step length and the second target type identifier corresponding to the gait of the person in the same row in the image;

[0166] According to the step length and type identifier corresponding to each person profile recorded in the database, search in the database for the candidate person profile of the person in the same row that matches the second target step length and the second target type identifier of the person in the same row;

[0167] Obtain the candidate person profiles that are the same in the candidate person profile corresponding to the target person and the candidate person profile of the person in the same row corresponding to the person in the same row, and perform subsequent operations on the same candidate person profiles.

[0168] In a possible implementation, the process of training the generative adversarial model including the classification model includes:

[0169] For each sample image in the sample set, where each sample image corresponds to a target label, and the target label is used to identify the identity information of the target person included in the sample image; input the sample image into the classification model in the original generative adversarial model, obtain the type identifier corresponding to the step length and gait output by the classification model, and other sub-models of the original generative adversarial model determine the recognition label of the sample image according to the type identifier corresponding to the step length and gait;

[0170] Determine the loss value corresponding to the sample image according to the recognition label of the sample image and the target label of the sample image, and adjust the parameters of the original classification model and other sub-models of the original generative adversarial model according to the loss value.

[0171] Since the principle of problem-solving of the computer-readable medium provided above is similar to that of the personal file recognition method, after the processor executes the computer program in the above computer-readable medium, the implemented steps can refer to the above embodiments, and the repeated parts will not be elaborated.

[0172] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0173] For the system / device embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiments.

[0174] The present application is described with reference to the flowcharts and / or block diagrams of the method, device (system), and computer program product according to the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0175] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured product including an instruction device, and the instruction device implements the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0176] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0177] Although the preferred embodiments of the present application have been described, those skilled in the art can make additional changes and modifications to these embodiments once they learn the basic creative concepts. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications that fall within the scope of the present application.

[0178] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application is also intended to include these modifications and variations.

Claims

1. A method for identifying a person's profile, characterized in that, the method includes: Receiving an image collected by an image acquisition device, inputting the image into a pre-trained classification model, and determining a first target step length of a target person in the image and a first target type identifier corresponding to the gait; According to the step lengths and type identifiers corresponding to the gaits recorded in the database for each person's profile, searching in the database for candidate person profiles that match the first target step length and / or the first target type identifier; Obtaining the time and location when the image was captured; inputting the captured location into a pre-trained spatio-temporal graph modeling model to obtain a target location identifier output by the spatio-temporal graph modeling model; determining, according to the captured time and the identification information of each time range corresponding to the target location identifier, a target time identifier corresponding to the time range to which the captured time belongs; searching in the database for candidate person profiles that match the target time identifier and the target location identifier according to the time identifiers and location identifiers recorded in the database for each person's profile; wherein, each time range corresponding to the target location identifier is determined according to the tidal law of human activities; If there is a standard image in the candidate person profiles whose similarity to the image is greater than a set similarity threshold, output the person profile to which the standard image with a similarity greater than the set similarity threshold belongs.

2. The method according to claim 1, characterized in that, before the step of, if there is a comparison image in the candidate person profiles whose similarity to the image is greater than a set similarity threshold, outputting the person profile to which the comparison image with a similarity greater than the set similarity threshold belongs, the method includes: Obtaining a target identifier corresponding to the target orientation and posture of the target person output by the pre-trained classification model; If it is determined that the target identifier corresponding to the target orientation and posture of the target person is consistent with the identifier corresponding to the orientation and posture included in other information of a certain person profile recorded in the database, then determine this person profile as a candidate person profile of the target person.

3. The method according to claim 2, characterized in that, after determining the candidate person profile corresponding to the target person and before the step of outputting the person profile to which the comparison image with a similarity greater than the set similarity threshold belongs, the method includes: For each of the candidate person profiles, determining the total number of successful matches of the candidate person profile according to the number of information matches between the candidate person profile and each of the first target step length, the first target type identifier, the target time identifier, and the target location identifier; Sorting each of the candidate person profiles according to the total number of successful matches of each candidate person profile, and performing subsequent operations on the sorted candidate person profiles.

4. The method according to claim 1, characterized in that, after determining the candidate person profile corresponding to the target person and before the step of outputting the person profile to which the comparison image with a similarity greater than the set similarity threshold belongs, the method includes: Based on the pre-trained classification model, determine the second target step length and the second target type identifier corresponding to the gait of the person in the same row in the image; According to the step length and type identifier corresponding to each person file recorded in the database, search in the database for the candidate person file of the person in the same row that matches the second target step length and the second target type identifier of the person in the same row; Obtain the candidate person files that are the same in the candidate person file corresponding to the target person and the candidate person file corresponding to the person in the same row, and perform subsequent operations on the same candidate person files.

5. The method according to claim 1, characterized in that The process of training the generative adversarial model including the classification model includes: For each sample image in the sample set, where each sample image corresponds to a target label, and the target label is used to identify the identity information of the target person included in the sample image; input the sample image into the classification model in the original generative adversarial model, obtain the type identifier corresponding to the step length and gait output by the classification model, and other sub-models of the original generative adversarial model determine the recognition label of the sample image according to the type identifier corresponding to the step length and gait; According to the recognition label of the sample image and the target label of the sample image, determine the loss value corresponding to the sample image, and adjust the parameters of the classification model and other sub-models of the original generative adversarial model according to the loss value.

6. A person file recognition device, characterized in that The device includes: A determination module, configured to receive an image collected by an image acquisition device, input the image into a pre-trained classification model, and determine a first target step length and a first target type identifier corresponding to the gait of the target person in the image; A search module, configured to search in the database for a candidate person file that matches the first target step length and / or the first target type identifier according to the step length and type identifier corresponding to each person file recorded in the database; The determination module is further configured to obtain the time and location when the image is captured; input the captured location into a pre-trained spatio-temporal graph modeling model, and obtain a target location identifier output by the spatio-temporal graph modeling model; according to the captured time and the label information of each time range corresponding to the target location identifier, determine the target time identifier corresponding to the time range to which the captured time belongs; The search module is further configured to search in the database for a candidate person file that matches the target time identifier and the target location identifier according to the time identifier and location identifier corresponding to each person file recorded in the database; wherein, each time range corresponding to the target location identifier is determined according to the tidal law of human activities; An identification module, configured to output the person file to which the standard image with a similarity greater than the set similarity threshold in the candidate person file belongs if there is a standard image in the candidate person file with a similarity greater than the set similarity threshold to the image.

7. The device according to claim 6, characterized in that The device further includes: A training module, for each sample image in a sample set, wherein each sample image corresponds to a target label, and the target label is used to identify the identity information of the target person included in the sample image; input the sample image into a classification model in an original generative adversarial model, obtain the type identifier corresponding to the step length and gait output by the classification model, and other sub-models of the original generative adversarial model determine the recognition label of the sample image according to the type identifier corresponding to the step length and gait; determine the loss value corresponding to the sample image according to the recognition label of the sample image and the target label of the sample image, and adjust the parameters of the classification model and other sub-models of the original generative adversarial model according to the loss value.

8. An electronic device characterized in that the electronic device at least includes a processor and a memory, and the processor is configured to implement the steps of the person file recognition method according to any one of claims 1-5 when executing a computer program stored in the memory.

9. A computer-readable storage medium characterized in that it stores a computer program, and when the computer program is executed by a processor, it implements the steps of the person file recognition method according to any one of claims 1-5.

Images