Face aggregation index statistical method, device, equipment and storage medium

By labeling and using the image search technology to automatically count the indicators of the face clustering algorithm, the problem of time-consuming and labor-consuming indicator statistics in the optimization of the face clustering algorithm is solved, and efficient and reliable indicator statistics are achieved in open scenarios.

CN118711020BActive Publication Date: 2024-11-19BEIJING SUREKAM CORP
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411186390.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-28
Publication Date
2024-11-19
Estimated Expiration
2044-08-28

AI Technical Summary

Technical Problem

In the process of optimizing face clustering algorithms, index statistics are time-consuming and labor-intensive and prone to calculation errors, resulting in a lack of credibility in data. Especially in open scenarios, automated processes are needed to reduce manpower investment and avoid index deviations.

Method used

By obtaining the profile trajectory data of the personnel to be marked, marking the captured images of the target personnel, and obtaining other captured images by searching pictures through pictures, and correlating and counting the diffusion rate, accuracy and recall rate in standard modes and strict modes. Implement automated indicator statistics.

Benefits of technology

It realizes automated indicator statistics in a local point environment, simplifies the data processing process, improves statistical efficiency and result reliability, avoids manual calculation errors, and is suitable for real-time deployment of open scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118711020B_ABST
    Figure CN118711020B_ABST
Patent Text Reader

Abstract

The present application discloses a method, device, equipment and storage medium for counting face clustering indexes, including: obtaining archive trajectory data corresponding to the archive of the personnel to be marked, marking the archive trajectory data, marking the trajectory snapshot image confirmed to be the target personnel as confirmed; based on the confirmed snapshot image, obtaining other snapshot images of the target personnel by image search; associating the other snapshot images obtained by image search with the confirmed snapshot image to obtain the image search associated field, writing the archive number corresponding to the other snapshot image and the image search associated field into the trajectory marking table; and counting the diffusion rate, accuracy rate and recall rate of the target personnel archive in the standard mode and the strict mode respectively according to the trajectory marking table. This scheme can be applied to open scenarios to automatically count face clustering indexes, thereby improving statistical efficiency and accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of computer software testing, and in particular to a method, device, equipment and storage medium for statistical analysis of facial profile indicators. Background Art

[0002] In the process of optimizing the face clustering algorithm, in order to measure the quality of the algorithm, it is necessary to evaluate it through indicators. Especially when performing indicator statistics in open scenarios, the statistical process consumes a lot of time and manpower. According to previous experience, it takes 4 people / day to count the indicators of 5 people in a 10 million data set. There may also be problems such as calculation errors in the statistical process, resulting in a lack of credibility of the data. The manpower and time costs consumed in the indicator statistics process will continue to increase with the iterative testing of the clustering algorithm. If it is executed one more time, the investment will double. Therefore, it is very necessary to solidify the statistical process into an automated process. This method can not only greatly reduce manpower investment, but also avoid the problem of indicator deviation due to manual calculation errors.

[0003] The existing automated indicator testing process requires first preparing calibrated test sets and interference sets as sample sets, then clustering and grading faces, and finally calculating indicators based on the initial calibrated test sets. This method has many limitations when applied to actual site environments. For example, a certain number of sample sets and interference sets need to be obtained, and these sample sets need to be pre-labeled. After copying these data to the laboratory environment and then calculating indicators, these data may involve privacy and pose certain legal risks. Summary of the invention

[0004] The embodiments of the present application provide a method, device, equipment and storage medium for counting facial clustering indicators, so as to at least solve the technical problem in the related art that it is difficult to automatically analyze the advantages and disadvantages of facial clustering algorithms.

[0005] According to one aspect of an embodiment of the present application, a method for collecting face index statistics is provided, including:

[0006] Obtain the archive trajectory data corresponding to the personnel file to be annotated, annotate the archive trajectory data, and annotate the trajectory snapshot image confirmed to be the target person as confirmed;

[0007] Based on the confirmed captured images, other captured images of the target person are obtained by image search;

[0008] Associating other captured images obtained through image search with the confirmed captured images to obtain image search associated fields, and writing the file numbers corresponding to other captured images and image search associated fields into the track annotation table;

[0009] According to the trajectory annotation table, the diffusion rate, accuracy rate and recall rate of the target personnel files in the standard mode and the strict mode are respectively counted.

[0010] In one embodiment, obtaining the file trajectory data corresponding to the personnel file to be labeled includes:

[0011] Query the personnel files to be marked according to one or more of the pre-set file gathering tasks, file generation time and track number;

[0012] Get the maximum trajectory time and minimum time index value of the personnel file to be marked;

[0013] The larger time between the maximum trajectory time and the minimum time index value is used as the query trajectory start time;

[0014] Obtain the trajectory data corresponding to the personnel file to be labeled according to the query trajectory start time.

[0015] In one embodiment, based on the confirmed captured image, other captured images of the target person are obtained by image search, including:

[0016] Extracting facial feature values ​​of multiple confirmed captured images at different angles respectively;

[0017] Traverse and extract facial feature values ​​of other images outside the files to be annotated;

[0018] Based on the facial feature values, similarity calculation is performed, and other images whose similarity with the confirmed captured image is greater than a threshold are recalled to obtain other captured images of the target person.

[0019] In one embodiment, the diffusion rate, accuracy rate, and recall rate of the target personnel profile in the statistical standard mode include:

[0020] According to the trajectory marking table, determining the number of splitting files in the standard mode, and obtaining the diffusion rate according to the number of splitting files in the standard mode;

[0021] Obtain the file with the largest number of confirmed snapshot images in the split files under the standard mode as the main file, and obtain the accuracy rate according to the ratio of the total number of confirmed snapshot images in the main file to the number of all snapshot images in the main file;

[0022] Obtaining a recall rate according to a ratio of the total number of confirmed snapshot images in the master file to the total number of all confirmed snapshot images of the target person in the trajectory annotation table;

[0023] Among them, in the standard mode, when there are multiple files of the captured images of the target person, and the number of captured images of the target person in a certain file accounts for less than the preset threshold of the total number of captured images in the file, it is considered that the file is not a split file.

[0024] In one embodiment, the diffusion rate, precision rate, and recall rate of the target personnel profile in the strict mode are statistically analyzed, including:

[0025] According to the trajectory marking table, determining the number of split files in the strict mode, and obtaining the diffusion rate according to the number of split files in the strict mode;

[0026] The accuracy rate is obtained according to the ratio of the total number of confirmed snapshot images in all split files to the total number of snapshot images in all split files in the strict mode;

[0027] The recall rate is obtained based on the ratio of the total number of confirmed snapshot images in all split files to the total number of all snapshot images of the target person in the trajectory annotation table;

[0028] Among them, in strict mode, when a file contains a captured image of a target person, the file is considered to be a split file.

[0029] In one embodiment, it further includes:

[0030] According to the ratio of the sum of the number of files of all personnel in the standard mode under the file gathering task to the total number of personnel, the diffusion rate of the file gathering task is obtained;

[0031] According to the archive with the most captured images of all personnel in the archive gathering task, the accuracy of the archive gathering task is obtained by taking the ratio of the total number of confirmed captured images in all the main archives to the total number of captured images in all the main archives;

[0032] The recall rate of the clustering task is obtained by taking the ratio of the total number of confirmed snapshot images in all master archives to the total number of confirmed snapshot images of all personnel;

[0033] The clustering rate of the archiving task is obtained based on the ratio of the sum of the confirmed snapshot images in all archives in the standard mode under the archiving task to the sum of the confirmed snapshot images of all annotated personnel under the task in the trajectory annotation table.

[0034] In one embodiment, it further includes:

[0035] According to the ratio of the sum of the number of files of all personnel in the strict mode under the file gathering task to the total number of personnel, the diffusion rate of the file gathering task is obtained;

[0036] According to the archive with the most captured images of all personnel in the archive gathering task, the accuracy of the archive gathering task is obtained by taking the ratio of the total number of confirmed captured images in all the main archives to the total number of captured images in all the main archives;

[0037] The recall rate of the clustering task is obtained by taking the ratio of the total number of confirmed snapshot images in all master archives to the total number of confirmed snapshot images of all personnel;

[0038] The clustering rate of the clustering task is obtained based on the ratio of the sum of the confirmed snapshot images in all archives in strict mode under the clustering task to the sum of the confirmed snapshot images of all annotators under the task in the trajectory annotation table.

[0039] According to another aspect of the embodiment of the present application, a face aggregation index statistics device is also provided, including:

[0040] The labeling module is used to obtain the file trajectory data corresponding to the personnel file to be labeled, label the file trajectory data, and label the trajectory snapshot image confirmed to be the target person as confirmed;

[0041] An image search module is used to obtain other captured images of the target person by using an image search method based on the confirmed captured images;

[0042] An association module is used to associate other captured images obtained by image search with the confirmed captured images, obtain image search association fields, and write the file numbers corresponding to other captured images and the image search association fields into the trajectory annotation table;

[0043] The statistical module is used to count the diffusion rate, accuracy rate and recall rate of the target personnel files in the standard mode and the strict mode respectively according to the trajectory annotation table.

[0044] According to another aspect of an embodiment of the present application, there is also provided an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to execute the above-mentioned face clustering index statistical method through the computer program.

[0045] According to another aspect of the embodiments of the present application, a computer-readable storage medium is provided, in which a computer program is stored, wherein the computer program is configured to execute the above-mentioned face clustering index statistical method when running.

[0046] The technical solution provided by the embodiments of the present application may have the following beneficial effects:

[0047] The face clustering index statistical method provided in the embodiment of the present application finally obtains all the files of the labeled personnel and all the captured data for index calculation. This scheme can select any file as the labeled personnel, and search for other captured trajectories of the labeled personnel in a divergent manner according to the captured trajectories in the file, and then obtain the classification to which it belongs according to the recalled trajectories, and finally obtain all the files of the labeled personnel and all the captured data for index calculation. It can achieve the effect of being deployed and available at any time in the local environment. This method can be applied to open scenarios, and does not require the pre-preparation of calibrated sample sets and interference sets, which simplifies the pre-preparation process of statistics, improves data processing efficiency, and improves the reliability and statistical efficiency of the index statistical results. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0049] Figure 1 It is a flow chart of a method for counting face clustering indexes according to an embodiment of the present application;

[0050] Figure 2 This is a schematic diagram of an interface of a method for statistically analyzing aggregate file indicators according to an embodiment of the present application;

[0051] Figure 3 is a schematic diagram of a method for obtaining files to be annotated according to an embodiment of the present application;

[0052] Figure 4 is a schematic diagram of obtaining archive trajectory data according to an embodiment of the present application;

[0053] Figure 5 is a schematic diagram of a trajectory capture image annotation method according to an embodiment of the present application;

[0054] Figure 6 is a schematic diagram of an image search method according to an embodiment of the present application;

[0055] Figure 7 It is a schematic diagram of a method for querying graph search result data according to an embodiment of the present application;

[0056] Figure 8 is a schematic diagram of a personnel file according to an embodiment of the present application;

[0057] Fig. 9 is a schematic diagram of an indicator statistics method according to an embodiment of the present application;

[0058] Fig.10 It is a schematic diagram of a face aggregation index statistics device according to an embodiment of the present application;

[0059] Fig.11 It is a schematic diagram of the structure of an optional electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0060] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0061] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0062] The following is combined with Figure 1 The face clustering index statistical method of the embodiment of the present application is described in detail. Figure 1 As shown, the method mainly includes the following steps:

[0063] S101 obtains the file trajectory data corresponding to the personnel file to be marked, marks the file trajectory data, and marks the trajectory captured image of the target person as confirmed.

[0064] In one embodiment, first, according to one or more pre-set conditions of the archive gathering task, archive generation time, and track number, the personnel archive to be marked is queried.

[0065] Construct a statistical task of aggregated indexes to describe a statistical process. Figure 2 As shown, it mainly includes the task name, creation time, number of labeled people, accuracy, recall rate, diffusion rate, and clustering rate in standard mode and strict mode.

[0066] Click New Annotation on the task interface to query the personnel files to be annotated based on one or more screening conditions such as specific archive task, file generation time, number of tracks, etc. The user clicks on the file of the person to be counted to annotate.

[0067] like Figure 3 As shown, the file query conditions are issued, including the file gathering task, file generation time, number of tracks, number of pages per page, and page number. Further, the query conditions are encapsulated, the file table and track table in the database are queried, and the corresponding query results are returned to the interface for display to obtain the files of the personnel to be marked.

[0068] Furthermore, the maximum trajectory time and the minimum time index value of the personnel file to be marked are obtained; the larger time of the maximum trajectory time and the minimum time index value is used as the query trajectory start time; and the trajectory data corresponding to the personnel file to be marked is obtained according to the query trajectory start time.

[0069] Click on the file to be annotated on the interface and send the file ID. The background queries the maximum track time traceTime of the file in the track annotation table, and obtains the minimum time index value indexTime in the memory. The larger time time = max(traceTime, indexTime) is selected as the start time of the query track. Query the corresponding file track table for tracks with a time greater than time. Copy the file tracks within this interval to the track annotation table. The default image search similarity is 0.0, the default track annotation status is 0, and the default image search association ID is -1.

[0070] In order to avoid some early trajectories in the archive not existing in the memory, the data whose trajectories have a time greater than max (the maximum time of the trajectories marked in the trajectory marking table and the minimum time index of the snapshot in the memory) are copied.

[0071] like Figure 4 As shown in the figure, the client sends the file number to be annotated to the server, and the server queries the trajectory annotation table in the database to obtain the maximum trajectory time of the file, and the database returns the maximum trajectory time of the corresponding file. The server calls the big data service interface to obtain the minimum time index of the memory capture, and queries the database for the data whose corresponding file trajectory time is greater than the larger value of (the maximum time of the annotated trajectory in the trajectory annotation table, and the minimum time index of the capture in the memory). The trajectory data that meets the conditions is returned to the server, and the trajectory data is written to the trajectory annotation table of the database, and the operation result is returned to the front end.

[0072] Furthermore, the archive trajectory data is annotated, the trajectory snapshot image confirmed to be the target person is marked as confirmed, and the archive number corresponding to the confirmed trajectory snapshot image is assigned to the trajectory snapshot image number of the target person; the trajectory snapshot image confirmed not to be the target person is marked as excluded; and the trajectory snapshot image to be confirmed is marked as to be confirmed.

[0073] The interface submits the confirmed trajectory data of the file in batches. The background sets the corresponding trajectory annotation status of the trajectory annotation table in the database to confirmed (pending confirmation, excluded) according to the set of snapshot IDs issued. (The annotation status of the confirmed trajectory is 1, the annotation status of the pending confirmation trajectory is 0, and the annotation status of the excluded trajectory is 2). If a trajectory is in the confirmed status, the value of the file ID corresponding to the trajectory is assigned to the number of the target person.

[0074] like Figure 5 As shown, the client submits the confirmed trajectory data of a specific file, or the annotation data to be confirmed or excluded, and updates the status of the corresponding trajectory captured image in the trajectory annotation table in the database.

[0075] S102 acquires other captured images of the target person by image search based on the confirmed captured images.

[0076] Specifically, facial feature values ​​of multiple confirmed snapshot images taken at different angles are extracted respectively; facial feature values ​​of other images outside the archives to be labeled are extracted traversally; similarity calculation is performed based on the facial feature values, and other images whose similarity with the confirmed snapshot images is greater than a threshold are recalled to obtain other snapshot images of the target person.

[0077] Click on a confirmed snapshot image on the interface to send the snapshot image for image search. The background service obtains the facial feature value of the snapshot image based on the snapshot image number, and uses image search to obtain other snapshot data of the person. Select snapshots of the person in actual situations from multiple angles, and all snapshots of the person can be recalled through image search.

[0078] S103 associates other captured images obtained by image search with the confirmed captured images to obtain image search associated fields, and writes the file numbers corresponding to other captured images and the image search associated fields into the track annotation table.

[0079] In one embodiment, other captured images obtained through image search are associated with the confirmed captured images to obtain an image search associated field. The image search associated field defaults to -1, indicating that this image is not image search data. If it is image search data, the image search associated field of the captured image obtained through image search is set to the ID of the original image searched.

[0080] Furthermore, if the data found by the image search exists in the track table of the archive, the file number corresponding to the captured image and the image search associated field are written into the track annotation table. If the image found by the image search does not exist in other archives and is an image outside the archive, the file number of the image is set to -1, and finally the image search data is written into the track annotation table.

[0081] In one embodiment, the image search result data is not a snapshot of the marked person by default, that is, the marking status is excluded. You can click the pending confirmation and confirmed buttons to mark the snapshot as pending confirmation and confirmed.

[0082] like Figure 6 As shown in the figure, the captured image number to be searched is sent, and the image search is performed after obtaining the feature value and algorithm version according to the number, and the result is obtained. The image search result is associated with the trajectory table and the query result is written into the trajectory annotation table in the database. The operation result is returned to the front end.

[0083] When querying graph search result data, if Figure 7 As shown, the snapshot number to be searched is issued, the detailed information of the trajectory is obtained according to the number and written into the historical image search details table, the image search data that has been initialized and excluded is obtained from the trajectory annotation table of the database according to the snapshot number, and the corresponding image search data is returned.

[0084] After the end of the labeling operation is triggered, the background service obtains the corresponding captured data from the trajectory labeling table according to the issued file id (with the person's primary file id as the person's id. Note: The person's primary file id is the file id to be labeled selected by the manual interface. This id is subsequently used as the id representing the labeled person.

[0085] This application selects any file as the labeled person, and searches for other captured trajectories of the labeled person in a divergent manner based on the captured trajectories in the file, and then obtains the file to which it belongs based on the recalled trajectories, and finally obtains all the files of the labeled person.

[0086] S104 counts the diffusion rate, accuracy rate and recall rate in the standard mode and the strict mode respectively according to the trajectory annotation table.

[0087] In one embodiment, the diffusion rate, accuracy rate, and recall rate in the standard mode and the strict mode are respectively counted according to the trajectory annotation table.

[0088] Determine whether the current statistical mode belongs to the standard mode or the strict mode; in the standard mode, if there are multiple archives of the captured images of the target person, and the number of captured images of the target person in a certain archive accounts for less than the preset threshold of the total number of captured images in the archive, then the archive is considered not to be a split file.

[0089] In strict mode, when a file contains a captured image of a target person, the file is considered to be a split file.

[0090] When the threshold in standard mode is 0, the logic of standard mode and strict mode is the same. The indicators of strict mode are helpful for analyzing the reasons why some snapshots are out of the main file.

[0091] In one embodiment, the diffusion rate, accuracy rate, and recall rate of the target personnel profile in the standard mode are statistically analyzed.

[0092] Specifically, according to the trajectory marking table, the number of splitting files in the standard mode is determined, and the diffusion rate is obtained according to the number of splitting files in the standard mode.

[0093] Query the data in the trajectory annotation table where the person image number is the file number, and group them according to the file number in the trajectory annotation table. Traverse the groups. If the file number of the trajectory is not equal to -1 and is inconsistent with the issued file number, record the number of confirmed trajectories in the file id group as m. Check the total number of trajectories for the file and record it as n. Set the threshold threshold. When m / n>=threshold, the file is considered to be a split file of the labeled person in the standard mode and is added to the diffusion rate calculation. Traverse all groups to obtain the final diffusion rate, that is, the number of split files in the standard mode is the diffusion rate.

[0094] like Figure 8 As shown in the figure, file 1 is the file selected during labeling, which is used as the main file, and the threshold is 60%. Files 2 and 3 are the files associated with this person during the labeling process. In file 2, m=6, n=8, then m / n = 75%>60%, that is, file 2 is a split file of file 1. In file 3, m=2, n=6, then m / n = 33%<60%, so file 3 is not a split file of file 1. Finally, the diffusion rate of this person is diffusion = 200%.

[0095] Furthermore, the file with the largest number of confirmed snapshot images in the split files under the standard mode is obtained as the main file, and the accuracy rate is obtained according to the ratio of the total number of confirmed snapshot images in the main file to the number of all snapshot images in the main file.

[0096] by Figure 8 For example, the total number of snapshots of this person in file 1 is a = 8, and the total number of snapshots is b = 10, then the accuracy is precision = 8 / 10 = 80.0%.

[0097] Furthermore, the recall rate is obtained according to the ratio of the total number of confirmed snapshot images in the main file to the total number of all confirmed snapshot images of the target person in the trajectory annotation table.

[0098] by Figure 8 For example, the total number of snapshots of this person in file 1 is a = 8, and all snapshots of this person are c = 18, then the recall rate recall = 8 / 18 = 44%.

[0099] In one embodiment, the diffusion rate, precision rate, and recall rate of the target personnel profile in the strict mode are statistically analyzed.

[0100] Specifically, according to the trajectory marking table, the number of split files in the strict mode is determined, and the diffusion rate is obtained according to the number of split files in the strict mode.

[0101] Query the data in the trajectory annotation table where the person image ID is equal to the file ID, and group them according to the file ID in the trajectory annotation table. Remove the free snapshots except for the file with the file number -1, and the number of groups is the diffusion rate of this person's file. That is, the number of split files in strict mode is the diffusion rate.

[0102] like Figure 8 As shown in the figure, file 1 is the file selected in the interface and is used as the main file. Files 2 and 3 are the files associated during the annotation process. The following three files all contain snapshots of this person, and the final diffusion rate is diffusion = 300%.

[0103] The accuracy rate is obtained based on the ratio of the total number of confirmed snapshot images in all split files to the total number of snapshot images in all split files in the strict mode.

[0104] by Figure 8 For example, the total number of snapshots of this person in the three files is a = 16, and the total number of snapshots is b = 24, then the accuracy is precision = 16 / 24 = 66.7%.

[0105] Furthermore, the recall rate is obtained according to the ratio of the total number of confirmed captured images in all split files to the total number of all captured images of the target person in the trajectory annotation table.

[0106] by Figure 8 For example, the total number of confirmed trajectories of this person in the three archives is a = 16, and all snapshots of this person are c = 18, including snapshots in the archives and snapshots outside the archives, then the recall rate recall = 16 / 18 = 89%.

[0107] After the calculation of the various indicators of the annotated personnel is completed, the information of this person needs to be added to the personnel annotation details table. The personnel ID in this table is set to the file ID selected for annotation on the interface. All files associated with this person also need to be added to the file annotation details table, and the indicator mode is used to indicate whether the file belongs to the standard mode or the strict mode.

[0108] In one embodiment, the method further includes calculating a clustering task index.

[0109] Specifically, the aggregation task index in the standard mode is calculated.

[0110] The diffusion rate of the aggregation task is obtained by the ratio of the sum of the number of files of all personnel in the standard mode under the aggregation task to the total number of personnel. The sum of the number of files of all personnel in the standard mode under the task is queried and recorded as a, and the number of personnel is recorded as b. Then the diffusion rate diffusion = a / b.

[0111] According to the archive with the most captured images of all personnel under the clustering task, the accuracy of the clustering task is obtained by taking the ratio of the total number of confirmed captured images in all main archives to the total number of captured images in all main archives. Query the archive with the top 1 number of captured images of all personnel under this task, and record the sum of the confirmed captured images in these archives as m, and the sum of all captured images in these archives as n, then the accuracy precision = m / n.

[0112] The recall rate is the ratio of the total number of confirmed snapshot images in all main archives to the total number of confirmed snapshot images of all personnel. Based on the sum of confirmed snapshot images in all main archives, denoted as m, all confirmed images of each labeled person (including images outside the archives) are queried and summed, denoted as s, and the recall rate recall = m / s.

[0113] The clustering rate is obtained by the ratio of the sum of the confirmed snapshot images in all archives in the standard mode under the clustering task to the sum of the confirmed snapshot images of all labeled personnel under this task in the trajectory annotation table. Traverse all archives in the archive annotation table under this task whose index mode is the standard mode. Query the number of labeled snapshot images from the trajectory annotation table based on these archive IDs. After traversing all archives, calculate the sum of confirmed trajectories, denoted as m. Count the sum of confirmed snapshot images of all labeled personnel under this task in the trajectory annotation table, denoted as n. Then the clustering rate clusterRate = m / n.

[0114] In one embodiment, it also includes counting the index of the clustering task in strict mode. Similar to the standard mode method, specifically, according to the ratio of the sum of the number of archives of all personnel in strict mode under the clustering task to the total number of personnel, the diffusion rate of the clustering task is obtained. According to the archive with the largest number of captured images of all personnel under the clustering task as the main archive, the accuracy of the clustering task is obtained by the ratio of the total number of confirmed captured images in all main archives to the total number of all captured images of all main archives. The recall rate is obtained by the ratio of the total number of confirmed captured images in all main archives to the total number of confirmed captured images of all personnel. According to the sum of the confirmed captured images in all archives in strict mode under the clustering task and the sum of the confirmed captured images of all labeled personnel under the task in the trajectory labeling table, the clustering rate is obtained.

[0115] Select a marked person to display the marking results of this person. The results include the number of files involved in the person's file, the number of confirmed, pending, and excluded trajectories. You can also switch the indicator calculation mode to observe the files and trajectories in a specific mode. The mode can be switched to the out-of-file mode to observe whether there is captured data outside the file. The study of such data is conducive to the iterative optimization of the subsequent face clustering algorithm.

[0116] like Fig. 9 The figure shows a schematic diagram of an indicator statistical method.

[0117] First, select File Marking and mark the tracks under File 1.

[0118] Furthermore, select the confirmed snapshots for image search. After the labeling is completed, select snapshots from various angles of the confirmed trajectory multiple times for image search and label the recalled data. Perform an image search on the trajectory of the target person in file 1 in the figure to obtain other snapshots of the person. After submitting the labeling results, the background program will query whether the currently confirmed snapshot belongs to a certain file. If so, the file trajectory involved will be initialized and requires secondary labeling. Label the files involved in the recalled snapshots again, and after completing multiple iterations, all snapshots of the person can be obtained.

[0119] Finally, the index of the marked person is calculated according to the index statistical method of step S104 to obtain the final index status of the person.

[0120] This application is extended from restrictive scenarios to open scenarios, and non-invasive real-time deployment and real-time statistics can be implemented in actual production environments, breaking environmental restrictions and greatly improving statistical flexibility. The reliability of indicator statistical results is improved by using all snapshots of a person's file to search for its files instead of the traditional method of searching files by image. The process is solidified by using a program to reduce accidental errors and labor costs caused by manual statistics in the statistical process. Using memory for image search and comparison to obtain all snapshots instead of extracting all snapshot data from the traditional database for analysis and comparison improves the efficiency of personnel statistics.

[0121] According to another aspect of the embodiment of the present application, a face file index statistics device for implementing the above-mentioned face file index statistics method is also provided. Fig.10 As shown, the device comprises:

[0122] The marking module 1001 is used to obtain the file trajectory data corresponding to the personnel file to be marked, mark the file trajectory data, and mark the trajectory snapshot image confirmed to be the target person as confirmed;

[0123] The image search module 1002 is used to obtain other captured images of the target person by using an image search method based on the confirmed captured images;

[0124] The association module 1003 is used to associate other captured images obtained by the image search with the confirmed captured images, obtain the image search association field, and write the file number corresponding to the other captured images and the image search association field into the track annotation table;

[0125] The statistics module 1004 is used to calculate the diffusion rate, accuracy rate and recall rate in the standard mode and the strict mode respectively according to the trajectory annotation table.

[0126] It should be noted that the face clustering index statistical device provided in the above embodiment only uses the division of the above-mentioned functional modules as an example when executing the face clustering index statistical method. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the face clustering index statistical device provided in the above embodiment and the face clustering index statistical method embodiment belong to the same concept, and the implementation process thereof is detailed in the method embodiment, which will not be repeated here.

[0127] According to another aspect of the embodiments of the present application, an electronic device corresponding to the face clustering index statistical method provided in the aforementioned embodiments is also provided to execute the aforementioned face clustering index statistical method.

[0128] Please refer to Fig.11, which shows a schematic diagram of an electronic device provided by some embodiments of the present application. Fig.11 As shown, the electronic device includes: a processor 1100, a memory 1101, a bus 1102 and a communication interface 1103, and the processor 1100, the communication interface 1103 and the memory 1101 are connected via the bus 1102; the memory 1101 stores a computer program that can be run on the processor 1100, and when the processor 1100 runs the computer program, it executes the face clustering index statistical method provided in any of the aforementioned embodiments of the present application.

[0129] The memory 1101 may include a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk storage. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 1103 (which may be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. may be used.

[0130] The bus 1102 may be an ISA bus, a PCI bus, or an EISA bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. Among them, the memory 1101 is used to store programs, and the processor 1100 executes the program after receiving the execution instruction. The face aggregation index statistical method disclosed in any implementation of the embodiment of the present application may be applied to the processor 1100, or implemented by the processor 1100.

[0131] The processor 1100 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the processor 1100. The above processor 1100 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a readily available programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiments of the present application can be directly embodied as a hardware decoding processor to be executed, or the hardware and software modules in the decoding processor can be executed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 1101, and the processor 1100 reads the information in the memory 1101 and completes the steps of the above method in combination with its hardware.

[0132] The electronic device provided in the embodiment of the present application and the face clustering index statistical method provided in the embodiment of the present application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented therein.

[0133] According to another aspect of the embodiments of the present application, a computer-readable storage medium corresponding to the face clustering index statistical method provided in the aforementioned embodiments is also provided, on which a computer program (i.e., a program product) is stored. When the computer program is run by a processor, it will execute the face clustering index statistical method provided in any of the aforementioned embodiments.

[0134] It should be noted that examples of computer-readable storage media may also include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical or magnetic storage media, which are not listed here one by one.

[0135] The computer-readable storage medium provided in the above-mentioned embodiments of the present application and the face clustering index statistical method provided in the embodiments of the present application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application programs stored therein.

[0136] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0137] The above embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present invention. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention patent shall be subject to the attached claims.

Claims

1. A statistical method for face clustering index, It is characterized in that include: Obtaining the file trajectory data corresponding to the personnel file to be marked, marking the file trajectory data, and marking the trajectory snapshot image confirmed to be the target person as confirmed; Based on the confirmed captured image, other captured images of the target person are acquired by image search; Associating the other captured images obtained by image search with the confirmed captured images to obtain image search associated fields, and writing the file numbers corresponding to the other captured images and the image search associated fields into the track annotation table; According to the trajectory annotation table, the diffusion rate, accuracy rate and recall rate of the target personnel files in the standard mode and the strict mode are respectively counted.

2. The method according to claim 1, It is characterized in that Obtain the file trajectory data corresponding to the personnel file to be labeled, including: Query the personnel files to be marked according to one or more of the pre-set file gathering tasks, file generation time and track number; Obtain the maximum trajectory time and the minimum time index value of the personnel file to be marked; The larger time between the maximum trajectory time and the minimum time index value is used as the query trajectory start time; The trajectory data corresponding to the personnel file to be marked is obtained according to the query trajectory start time.

3. The method according to claim 1, It is characterized in that Based on the confirmed captured image, other captured images of the target person are obtained by image search, including: Respectively extracting facial feature values ​​of the confirmed captured images at different angles; Traverse and extract facial feature values ​​of other images outside the files to be annotated; Similarity calculation is performed based on the facial feature value, and other images whose similarity with the confirmed captured image is greater than a threshold are recalled to obtain other captured images of the target person.

4. The method according to claim 1, It is characterized in that The diffusion rate, accuracy rate and recall rate of the target personnel files under the statistical standard mode include: According to the trajectory marking table, determining the number of splitting files in the standard mode, and obtaining the diffusion rate according to the number of splitting files in the standard mode; Obtain the file with the largest number of confirmed snapshot images in the split files under the standard mode as the main file, and obtain the accuracy rate according to the ratio of the total number of confirmed snapshot images in the main file to the number of all snapshot images in the main file; Obtaining a recall rate according to a ratio of the total number of confirmed snapshot images in the master file to the total number of all confirmed snapshot images of the target person in the trajectory annotation table; Among them, in the standard mode, when there are multiple files of the captured images of the target person, and the number of captured images of the target person in a certain file accounts for less than the preset threshold of the total number of captured images in the file, it is considered that the file is not a split file.

5. The method according to claim 1, It is characterized in that Statistics on the diffusion rate, accuracy and recall rate of the target personnel files in strict mode, including: According to the trajectory marking table, determining the number of split files in the strict mode, and obtaining the diffusion rate according to the number of split files in the strict mode; The accuracy rate is obtained according to the ratio of the total number of confirmed snapshot images in all split files to the total number of snapshot images in all split files in the strict mode; The recall rate is obtained based on the ratio of the total number of confirmed snapshot images in all split files to the total number of all snapshot images of the target person in the trajectory annotation table; Among them, in strict mode, when a file contains a captured image of a target person, the file is considered to be a split file.

6. The method according to claim 1, It is characterized in that Also includes: According to the ratio of the sum of the number of files of all personnel in the standard mode under the file gathering task to the total number of personnel, the diffusion rate of the file gathering task is obtained; According to the archive with the most captured images of all personnel in the archive gathering task, the accuracy of the archive gathering task is obtained by taking the ratio of the total number of confirmed captured images in all the main archives to the total number of captured images in all the main archives; The recall rate of the clustering task is obtained by taking the ratio of the total number of confirmed snapshot images in all master archives to the total number of confirmed snapshot images of all personnel; The clustering rate of the archiving task is obtained based on the ratio of the sum of the confirmed snapshot images in all archives in the standard mode under the archiving task to the sum of the confirmed snapshot images of all annotated personnel under the task in the trajectory annotation table.

7. The method according to claim 1, It is characterized in that Also includes: According to the ratio of the sum of the number of files of all personnel in the strict mode under the file gathering task to the total number of personnel, the diffusion rate of the file gathering task is obtained; According to the archive with the most captured images of all personnel in the archive gathering task, the accuracy of the archive gathering task is obtained by taking the ratio of the total number of confirmed captured images in all the main archives to the total number of captured images in all the main archives; The recall rate of the clustering task is obtained by taking the ratio of the total number of confirmed snapshot images in all master archives to the total number of confirmed snapshot images of all personnel; The clustering rate of the archiving task is obtained based on the ratio of the sum of the confirmed snapshot images in all archives in strict mode under the archiving task to the sum of the confirmed snapshot images of all annotators under the task in the trajectory annotation table.

8. A face aggregation index statistical device, It is characterized in that include: A labeling module is used to obtain the file trajectory data corresponding to the personnel file to be labeled, label the file trajectory data, and label the trajectory snapshot image confirmed to be the target person as confirmed; An image search module is used to obtain other captured images of the target person by image search based on the confirmed captured image; An association module, used to associate the other captured images obtained by image search with the confirmed captured images, obtain image search association fields, and write the file numbers corresponding to the other captured images and the image search association fields into the track annotation table; The statistical module is used to count the diffusion rate, accuracy rate and recall rate of the target personnel files in the standard mode and the strict mode respectively according to the trajectory annotation table.

9. An electronic device, It is characterized in that It includes a processor and a memory storing program instructions, and the processor is configured to execute the face clustering index statistical method as described in any one of claims 1 to 7 when executing the program instructions.

10. A computer readable medium, It is characterized in that Computer-readable instructions are stored thereon, and the computer-readable instructions are executed by a processor to implement a face clustering index statistical method as described in any one of claims 1 to 7.