A face clustering method, device and storage medium

Abstract

The present disclosure relates to the field of image clustering, and discloses a face clustering method, device and storage medium. The method comprises the following steps: determining that the similarity between a target face image and a preset first face template is between a dissimilarity judgment value and a similarity judgment value, the dissimilarity judgment value representing that the target face image cannot be clustered into the identity corresponding to the first face template, and the similarity judgment value representing that the target face image can be clustered into the identity corresponding to the first face template; if the target face image has at least one corresponding second face image in a target same-trip period, then taking the identity corresponding to the first face template as the identity corresponding to the target face image, and the identity corresponding to a second face template into which the at least one second face image is clustered and the identity corresponding to the first face template are in a same-trip relationship, the same-trip relationship being determined by the snapshot records of the identity corresponding to the first face template and the identity corresponding to the second face template. With the aid of the same-trip relationship, the accuracy of face clustering is improved.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of image clustering, and provides a face clustering method, device and storage medium. BACKGROUND

[0002] With the continuous innovation and development of face recognition technology, relevant industries and departments have built monitoring technology based on face recognition. At present, the monitoring system based on face recognition technology can not only perform face recognition to extract attribute information such as age and gender of the face from the picture, but also can cluster the face to depict the travel trajectory of the person.

[0003] In use, high-accuracy face clustering technology can provide great convenience for subsequent work, otherwise, high-miss rate leads to the inability to completely depict the travel trajectory of the person. The existing face clustering technology usually only uses the feature information of the face for comparison, and often appears the situation that the clustering effect is poor due to the monitoring angle and the picture capture quality.

[0004] In summary, there is no effective solution to guarantee the accuracy of the face clustering process at present. SUMMARY

[0005] The embodiments of the present disclosure provide a face clustering method, device and storage medium to guarantee the accuracy of the face clustering process.

[0006] The specific technical solutions provided by the present disclosure are as follows:

[0007] In a first aspect, the embodiments of the present disclosure provide a face clustering method, comprising:

[0008] determining that the similarity between the target face image and the preset first face template is located between a dissimilarity determination value and a similarity determination value; wherein the dissimilarity determination value represents that the target face image cannot be clustered into the identity corresponding to the first face template, and the similarity determination value represents that the target face image can be clustered into the identity corresponding to the first face template;

[0009] if the target face image has at least one second face image corresponding thereto within the target travel period, taking the identity corresponding to the first face template as the identity corresponding to the target face image; wherein the identity corresponding to the second face template to which the at least one second face image is clustered is in a travel relationship with the identity corresponding to the first face template, and the travel relationship is determined by the capture records of the identity corresponding to the first face template and the identity corresponding to the second face template.

[0010] Optionally, the first face template is an ID photo of the identity corresponding to the first face template; and / or

[0011] The second face template is an ID photo of an identity corresponding to the second face template.

[0012] Optionally, the same-trip relationship is determined by the following manner:

[0013] In each of the plurality of same-trip time periods, a snapshot record of the identity corresponding to the first face template and the identities corresponding to the at least one preselected face template is obtained, wherein the snapshot record includes the first face image and a plurality of preselected face images, and the time periods corresponding to the plurality of same-trip time periods are earlier than the time period corresponding to the target same-trip time period;

[0014] In each of the same-trip time periods, if a time difference between the first face image and the preselected face image in the snapshot record is greater than a set time difference threshold, and the number of times that the time difference between the first face image and the preselected face image in each of the same-trip time periods is greater than the time difference threshold exceeds a preset number of times, it is determined that the identity corresponding to the first face template and the identities corresponding to the plurality of preselected face templates are in a same-trip relationship, wherein the second face template is one of the plurality of preselected face templates, and the smaller the time difference threshold is, the closer the time points at which the first face image and the preselected face image appear in the same location.

[0015] Optionally, obtaining the snapshot record of the identity corresponding to the first face template and the identities corresponding to the at least one preselected face template comprises:

[0016] determining a shooting device for obtaining the first face image, and determining an obtaining time of the first face image; and

[0017] determining, based on the same shooting device, a plurality of face images in a candidate time interval with the first face image;

[0018] taking the first face image as a first snapshot record corresponding to the first face template, taking the plurality of face images as a second snapshot record corresponding to the second face template, and taking a union of the first snapshot record and the second snapshot record as the snapshot record of the identity corresponding to the first face template and the identities corresponding to the at least one second face template.

[0019] Optionally, the method further comprises:

[0020] if there is no second face image corresponding to the target face image in the target same-trip time period, selecting any one of the preselected face templates other than the second face template from the preselected face templates, and updating the identity corresponding to the second face template to the identity corresponding to the selected preselected face template;

[0021] based on the identity corresponding to the preselected face template, obtaining a preselected second face image whose shooting time is within a preset time difference threshold from the obtaining time;

[0022] If the preselected similarity value between the preselected second face image and the identity corresponding to the second face template is greater than the preselected similarity determination value, the identity corresponding to the first face template at the acquisition time is taken as the identity corresponding to the target face image; wherein the preselected similarity determination value represents that the preselected second face image can be clustered into the identity corresponding to the second face template.

[0023] Optionally, the set time difference threshold is determined based on a shooting interval of the shooting device.

[0024] The pre-set number of times is determined based on the density of the shooting device.

[0025] Optionally, the density is determined in the following manner:

[0026] The total length corresponding to the action trajectory of the first face image in the first snapshot record is obtained.

[0027] The number of first face images included in the first snapshot record is determined.

[0028] Based on the total length and the number, the density is determined.

[0029] Optionally, if the plurality of first face images are obtained by a plurality of shooting devices in one shooting device group, the first face image is determined in the following manner, comprising:

[0030] The association relationship between the plurality of shooting devices in one shooting device group is obtained.

[0031] Based on the association relationship, a plurality of preselected first face images are obtained by each shooting device respectively.

[0032] The plurality of preselected first face images are taken as the first face image.

[0033] Optionally, if the plurality of first face images are obtained by a plurality of shooting devices in one shooting device group, based on the same shooting device, a plurality of face images in the first face image within the alternative time interval are determined, comprising:

[0034] Based on the association relationship, the device identifier of the shooting device corresponding to the first face image to be compared is determined among the plurality of shooting devices in one shooting device group.

[0035] Based on the device identifier, a plurality of face images in the first face image within the alternative time interval are determined.

[0036] Optionally, if the target face image has at least one corresponding second face image in the target same period, the identity corresponding to the first face template is taken as the identity corresponding to the target face image, comprising:

[0037] determine the acquisition time of the target face image, and determine the identity corresponding to the second face template as the identity corresponding to the first face template in the target same-trip period;

[0038] based on the identity corresponding to the second face template, acquire any one of the second face images captured at the acquisition time and the acquisition time within the preset time difference threshold;

[0039] If the similarity value between the any one of the second face images and the identity corresponding to the second face template is greater than the similarity determination value, the identity corresponding to the first face template at the acquisition time is taken as the identity corresponding to the target face image; wherein the similarity determination value represents that the second face image can be clustered into the identity corresponding to the second face template.

[0040] In a second aspect, the embodiments of the present disclosure further provide a device for face clustering, comprising:

[0041] a similarity determination unit configured to determine that the similarity between the target face image and the preset first face template is between the non-similarity determination value and the similarity determination value; wherein the non-similarity determination value represents that the target face image cannot be clustered into the identity corresponding to the first face template, and the similarity determination value represents that the target face image can be clustered into the identity corresponding to the first face template;

[0042] an identity determination unit configured to, if the target face image has at least one corresponding second face image in the target same-trip period, take the identity corresponding to the first face template as the identity corresponding to the target face image; wherein the identity corresponding to the second face template to which the at least one second face image is clustered is in a same-trip relationship with the identity corresponding to the first face template, and the same-trip relationship is determined by the snapshot record of the identity corresponding to the first face template and the identity corresponding to the second face template.

[0043] Optionally, the first face template is an ID photo of the identity corresponding to the first face template; and / or

[0044] the second face template is an ID photo of the identity corresponding to the second face template.

[0045] Optionally, the same-trip relationship is determined by the following way:

[0046] In each of the plurality of same-trip periods, the snapshot record of the identity corresponding to the first face template and the identity corresponding to at least one preselected face template is acquired, wherein the snapshot record includes the first face image and a plurality of preselected face images, and the time corresponding to each of the plurality of same-trip periods is earlier than the time corresponding to the target same-trip period;

[0047] If the time difference between the first face image and the preselected face image in the snapshot record in each same-trip period is greater than a set time difference threshold, and the number of times that the time difference between the first face image and the preselected face image in each same-trip period is greater than the time difference threshold exceeds a preset number of times, the identity corresponding to the first face template and the identities corresponding to the plurality of preselected face templates are determined to be in a same-trip relationship, wherein the second face template is one of the plurality of preselected face templates, and the smaller the time difference threshold is, the closer the time points at which the first face image and the preselected face image appear in the same location are.

[0048] Optionally, the snapshot record of the identity corresponding to the first face template and the identity corresponding to at least one preselected face template is obtained, including:

[0049] The shooting device for obtaining the first face image is determined, and the time point for obtaining the first face image is determined.

[0050] Based on the same shooting device, a plurality of face images in a candidate time interval with the first face image are determined.

[0051] The first face image is taken as a first snapshot record corresponding to the first face template, the plurality of face images are taken as second snapshot records corresponding to second face templates, and the union of the first snapshot record and the second snapshot record is taken as the snapshot record of the identity corresponding to the first face template and the identity corresponding to at least one second face template.

[0052] Optionally, the method further includes:

[0053] If there is no second face image corresponding to the target face image in the target same-trip period, any one of the preselected face templates different from the second face template is selected from the preselected face templates, the preselected face template in a same-trip relationship with the first face template is selected, and the identity corresponding to the second face template is updated to the identity corresponding to the selected preselected face template.

[0054] Based on the identity corresponding to the preselected face template, a preselected second face image is obtained, whose shooting time point and obtaining time point are within a preset time difference threshold.

[0055] If the preselected similarity value between the preselected second face image and the identity corresponding to the second face template is greater than a preselected similarity determination value, the identity corresponding to the first face template at the obtaining time point is taken as the identity corresponding to the target face image, wherein the preselected similarity determination value represents that the preselected second face image can be clustered into the identity corresponding to the second face template.

[0056] Optionally, the set time difference threshold is determined based on a shooting interval of the shooting device.

[0057] The preset number of times is determined based on the density of the shooting device.

[0058] Optionally, the density is determined by the following manner:

[0059] Obtaining a total length corresponding to the action track of the first face image in the first snapshot record;

[0060] Determining the number of the first face images included in the first snapshot record;

[0061] Determining the density based on the total length and the number.

[0062] Optionally, if the plurality of first face images are obtained by a plurality of shooting devices in one shooting device group, the first face images are determined by the following manner, comprising:

[0063] Obtaining the association relationship between the plurality of shooting devices in one shooting device group;

[0064] Based on the association relationship, obtaining a plurality of pre-selected first face images by each shooting device respectively;

[0065] Taking the plurality of pre-selected first face images as the first face images.

[0066] Optionally, if the plurality of first face images are obtained by a plurality of shooting devices in one shooting device group, the plurality of face images in the alternative time interval with the first face image are determined based on the same shooting device, comprising:

[0067] Based on the association relationship, determining the device identifier of the shooting device corresponding to the first face image to be compared among the plurality of shooting devices in one shooting device group;

[0068] Based on the device identifier, determining the plurality of face images in the alternative time interval with the first face image.

[0069] Optionally, if the target face image has at least one second face image corresponding thereto in the target same-travel time period, the identity corresponding to the first face template is taken as the identity corresponding to the target face image, and the identity determining unit is configured to:

[0070] In the target same-travel time period, determining the acquisition time of the target face image, and determining the identity corresponding to the first face template as the identity corresponding to at least one second face template of the same-travel relationship;

[0071] Based on the identity corresponding to the second face template, obtaining any one of the at least one second face image whose shooting time is within a preset time difference threshold from the acquisition time;

[0072] If the similarity value between any one second face image and the identity corresponding to the second face template is greater than the similarity determination value, the identity corresponding to the first face template at the acquisition time is taken as the identity corresponding to the target face image; wherein the similarity determination value represents that the second face image can be clustered into the identity corresponding to the second face template.

[0073] In a third aspect, an intelligent terminal comprises:

[0074] a memory for storing executable instructions;

[0075] a processor for reading and executing the executable instructions stored in the memory to implement the method of any one of the first aspect.

[0076] In a fourth aspect, a computer readable storage medium, when the instructions in the storage medium are executed by a processor, enable the processor to execute the method of any one of the first aspect.

[0077] The present disclosure has the following beneficial effects:

[0078] In summary, in the embodiments of the present disclosure, the method, device and storage medium for face clustering are provided, and the method comprises: determining that the similarity between a target face image and a preset first face template is located between a dissimilarity determination value and a similarity determination value, i.e. the identity to which the target face image belongs cannot be directly determined by the similarity determination value, wherein the dissimilarity determination value represents that the target face image cannot be clustered into the identity corresponding to the first face template, and the similarity determination value represents that the target face image can be clustered into the identity corresponding to the first face template; if the target face image has at least one second face image corresponding thereto within a target same-trip time period, the identity corresponding to the first face template is taken as the identity corresponding to the target face image, wherein the at least one second face image is clustered into a second face template corresponding to an identity, and the identity corresponding to the first face template and the identity corresponding to the second face template have a same-trip relationship, and the same-trip relationship is determined by the snapshot record of the identity corresponding to the first face template and the identity corresponding to the second face template. In this way, in the case that the identity to which the target face image belongs cannot be directly determined by the similarity determination value, the identity to which the target face image belongs can be determined by the same-trip relationship, thereby improving the accuracy of face clustering.

[0079] Other features and advantages of the present disclosure will be set forth in the following description, and in part will become apparent to those skilled in the art from the description, or can be learned by practice of the present disclosure. The objects and other advantages of the present disclosure can be achieved and obtained by the structures particularly pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF DRAWINGS

[0080] The accompanying drawings, which are included to provide a further understanding of the disclosure and constitute a part of the disclosure, illustrate the preferred embodiments of the disclosure and serve to explain the principles of the disclosure, and do not constitute improper limitations on the disclosure. In the drawings:

[0081] Figure 1 A system architecture diagram for clustering faces in an embodiment of the disclosure;

[0082] Figure 2 A flow diagram for clustering faces in an embodiment of the disclosure;

[0083] Figure 3 A flow diagram for determining a same-row relationship in an embodiment of the disclosure;

[0084] Figure 4 A flow diagram for determining a snapshot record in an embodiment of the disclosure;

[0085] Figure 5 A flow diagram for determining an identity corresponding to a target face image in an embodiment of the disclosure;

[0086] Figure 6 A logical architecture diagram of a face clustering device in an embodiment of the disclosure;

[0087] Figure 7 An entity architecture diagram of a computing device in an embodiment of the disclosure. DETAILED DESCRIPTION

[0088] To make the purposes, technical solutions and advantages of the embodiments of the disclosure clearer, the technical solutions of the disclosure will be described clearly and completely below with reference to the drawings in the embodiments of the disclosure. Obviously, the described embodiments are part of the technical solutions of the disclosure, rather than all the embodiments. Based on the embodiments described in the disclosure document, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the scope of protection of the technical solutions of the disclosure.

[0089] The terms "first", "second", and the like in the specification and claims of the disclosure and the above drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the disclosure described herein can be implemented in an order other than that illustrated or described herein.

[0090] The preferred embodiments of the disclosure will be described in detail below with reference to the accompanying drawings.

[0091] Referring to Figure 1 In an embodiment of the disclosure, the system includes at least one computing device and a shooting device, and the system is configured to Figure 1In the embodiment, the photographing device is exemplarily represented by a camera. It should be noted that, in order to obtain the snapshot record of the identity corresponding to the first face template and the identity corresponding to the second face template, the photographing device is usually a network camera, and the number of photographing devices is multiple, and each of the multiple photographing devices has a unique network number, which is used to obtain the snapshot record of the identity corresponding to the first face template and the identity corresponding to the second face template, that is, the first face image, the multiple face images, and the like are continuously photographed by multiple network cameras.

[0092] In the embodiment of the present disclosure, the method for face clustering is mainly implemented on a computing device, that is, the computing device performs face clustering processing after obtaining the image photographed by the camera device. Details are described below. Referring to Figure 2 As shown in the figure, in the embodiment of the present disclosure, the specific process of face clustering is as follows:

[0093] Step 201: determining that the similarity between the target face image and the preset first face template is located between the non-similarity determination value and the similarity determination value; wherein the non-similarity determination value represents that the target face image cannot be clustered into the identity corresponding to the first face template, and the similarity determination value represents that the target face image can be clustered into the identity corresponding to the first face template.

[0094] First of all, it should be noted that the first face template is an ID photo of the identity corresponding to the first face template; and / or, the second face template is an ID photo of the identity corresponding to the second face template. In the implementation process, in order to accurately determine the identity of the same person, the first face template is set as the corresponding ID photo in advance, for example, the ID photo of the identity corresponding to the first face template.

[0095] In the similarity determination process, if the similarity between the target face image and the ID photo of the identity corresponding to the first face template is greater than or equal to the similarity determination value, it can be directly determined that the target face image can be clustered into the identity corresponding to the first face template; if the similarity between the target face image and the ID photo of the identity corresponding to the first face template is less than the non-similarity determination value, that is, the similarity between the target face image and the ID photo of the identity corresponding to the first face template is very low (even the similarity is zero), it can be directly determined that the target face image cannot be clustered into the identity corresponding to the first face template. Obviously, the value of the similarity determination value is greater than the value of the non-similarity determination value.

[0096] In the prior art, if the similarity between the target face image and the first face template is less than the similarity determination value, it can also be directly determined that the target face image cannot be clustered into the identity corresponding to the first face template, i.e., the target face image is not the identity corresponding to the first face template. However, in fact, the similarity between the target face image and the first face template is relatively high, and there may be a case that the identity corresponding to the target face image is the identity corresponding to the first face template. Obviously, in the processing process of the prior art, a misjudgment will occur.

[0097] In the embodiment of the present application, in order to avoid the occurrence of the above misjudgment, after it is determined that the similarity between the target face image and the preset first face template is between the dissimilarity determination value and the similarity determination value, the identity corresponding to the target face image is further determined by means of the same-traveling relationship, so as to improve the accuracy of identity determination.

[0098] It should be noted that, as shown in Figure 3 The same-traveling relationship is determined by the following method:

[0099] Step 101: In each of the plurality of same-traveling time periods, obtain the snapshot record of the identity corresponding to the first face template and the identity corresponding to at least one preselected face template, wherein the snapshot record includes the first face image and a plurality of preselected face images, and the time corresponding to the plurality of same-traveling time periods is earlier than the time corresponding to the target same-traveling time period;

[0100] Considering that in actual application scenarios, there can be multiple objects traveling with the identity corresponding to the first face template, and the determination of the object traveling with the identity corresponding to the first face template is related to the length of time. Therefore, in the embodiment of the present application, whether the first face template and the at least one preselected face template travel together is determined according to the snapshot record of the first face template and the at least one preselected face template in the plurality of same-traveling time periods earlier than the target same-traveling time period.

[0101] Specifically, the snapshot record of the identity corresponding to the first face template and the identity corresponding to at least one preselected face template is obtained, as shown in Figure 4 , which includes:

[0102] Step 1011: Determine the shooting device for obtaining the first face image, and determine the time for obtaining the first face image; and

[0103] Since the same period is a period of time, and in fact, there can be multiple shooting devices set at the intersection, and the basis for determining whether two or more objects are in the same period is that they appear in the video captured by the shooting device in the same frame (or a very short time apart). Therefore, in the implementation process, in order to accurately lock other images in the same period as the first face image, the shooting device that obtains the first face image is determined, that is, the identity of the shooting device is determined, and on this basis, the time when the first face image is obtained is determined. Usually, the above-mentioned acquisition time is accurate to the second level.

[0104] Step 1012: Based on the same shooting device, determine multiple face images in the first face image within the candidate time interval.

[0105] It should be noted that the candidate time interval is a maximum value of the time interval that can be used to determine whether two or more objects are in the same period, which is set in advance according to statistics. The same camera device here is the shooting device that obtains the first face image. In the implementation process, after the shooting device is determined, multiple face images in the first face image within the candidate time interval can be obtained through the shooting device, so as to pre-screen multiple face images as candidates for determining the same period as the first face image.

[0106] Step 1013: Take the first face image as the first snapshot record corresponding to the first face template, take the multiple face images as the second snapshot record corresponding to the second face template, and take the union of the first snapshot record and the second snapshot record as the snapshot record of the identity corresponding to the first face template and the identity corresponding to at least one second face template.

[0107] Considering that whether two or more objects are in the same period involves a long time, after steps 1011 and 1012 are performed for each same period, the first face image is taken as the first snapshot record corresponding to the first face template, the multiple face images are taken as the second snapshot record corresponding to the second face template, and the union of the first snapshot record and the second snapshot record is taken as the snapshot record of the identity corresponding to the first face template and the identity corresponding to at least one second face template.

[0108] In this way, after the above operations are performed in multiple same periods, all the basis for determining whether the identity corresponding to the first face template and the identity corresponding to at least one second face template are in the same period is obtained.

[0109] Step 102: In each co-travel period, if the time difference between the first face image and the preselected face image in the snapshot record is greater than the set time difference threshold, and the number of times that the time difference between the first face image and the preselected face image in each co-travel period is greater than the time difference threshold exceeds the preset number of times, it is determined that the identity corresponding to the first face template and the identities corresponding to the plurality of preselected face templates are in a co-travel relationship, wherein the second face template is one of the plurality of preselected face templates, and the smaller the time difference threshold, the closer the time points at which the first face image and the preselected face image appear in the same location.

[0110] In the implementation process, for the snapshot record corresponding to each co-travel period, the identities of the plurality of preselected face templates in the true co-travel relationship are again screened, and the plurality of preselected face templates are all provided with corresponding ID photos. The second face template is one of the plurality of preselected face templates.

[0111] Specifically, two conditions are set for the determination of the co-travel relationship: 1) the set time difference threshold. The smaller the time difference threshold, the closer the time points at which the first face image and the preselected face image appear in the same location. If the time difference threshold is zero, it means that the first face image and the preselected face image are in the same frame (or separated by a very short time); 2) the preset number of times. If the above snapshot record shows that the first face image and the preselected face image are in the same frame only once, it cannot be determined that the first face image and the preselected face image are in a co-travel relationship.

[0112] Specifically, the set time difference threshold is determined based on the shooting interval of the shooting device. Obviously, when the shooting interval of the shooting device is large, the probability of the first face image and the preselected face image being in the same frame in the captured video is small, or in other words, the time interval between the first face image and the preselected face image in the captured video will be large, in which case the set time difference threshold should also be set to a longer interval; when the shooting interval of the shooting device is small, the probability of the first face image and the preselected face image being in the same frame in the captured video is large, or in other words, the time interval between the first face image and the preselected face image in the captured video will be small, in which case the set time difference threshold should also be set to a shorter interval.

[0113] The above-mentioned preset number of times is determined based on the density of the shooting device.

[0114] First, an example is given. If there are only two shooting devices on the same road, the above-mentioned preset number of times is set to a maximum of 2, because if A and B are in a co-travel relationship, the shooting device can at most capture A and B twice, and in actual application, when a large number of shooting devices (for example, 100) are set on the same road, it is obviously inappropriate to determine that A and B are in a co-travel relationship only by 2 times.

[0115] To this end, the preset number of times is determined by the density of the shooting device, where the density is determined in the following manner:

[0116] (1) Obtain the total length corresponding to the action trajectory of the first face image in the first snapshot record, and in the implementation process, the total length corresponding to the action trajectory in the first snapshot record is determined by summing the straight-line distance of the first face image movement of each two adjacent same period, and the total length represents the length of the total distance of the same period.

[0117] (2) Determine the number of first face images included in the first snapshot record, and in the implementation process, the number of first face images can be counted in the snapshot record according to a fixed time interval, and the number of first face images represents the frequency of the first face image appearing in the first snapshot record.

[0118] (3) Determine the density based on the total length and the number, and in the implementation process, the total length and the number are usually used as a quotient, and the quotient is taken as the density.

[0119] Of course, since the first snapshot record of the first face image and the second snapshot record corresponding to the preselected face image are obtained by the same shooting device, the acquisition of the density is only described in the first snapshot record.

[0120] In another embodiment, if multiple first face images are obtained by multiple shooting devices in a shooting device group, the first face image is determined in the following manner:

[0121] [1] Obtain the association relationship between multiple shooting devices in a shooting device group.

[0122] In the implementation process, in order to obtain more accurate three-dimensional first face images, a shooting device group can be set, that is, multiple shooting devices are used to obtain first face images at the intersection, for example, four shooting devices are set in a shooting device group, and the four shooting devices obtain the first face image from the upper, lower, left and right directions. In this case, the association relationship between the multiple shooting devices is the directional relationship of the upper, lower, left and right directions.

[0123] Of course, the association relationship between the multiple shooting devices can also be flexibly set according to the use scene.

[0124] [2] Based on the association relationship, multiple preselected first face images are obtained by each shooting device.

[0125] Continuing to take the example of setting four shooting devices in one shooting device group, which acquire the first face image from the upper, lower, left, and right four directions, in the implementation process, at the same acquisition time, multiple preselected first face images in the upper, lower, left, and right four directions are acquired by the respective shooting devices.

[0126] [3] The multiple preselected first face images are taken as the first face image.

[0127] Continuing to take the example of setting four shooting devices in one shooting device group, which acquire the first face image from the upper, lower, left, and right four directions, after obtaining the multiple preselected first face images, the multiple preselected first face images are processed (for example, merged, etc.) according to the upper, lower, left, and right four directions, so that a more accurate first face image can be obtained from the multiple preselected first face images, thereby avoiding the situation of unclear first face image, omission, etc. caused by the angle of the shooting device.

[0128] Further, if the multiple first face images are acquired by multiple shooting devices in one shooting device group, based on the same shooting device, the multiple face images in the candidate time interval with the first face image are determined, including:

[0129] 1] Based on the association relationship, in the multiple shooting devices in one shooting device group, the device identifier of the shooting device corresponding to the first face image to be compared is determined.

[0130] In the implementation process, in order to make the comparison of whether the identity is a same-trip relationship more accurate, in the case where the first face image is obtained from multiple preselected first face images, when determining the multiple face images in the candidate time interval with the first face image, first, based on the association relationship, the device identifier of the shooting device corresponding to the first face image to be compared is determined in one shooting device group. Specifically, in the multiple shooting devices in one shooting device group, the device identifiers of the respective shooting devices that acquire the above preselected first face images are sequentially searched.

[0131] 2] Based on the device identifier, the multiple face images in the candidate time interval with the first face image are determined.

[0132] After obtaining the device identifier, multiple face images that may be in a same-trip relationship with the first face image in time, i.e., multiple preselected second face images in the candidate time interval with the first face image, are screened, and after processing the multiple preselected second face images based on the device identifier and the association relationship, the multiple face images are obtained.

[0133] Step 202: If there is at least one second face image corresponding to the target face image in the target co-travel time period, the identity corresponding to the first face template is taken as the identity corresponding to the target face image; wherein the identity corresponding to the second face template clustered by the at least one second face image is in a co-travel relationship with the identity corresponding to the first face template, and the co-travel relationship is determined by the snapshot records of the identity corresponding to the first face template and the identity corresponding to the second face template.

[0134] It should be noted that the identity corresponding to the second face template clustered by the at least one second face image is in a co-travel relationship with the identity corresponding to the first face template, that is, by default, the similarity between the second face image and the identity corresponding to the second face template is greater than the similarity determination value, that is, on the basis of the identity corresponding to the second face template clustered by the second face image, the identity corresponding to another target face image in a co-travel relationship is determined as the identity corresponding to the first face template.

[0135] Specifically, if there is at least one second face image corresponding to the target face image in the target co-travel time period, the identity corresponding to the first face template is taken as the identity corresponding to the target face image, as shown in FIG. 2, which includes the following steps: Figure 5

[0136] Step 2021: In the target co-travel time period, the acquisition time of the target face image is determined, and the identity corresponding to the first face template is determined as the identity corresponding to at least one second face template in a co-travel relationship.

[0137] In order to facilitate the search for objects in a co-travel relationship with the target face image, the time range of the search is further narrowed in the target co-travel time period, that is, the acquisition time of the target face image is determined. In addition, the identity corresponding to at least one second face template in a co-travel relationship with the identity corresponding to the first face template needs to be obtained in advance, and the co-travel relationship is determined by the snapshot records of multiple co-travel time periods earlier than the target co-travel time period.

[0138] Step 2022: Based on the identity corresponding to the second face template, any one of the at least one second face image whose shooting time and acquisition time are within a preset time difference threshold is obtained.

[0139] After the identity corresponding to the second face template is determined, any one of the at least one second face image whose shooting time and acquisition time are within a preset time difference threshold in the second snapshot record corresponding to the identity is further searched, that is, one second face image in a co-travel relationship with the above-mentioned target face image in the same co-travel time period is found.

[0140] ​Step 2023: If the similarity value between any one of the second face images and the identity corresponding to the second face template is greater than the similarity determination value, the identity corresponding to the first face template at the acquisition time is taken as the identity corresponding to the target face image; wherein the similarity determination value represents that the second face image can be clustered into the identity corresponding to the second face template.

[0141] In the implementation process, if the similarity value between one of the second face images found in the above step 2022 and the identity corresponding to the second face template pre-acquired in step 2021 is greater than the similarity determination value, i.e., the similarity between the second face image and the identity corresponding to the second face template can enable the second face image to be clustered into the identity corresponding to the second face template, in this case, the identity corresponding to the first face template at the acquisition time is taken as the identity corresponding to the target face image.

[0142] For example, the target face image A and the second face image B are in the same relationship, i.e., it is known in advance that the identity 1 corresponding to A and the identity 2 corresponding to B are in the same relationship, and on the basis of the clustering of the second face image B into the identity 2 in the target same relationship time period, it is determined that the target face image A is clustered into the identity 1.

[0143] In addition, considering the situation that the shooting device may suddenly malfunction and be unable to acquire a second face image, etc., the above method further includes:

[0144] 1) If there is no corresponding second face image of the target face image in the target same relationship time period, any one of the pre-selected face templates different from the second face template and the pre-selected face template corresponding to the identity in the same relationship with the first face template is selected, and the identity corresponding to the second face template is updated to the identity corresponding to the selected pre-selected face template.

[0145] In the implementation process, if there is no corresponding second face image of the target face image in the target same relationship time period, in this case, the existing second face template needs to be replaced, since the identity corresponding to the first face template and the identities corresponding to the plurality of pre-selected face templates are in the same relationship, and the second face template is one of the plurality of pre-selected face templates. In this way, any one of the pre-selected face templates different from the second face template and the pre-selected face template corresponding to the identity in the same relationship with the first face template is selected from the pre-selected face templates, and the identity corresponding to the second face template is updated to the identity corresponding to the selected pre-selected face template.

[0146] Still taking the above example, the target face image A and the second face image B are in the same relationship, i.e., it is known in advance that the identity 1 corresponding to A and the identity 2 corresponding to B are in the same relationship, and in the target same relationship time period, there is no corresponding second face image B, identity 3 in the same relationship with the identity 1 corresponding to A is selected, and the second face template corresponding to the identity 2 is updated to the second face template corresponding to the identity 3.

[0147] 2) based on the identity corresponding to the pre-selected face template, obtaining a pre-selected second face image whose shooting time and acquisition time are within a preset time difference threshold.

[0148] After replacing the existing second face template, the face image in the snapshot record matched with the updated identity needs to be replaced and updated. Specifically, based on the identity corresponding to the pre-selected face template, a pre-selected second face image whose shooting time and acquisition time are within a preset time difference threshold is obtained in the target co-travel period, that is, a pre-selected second face image that meets the co-travel relationship is found.

[0149] Still taking the above example, the pre-selected second face image C that meets the co-travel relationship is found, and it is determined that the identity corresponding to the pre-selected face template corresponding to the pre-selected second face image C is identity 3.

[0150] 3) If the pre-selected similarity value between the pre-selected second face image and the identity corresponding to the second face template is greater than the pre-selected similarity determination value, the identity corresponding to the first face template at the acquisition time is taken as the identity corresponding to the target face image; wherein the pre-selected similarity determination value represents that the pre-selected second face image can be clustered into the identity corresponding to the second face template.

[0151] In the specific implementation process, after the pre-selected second face image is found, it is further judged whether the pre-selected similarity value between the pre-selected second face image and the identity corresponding to the second face template is greater than the pre-selected similarity determination value, that is, whether the pre-selected second face image can be clustered into the identity corresponding to the updated second face template. If the pre-selected similarity value between the pre-selected second face image and the identity corresponding to the second face template is greater than the pre-selected similarity determination value, the identity corresponding to the first face template at the acquisition time is taken as the identity corresponding to the target face image.

[0152] Still taking the above example, after it is determined that the pre-selected second face image C can be clustered into the identity 3 corresponding to the pre-selected face template in the co-travel relationship, it can be determined that the target face image A is clustered into identity 1.

[0153] Based on the same inventive concept, referring to Figure 6 The embodiment of the present disclosure provides a face clustering device, which comprises:

[0154] The similarity determination unit 601 is configured to determine that the similarity between the target face image and the preset first face template is located between the non-similarity determination value and the similarity determination value; wherein the non-similarity determination value represents that the target face image cannot be clustered into the identity corresponding to the first face template, and the similarity determination value represents that the target face image can be clustered into the identity corresponding to the first face template.

[0155] The identity determination unit 602 is configured to, if there is at least one second face image corresponding to the target face image in the target co-occurrence period, take the identity corresponding to the first face template as the identity corresponding to the target face image; wherein the identity corresponding to the second face template clustered by the at least one second face image is in a co-occurrence relationship with the identity corresponding to the first face template, and the co-occurrence relationship is determined by the snapshot records of the identity corresponding to the first face template and the identity corresponding to the second face template.

[0156] Optionally, the first face template is an ID photo of the identity corresponding to the first face template; and / or

[0157] The second face template is an ID photo of the identity corresponding to the second face template.

[0158] Optionally, the co-occurrence relationship is determined by the following manner:

[0159] In each of the plurality of co-occurrence periods, the snapshot records of the identity corresponding to the first face template and the identities corresponding to the at least one preselected face template are obtained, wherein the snapshot records include the first face image and a plurality of preselected face images, and the time points corresponding to the plurality of co-occurrence periods are earlier than the time point corresponding to the target co-occurrence period;

[0160] In each of the co-occurrence periods, if the time difference between the first face image and the preselected face image in the snapshot records is greater than a set time difference threshold, and the number of times that the time difference between the first face image and the preselected face image in each of the co-occurrence periods is greater than the time difference threshold exceeds a preset number of times, it is determined that the identity corresponding to the first face template and the identities corresponding to the plurality of preselected face templates are in a co-occurrence relationship, wherein the second face template is one of the plurality of preselected face templates, and the smaller the time difference threshold is, the closer the time points at which the first face image and the preselected face image appear in the same location.

[0161] Optionally, obtaining the snapshot records of the identity corresponding to the first face template and the identities corresponding to the at least one preselected face template comprises:

[0162] determining a shooting device for obtaining the first face image, and determining a time point for obtaining the first face image; and

[0163] based on the same shooting device, determining a plurality of face images of the first face image in a candidate time interval;

[0164] taking the first face image as a first snapshot record corresponding to the first face template, taking the plurality of face images as a second snapshot record corresponding to the second face template, and taking the union of the first snapshot record and the second snapshot record as the snapshot records of the identity corresponding to the first face template and the identities corresponding to the at least one second face template.

[0165] Optionally, the method further comprises:

[0166] If the target face image does not have a corresponding second face image in the target co-travel period, any one of the pre-selected face templates different from the second face template is selected, the identity corresponding to the first face template is updated to the identity corresponding to the selected pre-selected face template, and the identity corresponding to the second face template is updated to the identity corresponding to the selected pre-selected face template;

[0167] Based on the identity corresponding to the pre-selected face template, a pre-selected second face image is obtained, which is captured at a time point within a preset time difference threshold from the acquisition time point;

[0168] If the pre-selected similarity value between the pre-selected second face image and the identity corresponding to the second face template is greater than the pre-selected similarity determination value, the identity corresponding to the first face template at the acquisition time point is taken as the identity corresponding to the target face image; wherein the pre-selected similarity determination value represents that the pre-selected second face image can be clustered into the identity corresponding to the second face template.

[0169] Optionally, the preset time difference threshold is determined based on a shooting interval of the shooting device;

[0170] The preset number of times is determined based on the density of the shooting device.

[0171] Optionally, the density is determined in the following manner:

[0172] The total length corresponding to the action trajectory of the first face image in the first snapshot record is obtained;

[0173] The number of first face images included in the first snapshot record is determined;

[0174] Based on the total length and the number, the density is determined.

[0175] Optionally, if the plurality of first face images are obtained by a plurality of shooting devices in one shooting device group, the first face images are determined in the following manner, including:

[0176] An association relationship between the plurality of shooting devices in one shooting device group is obtained;

[0177] Based on the association relationship, a plurality of pre-selected first face images are obtained by each shooting device, respectively;

[0178] The plurality of pre-selected first face images are taken as the first face images.

[0179] Optionally, if the plurality of first face images are obtained by a plurality of shooting devices in one shooting device group, based on the same shooting device, a plurality of face images within a candidate time interval with the first face image are determined, including:

[0180] Based on the association relationship, in a plurality of shooting devices in a shooting device group, a device identifier of a shooting device corresponding to the first face image to be compared is determined;

[0181] Based on the device identifier, a plurality of face images in the alternative time interval with the first face image are determined.

[0182] Optionally, if there is at least one second face image corresponding to the target face image in the target same-traveling time period, the identity corresponding to the first face template is taken as the identity corresponding to the target face image, and the identity determination unit 602 is configured to:

[0183] In the target same-traveling time period, the acquisition time of the target face image is determined, and the identity corresponding to the first face template is determined as the identity corresponding to at least one second face template of the same-traveling relationship;

[0184] Based on the identity corresponding to the second face template, at least one second face image of the shooting time and the acquisition time within a preset time difference threshold is obtained.

[0185] If the similarity value between the any one second face image and the identity corresponding to the second face template is greater than the similarity determination value, the identity corresponding to the first face template at the acquisition time is taken as the identity corresponding to the target face image; wherein the similarity determination value represents that the second face image can be clustered into the identity corresponding to the second face template.

[0186] Based on the same inventive concept, refer to Figure 7 The embodiment of the present disclosure provides a kind of computing device, comprising: memory 701, for storing executable instruction;Processor 702, for reading and executing the executable instruction stored in memory, and any one method of the first aspect is executed.

[0187] Based on the same inventive concept, the embodiment of the present disclosure provides a kind of computer readable storage medium, when the instruction in the storage medium is executed by processor, so that the processor can execute the method described in any one of the first aspect.

[0188] In summary, in the embodiments of the present disclosure, the method, device and storage medium for face clustering are provided, the method comprises: determining that the similarity between a target face image and a preset first face template is between a dissimilarity determination value and a similarity determination value, that is, the identity to which the target face image belongs cannot be directly determined by the similarity determination value, wherein the dissimilarity determination value represents that the target face image cannot be clustered into the identity corresponding to the first face template, and the similarity determination value represents that the target face image can be clustered into the identity corresponding to the first face template; if the target face image has at least one second face image corresponding thereto within a target same-trip period, the identity corresponding to the first face template is taken as the identity corresponding to the target face image, wherein the identity corresponding to a second face template to which the at least one second face image is clustered is in a same-trip relationship with the identity corresponding to the first face template, and the same-trip relationship is determined by the snapshot records of the identity corresponding to the first face template and the identity corresponding to the second face template. In this way, in the case that the identity to which the target face image belongs cannot be directly determined by the similarity determination value, the identity to which the target face image belongs is determined by means of the same-trip relationship, thereby improving the accuracy of face clustering.

[0189] Those skilled in the art will appreciate that embodiments of the present disclosure can be provided as methods, systems, or computer program product systems. Accordingly, the present disclosure can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present disclosure can take the form of a computer program product system implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) having computer usable program code embodied therein.

[0190] The present disclosure is described with reference to the flowcharts and / or block diagrams according to the methods, devices (systems), and computer program product systems of the present disclosure. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce an apparatus that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in a flow or multiple flows and / or blocks Figure 1 The functions specified in a flow or multiple flows and / or blocks

[0191] These computer program instructions can also be stored in a computer-readable memory that can direct the computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including instruction apparatus, which implements the functions specified in the flowcharts and / or block diagrams. Figure 1one or more processes and / or blocks Figure 1 the function specified in the one or more blocks.

[0192] These computer program instructions can also be loaded into computer or other programmable data processing devices, so that a series of operation steps are performed on the computer or other programmable data processing devices to generate computer-implemented processes, so that the instructions executed on the computer or other programmable data processing devices provide processes for implementing the flow Figure 1 one or more processes and / or blocks Figure 1 the function specified in the one or more blocks.

[0193] Obviously, those skilled in the art can make various modifications and variations to the present disclosure without departing from the spirit and scope of the present disclosure. Thus, if these modifications and variations of the present disclosure belong to the scope of the claims of the present disclosure and their equivalent technologies, the present disclosure also intends to include these modifications and variations.

Claims

1. A method of face clustering, characterized in that, The method comprises: determining that the similarity between the target face image and the preset first face template is between a non-similarity determination value and a similarity determination value; wherein the non-similarity determination value represents that the target face image cannot be clustered into the identity corresponding to the first face template, and the similarity determination value represents that the target face image can be clustered into the identity corresponding to the first face template; if the target face image has at least one corresponding second face image in the target same-travel time period, then the identity corresponding to the first face template is taken as the identity corresponding to the target face image; wherein the identity corresponding to the second face template to which the at least one second face image is clustered is in a same-travel relationship with the identity corresponding to the first face template, the same-travel relationship is determined by the snapshot records of the identity corresponding to the first face template and the identity corresponding to the second face template, and the same-travel relationship is determined by the following method: in each of a plurality of same-travel time periods, the snapshot records of the identity corresponding to the first face template and the identity corresponding to at least one preselected face template are obtained, wherein the snapshot records include a first face image and a plurality of preselected face images, the time points corresponding to the plurality of same-travel time periods are earlier than the time point corresponding to the target same-travel time period, in each same-travel time period, if the time difference between the first face image and the preselected face image in the snapshot records is greater than a set time difference threshold, and the number of times that the time difference between the first face image and the preselected face image in each same-travel time period is greater than the time difference threshold exceeds a preset number of times, then it is determined that the identity corresponding to the first face template and the identity corresponding to the plurality of preselected face templates are in a same-travel relationship, wherein the second face template is one of the plurality of preselected face templates, and the smaller the time difference threshold is, the closer the time points at which the first face image and the preselected face image appear in the same location.

2. The method of claim 1, wherein, The first face template is an ID photo of the identity corresponding to the first face template; and / or The second face template is an ID photo of the identity corresponding to the second face template.

3. The method of claim 1, wherein, The method further comprises: determining the photographing device that obtains the first face image, and determining the time point at which the first face image is obtained; and based on the same photographing device, determining a plurality of face images of the first face image in a candidate time interval; taking the first face image as the first snapshot record corresponding to the first face template, taking the plurality of face images as the second snapshot record corresponding to the second face template, and taking the union of the first snapshot record and the second snapshot record as the snapshot records of the identity corresponding to the first face template and the identity corresponding to at least one second face template.

4. The method of claim 1, wherein, The method further comprises: if the target face image does not have a corresponding second face image in the target same-trip period, selecting any one of the preselected face templates different from the second face template, the identity corresponding to the first face template being the preselected face template of a same-trip relationship, and updating the identity corresponding to the second face template to the identity corresponding to the selected preselected face template; based on the identity corresponding to the preselected face template, obtaining a preselected second face image whose shooting time and acquisition time are within the preset time difference threshold; if the preselected similarity value between the preselected second face image and the identity corresponding to the second face template is greater than a preselected similarity determination value, the identity corresponding to the first face template at the acquisition time is taken as the identity corresponding to the target face image; wherein the preselected similarity determination value represents that the preselected second face image can be clustered into the identity corresponding to the second face template.

5. The method of claim 3, wherein, The set time difference threshold is determined based on the shooting interval of the shooting device; The preset number of times is determined based on the density of the shooting device.

6. The method of claim 5, wherein, The density is determined in the following way: obtaining the total length corresponding to the movement trajectory of the first face image in the first snapshot record; determining the number of the first face images included in the first snapshot record; based on the total length and the number, determining the density.

7. The method of claim 3, wherein, If multiple first face images are obtained by multiple shooting devices in one shooting device group, the first face images are determined in the following way, including: obtaining the association relationship between the multiple shooting devices in the one shooting device group; based on the association relationship, obtaining multiple preselected first face images through the respective shooting devices; the multiple preselected first face images are taken as the first face images.

8. The method of claim 7, wherein, If multiple first face images are obtained by multiple shooting devices in one shooting device group, the multiple face images within the alternative time interval with the first face image are determined based on the same shooting device, including: based on the association relationship, determining the device identifier of the shooting device corresponding to the first face image to be compared among the multiple shooting devices in the one shooting device group; based on the device identifier, determining the multiple face images within the alternative time interval with the first face image.

9. The method of claim 4, wherein, If the target face image has at least one corresponding second face image in the target same-trip period, the identity corresponding to the first face template is taken as the identity corresponding to the target face image, including: in the target same-trip period, determining the acquisition time of the target face image, and determining at least one identity corresponding to the second face template of the first face template of a same-trip relationship; based on the identity corresponding to the second face template, obtaining any one of the at least one second face image whose shooting time and acquisition time are within the preset time difference threshold; If a similarity value between the identity corresponding to the second face template and the identity corresponding to the arbitrary one of the second face images is greater than a similarity determination value, the identity corresponding to the first face template at the acquisition time is taken as the identity corresponding to the target face image; wherein the similarity determination value represents that the second face images can be clustered into the identity corresponding to the second face template.

10. An apparatus for face clustering, the apparatus comprising: The method comprises: a similarity determination unit configured to determine that a similarity between a target face image and a preset first face template is between a non-similarity determination value and a similarity determination value; wherein the non-similarity determination value represents that the target face image cannot be clustered into the identity corresponding to the first face template, and the similarity determination value represents that the target face image can be clustered into the identity corresponding to the first face template; an identity determination unit configured to, if the target face image has at least one second face image corresponding thereto within a target same-trip time period, take the identity corresponding to the first face template as the identity corresponding to the target face image; wherein the second face template corresponding to the at least one second face image and the first face template are in a same-trip relationship, the same-trip relationship is determined by a snapshot record of the identity corresponding to the first face template and the identity corresponding to the second face template, and the same-trip relationship is determined by: in each of a plurality of same-trip time periods, acquiring a snapshot record of the identity corresponding to the first face template and the identity corresponding to at least one preselected face template, wherein the snapshot record comprises a first face image and a plurality of preselected face images, the time corresponding to the plurality of same-trip time periods is earlier than the time corresponding to the target same-trip time period, in each of the same-trip time periods, if a time difference between the first face image and the preselected face image in the snapshot record is greater than a set time difference threshold, and the number of times that the time difference between the first face image and the preselected face image in each same-trip time period is greater than the time difference threshold exceeds a preset number of times, it is determined that the identity corresponding to the first face template and the identity corresponding to the plurality of preselected face templates are in a same-trip relationship, wherein the second face template is one of the plurality of preselected face templates, and the smaller the time difference threshold, the closer the time points at which the first face image and the preselected face image appear in the same location.

11. A computing device, comprising: The method comprises: a memory configured to store executable instructions; a processor configured to read and execute the executable instructions stored in the memory to implement the method of any one of claims 1-9.

12. A computer-readable storage medium, characterized in that, When the instructions in the storage medium are executed by the processor, the processor can execute the method of any one of claims 1-9.

Images