A face confidence method, system, device and medium based on space-time collision

Abstract

The present invention proposes a face confidence method, system, device and medium based on spatiotemporal collision, including: obtaining face images in a capture scene, classifying the face images to construct a face file; associating the corresponding capture device with a target place according to the point information of each capture device in the capture scene, and presetting the spatiotemporal collision score of each capture device; obtaining the visit record of the target object in the target place, and determining the target capture device corresponding to the target object according to the time node of the visit record; obtaining the cumulative spatiotemporal collision score and the cumulative spatiotemporal collision frequency of the face file and the target capture device within a preset time period, and associating the face file that meets the score threshold and frequency threshold of the cumulative spatiotemporal collision score with the target object, wherein the spatiotemporal collision represents the spatiotemporal intersection of the face file and the target capture device.

Description

Technical Field

[0001] The present invention relates to the field of intelligent security, and in particular to a face confidence method, system, device and medium based on time-space collision. Background Art

[0002] With the continuous development of artificial intelligence, the application of association between faces and scenes is becoming more and more extensive, such as trajectory analysis and crowd counting through the characteristic information carried by faces. However, in some practical application scenarios, such as anonymous consumption in stores, it is difficult to accurately associate store and public area information, resulting in the inability to integrate the trajectory of anonymous people with other information. For example, in retail, anonymous trajectories can only tell whether customers have been to the mall, but not whether the customers have actually made purchases. How to accurately perform statistical analysis on anonymous information has become a major problem that needs to be solved urgently. Summary of the invention

[0003] In view of the above problems existing in the prior art, the present invention proposes a face confidence method, system, device and medium based on spatiotemporal collision, which mainly solves the problem that the existing methods are difficult to effectively control anonymous information in specific places and the utilization rate of corresponding place data is low.

[0004] In order to achieve the above purpose and other purposes, the technical solution adopted by the present invention is as follows.

[0005] A face confidence method based on spatiotemporal collision, comprising:

[0006] Acquire facial images in the captured scene, classify the facial images and build facial files;

[0007] According to the position information of each capture device in the capture scene, the corresponding capture device is associated with the target location, and a spatiotemporal collision score of each capture device is preset;

[0008] Obtaining a visit record of a target object in the target place, and determining a target capture device corresponding to the target object according to a time node of the visit record;

[0009] The cumulative spatiotemporal collision score and the cumulative spatiotemporal collision frequency of the face file and the target capture device within a preset time period are obtained, and the face file that meets the score threshold and frequency threshold of the cumulative spatiotemporal collision score is associated with the target object, wherein the spatiotemporal collision represents the spatiotemporal intersection of the face file and the target capture device.

[0010] Optionally, the temporal and spatial collision scores of the capture devices are preset, including:

[0011] Obtaining quality scores of images captured by each of the capture devices;

[0012] According to the pre-established correspondence between the quality score and the spatiotemporal collision score, the spatiotemporal collision score of the corresponding capture device is set.

[0013] Optionally, obtaining a cumulative spatiotemporal collision score and a cumulative spatiotemporal collision frequency between the face file and the target capture device within a preset time period, and associating a face file that meets a score threshold and a frequency threshold of the cumulative spatiotemporal collision score with the target object includes:

[0014] Preset the time-space collision screening durations of different target places, and determine the upper and lower limits of the preset time period according to the visit time nodes of the target objects of the corresponding target places and the corresponding time-space collision screening durations;

[0015] Determine the facial image associated with the target object according to the capture record of the target capture device;

[0016] Retrieving from the corresponding face file the number of times the associated face image is captured by the target capture device within the preset time period as the accumulated spatiotemporal collision frequency;

[0017] The cumulative spatiotemporal collision score is the product of the cumulative spatiotemporal collision frequency and the spatiotemporal collision score of the target capture device.

[0018] Optionally, when there are multiple associated face images, the cumulative spatiotemporal collision score and the cumulative spatiotemporal collision frequency of each corresponding face file are obtained respectively;

[0019] If the face files corresponding to the plurality of associated face images all meet the score threshold and the frequency threshold, then selecting the face file with a higher cumulative spatiotemporal collision frequency to associate with the corresponding target object;

[0020] If the spatiotemporal collision frequencies of the plurality of face files are consistent, the face file with the highest cumulative spatiotemporal collision score is selected and associated with the corresponding target object.

[0021] Optionally, after associating the face profile that meets the score threshold and frequency threshold of the cumulative spatiotemporal collision score with the target object, the method further includes:

[0022] Acquire the peer target of the corresponding target object according to the associated face file, and acquire the corresponding peer face file through peer target retrieval;

[0023] According to the peer face file, a related face map of the corresponding target object and the activity track of each peer target are constructed.

[0024] Optionally, after associating the face profile that meets the score threshold and frequency threshold of the cumulative spatiotemporal collision score with the target object, the method further includes:

[0025] The association result is output to the management terminal of the target location, and feedback results are obtained by interacting with the management terminal. The accuracy of the association between the face file and the target object is counted according to the feedback results. If the association result is wrong, the association result is obtained again.

[0026] Optionally, after the correct rate of association between the face file and the target object is counted, it includes:

[0027] Obtain the distribution of correct spatiotemporal collisions of each of the capture devices, obtain the time interval of the correct spatiotemporal collision concentration under a preset first confidence level, and correct the time threshold of the spatiotemporal collision corresponding to the target location according to the time interval; and / or,

[0028] According to the correct spatiotemporal collision distribution of the capture device, the point information of the capture device corresponding to the correct spatiotemporal collision under a preset second confidence level is obtained, and the spatiotemporal collision score of the corresponding capture device is corrected.

[0029] A face confidence method based on spatiotemporal collision, comprising:

[0030] A face archive building module is used to obtain face images in the captured scene, classify the face images and build a face archive;

[0031] A device association module, used to associate the corresponding capture device with the target location according to the point information of each capture device in the capture scene, and preset the spatiotemporal collision score of each capture device;

[0032] A target capture device determination module is used to obtain a visit record of a target object in the target place, and determine a target capture device corresponding to the target object according to a time node of the visit record;

[0033] The face file association module obtains the cumulative spatiotemporal collision score and the cumulative spatiotemporal collision frequency of the face file and the target capture device within a preset time period, and associates the face file that meets the score threshold and frequency threshold of the cumulative spatiotemporal collision score with the target object, wherein the spatiotemporal collision represents the spatiotemporal intersection of the face file and the target capture device.

[0034] A device comprising:

[0035] one or more processors; and

[0036] One or more machine-readable media having instructions stored thereon, when executed by the one or more processors, enable the device to perform the face confidence method based on spatiotemporal collision.

[0037] One or more machine-readable media store instructions thereon, which, when executed by one or more processors, enable the device to perform the face confidence method based on spatiotemporal collision.

[0038] As described above, the face confidence method, system, device and medium based on spatiotemporal collision of the present invention have the following beneficial effects.

[0039] By performing spatiotemporal collision between the facial archive of the corresponding scene and the capture device associated with the target place in the scene, the target object in the target place is associated with the facial archive, so that the activity data of the target object in the target place can be effectively managed based on the associated facial archive, and the target object can be real-named based on the face, which is conducive to trajectory analysis of anonymous data and improves the utilization rate of anonymous data. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 Schematic diagram of the process of a face confidence method based on spatiotemporal collision in one embodiment of the present invention.

[0041] Figure 2 It is a module diagram of a face confidence system based on spatiotemporal collision in one embodiment of the present invention.

[0042] Figure 3 It is a schematic diagram of the structure of a device in one embodiment of the present invention. DETAILED DESCRIPTION

[0043] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0044] It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention, and thus the drawings only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0045] See also Figure 1 The present invention provides a face confidence method based on spatiotemporal collision, which includes the following steps.

[0046] Step S1, obtaining a face image in a captured scene, and classifying the face image to construct a face file;

[0047] Step S2, associating the corresponding capture device with the target location according to the position information of each capture device in the capture scene, and presetting the spatiotemporal collision score of each capture device;

[0048] Step S3, obtaining a visit record of a target object in the target place, and determining a target capture device corresponding to the target object according to a time node of the visit record;

[0049] Step S4, obtaining the cumulative spatiotemporal collision score and the cumulative spatiotemporal collision frequency between the face file and the target capture device within a preset time period, and associating the face file that meets the score threshold and frequency threshold of the cumulative spatiotemporal collision score with the target object, wherein the spatiotemporal collision represents the spatiotemporal intersection of the face file and the target capture device.

[0050] In one embodiment, the capture device may include a capture camera installed in a corresponding capture scene.

[0051] In one embodiment, constructing a face profile includes the following steps:

[0052] Step 1: Face image acquisition, including

[0053] Step 1.1 The captured facial image information obtained by the capture camera includes the actual facial image, device ID, capture time, etc. The facial image is identified and obtained from the capture scene image through a face detection algorithm.

[0054] The face detection algorithm in step 1.2 is generally a neural network model, such as MTCNN, YOLO series and other target detection models. The captured scene graph may contain one or more faces, and the face detection algorithm needs to identify all possible face images in the captured scene graph.

[0055] Step 2: Face feature extraction

[0056] Step 2.1: Extract facial feature vectors and structured information from the face images to be clustered through the face analysis service.

[0057] The face parsing service described in step 2.2 is generally one or more multi-task neural network parsing models, and the multi-task neural network parsing model means that a model can simultaneously recognize certain types or all structured information and facial feature information of a face image. For example, if the face parsing model is a multi-task neural network model, the model can parse the facial features and regress and predict all structured information values, such as the face pitch angle, horizontal angle, face quality score, gender, age, whether wearing a mask, etc.

[0058] The face structured information in step 2.3 includes face pitch angle, horizontal angle, face quality score, gender, age, whether wearing a mask, etc. For example, face pitch angle: 0, horizontal angle: 10, face quality score: 80, gender: 0, age: 30, whether wearing a mask: 1, etc.

[0059] Step 3: Collect face files

[0060] Step 3.1 screens the faces to be clustered based on the structured face information, removes low-quality faces, and obtains the actual faces to be clustered. For example, faces to be clustered that have a large face pitch angle, a large horizontal angle, and a low face quality score due to factors such as the relative capture orientation of the capture camera and lighting, usually have low recognizability and will result in poor quality of the corresponding facial features, which has a large and lasting negative impact on the final clustering effect. They can all be considered as dirty data for filtering and do not participate in face clustering, which is beneficial to improving clustering efficiency and clustering accuracy. For example, it can be set that when the face pitch angle>40 or the horizontal angle>50 or the face quality data<30, the face image is directly placed in the waste film library without clustering.

[0061] Step 3.2 compares the similarity between the actual face feature vector to be clustered and the class feature center of the existing face file. If the maximum similarity between a face and an existing face file meets the set similarity threshold, the face is classified into the face file corresponding to the maximum similarity, the face is clustered, and the age, gender, and class feature center of the corresponding face file are updated. If the maximum similarity between a face and an existing face file does not meet the set similarity threshold, a new face file is created and the face is classified into the new face file. For example, let the existing face file be denoted as F h ={a i :[fc i , (p i1 , p i2 ,…p in )]}, where a i Indicates the unique sign of clustering results, fc i Indicates a i The class feature center, p in Indicates the unique mark of the face image, (p i1 , p i2 ,…p in ) represents the clustering result a i For example, there is currently a face image p k , the corresponding facial feature is J k, the key structural information of the corresponding face is pitch angle: 5, horizontal angle: 10, face quality score: 80, gender: 0, age: 30, whether wearing a mask: 1, which obviously meets the entry conditions for clustering. Calculate f k With all fc i The cosine similarity of face p is set to 90%, and the similarity threshold is set to 90%. Case 1: The similarity with fc1 is the largest and is 95%. Obviously, face p k If the maximum similarity of 95% with the existing clustering results is greater than the set similarity threshold of 90%, the face is classified into the file corresponding to the maximum similarity, that is, p1 is classified into the face image set in the clustering result a1 (p i1 , p i2 ,…p in ), recalculate the age and gender corresponding to the face file, and complete the clustering and archiving of the face; Case 2: Get the same age and gender as fc i The maximum similarity in is 50%, obviously the face p k If the maximum similarity of 50% with the existing clustering results is less than the set similarity threshold of 90%, a new face file a is created. i+1 , and p k Classified into a i+1 In the face image collection, at this time a i+1 The face image collection contains only p k , a i+1 The class feature center is fc i+1 =f k .

[0062] The existing face files in step 3.3 are existing historical clustering results, each file is associated with at least one face image, and the class feature center is the representative feature vector of each face file.

[0063] Step 3.4 updates the age of the face file by updating the mean after removing outliers, and updates the gender of the face file by using the gender with a higher proportion as the gender of the face file. Specifically, the face file age is calculated as the mean of all the ages of the faces belonging to the file after removing outliers. The outlier measurement method can be the Grubbs test method, normal distribution judgment (3σ criterion), etc. For example, it is advisable to use normal distribution judgment, that is, when a face age value a satisfies Then the age is an outlier, where δ is the standard deviation of the age value in the file, If it is the mean age of all faces in the file, there is no need to update the age; otherwise, calculate the mean and update the file age value.

[0064] Step 3.5 updates the central features of the face file class by multi-stage selection averaging. Specifically, when the number of faces in the face file is less than the quantity threshold, the mean of all face feature vectors is selected for update; when the number of faces in the face file is equal to an integer multiple of the quantity threshold, weighted average selection is used for update.

[0065] Step 3.6 For example, let us set the number threshold to N. If the total number of faces in the file to be classified is , then the total number of faces in the file after classification is . Assuming that the face feature is , and the file feature before classification is , then the file feature after classification can be quickly iterated and calculated as follows:

[0066] Step 3.7 For example, if the total number of faces in the file to be classified into is N-1, then the total number of faces in the file after the face is classified is N, which is an integer multiple of the number threshold N. Then the weighted average method is selected. Generate a similarity matrix based on all face feature vectors in the face file, which may be set to S. Each element in the matrix can be expressed as S ij , S ij It also means the face p in the face file i With p j The similarity is cosine similarity, i, j are the indexes of the faces in the file, and 0=<i, j<N. For each element S in the similarity matrix S ij , assuming the second similarity threshold is 93, if S ij is greater than the set second similarity threshold, then establish p i With p j , that is, they are classified into a face subset, and finally a face subset set is formed. Obviously, the face subset set contains at least one face subset. For each face subset, the feature vector with the highest face quality score in the subset is selected as the representative feature vector of each subset, and then the representative feature vectors of all subsets are weighted averaged to finally obtain the representative feature vector of each file, that is, the class feature center. The weighting coefficient in the weighted average is the number of faces in each subset. For example, assuming that the three subsets are obtained, assuming that the face with the highest face quality score in each subset is p i 、p j 、p k , then the corresponding facial features are selected as f i 、f j 、f k As the representative feature of each subset, in addition, the number of faces contained in each subset can be calculated as p, q, r. The class center fc calculated by weighted average N , the specific calculation formula is as follows:

[0067] Step 4 Face Archive Cleaning

[0068] Step 4.1 Clean invalid archives based on the statistical information of the capture trajectories of face archives. The specific cleaning calculation rules include three methods: based on unbalanced gender distribution, based on inability to merge archives, and based on rapid growth of trajectory quantity.

[0069] The cleaning based on unbalanced gender distribution described in Step 4.2 means that since there may be a situation where a person is captured multiple times by a camera in a short period, it is first necessary to remove duplicates from the capture trajectories of face archives, and then count according to the male-female ratio of each capture trajectory. When the system identifies an unbalanced male-female ratio, the corresponding face archive will be deleted. For example, if the total number of captures of an archive is N, and the total number of captures N is scaled to R according to the condition that only 1 photo is taken within 1 minute by the same camera for statistical calculation. According to Step 2, the male-female characteristics of each captured photo have been identified, and the total number of males and females in the set R is calculated. The total number of males identified is R1, and the total number of females identified is R2. Compare the sizes of R1 and R2, with the larger number as the denominator and the smaller number as the numerator. In this example, assume that R1 < R2. Then when R1 / R2 > 3 / 7, the archive will be deleted.

[0070] The cleaning based on inability to merge archives described in Step 4.3 means that due to some special captures, such as special capture angles, etc., the archive of a person may never be merged. However, the real face of such an archive actually exists, so this type of archive belongs to invalid archives. For example, if the total number of captures of an archive is N (N < 3), and the total number of this archive has been less than 3 for 15 consecutive days, then this type of archive will be deleted.

[0071] The cleaning based on abnormal growth of trajectory quantity described in Step 4.4 means that as described in Step 3, as the number of trajectories of an archive increases, the features of the archive will be weighted. When a clustering error occurs in an archive, with the generation of weighting, the new eigenvalue features will tend to be averaged, that is, they will no longer have feature effects, which will cause the captured photos of different people to be clustered into this archive. Similarly, as the number of trajectories of different people increases, the features will become more and more averaged, forming a negative cycle, ultimately resulting in a large number of captures being wrongly merged and affecting the statistical effect. Therefore, it is necessary to delete archives with too rapid growth of trajectories. For example, assume that after the system has been running stably for 30 days, the maximum growth trajectory quantity of face archives within 1 day is screened out as A. When the system identifies that the trajectory growth quantity of a certain archive within 1 day is 2A, then this archive will be deleted.

[0072] Furthermore, all camera location information can be obtained, and cameras can be classified according to actual business needs. For example, cameras a1-a5 monitor the electrical equipment area, and cameras a6-a11 monitor the dining area. The specific classification method can be set according to actual application needs and is not limited here.

[0073] In one embodiment, the location information of the snapshot camera includes but is not limited to the unique mark of the snapshot camera, longitude and latitude coordinate information, geographic location information, tag information, etc. The tag information may be a business district, a street, a community, an administrative division, a store, etc. For example, the location information of the snapshot camera is generally obtained from the snapshot device management database, and the unique mark of the snapshot camera is generally the device ID of the snapshot camera. For example, the three cameras A1 / A2 / A3 are associated with store a.

[0074] In one embodiment, the spatial and temporal information of personnel can be obtained by connecting with a third-party system. Specifically, the system provides a standard interface or a method of accessing a third-party database to obtain external data. The data content obtained includes but is not limited to the personnel ID of anonymous personnel, the time information and location information of the confirmed appearance according to actual business needs. For example, in a commercial environment, it is the mobile phone number of an unknown member and the time of consumption in a certain store. The spatial and temporal information corresponds to the visit record of the target object, and the visit record can also include the identity association information of the corresponding target object, such as whether it is a store member.

[0075] In one embodiment, after associating the snapshot camera at the corresponding position with the store, the spatiotemporal collision score of each of the snapshot devices is preset, including: obtaining the quality score of the captured image of each of the snapshot devices; setting the spatiotemporal collision score of the corresponding snapshot device according to the pre-established correspondence between the quality score and the spatiotemporal collision score. Specifically, the quality score corresponding to the captured image of each snapshot device can be obtained according to the image quality scoring method of the aforementioned steps, the quality score interval can be divided, and the correspondence between the quality score and the spatiotemporal collision score can be established. Exemplarily, the spatiotemporal collision score of the snapshot device with a quality score greater than 70 points can be set to 5 points, and the spatiotemporal collision score of the snapshot device with a quality score less than 70 points can be set to 3 points. The specific corresponding method can be adjusted according to the actual application requirements and is not limited here. Since the entire portrait file is affected by the camera installation angle, some points have good capture effects and some points have poor capture effects. Therefore, first, the score of successful collision is configured according to the actual installation situation of the point. For example, the point with good capture effect is set to 5 points, and the point with poor capture effect is set to 3 points.

[0076] In one embodiment, the cumulative spatiotemporal collision score and the cumulative spatiotemporal collision frequency of the face file and the target capture device within a preset time period are obtained, and the face file that meets the score threshold and frequency threshold of the cumulative spatiotemporal collision score is associated with the target object, including: presetting the spatiotemporal collision screening time of different target places, determining the upper and lower limits of the preset time period according to the visit time nodes of the target objects of the corresponding target places and the corresponding spatiotemporal collision screening time; determining the face image associated with the target object according to the capture record of the target capture device; retrieving the number of times the associated face image is captured by the target capture device within the preset time period from the corresponding face file as the cumulative spatiotemporal collision frequency; the cumulative spatiotemporal collision score is the product of the cumulative spatiotemporal collision frequency and the spatiotemporal collision score of the target capture device.

[0077] Specifically, time-space collision means that when a person is sure to appear, there will be a camera capture record within a certain period of time. Therefore, the time range threshold must be configured first. If the time range is within the time range, the collision is considered successful and a collision success score of the point can be obtained. In actual situations, due to different camera installation conditions, the time range threshold should be set separately according to the point.

[0078] Configure the score threshold for space-time collisions. After multiple collisions, the system will notify you when the score and the frequency mentioned below reach a certain threshold.

[0079] In addition to considering the successful collision score when data collides, the frequency of successful collision also needs to be considered. When the frequency reaches the threshold, the system can send notification information to the terminal of the corresponding store and interact with the store manager.

[0080] In one embodiment, the face trajectory can be obtained through the face file and subjected to a spatiotemporal collision with the spatiotemporal information of the target object.

[0081] Specifically, the system can obtain the identity identification, time point information, and location information of anonymous persons. According to the special camera associated with the corresponding store, the system can convert the location information into a camera ID. Therefore, the space-time collision is actually to search for tracks in the database based on the time range and device ID. When a track is satisfied, the cumulative space-time collision score is calculated. For example, the collision threshold is set to 10 seconds before and after, and the space-time collision score of camera A1 is 3 points. The order data provided by a third party is in store a. The member with id 138888888 completed an order at 13:30:25. The file with file id 291919239912 was retrieved from the database. There is a snapshot record at 13:30:20 for the A1 camera associated with store a. Therefore, the correlation between the 291919239912 portrait file and the member 138888888 is 3 points.

[0082] In one embodiment, when there are multiple spatiotemporal information for the same person ID, each spatiotemporal information is retrieved according to the above steps, and the points of the face files that meet the conditions are accumulated. After all the spatiotemporal information are collided, if there is a face file that meets the score threshold and frequency threshold configured in the step, the system can consider that the person ID and the face file are successfully matched, and the face file is associated with the person ID to complete the real-name registration.

[0083] In one embodiment, in actual business, due to the existence of peer situations, there will often be a situation where multiple portrait files meet the score threshold and frequency threshold at the same time. The strategy generally adopted is frequency priority. When the frequencies are the same, the face file with a higher score is taken.

[0084] In one embodiment, after associating the face profile that meets the score threshold and frequency threshold of the cumulative spatiotemporal collision score with the target object, the method includes:

[0085] Acquire the peer target of the corresponding target object according to the associated face file, and acquire the corresponding peer face file through peer target retrieval;

[0086] According to the peer face file, a related face map of the corresponding target object and the activity track of each peer target are constructed.

[0087] In one embodiment, after the face profile that meets the score threshold and frequency threshold of the cumulative spatiotemporal collision score is associated with the target object, the association result can also be output to the management terminal of the target place, and feedback results can be obtained by interacting with the management terminal. The accuracy of the association between the face profile and the target object is counted based on the feedback results. If the association result is wrong, the association result is obtained again.

[0088] In one embodiment, after the correct rate of association between the face file and the target object is calculated, the following steps are included:

[0089] Obtain the distribution of correct spatiotemporal collisions of each of the capture devices, obtain the time interval of the correct spatiotemporal collision concentration under a preset first confidence level, and correct the time threshold of the spatiotemporal collision corresponding to the target location according to the time interval; and / or,

[0090] According to the correct spatiotemporal collision distribution of the capture device, the point information of the capture device corresponding to the correct spatiotemporal collision under a preset second confidence level is obtained, and the spatiotemporal collision score of the corresponding capture device is corrected.

[0091] Specifically, after the real-name system is completed, the trajectory of a specific person can be accurately restored. In the commercial scenario, through real-name system, the operator can clearly understand which stores a member has visited, how long they have stayed, and what the activity trajectory is in the entire mall. At the same time, combined with the analysis of peers, it can be determined whether they have children, whether they are married, their age, gender, etc., to form a complete member portrait, and the operator can provide targeted services. In the medical insurance scenario, through real-name system, the medical insurance regulatory agency can analyze whether a medical insurance settlement order has insurance fraud, such as a portrait file associated with a hospitalization reimbursement order. Through the trajectory analysis of the portrait file, it is found that the trajectory of the file regularly enters and leaves the hospital, and has never entered and left the inpatient ward. It can be considered that the order is suspected of hanging a bed for hospitalization. Further verification can be carried out.

[0092] In actual use, the user's use can play a role in calibration. The terminal can calibrate the associated target object and face file, and the collected calibration data can automatically adjust the collision threshold. The adjustment method may include:

[0093] Tuning based on time rules, for example, the time threshold of each camera under the conditions of correct or incorrect collision can be counted, and its distribution pattern can be counted. When 95% of the correct collisions are concentrated in the first 8 seconds and the last 7 seconds, and when 95% of the incorrect collisions are concentrated in the first 10 seconds to the first 9 seconds, and the last 8 seconds to 10 seconds, the time collision threshold of the camera is adjusted from the first 10 seconds and the last 10 seconds to the first 8 seconds and the last 8 seconds.

[0094] Tuning is performed based on position. By analyzing and comparing the camera points of correct collisions with those of incorrect collisions, if 80% of correct collisions involve camera A, and 80% of incorrect collisions do not involve camera A, the collision score of camera A will be increased accordingly.

[0095] See also Figure 2 This embodiment provides a face confidence system based on time-space collision, which is used to execute the face confidence method based on time-space collision described in the above method embodiment. Since the technical principle of the system embodiment is similar to the technical principle of the above method embodiment, the same technical details will not be repeated.

[0096] In one embodiment, a face confidence method based on spatiotemporal collision includes: a face profile construction module 10, which is used to obtain face images in a capture scene, classify the face images and construct a face profile; a device association module 11, which is used to associate the corresponding capture device with a target place according to the point information of each capture device in the capture scene, and preset the spatiotemporal collision score of each capture device; a target capture device determination module 12, which is used to obtain a visit record of a target object in the target place, and determine the target capture device corresponding to the target object according to the time node of the visit record; a face profile association module 13, which obtains the cumulative spatiotemporal collision score and the cumulative spatiotemporal collision frequency of the face profile and the target capture device within a preset time period, and associates a face profile that meets the score threshold and frequency threshold of the cumulative spatiotemporal collision score with the target object, wherein the spatiotemporal collision represents the spatiotemporal intersection of the face profile and the target capture device.

[0097] The embodiment of the present application also provides a face confidence device based on spatiotemporal collision, the device may include: one or more processors; and one or more machine-readable media having instructions stored thereon, which, when executed by the one or more processors, enable the device to execute Figure 1 In practical applications, the device can be used as a terminal device or a server. Examples of terminal devices may include: smart phones, tablet computers, e-book readers, MP3 (Moving Picture Experts Group Audio Layer III) players, MP4 (Moving Picture Experts Group Audio Layer IV) players, laptop computers, car computers, desktop computers, set-top boxes, smart TVs, wearable devices, etc. The embodiments of this application do not limit specific devices.

[0098] The present application also provides a machine-readable medium in which one or more modules (programs) are stored. When the one or more modules are applied to a device, the device can execute the embodiment of the present application. Figure 1 Instructions for the steps of the face confidence method based on spatiotemporal collision in the machine-readable medium. The machine-readable medium can be any available medium that can be stored by a computer or a data storage device such as a server or a data center that includes one or more available media. The available medium can be a magnetic medium (such as a floppy disk, a hard disk, a tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)).

[0099] See also Figure 3 This embodiment provides a device 80, which may be a desktop computer, a portable computer, a smart phone, etc. In detail, the device 80 at least includes: a memory 82 and a processor 83 connected via a bus 81, wherein the memory 82 is used to store a computer program, and the processor 83 is used to execute the computer program stored in the memory 82 to execute all or part of the steps in the aforementioned method embodiment.

[0100] The system bus mentioned above can be a Peripheral Pomponent Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The system bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus. The communication interface is used to realize the communication between the database access device and other devices (such as clients, read-write libraries, and read-only libraries). The memory may include random access memory (RAM), and may also include non-volatile memory, such as at least one disk storage.

[0101] The above-mentioned processor can 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 field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0102] In summary, the face confidence method, system, device and medium based on time-space collision of the present invention can accurately associate anonymous information with faces in combination with different application scenarios, which is conducive to targeted behavior analysis and trajectory analysis of associated objects, and enhances user experience. Therefore, the present invention effectively overcomes various shortcomings in the prior art and has high industrial utilization value.

[0103] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by a person of ordinary skill in the art without departing from the spirit and technical concept disclosed by the present invention shall still be covered by the claims of the present invention.

Claims

1. A face confidence method based on spatiotemporal collision, characterized in that: include: Acquire facial images in the captured scene, classify the facial images and build facial files; According to the position information of each capture device in the capture scene, the corresponding capture device is associated with the target location, and a spatiotemporal collision score of each capture device is preset; Obtaining a visit record of a target object in the target place, and determining a target capture device corresponding to the target object according to a time node of the visit record; The method comprises the following steps: obtaining a cumulative spatiotemporal collision score and a cumulative spatiotemporal collision frequency between the face file and the target capture device within a preset time period, and associating a face file that meets a score threshold and a frequency threshold of the cumulative spatiotemporal collision score with the target object, wherein the spatiotemporal collision represents a spatiotemporal intersection of the face file and the target capture device; obtaining a cumulative spatiotemporal collision score and a cumulative spatiotemporal collision frequency between the face file and the target capture device within a preset time period, and associating a face file that meets a score threshold and a frequency threshold of the cumulative spatiotemporal collision score with the target object, including: presetting a spatiotemporal collision screening time for different target places, determining an upper limit and a lower limit of the preset time period according to a visit time node of a target object at a corresponding target place and a corresponding spatiotemporal collision screening time; determining a face image associated with the target object according to a capture record of the target capture device; retrieving from the corresponding face file the number of times the associated face image is captured by the target capture device within the preset time period as the cumulative spatiotemporal collision frequency; and the cumulative spatiotemporal collision score is the product of the cumulative spatiotemporal collision frequency and the spatiotemporal collision score of the target capture device.

2. The face confidence method based on spatiotemporal collision according to claim 1, characterized in that: The preset time-space collision scores of each of the capture devices include: Obtaining quality scores of images captured by each of the capture devices; According to the pre-established correspondence between the quality score and the spatiotemporal collision score, the spatiotemporal collision score of the corresponding capture device is set.

3. The face confidence method based on spatiotemporal collision according to claim 1, characterized in that: When there are multiple associated face images, respectively obtain the cumulative spatiotemporal collision score and the cumulative spatiotemporal collision frequency of each corresponding face file; If the face files corresponding to the plurality of associated face images all meet the score threshold and the frequency threshold, then selecting the face file with a higher cumulative spatiotemporal collision frequency to associate with the corresponding target object; If the spatiotemporal collision frequencies of the plurality of face files are consistent, the face file with the highest cumulative spatiotemporal collision score is selected and associated with the corresponding target object.

4. The face confidence method based on spatiotemporal collision according to claim 1, characterized in that: After associating the face profile that meets the score threshold and frequency threshold of the cumulative spatiotemporal collision score with the target object, the method includes: Acquire the peer target of the corresponding target object according to the associated face file, and acquire the corresponding peer face file through peer target retrieval; According to the peer face file, a related face map of the corresponding target object and the activity track of each peer target are constructed.

5. The face confidence method based on spatiotemporal collision according to claim 1, characterized in that: After associating the face profile that meets the score threshold and frequency threshold of the cumulative spatiotemporal collision score with the target object, the method includes: The association result is output to the management terminal of the target location, and feedback results are obtained by interacting with the management terminal. The accuracy of the association between the face file and the target object is counted according to the feedback results. If the association result is wrong, the association result is obtained again.

6. The face confidence method based on spatiotemporal collision according to claim 5, characterized in that: After counting the correct rate of association between face files and target objects, it includes: Obtain the distribution of correct spatiotemporal collisions of each of the capture devices, obtain the time interval of the correct spatiotemporal collision concentration under a preset first confidence level, and correct the time threshold of the spatiotemporal collision corresponding to the target location according to the time interval; and / or, According to the correct spatiotemporal collision distribution of the capture device, the point information of the capture device corresponding to the correct spatiotemporal collision under a preset second confidence level is obtained, and the spatiotemporal collision score of the corresponding capture device is corrected.

7. A face confidence system based on spatiotemporal collision, characterized in that: include: A face archive building module is used to obtain face images in the captured scene, classify the face images and build a face archive; A device association module, used to associate the corresponding capture device with the target location according to the point information of each capture device in the capture scene, and preset the spatiotemporal collision score of each capture device; A target capture device determination module is used to obtain a visit record of a target object in the target place, and determine a target capture device corresponding to the target object according to a time node of the visit record; The face file association module obtains the cumulative space-time collision score and the cumulative space-time collision frequency between the face file and the target capture device within a preset time period, and associates the face file that meets the score threshold and frequency threshold of the cumulative space-time collision score with the target object, wherein the space-time collision represents the space-time intersection of the face file and the target capture device; obtains the cumulative space-time collision score and the cumulative space-time collision frequency between the face file and the target capture device within a preset time period, and associates the face file that meets the score threshold and frequency threshold of the cumulative space-time collision score with the target object Association, including: presetting the spatiotemporal collision screening time of different target places, determining the upper and lower limits of the preset time period according to the visiting time nodes of the target objects of the corresponding target places and the corresponding spatiotemporal collision screening time; determining the facial image associated with the target object according to the capture records of the target capture device; retrieving the number of times the associated facial image is captured by the target capture device within the preset time period from the corresponding facial file as the cumulative spatiotemporal collision frequency; the cumulative spatiotemporal collision score is the product of the cumulative spatiotemporal collision frequency and the spatiotemporal collision score of the target capture device.

8. An electronic device, characterized in that: include: one or more processors; and One or more machine-readable media having instructions stored thereon, when the one or more processors execute, enable the device to execute the face confidence method based on spatiotemporal collision as described in any one of claims 1-6.

9. One or more machine-readable media, characterized in that Instructions are stored thereon, which, when executed by one or more processors, enable the device to execute the face confidence method based on spatiotemporal collision as described in any one of claims 1-6.

Images