An analysis method for intelligent association between scenes and people

Abstract

The present invention discloses an analysis method for intelligent association between scenes and personnel. In combination with actual public security control needs, key scenes and personnel are classified and coding rules are formulated respectively, and different thematic libraries are established in combination with personnel categories; specific places under various key scenes are marked on the map, and the map grid is used to screen the people card snapshot cameras that are closer; by comparing the people card snapshot data around the specific place with the thematic library of all personnel, the activity frequency and companion information of the people who are successfully matched are recorded; the active personnel in all places under the same type of scene are clustered, and an automatic association relationship between key scenes and personnel is established. This method solves the problem of indiscriminate control of existing key personnel, establishes a dynamic association relationship between personnel and scenes, and realizes dynamic control of specific personnel in specific scenes; solves the problem of automatic discovery and association of personnel groups, and builds group relationships through the activity characteristics between personnel, thereby realizing effective monitoring of groups.

Description

Technical Field

[0001] The present invention relates to the field of intelligent management and control technology, and particularly to a method for analyzing the intelligent association between scenes and personnel. Background Art

[0002] Video surveillance has covered key units, important places and major traffic arteries in the city, but the application of video-based regional security control and urban governance is still insufficient. There is an urgent need to achieve all-round control of key areas in the city.

[0003] Comprehensive control of key urban areas focuses on five key aspects: people, places, events, objects, and organizations. As fundamental elements of control, people and positions must fully consider the individual needs of position control and the targeted needs of personnel control. Previous video-based applications implemented indiscriminate early warning and response based on key personnel. This model implemented facial recognition within a selected range, and responded to any early warning. This model was unable to efficiently achieve accurate early warning of key locations, failed to effectively link positions and personnel, and lacked in-depth, segmented management of personnel, resulting in significant shortcomings. Summary of the Invention

[0004] The purpose of the present invention is to provide an analysis method for the intelligent association between scenes and personnel, to solve the problem of association matching between specific scenes and personnel and intelligent management of personnel in key parts of the city, and to establish an association relationship between scenes and personnel on the basis of the classification of scenes and personnel involved in key parts of the city, to realize intelligent management of scenes and hierarchical classification management of personnel, and at the same time improve the overall management efficiency and realize more efficient, accurate and intelligent management.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] A method for intelligent correlation analysis of scenes and people, wherein the system implementing the method includes: a front-end capture module, an analysis and processing module, and a display application module, including the following steps:

[0007] S1, classify key scenarios of regional public security control to form a scenario category library including scenario 1, scenario 2, scenario 3, scenario 4, scenario 5, scenario 6, scenario 7 and scenario 8, and each scenario is coded with two digits according to its name;

[0008] In combination with the actual control scope of key personnel, a key control personnel category database is established, and each category of personnel is coded with four digits according to its name; at the same time, a corresponding personnel subject database is established according to the personnel category, and the same personnel category can correspond to multiple personnel subject databases;

[0009] The personnel database includes the personnel name, personnel category code, ID number, current residential address, current address latitude and longitude, contact information, and ID photo.

[0010] S2: Various scenarios actually contain multiple specific locations. Map each location and record its center location latitude and longitude. Combined with the distribution of human card capture cameras, select the human card capture cameras that are closest to the specific location. The selection method is as follows:

[0011] Grid the map into multiple square areas of equal size, number each square area, and record the longitude and latitude of the four vertices of the area;

[0012] Locate the square area where the venue is located and obtain the latitude and longitude of the people card capture camera in the square area;

[0013] Based on the fact that at least one vertex has the same longitude and latitude, the eight areas closest to the square area are selected, and the longitude and latitude of the people card capture cameras in the eight areas are obtained;

[0014] Calculate the Euclidean distance between the longitude and latitude of the center of the venue and the longitude and latitude of the pedestrian card capture cameras in this area and the nearest 8 areas.

[0015] Sort all calculated Euclidean distances from small to large;

[0016] Get the front L personal card snapshot camera with smaller distance.

[0017] S3: The front-end capture module collects face capture data from L personal card capture cameras, and compares facial features with all personnel databases through the analysis and processing module. Data that exceeds the set threshold is saved, including the ID number of the successfully matched person, the matching captured person card camera, the capture time, the ID of the captured small face image, and the ID of the captured panoramic large image. The activities of the successful person are compared and analyzed with those of their companions. The analysis process is as follows:

[0018] Map the longitude and latitude of the current address of the successfully matched person to the map grid. If the person appears in the square area where the L personal card cameras are distributed, the person will be filtered out.

[0019] Set a time period T, and generate successful matching data based on each person's card capture camera within this time period. For the same person captured and matched successfully multiple times in a short period of time by the same person's card camera, set a monitoring time period, and the activity frequency within this filtering time period is considered to be 1;

[0020] Clustering is performed based on a single person, generating the number of successful matches (i.e., activity frequency) for the person under L personal card capture cameras;

[0021] By comparing successfully captured images for secondary recognition, we obtain facial information of people in the same frame in the image and compare it with all personnel subject databases. People whose comparison exceeds the set threshold are marked as companions of the activity personnel, and the personnel category code, companion location and time are recorded to generate the companion correlation between the two. Every time the two are captured together, the companion correlation degree increases by 1.

[0022] S4: Cluster the people in all places under the same category of scenes, and automatically obtain the person categories corresponding to the scenes through analysis. The clustering method is as follows:

[0023] Personnel clustering: merging active personnel from multiple locations and accumulating the activity frequencies of the same active personnel; extracting the category codes of the merged active personnel and establishing a dynamic correspondence between scene category codes and personnel category codes. One scene code can correspond to multiple personnel category codes, thus achieving automatic association between scenes and personnel categories. This association can be directly called during use and dynamically adjusted according to actual conditions;

[0024] Clustering of fellow travelers: merging the fellow travelers obtained from multiple locations to generate pairs of fellow travelers, including the number of fellow travelers’ locations and the cumulative degree of fellow travelers’ relevance.

[0025] S5: For the same category of scenes, clustered active personnel are classified according to activity frequency, thereby achieving hierarchical monitoring of personnel. The classification method is as follows:

[0026] Calculate the mean and standard deviation of the activity frequencies of all active personnel in the same category of scenes to obtain their evaluation range;

[0027] When the number of comparisons is greater than the maximum value of the interval, the person is defined as a type of person;

[0028] When the number of comparisons is within the confidence interval, the person is defined as a Class II person;

[0029] When the number of comparisons is less than the minimum interval, the person is defined as a third-category person;

[0030] The successfully matched personnel and their personnel levels are displayed in the display application module. For the first category of personnel, they are the key focus personnel, and their activities are monitored in real time; for the second category of personnel, their activities within a certain period of time are monitored; for the third category of personnel, their activities are only monitored;

[0031] S6, based on the corresponding set of companions of all the participants in the same scenario, group analysis is performed by traversal. The specific analysis method is as follows:

[0032] All participants in the same scenario are sorted from highest to lowest according to their activity frequency. The person with the highest activity frequency is selected and their companion set is used as the basic companion set. Starting from the first person in the set, the corresponding companion set is obtained. This companion set is compared with the basic companion set to merge the companions. During the merging process, the number of companion pairs is first compared based on the number of places they travel to. The larger number of companion pairs is obtained. If they are the same, the larger number of companion pairs is obtained based on the cumulative association degree.

[0033] Traverse and compare the peers in the basic peer set in sequence. Once the traversal is complete, a new group is created and a group label is generated. The label information is determined by the categories of all people in the group.

[0034] Select the basic peer group according to the activity frequency and start a new round of traversal. If the peer group has been merged into the existing group, skip it. If not, generate a new group.

[0035] The above group data is displayed in the display application module.

[0036] S7, calculate the recommendation degree of the generated internal members of the group, the specific method is as follows:

[0037] In a group, the maximum values of the number of peer locations and the peer correlation are determined respectively, and the number of peer locations and the peer correlation in different pairs are normalized by the maximum values to form normalized peer pairs;

[0038] Normalized peer numbers are multiplied internally and converted into percentages to obtain specific recommendation levels;

[0039] Arrange in descending order according to the degree of recommendation to understand the credibility of people in the group.

[0040] As a further solution of the present invention: In S1, the classification of scenes and personnel and the establishment of corresponding category libraries are specifically classified and coded as follows:

[0041] The venues are classified according to the actual security control focus, and a scene category library is established. Each scene type is coded with two digits according to its name. Duplication is avoided during the coding process. Scene categories can be added or adjusted according to actual control needs, as shown in Table 1;

[0042] Table 1 Scenario category library

[0043] Scene Category coding Scene Category coding Scene 1 XX Scene 5 YY Scene 2 ZF Scene 6 SY Scene 3 YL Scene 7 JQ Scene 4 ZJ Scene 8 JT

[0044] In combination with the actual control scope of key personnel, a key control personnel category database is established. Each category of personnel is coded with four digits according to their name to avoid duplication during the coding process. Personnel categories can be added or adjusted according to actual control needs, as shown in Table 2.

[0045] Table 2 Scenario category library

[0046] Personnel Category coding Personnel Category coding Group A SDRY Crowd H FFRY Group B SHRY Crowd I SJRY Crowd C SXRY Crowd J JSBR Crowd D SKRY Crowd K ZTRY Crowd E SQRY Crowd L DBRY Crowd F JSJZ Crowd M GJR Crowd G WWRY Crowd N ZDRK

[0047] As a further solution of the present invention: in S2, the map is gridded and divided into a plurality of square areas of equal size, each square area is numbered, and the longitude and latitude of the four vertices of the area are recorded;

[0048] Locate the square area where the venue is located and obtain the latitude and longitude of the people card capture camera in the square area;

[0049] Based on the fact that at least one vertex has the same longitude and latitude, the eight areas closest to the square area are selected, and the longitude and latitude of the people card capture cameras in the eight areas are obtained;

[0050] Calculate the Euclidean distance between the longitude and latitude of the center of the venue and the longitude and latitude of the pedestrian card capture cameras in this area and the nearest 8 areas.

[0051] Sort all calculated Euclidean distances from small to large;

[0052] Get the front L personal card snapshot camera with smaller distance.

[0053] As a further solution of the present invention: in S3, the facial feature matching success data and its associated data include the ID number of the successfully matched person, the matching captured person card camera, the capture time, the captured face small picture ID and the captured panoramic large picture ID;

[0054] Cluster the number of successful comparisons of the same person under L personal card capture cameras to obtain their activity frequency;

[0055] As a further solution of the present invention: for analysis of the companions of the successfully matched persons, secondary recognition is performed on all successfully matched snapshots to obtain facial feature information of the people in the same frame in the pictures, and the facial feature information is compared with the subject database of all persons respectively. Personnel whose comparison exceeds the set threshold are marked as companions of the active person and added to the array of companions of the person, and the person category code, the place and time of the companionship are recorded to generate a companion correlation degree between the two. Every time the two are captured together, the companion correlation degree is increased by 1, and the number of places of companionship at this time defaults to 1.

[0056] As a further solution of the present invention: In S4, the clustering of personnel is based on selecting one of the multiple places in the same scene as the others. The personnel in the place are sequentially compared with the personnel in other places according to the ID card number index. If the comparison is successful, the activity frequency of the person is the sum of the multiple activity frequencies, and only one is retained.

[0057] If the comparison is unsuccessful, the activity frequency of the active personnel remains unchanged, the personnel are retained, and finally an array of active personnel corresponding to the scene is formed;

[0058] The clustering of people traveling together is to select a place, take the people who are active in the place in turn, and compare them with the people who are active in other places according to the ID number index. If the comparison is successful, the group of people traveling together is merged, and the number of places traveling together and the degree of relevance of traveling together in the number of people traveling together are updated;

[0059] If the match is unsuccessful, skip it; after all places of the person are traversed, select other people in turn until all places of the person are traversed;

[0060] Next, go to the next place and repeat the above traversal for the people who were not successfully matched.

[0061] As a further solution of the present invention: the association between personnel category and scene category is to extract personnel codes from the array of active personnel formed by the scene, and establish a dynamic correspondence between the scene category code and the personnel category code. One scene code can correspond to multiple personnel category codes. This association relationship can be directly called during use and dynamically adjusted according to actual conditions.

[0062] As a further solution of the present invention: in S5, the classification is to calculate the average and standard deviation of the activity frequencies of all active persons in the same category of scenes to obtain their evaluation intervals;

[0063] When the number of comparisons is greater than the maximum value of the interval, the person is defined as a type of person;

[0064] When the number of comparisons is within the confidence interval, the person is defined as a Class II person;

[0065] When the number of comparisons is less than the minimum interval, the person is defined as a third-category person;

[0066] The successfully matched personnel and their personnel levels are displayed in the display application module. For the first category of personnel, who are the key focus personnel, their activities are monitored in real time; for the second category of personnel, their activities within a certain period of time are monitored; for the third category of personnel, it is sufficient to know their activities.

[0067] As a further solution of the present invention: In S6, group analysis is to sort all active personnel in the same type of scenario from large to small according to the frequency of activity, screen out the personnel with the highest activity frequency, and use their peer set as the basic peer set. Starting from the first peer in the set, obtain the corresponding peer set, compare the peer set with the basic peer set, and realize the merging of peers. During the merging process, the number of peer pairs is first compared according to the number of peer locations to obtain the larger number of peer pairs. If they are the same, the larger number of peer pairs is obtained according to the accumulated peer correlation; the peers in the basic peer set are traversed and compared in turn, and when the traversal is completed, a new group is established and a group label is generated. The label information is determined by the categories of all personnel in the group; the peer set is screened in order from large to small according to the frequency of activity as the basic peer set, and a new round of traversal is carried out. If the peer set has been merged into the existing group, it is skipped. If not, a new group is generated; the above group data is displayed in the display application module.

[0068] As a further solution of the present invention: In S7, the recommendation degree of the internal personnel of the group is calculated by determining the maximum values of the number of places of travel together and the degree of relevance of the same company within the group, and normalizing the number of places of travel together and the degree of relevance of the same company in different number pairs by the maximum values to form normalized number pairs of travel together; the normalized number pairs of travel together are internally multiplied and converted into percentages to obtain specific recommendation degrees; and they are arranged in descending order according to the degree of recommendation to grasp the credibility of the people in the group.

[0069] Beneficial effects of the present invention:

[0070] (1) The present invention classifies key urban scenes and important control personnel in accordance with the actual needs of public security control, and encodes them one by one, establishes a scene category library and a personnel category library, realizes the classification management of scenes and personnel, and provides a basis for further analysis of scenes and personnel;

[0071] (2) The present invention combines map grid division and Euclidean distance to realize the automatic selection of face capture cameras around a specific place, thereby reasonably screening cameras that are closer to avoid omissions;

[0072] (3) The present invention relies on the relatively mature face recognition technology used in existing actual combat. It dynamically identifies the active personnel around the venue through successful face comparison data, obtains all active personnel in a specific scene through clustering, and manages the personnel in a hierarchical manner according to the frequency of their activities. It can be dynamically adjusted according to the actual situation to ensure that active key personnel can receive timely attention and disposal, realize the reasonable allocation of supervision resources in actual management and control, solve the indiscriminate management and control of personnel, and realize hierarchical monitoring based on the activity situation;

[0073] (4) The present invention can establish a correspondence between scenes and people by acquiring active people in specific scenes. This correspondence is dynamically adjusted according to the actual face capture situation, fully reflecting the distribution of key people in the current specific scene, realizing automatic association and analysis of people in the scene, and solving the problem of insufficient accuracy of manual association and inability to make real-time dynamic adjustments.

[0074] (5) The present invention makes full use of the personal information hidden in the large face capture image. Based on the secondary recognition, it mines the information of people in the same frame, analyzes potential groups based on this, and quantitatively analyzes the credibility of people in the group. It further analyzes the group organization associated with the people and establishes a personnel relationship network, thereby providing data support for crackdown and processing;

[0075] (6) The present invention solves a series of problems such as scene personnel classification definition, automatic personnel identification, dynamic association between scenes and personnel, hierarchical management and control of personnel, and analysis of associated groups, and builds a complete set of analysis methods to systematically achieve the purpose of associated management and control of specific personnel in specific scenes, meeting the needs of refined and efficient management of scenes and personnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] The present invention will be further described below with reference to the accompanying drawings.

[0077] Figure 1 This is an overall flow chart of a scene and personnel intelligent association analysis method of the present invention;

[0078] Figure 2 This is a flow chart for identifying people and their companions at a venue according to the present invention;

[0079] Figure 3 This is a flow chart of cluster analysis of people involved in the same scene according to the present invention;

[0080] Figure 4 Generate a corresponding flow chart of the group of people traveling together for the same scene activity personnel of the present invention;

[0081] Figure 5 A flowchart for hierarchical monitoring of active personnel in the same scenario of the present invention;

[0082] Figure 6 Customize the recognition flow chart for the same scene group of the present invention;

[0083] Figure 7 This is a flowchart of the credibility analysis of people in a group in the same scenario according to the present invention. DETAILED DESCRIPTION

[0084] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0085] See also Figure 1 As shown, the present invention is an analysis method for intelligent association between scenes and people, including a front-end capture module, an analysis and processing module, and a display application module, which specifically includes the following steps:

[0086] S1. Categorize key scenes and key personnel for regional public security control. Key scenes include scene 1, scene 2, scene 3, scene 4, scene 5, scene 6, scene 7, and scene 8. Establish a scene category library, and assign a two-digit code to each scene category according to its name. At the same time, divide personnel into groups A, B, C, D, E, F, G, H, I, J, K, L, M, and N. Establish a key control personnel category library, and assign a four-digit code to each group of personnel according to their name.

[0087] At the same time, a corresponding personnel database is established according to the personnel category. The same personnel category can correspond to multiple personnel databases.

[0088] The personnel database includes the personnel name, personnel category code, ID number, current residential address, current address latitude and longitude, contact information, and ID photo.

[0089] S2: Various scenarios actually contain multiple specific locations. Map each location and record its center location latitude and longitude. Combined with the distribution of human card capture cameras, select the human card capture cameras that are closest to the specific location. The selection method is as follows:

[0090] The map is gridded and divided into multiple square areas of equal size. The starting point O(0,0) is selected on the map, and each square area is numbered S from the starting point. i,j (i=0,1,2......N;j=0,1,2......M), and record the longitude and latitude coordinates of the four vertices of the area

[0091] Locate the square area where the specific place is located and obtain its number S k,g Mark the human card capture camera on the map and obtain the latitude and longitude of all human card capture cameras in the square area;

[0092] Based on the fact that at least one vertex has the same longitude and latitude, the 8 nearest areas around the square area are selected and numbered as S k,g-1 、S k,g+1 、S k-1,g-1 、S k-1,g+1 、S k-1,g 、S k+1,g 、S k+1,g-1 、S k+1,g+1 , and obtain the longitude and latitude coordinates of the human card capture cameras in the 8 areas;

[0093] The longitude and latitude of the center of the specific place The latitude and longitude of the human card capture camera in this area and the nearest 8 areas Perform Euclidean distance calculation in sequence;

[0094]

[0095] Sort all calculated Euclidean distances from small to large;

[0096] Get the front L personal card snapshot camera with smaller distance.

[0097] S3: The front-end capture module collects L facial capture data from the personal card capture camera, and compares the facial features with the all-person database through the analysis and processing module. The data exceeding the set threshold is saved, including the ID number of the successfully matched person, the matching captured person card camera, the capture time, the ID of the captured face small picture and the ID of the captured panoramic large picture. The activities of the successful person are compared and analyzed with those of their companions. The analysis process is as follows:

[0098] For the activities of the successfully matched persons, first filter out the worthless data and map the longitude and latitude of the current address of the successfully matched persons to the map grid. If the person appears in the square area where the L personal card cameras are distributed, the person will be filtered out.

[0099] Set a time period T, and generate successful matching data based on each person's card capture camera within this time period. For the same person captured and matched successfully multiple times in a short period of time by the same person's card camera, set a monitoring time period, and the activity frequency within this filtering time period is considered to be 1;

[0100] Taking a single person as the dimension and using the ID number as the index, we obtain the number of successful matches for that person, and perform clustering and accumulation to generate the number of successful matches for that person under L personal card capture cameras (i.e., activity frequency);

[0101] For the analysis of the companions of the successfully matched persons, secondary recognition is performed on all successfully matched snapshots to obtain the facial feature information of the people in the same frame in the picture, and the facial feature information is compared with the subject database of all persons respectively. The persons whose comparison exceeds the set threshold are marked as companions of the active person and added to the array of companions of the person. The person category code, the place and time of the companionship are recorded to generate the companionship correlation between the two. Every time the two are captured together, the companionship correlation degree is increased by 1. At this time, the number of places of companionship defaults to 1.

[0102] S4: Cluster the people in all places under the same category of scenes, and automatically obtain the person categories corresponding to the scenes through analysis. The clustering method is:

[0103] Clustering of active people: For multiple venues in the same scenario, one venue is selected as the basis, and the active people in the venue are compared with the active people in other venues in sequence according to the ID number index. If the comparison is successful, the activity frequency of the person is the sum of the multiple activity frequencies, and only one is retained;

[0104] If the comparison is unsuccessful, the activity frequency of the active personnel remains unchanged, the personnel are retained, and finally an array of active personnel corresponding to the scene is formed.

[0105] The clustering of people traveling together is to select a place, take the people who are active in the place in turn, and compare them with the people who are active in other places according to the ID number index. If the comparison is successful, the group of people traveling together is merged, and the number of places traveling together and the degree of relevance of traveling together in the number of people traveling together are updated;

[0106] If the matching is unsuccessful, it will be skipped; after all places of the person are traversed, other people will be selected in turn until all people in the place are traversed; then go to the next place and repeat the above traversal for the people who have not been matched successfully.

[0107] The association between personnel category and scene category is to extract the personnel code from the array of active personnel formed by the scene, and establish a dynamic correspondence between the scene category code and the personnel category code. One scene code can correspond to multiple personnel category codes. This association relationship can be directly called during use and dynamically adjusted according to actual conditions.

[0108] S5: For the same category of scenes, clustered active personnel are classified according to activity frequency, thereby achieving hierarchical monitoring of personnel. The classification method is:

[0109] Calculate the mean and standard deviation of the activity frequencies of all active personnel in the same category of scenes to obtain their evaluation range;

[0110] When the number of comparisons is greater than the maximum value of the interval, the person is defined as a type of person;

[0111] When the number of comparisons is within the confidence interval, the person is defined as a Class II person;

[0112] When the number of comparisons is less than the minimum interval, the person is defined as a third-category person;

[0113] The successfully matched personnel and their personnel levels are displayed in the display application module. For the first category of personnel, they are the key focus personnel, and their activities are monitored in real time; for the second category of personnel, their activities within a certain period of time are monitored; for the third category of personnel, their activities are only monitored;

[0114] S6, based on the corresponding set of companions of all the activity personnel in the same scenario, group analysis is performed by traversal. The specific analysis method is as follows:

[0115] All participants in the same scenario are sorted by activity frequency, and the participants with the highest activity frequency are selected. Their companion sets are used as the basic companion sets. Starting from the first companion in the set, the corresponding companion set is obtained. This companion set is compared with the basic companion set to merge the companions. During the merging process, the number of companion pairs is first compared based on the number of places of travel, and the larger number of companion pairs is obtained. If they are the same, the larger number of companion pairs is obtained based on the accumulated companion correlation. The companions in the basic companion set are traversed and compared in sequence. When the traversal is completed, a new group is established and a group label is generated. The label information is determined by the categories of all people in the group.

[0116] The peer set is selected in descending order based on the frequency of activities as the basic peer set, and a new round of traversal is carried out. If the peer set has been merged into the existing group, it is skipped. If not, a new group is generated;

[0117] The above group data is displayed in the display application module.

[0118] S7, calculate the recommendation degree of the generated internal members of the group, specifically:

[0119] In a group, the maximum values of the number of peer locations and the peer correlation are determined respectively, and the number of peer locations and the peer correlation in different pairs are normalized by the maximum values to form normalized peer pairs;

[0120] Normalized peer numbers are multiplied internally and converted into percentages to obtain specific recommendation levels;

[0121] Arrange in descending order according to the degree of recommendation to understand the credibility of people in the group.

[0122] Depend on Figure 2 As shown, Example 1:

[0123] The identification of participants and their companions at various locations specifically includes the following steps:

[0124] S21, set a time period T, collect L personal card capture camera face capture data;

[0125] S22, comparing facial features of the face capture data of each camera with the personnel database to generate comparison success data;

[0126] S23, filtering out worthless data based on all successfully matched data;

[0127] First, considering the relationship between the current address of the person and the analysis location, activities around the residence are excluded. The longitude and latitude of the current address of the successfully matched person are mapped to the map grid, and the square area number of the square area is obtained. This number is compared with the square area numbers of the L personal card snapshot cameras. If they are the same, the person is filtered out and activity frequency analysis is not performed. If they are different, activity frequency analysis is performed and the process jumps to step S24;

[0128] Secondly, consider the situation where the same person wanders or stays under the same person card camera for a certain period of time. This situation will cause a surge in activity frequency, but the actual application value is not great. Therefore, a monitoring time period t is set, and the number of successful comparisons of the same person under the same person card camera within the time period t is filtered, recorded as 1, and jump to step S24;

[0129] S24, taking a single person as the dimension, cluster all the successful comparisons using the ID number as the index to generate the person R m Activity frequency P m (m=1,2,3......K), K is the total number of successful matches;

[0130] S25, performing secondary face recognition on the successfully captured panoramic image, obtaining facial feature information of the people in the same frame in the image, and comparing it with all personnel subject databases respectively, and setting a comparison threshold;

[0131] S26, if person R r The match was successful and marked as active person R m colleagues, joined R m The number of companions Λ m middle.

[0132] Record the category, location and time of the traveling companions and generate the personnel R r Relative to personnel R m Peer correlation R r With R m The relationship and relevance of fellow travelers obtained based on the analysis of different snapshots of panoramic images Automatically add 1, the number of places you go together is D m The default is 1.

[0133] Depend on Figure 3 As shown, Example 2:

[0134] Cluster analysis is performed on the participants in different places under the same scene to obtain the participants associated with the scene. Specifically, the following steps are included:

[0135] S31, for the same type of scene (taking scene 2 as an example), for Y places, they can be marked as ZF1, ZF2...ZF Y , for each location ZF d (d=1,2,3......Y) Create an array for the analyzed active personnel

[0136] S32, based on location ZF d As a basis, usually d starts from 1, and the location ZF d In sequence with ZF d+i (i=1,2......,Yd) comparison, the corresponding array The people in the All the people in are traversed and compared, and after the comparison is completed, i=i+1 is incremented sequentially;

[0137] S33, if there is a person with a matching ID number, The activity frequency corresponding to the person in the update is the sum of the activity frequencies in the two places, The corresponding person in the array is deleted and becomes Jump to step S35;

[0138] S34, if the match is inconsistent, the activity personnel are in the original array remains unchanged, jump to step S35;

[0139] S35, and so on, finally The active people in the array are There is no overlap, The original personnel will remain unchanged. The number of people in and retain its original activity frequency, jumping to step S32;

[0140] S36, until all Y arrays are completed, and finally the activity personnel array corresponding to the scene is formed

[0141] Depend on Figure 4 As shown, Example 3:

[0142] Performing a group analysis of people involved in the same scene includes the following steps:

[0143] S41, for the same type of scene (taking scene 2 as an example), for Y places, they can be marked as ZF1, ZF2...ZF Y , for each location ZF d (d=1,2,3......Y) Create an array for the analyzed active personnel

[0144] S42, get the array of active personnel Middle activity staff R m , m starts from 1 and increases in sequence;

[0145] S43, R m An array of active people with another venue The people in the are matched by their ID numbers as indexes, and after the matching is completed, i=i+1 increases in sequence;

[0146] S44, if the match is inconsistent, jump to step S43;

[0147] S45, if the match is consistent, then Middle staff R m The corresponding set of companions Λ m Merge into Middle staff R m The corresponding set of companions Λ m middle;

[0148] S46, judge R m R tr The number of people in the group;

[0149] S47, if R m R's fellow tr Only appears in one set of peers, and the number of peer locations corresponding to it in the new set of peers is D m 1, peer correlation is the correlation degree of the original set of companions, jump to step S43;

[0150] S48, if the accompanying person R tr If it appears in both sets of peers, the number of its corresponding peer locations in the new set of peers is D m Plus 1, peer correlation is the sum of the corresponding correlations of the two peers. The person is deleted and the process goes to step S43;

[0151] S49, for each location R m The set of companions is matched and merged, and finally the activity personnel R is generated m The set of people who travel with you in this type of scenario Λ'm , where Λ' m =Λ m , and obtain the R tr Relative to R m The maximum number of peer locations and peer relevance in this scenario.

[0152] Depend on Figure 5 As shown, Example 4:

[0153] Conduct hierarchical monitoring of people in the same scene, including the following steps:

[0154] S51, the final analysis of the same category of scenes (taking scene 2 as an example) results in an array of active personnel Calculate the mean M and standard deviation δ based on the activity frequency;

[0155] S52, calculate the evaluation interval [M-δ, M+δ];

[0156] S53, will The activity frequency P of each person in m Compare with the evaluation interval;

[0157] S54, when the activity frequency P m >M+δ, the personnel is defined as a type of personnel;

[0158] S55, when the activity frequency P m ∈[M-δ,M+δ], the person is defined as a second-class person;

[0159] S56, when the activity frequency P m When <M-δ, the personnel is defined as the third category personnel;

[0160] Depend on Figure 6 As shown, Example 5:

[0161] Analyze existing groups based on the people involved in the same scenario. This includes the following steps:

[0162] S61, array of active personnel for the same type of scene Sort by activity frequency from large to small, and give priority to the person with the highest activity frequency R max , with R max The corresponding set of companions Λ max As a basic peer set;

[0163] S62, select the companion R from the basic companion set r (r=1,2......,g), r usually starts from 1, and obtains its corresponding set of peers Λ r ;

[0164] S63, collect the companions into Λ r With the basic people set Λ max The people in the list are compared one by one through the ID card number index;

[0165] S64, if the same, the person will be placed in Λ max and Λ r Compare the number of peers in the same industry, and give priority to comparing the number of peer locations. If they are inconsistent, take the number of peers corresponding to the larger number of peer locations. If they are consistent, compare according to the peer correlation, take the number of peers corresponding to the larger value, and update the number of peers determined above to Λ max The number of the person's peers in the process is matched, and the process goes to step S66;

[0166] S65, if different, in Λ max There is Λ r No personnel in Λ max Continue to retain, and the number of pairs in the same row remains unchanged, jump to step S66; in Λ r There is Λ max If there is no person in the list, add the person to Λ max and bring in the original number of pairs of the same line, and jump to step S66;

[0167] S66: Traverse the group of people in sequence and perform the above-mentioned comparison analysis on their group of people with the basic group of people. After the traversal is completed, a new group G1 is established and a group label is generated. The label information consists of the categories of all people in the group, and duplicate categories are filtered out.

[0168] S67, sorting the set of companions by activity frequency;

[0169] S68, determine whether the peer set has been merged into the existing group; if so, jump to step S67; if not, use the peer set as the basic peer set to start traversal analysis to generate a new group, jump to step S62;

[0170] S69, establish a group set G = {G i}(i=1,2......L), where G i Represents a relatively independent group.

[0171] Depend on Figure 7 As shown, Example 6:

[0172] Conducting a credibility analysis on people in a group under the same scenario includes the following steps:

[0173] S71, in a group G i The total number of people is Q;

[0174] S72: Take the number of places and the degree of relevance of all pairs of people traveling together, and create a set of places. and peer correlation set

[0175] S73, determine the maximum number of locations D for each trip max The maximum correlation degree with peers θ max ;

[0176] S74, normalize the number of locations and the degree of relevance of the same industry in all pairs of numbers by using the two maximum values to form normalized number pairs of the same industry (D q / D max ,θ q / θ max )(q=1,2......Q);

[0177] S75, calculating the recommendation degree of each normalized peer number pair, where the recommendation degree is the product of the normalized peer location number and the normalized peer association degree within the normalized pair and converted into a percentage.

[0178] Θ q =(D q / D max )*(θ q / θ max )*100%

[0179] S76, all recommendation degrees within the group are arranged in descending order. The credibility of the people with high recommendation degrees as group members is higher, and the credibility of the people with low recommendation degrees as group members is lower.

[0180] In summary, the present invention solves a series of problems in scene personnel classification definition, automatic personnel identification, dynamic association between scenes and personnel, hierarchical management and control of personnel, and analysis of associated groups. It builds a complete set of analysis methods to systematically achieve the purpose of associated management and control of specific personnel in specific scenes, solve the problem of indiscriminate personnel management, and at the same time solve the problem of insufficient flexibility and poor real-time performance of manual settings. It realizes the automation of association, intelligent analysis, and dynamic setting in a systematic way, meeting the needs of social governance for refined and efficient management of scenes and personnel. The present invention can be used in the design and development of social security prevention and control system platforms and video surveillance application platforms, serving social governance and public safety.

[0181] The above is a detailed description of an embodiment of the present invention. However, the content described is only a preferred embodiment of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A method for analyzing the intelligent association between scenes and people, characterized in that: The system implementing this method includes a front-end capture module, an analysis and processing module, and a display application module. The specific steps are as follows: S1: Establish and code a scenario category library based on key scenarios of regional public security control. At the same time, establish a category library of key controlled personnel and code the personnel categories based on the actual control scope of key personnel. Establish a corresponding thematic library for each category of personnel. S2: Various scenarios actually contain multiple specific locations. Map each location and record its center location latitude and longitude. Combined with the geographical distribution of the personal card snapshot cameras, select L personal card snapshot cameras that are closest to the specific location. S3: The front-end capture module collects facial capture data from L personal card capture cameras, and compares facial features with the database of all personnel through the analysis and processing module. The data exceeding the set threshold and its related data are saved, and the successful personnel are compared for activity analysis and companion analysis; S4: Cluster the people in all places under the same category of scenes, and automatically obtain the person category corresponding to the scene through analysis; S5: For the same category of scenes, clustered active personnel are classified according to activity frequency, thereby achieving hierarchical monitoring of personnel; S6: Perform group analysis by traversing the corresponding set of peers for all participants in the same scenario; S7: Calculate the recommendation degree of the generated internal members of the group to determine their credibility; In S7, the recommendation degree of internal group members is calculated by determining the maximum values of the number of peer locations and the peer correlation within a group, and normalizing the number of peer locations and the peer correlation in different pairs by the maximum values to form normalized peer pairs; the normalized peer pairs are multiplied internally and converted into percentages to obtain the specific recommendation degree; and the pairs are arranged in descending order according to the recommendation degree to grasp the credibility of the people in the group.

2. The method for analyzing intelligent association between scenes and people according to claim 1, characterized in that: In S1, key scenarios include scenario 1, scenario 2, scenario 3, scenario 4, scenario 5, scenario 6, scenario 7, and scenario 8. Each scenario is coded with two digits according to its name. Duplication is avoided during the coding process, and scenario categories are added or adjusted according to actual management and control needs. Personnel categories are coded with four digits according to their names. Duplication is avoided during the coding process, and personnel categories are added or adjusted according to actual management and control needs. At the same time, corresponding personnel subject libraries are established according to personnel categories, and the same personnel category corresponds to multiple personnel subject libraries. The personnel subject library contains personnel name, personnel category code, ID number, current residential address, current address latitude and longitude, contact information, and ID photo, and the fields support custom extension.

3. The method for analyzing intelligent association between scenes and people according to claim 1, characterized in that: In S2, the map is gridded and divided into multiple square areas of equal size. Each square area is numbered, and the longitude and latitude of the four vertices of the area are recorded. Locate the square area where the venue is located and obtain the latitude and longitude of the people card capture camera in the square area; Based on the fact that at least one vertex has the same longitude and latitude, the eight areas closest to the square area are selected, and the longitude and latitude of the people card capture cameras in the eight areas are obtained; Calculate the Euclidean distance between the longitude and latitude of the center of the venue and the longitude and latitude of the pedestrian card capture cameras in this area and the nearest 8 areas. Sort all calculated Euclidean distances from small to large; Get the front L personal card snapshot camera with smaller distance.

4. The method for analyzing intelligent association between scenes and people according to claim 1, characterized in that: In S3, the facial feature matching success data and its associated data include the ID number of the matched person, the camera that matched the captured person card, the capture time, the ID of the captured face small picture, and the ID of the captured panoramic large picture; The number of successful comparisons of the same person under L personal card capture cameras is clustered to obtain their activity frequency.

5. The method for analyzing intelligent association between scenes and people according to claim 4, characterized in that: For the analysis of the companions of the successfully matched persons, secondary recognition is performed on all successfully matched snapshots to obtain the facial feature information of the people in the same frame in the picture, and the facial feature information is compared with the subject database of all persons respectively. The persons whose comparison exceeds the set threshold are marked as companions of the active person and added to the array of companions of the person. The person category code, the place and time of the companionship are recorded to generate the companionship correlation between the two. Every time the two are captured together, the companionship correlation degree is increased by 1. At this time, the number of places of companionship defaults to 1.

6. The method for analyzing intelligent association between scenes and people according to claim 1, characterized in that: In S4, the clustering of people is based on selecting one place from multiple places in the same scene. The people who are active in this place are compared with the people who are active in other places in sequence according to their ID number index. If the comparison is successful, the activity frequency of this person is the sum of the multiple activity frequencies, and only one is retained. If the comparison is unsuccessful, the activity frequency of the active personnel remains unchanged, the personnel are retained, and finally an array of active personnel corresponding to the scene is formed; The clustering of people traveling together is to select a place, take the people who are active in the place in turn, and compare them with the people who are active in other places according to the ID number index. If the comparison is successful, the group of people traveling together is merged, and the number of places traveling together and the degree of relevance of traveling together in the number of people traveling together are updated; If the match is unsuccessful, skip it; after all places of the person are traversed, select other people in turn until all places of the person are traversed; Next, go to the next place and repeat the above traversal for the people who were not successfully matched.

7. The method for analyzing intelligent association between scenes and people according to claim 6, characterized in that: The association between personnel category and scene category is to extract the personnel code from the array of active personnel formed by the scene, and establish a dynamic correspondence between the scene category code and the personnel category code. One scene code can correspond to multiple personnel category codes. This association relationship can be directly called during use and dynamically adjusted according to actual conditions.

8. The method for analyzing intelligent association between scenes and people according to claim 1, characterized in that: In S5, the classification is to calculate the mean and standard deviation of the activity frequency of all active personnel in the same category of scenes to obtain their evaluation range; When the number of comparisons is greater than the maximum value of the interval, the person is defined as a type of person; When the number of comparisons is within the confidence interval, the person is defined as a Class II person; When the number of comparisons is less than the minimum interval, the person is defined as a third-category person; The successfully matched personnel and their personnel levels are displayed in the display application module. For the first category of personnel, who are the key focus personnel, their activities are monitored in real time; for the second category of personnel, their activities within a certain period of time are monitored; for the third category of personnel, it is sufficient to know their activities.

9. The method for analyzing intelligent association between scenes and people according to claim 1, characterized in that: In S6, group analysis is to sort all the active personnel in the same type of scene according to the frequency of activities from large to small, screen out the personnel with the highest frequency of activities, and use their companion set as the basic companion set. Starting from the first companion in the set, obtain the corresponding companion set, compare the companion set with the basic companion set, and merge the companions. During the merging process, the number of companions is first compared according to the number of companion locations to obtain the larger number of companions. If they are the same, the larger number of companions is obtained according to the cumulative companion correlation. The companions in the basic companion set are traversed and compared in turn. When the traversal is completed, a new group is established and a group label is generated. The label information is determined by the categories of all the people in the group. The companion set is selected in order from large to small according to the frequency of activities as the basic companion set, and a new round of traversal is carried out. If the companion set has been merged into the existing group, it is skipped. If not, a new group is generated. The above group data is displayed in the display application module.

Images