Personnel roll call method and personnel roll call system

Abstract

The invention relates to a personnel roll call method and a personnel roll call system. The method comprises the following steps: based on roll call task information, acquiring image acquisition data acquired by at least one image acquisition device in a preset area corresponding to the roll call task information; the roll call task information comprises preset personnel feature information and preset target roll call information; based on a preset image recognition algorithm corresponding to the roll call task information, recognizing the image acquisition data to obtain personnel feature information; the preset image recognition algorithm comprises one or more of a face recognition algorithm, a gait recognition algorithm and a human body recognition algorithm; comparing the personnel characteristic information with preset personnel characteristic information, and determining identified personnel information; and matching the identified personnel information with preset target roll-call information, and determining a personnel roll-call result. By adopting the method, the supervision cost can be reduced while the roll call accuracy is ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image recognition, in particular to a personnel roll call method and a personnel roll call system. BACKGROUND

[0002] Supervision places need to count the detainees more frequently every day, which is called roll call operation. With the development of information technology, in order to improve the accuracy of roll call and reduce the missed detection rate, in addition to the manual verification method, fingerprint recognition technology, portable two-dimensional code scanning equipment, vital signs bracelet and other various special devices are also used for supervision.

[0003] However, although the supervision method using special devices can improve the effectiveness of personnel roll call to some extent, the special devices themselves are additional hardware devices, which inevitably leads to an increase in the construction cost of the supervision system.

[0004] Therefore, the roll call operation of the supervision place still has the problem of high cost. SUMMARY

[0005] Therefore, it is necessary to provide a personnel roll call method and a personnel roll call system which can reduce the supervision cost while ensuring the accuracy of roll call.

[0006] In a first aspect, the present application provides a personnel roll call method, which comprises:

[0007] Based on the roll call task information, image acquisition data collected by at least one image acquisition device in a preset area corresponding to the roll call task information is obtained; the roll call task information comprises preset personnel feature information and preset target roll call information;

[0008] Based on a preset image recognition algorithm corresponding to the roll call task information, the image acquisition data is recognized to obtain personnel feature information; the preset image recognition algorithm comprises one or more of a face recognition algorithm, a gait recognition algorithm and a human body recognition algorithm;

[0009] According to the comparison between the personnel feature information and the preset personnel feature information, the identification personnel information is determined;

[0010] The identification personnel information is matched with the preset target roll call information to determine the personnel roll call result.

[0011] In one embodiment, the step of obtaining the image acquisition data collected by the at least one image acquisition device in the preset area corresponding to the roll call task information based on the roll call task information comprises:

[0012] Obtaining real-time available resource information;

[0013] determine a target resource requirement corresponding to the roll call task information based on the roll call task information;

[0014] if the real-time available resource information meets the target resource requirement, acquire the image acquisition data;

[0015] if the real-time available resource information does not meet the target resource requirement, write the roll call task information into a waiting queue, and acquire the image acquisition data when the real-time available resource information meets the target resource requirement.

[0016] In one embodiment, the determining of the target resource requirement corresponding to the roll call task information based on the roll call task information comprises:

[0017] perform task splitting based on the roll call task information to obtain a roll call task group; the roll call task group comprises a plurality of task subunits;

[0018] determine a target resource requirement of each of the task subunits to determine whether the real-time available resource information meets the target resource requirement of the task subunit.

[0019] In one embodiment, the generating of the roll call task information comprises:

[0020] acquire legal personnel access information in a preset area;

[0021] determine preset target roll call information in the preset area based on the legal personnel access information;

[0022] acquire personnel feature information corresponding to the preset target roll call information from a personnel feature information library as preset personnel feature information;

[0023] generate the roll call task information based on the preset personnel feature information and the preset target roll call information.

[0024] In one embodiment, the determining of the recognized personnel information based on the comparison of the personnel feature information and the preset personnel feature information comprises:

[0025] match first personnel feature information corresponding to the face recognition algorithm with preset personnel feature information to obtain pending recognized personnel information corresponding to the first personnel feature information and a confidence level;

[0026] if the confidence level meets a preset threshold, the pending recognized personnel information is used as the recognized personnel information;

[0027] If the confidence degree does not satisfy the preset threshold, verifying the to-be-identified personnel information based on the second personnel feature information corresponding to the gait recognition algorithm and the third personnel feature information corresponding to the human body recognition algorithm, and calculating a comprehensive confidence degree of the to-be-identified personnel information;

[0028] If the comprehensive confidence degree satisfies the preset threshold, taking the to-be-identified personnel information as the identified personnel information.

[0029] In one of the embodiments, verifying the to-be-identified personnel information based on the second personnel feature information corresponding to the gait recognition algorithm and the third personnel feature information corresponding to the human body recognition algorithm, and calculating a comprehensive confidence degree of the to-be-identified personnel information includes:

[0030] Comparing the second personnel feature information corresponding to the gait recognition algorithm with the preset personnel feature information to obtain first verification information;

[0031] Comparing the third personnel feature information corresponding to the human body recognition algorithm with the preset personnel feature information to obtain second verification information;

[0032] Verifying the to-be-identified personnel information based on the first verification information, the second verification information, and a preset confidence degree weight combination;

[0033] Based on the confidence degree and the verification result, calculating a comprehensive confidence degree of the to-be-identified personnel information.

[0034] In one of the embodiments, the roll-call task information further includes a roll-call mode; the preset target roll-call information corresponds to the roll-call mode;

[0035] The personnel feature information is obtained by identifying the image collection data based on a preset image recognition algorithm corresponding to the roll-call task information includes: based on the roll-call mode, calling a preset image recognition algorithm corresponding to the roll-call mode to identify the image collection data and obtain the personnel feature information.

[0036] In one of the embodiments, after matching the identified personnel information with the preset target roll-call information to determine the personnel roll-call result, the method further includes:

[0037] Determining roll-call information in the preset target roll-call information that cannot be matched with the identified personnel information as absent personnel information;

[0038] Generating change clue information corresponding to the absent personnel information; the change clue information includes historical image collection data, last appearance time, and last appearance position corresponding to the absent personnel information.

[0039] In a second aspect, the present application provides a personnel roll call system, comprising a task publishing component, an image recognition component, a personnel recognition component, and at least one image acquisition device, wherein:

[0040] The task publishing component is configured to store roll call task information, wherein the roll call task information comprises preset personnel feature information and preset target roll call information.

[0041] The image recognition component is configured to acquire image acquisition data collected by at least one image acquisition device in a preset area corresponding to the roll call task information, and identify the image acquisition data based on a preset image recognition algorithm corresponding to the roll call task information to obtain personnel feature information, wherein the preset image recognition algorithm comprises one or more of a face recognition algorithm, a gait recognition algorithm, and a human body recognition algorithm.

[0042] The personnel recognition component is configured to compare the personnel feature information with the preset personnel feature information to determine recognized personnel information, and match the recognized personnel information with the preset target roll call information to determine a personnel roll call result.

[0043] In one embodiment, the personnel roll call system further comprises a message queue component, and the image recognition component is further configured to send the personnel feature information to the message queue component, and the message queue component is configured to send the personnel feature information to the personnel recognition component.

[0044] The personnel roll call method and the personnel roll call system described above acquire image acquisition data collected by at least one image acquisition device in a preset area corresponding to roll call task information based on the roll call task information, wherein the roll call task information comprises preset personnel feature information and preset target roll call information; identify the image acquisition data based on a preset image recognition algorithm corresponding to the roll call task information to obtain personnel feature information, wherein the preset image recognition algorithm comprises one or more of a face recognition algorithm, a gait recognition algorithm, and a human body recognition algorithm; compare the personnel feature information with the preset personnel feature information to determine recognized personnel information; and match the recognized personnel information with the preset target roll call information to determine a personnel roll call result. By integrating existing monitoring devices and algorithms, the hardware cost caused by using special devices for roll call operations in related technologies is reduced, different personnel features are recognized cooperatively by using multiple preset image recognition algorithms, the accuracy of recognition is effectively improved, the roll call operation time is shortened, and human errors are reduced, thereby achieving the effect of reducing the supervision cost while ensuring the accuracy of roll call. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 An application environment diagram of the personnel roll call method in one embodiment;

[0046] Figure 2 Flowchart of a personnel roll-call method in an embodiment;

[0047] Figure 3 Interaction diagram of a personnel roll-call system in an embodiment;

[0048] Figure 4 Flowchart of a roll-call task state machine management mechanism in an embodiment;

[0049] Figure 5 Structural block diagram of a multi-modal recognition system in an embodiment;

[0050] Figure 6 Structural block diagram of a personnel roll-call system in an embodiment;

[0051] Figure 7 Internal structural diagram of a computer device in an embodiment. DETAILED DESCRIPTION

[0052] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0053] It should be noted that the personnel information (including but not limited to user personal information, personnel characteristic information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc., such as image acquisition data after each stage of collection, storage, analysis, display, etc.) involved in the present application are all information and data authorized by the user or fully authorized by all parties.

[0054] The personnel roll-call method provided by the embodiments of the present application can be applied to, for example, Figure 1The application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data required by the server 104 to process. The data storage system can be integrated on the server 104, or placed on the cloud or other network servers. The terminal 102 can be through the communication with the server 104, obtain the roll call task information stored on the server 104, and based on the roll call task information, obtain the image acquisition data collected by at least one image acquisition device in the preset area corresponding to the roll call task information; the roll call task information includes preset personnel characteristic information and preset target roll call information; based on the preset image recognition algorithm corresponding to the roll call task information, the image acquisition data is identified, and the personnel characteristic information is obtained; the preset image recognition algorithm includes one or more of face recognition algorithm, gait recognition algorithm and human body recognition algorithm; according to the comparison between the personnel characteristic information and the preset personnel characteristic information, the identification personnel information is determined; the identification personnel information is matched with the preset target roll call information, and the personnel roll call result is determined. Among them, the terminal 102 can be but not limited to various personal computers, notebook computers, smart phones, tablet computers, Internet of Things devices and portable wearable devices, Internet of Things devices can be smart speakers, smart televisions, smart air conditioners, smart vehicle devices, etc. The portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, etc. The server 104 can be realized by an independent server or a server cluster composed of multiple servers.

[0055] In one embodiment, as Figure 2 shown, a personnel roll call method is provided, and the personnel roll call method comprises:

[0056] Step S100, based on the roll call task information, obtaining the image acquisition data collected by at least one image acquisition device in the preset area corresponding to the roll call task information.

[0057] Among them, the roll call task information can be a parameter set of a specific roll call task initiated by a supervision place. The roll call task can be triggered by system setting or manual input. The roll call task information can be obtained by receiving the roll call task instruction, or by accessing the device storing the roll call task information.

[0058] The roll call task information includes preset personnel characteristic information and preset target roll call information. Among them, the preset target roll call information can include a preset area that needs to be rolled call, and can also include the information of the person to be rolled call in the roll call work. Illustratively, the preset target roll call information can include the number of the person to be rolled call, can also include the personnel list of each person to be rolled call, and can also include the identity information of each person to be rolled call.

[0059] The preset personnel feature information can be feature information describing the called personnel in the image. In an exemplary embodiment, in addition to the called personnel, the image acquisition data acquired by the image acquisition device can also acquire images of the supervisor and images of the characters on the printed publications that can appear in the roll call scene. By describing the image features of the called personnel, the above-mentioned content can be effectively distinguished, so as to obtain the identification result of the called personnel.

[0060] In another exemplary embodiment, according to the preset image recognition algorithm corresponding to the roll call task information, the preset personnel feature information includes feature information matching the output format of the preset image recognition algorithm. It can be understood that the output feature formats of different image recognition algorithms can be different. By presetting the same format of feature information, the matchability of the personnel feature information and the preset personnel feature information can be improved.

[0061] The preset area can be a physical area previously delimited in the supervision place. For example, it can be a geofence or the coverage range of a camera. For example, the preset area can be a cell, an activity area, or a corridor.

[0062] The image acquisition device can be a hardware device for capturing images, including but not limited to a visible light camera, an infrared camera, a thermal imaging device, etc. The image acquisition data can be the original visual information output by the image acquisition device, which can include personnel face, body posture, clothing features, and environmental background information, etc.

[0063] The image acquisition data acquired by the image acquisition device can be all or part of the image acquisition data acquired within a certain time range. It can be understood that during the image acquisition process, there can be cases where face recognition fails due to personnel activities, light changes, etc. By image acquisition within a certain time range, accurate results can be obtained by face recognition algorithm for the images with good acquisition quality, so as to achieve introspection of the identification result.

[0064] In this embodiment, the image acquisition data is obtained by dynamically calling existing monitoring devices, rather than using special devices or hardware to deploy the acquired information, which can effectively reduce the additional hardware procurement cost, and at the same time realize the collaborative operation of multiple devices, cover the preset area, and thus improve the comprehensiveness of data acquisition.

[0065] In step S200, the image acquisition data is identified based on the preset image recognition algorithm corresponding to the roll call task information, and personnel feature information is obtained.

[0066] The preset image recognition algorithm can be a pre-configured computer image recognition model, including one or more of a face recognition algorithm, a gait recognition algorithm, and a human body recognition algorithm.

[0067] The face recognition algorithm can be a deep learning-based model for analyzing facial features such as facial contours, feature positions, etc. The gait recognition algorithm can be a time series analysis-based model for extracting human walking posture or limb movement features, for example, a gait sequence analysis model using LSTM, etc. The human body recognition algorithm can be a target detection framework for detecting static features such as human body contours or clothing colors, for example, a YOLO or Faster R-CNN model, etc.

[0068] The personnel feature information can be feature parameters of identifiable individuals extracted from image acquisition data, for example, one or more of face features, gait feature vectors, human body contour sizes, etc.

[0069] Based on the preset image recognition algorithm corresponding to the roll call task information, the image acquisition data is identified, which can be based on the roll call task information to determine the preset image recognition algorithm to be used. In one exemplary embodiment, when the work content of the roll call task is to count the number of people in the preset area, the number of human bodies in the preset area can be identified only by the human body recognition algorithm, thereby obtaining the personnel feature information. In another exemplary embodiment, when the work content of the roll call task is to determine whether the personnel on the list are all in the preset area, the face recognition algorithm, the gait recognition algorithm, and the human body recognition algorithm can be used for comprehensive identification and judgment.

[0070] When multiple preset image recognition algorithms are used to identify personnel feature information, the output features obtained by the multiple algorithms can also be subjected to multi-modal feature fusion or priority sorting, thereby obtaining more comprehensive personnel feature information.

[0071] Step S300, comparing the personnel feature information with the preset personnel feature information to determine the identified personnel information.

[0072] The preset personnel feature information can be personnel feature information pre-stored in the supervision system, which includes face features, gait features, or body features of each person to be called.

[0073] The comparison can be a mathematical or algorithmic measure of the similarity of the features, for example, calculating the Euclidean distance or cosine similarity of the feature vectors.

[0074] Step S400, matching the identified personnel information with the preset target roll call information to determine the personnel roll call result.

[0075] In an exemplary embodiment, the identified personnel information is the number of personnel in the preset area counted through image collection data, the preset target roll-call information includes the number of personnel to be called in the preset area, and the matching manner can be to determine whether the number of personnel in the identified personnel information is consistent with the number of personnel in the preset target roll-call information.

[0076] In another exemplary embodiment, the identified personnel information includes personnel information of personnel in the preset area identified through image collection data, the preset target roll-call information includes a personnel list of each called personnel in the current roll-call task, and can further include corresponding identity information of each called personnel, such as identification information such as number, and the matching can be to determine whether the identity of the personnel in the identified personnel information corresponds to the identity of the called personnel in the preset target roll-call information.

[0077] The personnel roll-call result can be obtained by summarizing and arranging the identified personnel information, the matching condition, and the like of the current roll-call task, to obtain a result for the user to view. Further, an alarm can be given according to the personnel roll-call result, and the personnel roll-call result can be visually processed to improve the automation degree and intuitiveness of the personnel roll-call result.

[0078] The personnel roll-call method provided in this embodiment obtains image collection data collected by at least one image collection device in a preset area corresponding to roll-call task information based on the roll-call task information; the roll-call task information includes preset personnel feature information and preset target roll-call information; personnel feature information is obtained by identifying the image collection data based on a preset image recognition algorithm corresponding to the roll-call task information; the preset image recognition algorithm includes one or more of a face recognition algorithm, a gait recognition algorithm, and a human body recognition algorithm; the identified personnel information is determined by comparing the personnel feature information with the preset personnel feature information; and the personnel roll-call result is determined by matching the identified personnel information with the preset target roll-call information. By integrating existing monitoring devices and algorithms, the hardware cost caused by using special devices for roll-call operations in related technologies is reduced. By using multiple preset image recognition algorithms to cooperatively recognize different personnel features, the accuracy of the recognition can be effectively improved, the roll-call operation time can be shortened, and human errors can be reduced, thereby achieving the effect of reducing the supervision cost while ensuring the accuracy of the roll-call.

[0079] In one embodiment, based on the roll-call task information, the image collection data collected by at least one image collection device in a preset area corresponding to the roll-call task information includes:

[0080] Obtaining real-time available resource information;

[0081] Based on the roll-call task information, determining a target resource requirement corresponding to the roll-call task information;

[0082] If the real-time available resource information meets the target resource requirement, the image acquisition data is acquired;

[0083] If the real-time available resource information does not meet the target resource requirement, the roll call task information is written into the waiting queue, and when the real-time available resource information meets the target resource requirement, the image acquisition data is acquired.

[0084] The real-time available resource information can be a set of system current callable computing, storage or network resource state parameters, which can be obtained by interacting with the system real-time monitoring module. For example, the real-time available resource information can include CPU usage, memory occupancy, remaining storage space or bandwidth availability and the like.

[0085] The target resource requirement can be the minimum resource threshold required to complete the task according to the roll call task information, and the target resource requirement can be generated by analyzing the task parameters through the task analysis module. For example, according to the roll call task, the corresponding preset area and the image recognition algorithm to be used, and the image acquisition data under the preset area, the required CPU computing power, memory capacity and the like can be calculated according to the occupied space, format and the like of the image acquisition data, and the image recognition algorithm used, so as to obtain the target resource requirement.

[0086] The waiting queue can be a temporary storage structure for temporarily storing the roll call task information that is temporarily suspended due to insufficient resources, for managing the task execution order according to priority or time sequence, and for avoiding permanent blocking of tasks caused by resource competition.

[0087] For example, the specific parameters of the roll call task (such as the image recognition algorithm used, the coverage range of the preset area, etc.) can be analyzed to determine the minimum requirement of the required computing, storage and network resources as the target resource requirement. Then, the real-time available resource information is compared with the target resource requirement. If the resources are sufficient, the image acquisition device is triggered to start data acquisition; if the resources are insufficient, the task information is stored in the waiting queue, and the resource state change is continuously monitored. When the system detects that the resources recover to meet the condition, the tasks are selected for execution from the waiting queue according to priority or time sequence, so that the dynamic evaluation of resource availability can be realized, and the failure of task execution or system overload caused by resource competition can be avoided.

[0088] In a specific embodiment, if multiple roll call tasks are initiated at the same time in a supervised place and the current computing resources are tight, the new task will be temporarily stored in the waiting queue, which can be a high-priority task to be processed after the resources are released, so as to improve the system resource utilization and the stability of task execution.

[0089] The personnel roll call method provided by the embodiment comprises the following steps: acquiring real-time available resource information; determining target resource requirements corresponding to the roll call task information based on the roll call task information; if the real-time available resource information meets the target resource requirements, acquiring image acquisition data; if the real-time available resource information does not meet the target resource requirements, writing the roll call task information into a waiting queue, and acquiring the image acquisition data when the real-time available resource information meets the target resource requirements. Through the comparison and decision of the real-time available resource information and the target resource requirements, it is ensured that the task is started only when the resources are sufficient, and the tasks with insufficient resources are temporarily stored in the waiting queue and executed as needed, so as to maintain the balance between the system resource utilization rate and the task execution reliability, thereby effectively avoiding the resource competition and system overload problem in the multi-task concurrent scene, and achieving the effect of reducing the hardware cost while improving the supervision efficiency.

[0090] In one of the embodiments, determining the target resource requirements corresponding to the roll call task information based on the roll call task information comprises:

[0091] Splitting the task based on the roll call task information to obtain a roll call task group; the roll call task group comprises a plurality of task subunits;

[0092] Determining the target resource requirements of each task subunit respectively to determine whether the real-time available resource information meets the target resource requirements of the task subunit.

[0093] The task splitting can be a technical means for decomposing a complex roll call task into subtasks that are related to each other but are executed independently. For example, the task splitting can be achieved by analyzing the size, number and other characteristics of the preset area covered by the task, and analyzing the required image recognition algorithm type and other characteristics. For example, the task splitting can include decomposing a roll call task that requires both face recognition algorithm and gait recognition algorithm into task subunits corresponding to different algorithms.

[0094] The roll call task group can be a set composed of a plurality of task subunits generated after task splitting, which can be directly generated through the task splitting process. For example, for a task of simultaneously calling prisoners in a dormitory and an activity area, two task subunits corresponding to different areas are formed after splitting. The task subunit can be the smallest execution unit after task splitting, which has a clear execution target and resource requirements. For example, a task subunit can be responsible only for face recognition in the dormitory area, and its resource requirements only need to meet the requirements of camera data processing and face recognition algorithm running in the area.

[0095] In an exemplary embodiment, the splitting can be performed according to the complexity of the roll-call task information, and a plurality of task sub-units are generated. For example, when the roll-call task needs to simultaneously call a face recognition algorithm and a gait recognition algorithm, two task sub-units can be split to correspond to the execution requirements of different algorithms. Secondly, for each task sub-unit, the system calculates independent target resource requirements according to its specific requirements. For example, the sub-unit corresponding to the face recognition algorithm can need higher CPU computing power to run a deep learning model, while the task sub-unit corresponding to the gait recognition algorithm can need larger memory space to process time series data. Then, the target resource requirements of each task sub-unit are summarized, so as to form the resource requirements of the whole task, and then it can be judged more accurately whether the real-time available resources meet the task execution conditions.

[0096] The personnel roll-call method provided by the embodiment splits a complex roll-call task into sub-units that can be independently evaluated, calculates specific resource requirements for each sub-unit, and dynamically matches available resources. Through a hierarchical resource requirement evaluation mechanism, the system's resource scheduling capability and task execution reliability in complex task scenarios are enhanced while maintaining the original cost advantage. The resource allocation flexibility and accuracy are improved, multi-sub-unit parallel processing is supported to shorten the task response time, and the effects of improving supervision efficiency and improving resource utilization efficiency are achieved.

[0097] In one of the embodiments, generating the roll-call task information includes:

[0098] Obtaining legal personnel access information in a preset area;

[0099] Based on the legal personnel access information, determining preset target roll-call information in the preset area;

[0100] From the personnel feature information library, obtaining personnel feature information corresponding to the preset target roll-call information as preset personnel feature information;

[0101] Based on the preset personnel feature information and the preset target roll-call information, generating the roll-call task information.

[0102] The legal personnel access information can be dynamic data recording the access time, identity, and location of legal personnel in the preset area, and can be obtained by accessing the access control system, sensors, or monitoring data. It can be understood that the legal personnel access information represents the current personnel record through the relevant access control equipment. In an exemplary embodiment, according to the legal personnel access information of the access control systems of area A and area B, it can be known that personnel A1 has left area A and entered area B, so in the legal personnel access information, personnel A1 should be called out in area B, and if no personnel A1 is found in area B through personnel feature comparison, it means that personnel A1 may have unauthorized activities without being recorded by the access control system.

[0103] The personnel feature information library can be a database storing pre-stored feature parameters of the supervised personnel, which includes face features, gait features, body contours, etc. The feature parameters can be quickly retrieved through database query operations. The preset target calling information can include a list of legal personnel and identity that should currently exist in the preset area. For example, the preset target calling information can be determined based on time window filtering (e.g., personnel entering within the last 3 hours and not leaving) or permission verification (e.g., whether the personnel have permission to enter the area).

[0104] From the personnel feature information library, the personnel feature information corresponding to the preset target calling information is obtained as the preset personnel feature information. The personnel feature information of each supervised personnel can be determined by searching the personnel feature information library according to the identity of each supervised personnel in the preset target calling information, and the preset personnel feature information is obtained by summarizing the personnel feature information of each supervised personnel.

[0105] The personnel calling method provided in this embodiment dynamically obtains legal personnel access information to reflect the personnel status in the preset area in real time, determines the target calling list based on time window filtering and / or permission verification, and accurately matches the feature parameters through the personnel feature information library to finally generate complete calling task information including the area range, target personnel features, and triggering method. This can reduce the risk of missed or false detection caused by static lists, and can achieve information integration only by using existing access control and monitoring system data, thereby reducing the hardware deployment cost and improving the supervision efficiency and resource utilization efficiency.

[0106] In one embodiment, the identified personnel can be identified mainly by a face recognition algorithm, and supplemented by a gait recognition algorithm and a body recognition algorithm. In some other embodiments, other multiple image recognition algorithms can also be used in a main and auxiliary combination recognition scheme.

[0107] According to the comparison between the personnel feature information and the preset personnel feature information, the identified personnel information includes:

[0108] match the first personnel feature information corresponding to the face recognition algorithm with the preset personnel feature information to obtain to-be-determined identification personnel information corresponding to the first personnel feature information and a confidence level;

[0109] If the confidence level meets a preset threshold, the to-be-determined identification personnel information is taken as the identification personnel information.

[0110] If the confidence level does not meet the preset threshold, the to-be-determined identification personnel information is verified based on second personnel feature information corresponding to a gait recognition algorithm and third personnel feature information corresponding to a human body recognition algorithm, and a comprehensive confidence level of the to-be-determined identification personnel information is calculated.

[0111] If the comprehensive confidence level meets the preset threshold, the to-be-determined identification personnel information is taken as the identification personnel information.

[0112] The first personnel feature information can be a set of face biometric feature parameters extracted by the face recognition algorithm, which can be obtained by face detection, key point positioning and feature vectorization processing on an input image. For example, it includes facial contour, facial feature distance, texture feature, etc. The preset personnel feature information can be a biometric feature template library of registered personnel stored in advance, for example, including facial feature vectors, gait feature sequences or human body contour models of different personnel.

[0113] The confidence level can be a numerical index for quantifying the reliability of the matching result of the first personnel feature information, for example, a value between 0 and 1 can be calculated by cosine similarity, Euclidean distance, etc. to represent the similarity between the current matching result and the preset personnel feature. The preset threshold can be a confidence level threshold determined in advance according to prior knowledge or experimental data, or a judgment criterion set by the system according to the identification scene requirement.

[0114] For example, when the confidence level meets the preset threshold, it means that the matching degree between the current first personnel feature information and the preset personnel feature information is high, and it can be determined that the supervised personnel corresponding to the first personnel feature information is the supervised personnel corresponding to the preset personnel feature information. If the confidence level does not meet the preset threshold, it may be due to insufficient accuracy of image acquisition data, insufficient light intensity, or the current state of the supervised personnel cannot be captured effective face features, etc. Then the gait recognition algorithm and the human body recognition algorithm can be enabled for further verification.

[0115] The second personnel feature information corresponding to the gait recognition algorithm can be a dynamic feature parameter extracted based on a human body motion pattern, such as a limb swing frequency, a stride ratio or a posture change trajectory obtained by analyzing consecutive video frames.

[0116] The third personnel feature information corresponding to the human body recognition algorithm can be a parameter set extracted based on static or semi-static features, such as a human body contour shape, clothing color distribution, or height ratio, and the like.

[0117] The comprehensive confidence can be a final determination basis generated by a multi-source feature fusion algorithm. For example, the comprehensive value of 0.66 can be obtained by weighting the confidence of 0.7 of the gait recognition and the confidence of 0.6 of the human body recognition according to the weights of 0.6 and 0.4.

[0118] For example, the gait feature of the to-be-verified target can be extracted first, the gait feature is matched with a sequence in a preset gait feature library to generate a second confidence, the human body contour or clothing feature is compared with a preset human body feature library to generate a third confidence, and the first, second, and third confidences are comprehensively calculated by a preset fusion strategy to form a comprehensive confidence. By introducing complementary feature sources, the dynamic features of the gait or the static features of the human body contour can be used to compensate for the shortcomings of a single algorithm in a face recognition limited scene (such as occlusion or low light), thereby improving the robustness of the recognition result.

[0119] The personnel calling method provided in this embodiment can match the first personnel feature information corresponding to the face recognition algorithm with preset personnel feature information to obtain to-be-determined personnel information corresponding to the first personnel feature information and a confidence; if the confidence meets a preset threshold, the to-be-determined personnel information is used as the recognized personnel information; if the confidence does not meet the preset threshold, the to-be-determined personnel information is verified based on the second personnel feature information corresponding to the gait recognition algorithm and the third personnel feature information corresponding to the human body recognition algorithm, and a comprehensive confidence of the to-be-determined personnel information is calculated; if the comprehensive confidence meets the preset threshold, the to-be-determined personnel information is used as the recognized personnel information. Through the hierarchical verification mechanism, a single algorithm with low calculation cost is preferentially used to complete preliminary judgment, and multi-algorithm collaborative verification is introduced only when necessary to improve the recognition reliability in a complex scene. The calculation efficiency and the recognition accuracy can be balanced, the complementary nature of multi-modal features is used to reduce the risk of misjudgment caused by environmental interference, and the dependence on manual review is reduced, thereby achieving the technical effect of improving the personnel recognition robustness while keeping the hardware cost controllable.

[0120] In one of the embodiments, verifying the to-be-determined personnel information based on the second personnel feature information corresponding to the gait recognition algorithm and the third personnel feature information corresponding to the human body recognition algorithm, and calculating the comprehensive confidence of the to-be-determined personnel information includes:

[0121] The second personnel feature information corresponding to the gait recognition algorithm is compared with the preset personnel feature information to obtain first verification information;

[0122] The third personnel feature information corresponding to the human body recognition algorithm is compared with the preset personnel feature information to obtain second verification information;

[0123] The to-be-identified personnel information is verified based on the first verification information, the second verification information, and a preset confidence weight combination.

[0124] Based on the confidence and the verification result, the comprehensive confidence of the to-be-identified personnel information is calculated.

[0125] The second personnel feature information can be a dynamic feature parameter set extracted by a gait recognition algorithm, such as a gait sequence feature vector, which is obtained by analyzing biological characteristics such as joint motion trajectory and stride period when a human body walks. For example, the second personnel feature information can include gait energy map, step frequency, and other parameters.

[0126] The first verification information can be a combination parameter of gait feature comparison result and confidence, which includes the matching similarity of gait features and the corresponding confidence value. For example, the first verification information can include the Euclidean distance calculation result or the cosine similarity value, and the confidence score based on the similarity threshold.

[0127] The third personnel feature information can be a static feature parameter set extracted by a human body recognition algorithm, such as body contour size, clothing color distribution, and other visual features, which are obtained by image segmentation or color histogram analysis. For example, the third personnel feature information can include body height ratio, clothing dominant tone RGB value, and other parameters.

[0128] The second verification information can be a combination parameter of human body feature comparison result and confidence, which includes the matching similarity of static features and the confidence value. For example, the second verification information can include clothing color matching degree, contour size difference value, and corresponding confidence score.

[0129] The preset confidence weight combination can be a set of confidence weight parameters for different algorithms, which are used to weight and fuse the confidence of different verification information. For example, the preset confidence weight combination can include a gait weight coefficient (such as 0.6) and a human body weight coefficient (such as 0.4), or a dynamically adjusted weight parameter sequence.

[0130] In an exemplary embodiment, after obtaining the first verification information and the second verification information, the confidence of the first verification information and the second verification information can be weighted and summed according to a preset confidence weight, for example, the face confidence x the face weight + the gait confidence x the gait weight + the body confidence x the body weight, to obtain a comprehensive confidence, and then the comprehensive confidence is compared with a preset threshold, and if the condition is met, the recognition result is confirmed to be valid. Through the weighted fusion of multiple source features, the complementary verification of different recognition algorithms can be realized, for example, when the clothing feature is interfered by shielding, the recognition reliability can be compensated by the high weight of the gait feature, so as to reduce the risk of missed detection.

[0131] The personnel roll-call method provided in the embodiment generates verification information by comparing the gait recognition and the body recognition feature information with the preset features respectively, combines the dynamic weight parameters to weight and fuse the multi-source verification results, and finally realizes identity confirmation through confidence threshold judgment. Through flexible configuration of the weight parameters, the weight proportion of different verification information can be dynamically adjusted according to the scene requirements, for example, the body feature weight can be reduced in the clothing changeable scene, the dominant role of the gait feature can be improved, the confidence of the first personnel feature information is complementarily verified by the dynamic features of the gait and the static features of the body, the overall recognition reliability is maintained, and the effect of improving the robustness and accuracy of personnel recognition in complex scenes is achieved.

[0132] In one of the embodiments, the roll-call task information further includes a roll-call mode; the preset target roll-call information corresponds to the roll-call mode;

[0133] The personnel feature information is obtained by identifying the image collection data based on the preset image recognition algorithm corresponding to the roll-call mode.

[0134] The roll-call mode can be a preset execution strategy or task type, and can be a specific requirement and / or priority defined for this roll-call. In some exemplary embodiments, the roll-call mode includes a quick statistics mode, an identity verification mode, a mixed verification mode, etc. The mode parameter can be generated through a task configuration interface or a system preset rule. For example, the quick statistics mode only needs to count the number of people, and the identity verification mode needs to verify the identity in combination with the face or gait features.

[0135] The correspondence between the preset target roll-call information and the roll-call mode can be a data configuration rule dynamically adjusted by the system according to the task requirements, for example, in the identity verification mode, the preset target roll-call information needs to contain the detailed identity feature data (such as face feature vector, gait feature parameter, etc.) of the person to be called, and in the quick statistics mode, only the total number threshold of the preset area is needed.

[0136] The personnel feature information can be a structured data set output by an image recognition algorithm, including but not limited to a personnel quantity statistical result, an identity matching confidence value, or an identity identifier after multi-modal feature fusion.

[0137] Based on the roll call mode, a preset image recognition algorithm corresponding to the roll call mode can be called. The mode parameter in the roll call task information can be analyzed to determine the specific demand roll call mode type, and then the corresponding algorithm combination can be selected from the algorithm library, or the required algorithm result can be selected from the results of multiple algorithms. For example, the fast statistical mode can use a human body recognition algorithm, and the identity verification mode can call a collaborative processing of a human face recognition algorithm and a gait recognition algorithm. Further, if multiple algorithms are involved, the output results can be fused according to a preset rule (such as weighted average or priority sorting), so as to achieve the technical effects of reducing the computing load, improving the recognition efficiency and accuracy by dynamically matching the task demand and the algorithm resource.

[0138] The personnel roll call method provided in the embodiment realizes intelligent adaptation of the image recognition process by establishing a corresponding relationship between the roll call mode parameter and the preset target roll call information and combining a dynamic algorithm calling mechanism, wherein the roll call mode defines a task execution strategy, the preset target roll call information adjusts the data demand according to the mode, and the image recognition algorithm combination is dynamically selected through the mode analysis and algorithm matching steps. The personnel feature information highly matched with the task target is output. Through on-demand allocation of computing resources, multi-algorithm collaborative processing and an extensible mode configuration mode, the technical effects of balancing the recognition efficiency and accuracy in different scenarios, reducing the system energy consumption and enhancing the functional expansibility can be achieved.

[0139] In one of the embodiments, after the identified personnel information is matched with the preset target roll call information to determine the personnel roll call result, the following steps are included:

[0140] The roll call information in the preset target roll call information that cannot be matched with the identified personnel information is determined as the absent personnel information;

[0141] The absent personnel information corresponding to the change clue information is generated;

[0142] The absent personnel information can be a personnel list or an identifier in the preset target roll call information that is not matched in the identified personnel information, and is used to identify the personnel that need to be further tracked. For example, the absent personnel information can be obtained by the system after the matching is completed, by screening the preset target roll call information that is not covered.

[0143] The change clue information can be a set of auxiliary information related to the absent personnel, used to track the dynamic of the absent personnel or analyze the reason for absence. Exemplarily, the change clue information can include historical image acquisition data corresponding to the absent personnel information, last appearance time, last appearance location, etc. The change clue information can be obtained by searching the database, and can be searched according to different information contents, or the historical roll-call result can be queried. The historical image acquisition data can be historical image records related to the absent personnel within the time range of the current roll-call task, and can be obtained by searching the stored image acquisition data. Exemplarily, the historical image acquisition data can include past image frames captured by a surveillance area camera. The last appearance time can be a specific time when the absent personnel was last captured by an image acquisition device, and can be determined by backtracking the timestamp information of the image acquisition data. The last appearance location can be a specific physical area or camera coverage range where the absent personnel last appeared, and can be determined by spatial positioning information of the image acquisition device. Exemplarily, the last appearance location can be a specific area such as a prison corridor or an activity area. Through the above information, a dynamic track of the absent personnel can be automatically generated to assist the supervisor to quickly locate or analyze abnormal situations, such as whether the personnel left the supervision area in advance, whether there is a border crossing behavior, etc.

[0144] The personnel roll-call method provided by the embodiment determines the un-matched preset target roll-call information as absent personnel information, generates change clue information in combination with historical image acquisition data, last appearance time and location, and can provide visual evidence for the supervisor, reduce the search range and reduce the investigation cost. By structurally storing the change clue information, long-term analysis of personnel behavior patterns can be realized, the future missed detection risk can be reduced, and the dependence on manual intervention can be reduced through automatic processing, so that the technical effects of improving the response efficiency of abnormal events, enhancing the safety of supervision and reducing the system expansion cost can be achieved.

[0145] In order to more clearly set forth the technical solutions of the present application, a detailed embodiment is further provided.

[0146] In one embodiment, as shown in Figure 3 A personnel roll-call method is provided, applied to an intelligent roll-call system of multi-modal video analysis, which can fully utilize existing monitoring equipment in prisons to realize efficient personnel counting. The system realizes intelligent management of the prisoners by fusing multiple computer vision technologies without the need for additional deployment of special equipment.

[0147] The system adopts a hierarchical processing architecture. The bottom layer is compatible with various existing monitoring equipment through a video access layer, including fixed cameras, dome cameras and patrol robots, etc. The core of the system adopts an analysis engine of multi-algorithm fusion, which integrates three key technologies of face recognition, gait feature extraction and human body detection.

[0148] Among them, the face recognition module adopts ArcFace algorithm, which optimizes the recognition performance under low light conditions for prison scenes; gait detection is based on spatio-temporal graph convolution network, which effectively identifies the walking characteristics of personnel; human body detection adopts YOLOv5 model to improve the detection accuracy in dense scenes.

[0149] ArcFace (Additive Angular Margin Loss) algorithm is a deep learning loss function for face recognition. It introduces an angular margin based on the Softmax loss function, making the face features more separable in the angle space, thereby improving the recognition accuracy.

[0150] Spatio-temporal graph convolution network is a deep learning model specifically designed for processing spatio-temporal graph data. It combines graph convolution network (GCN) and temporal convolution module, which can capture both spatial dependencies and temporal dynamics in data.

[0151] YOLOv5 model is a single-stage object detection algorithm that converts the object detection task into an end-to-end regression problem, directly predicting the bounding box and class of the target on the image. Unlike traditional two-stage detection methods such as R-CNN series, YOLO divides the entire image into grids, and each grid unit is responsible for predicting multiple bounding boxes and their confidence, as well as outputting class probabilities. This design allows YOLO to process images at an extremely fast speed, making it suitable for real-time detection scenarios.

[0152] The personnel roll call system supports multiple roll call modes, such as using timed and fixed face recognition mode in the labor workshop; combining gait recognition and human body detection to achieve dynamic counting in activity areas; and completing fast population statistics through human body detection in the dormitory corridor. The specially designed abnormal handling mechanism can automatically identify special situations such as occlusion and facing away from the camera, and improve the accuracy of roll call through multi-camera data fusion.

[0153] When the system starts the roll call task, it locks the target personnel list data, establishes complete roll call records in ElasticSearch, and caches key information to Redis to support real-time query. The system then accesses the specified monitoring video source and starts the intelligent analysis module to enter the fast frame mode.

[0154] ElasticSearch is an open-source distributed search and analysis engine based on Apache Lucene, which can provide near-real-time full-text search, structured retrieval, and aggregation analysis functions, suitable for log analysis, monitoring systems, enterprise search, and other scenarios.

[0155] The video analysis process integrates three recognition techniques: a face detection module extracts face images from the video stream, which are transmitted to a real-time processing engine via a Kafka message queue for 1:N identity comparison; a gait analysis module extracts features from the walking posture of personnel in the monitored area; and a human body detection module uses an improved YOLO algorithm to locate personnel and identify basic attributes. The three recognition techniques complement each other through a decision-level fusion strategy, with face recognition as the primary verification method, gait features for auxiliary verification, and human body detection for basic personnel presence confirmation.

[0156] The real-time processing engine analyzes the multi-modal recognition results received: first, it matches and verifies against a preset list, and then updates the personnel status in Redis; for targets that cannot be confirmed by face recognition, the system automatically calls the gait features from the last 5 minutes for secondary verification; and through human body detection, it ensures the accuracy of regional population statistics. The front-end interface obtains state updates in real-time through long connections and visually displays the roll call progress.

[0157] At the end of the task, the system automatically releases video resources, synchronizes the final state to ElasticSearch, and generates a roll call report containing multi-dimensional data analysis. For absent personnel, the system automatically assesses the risk level based on historical behavior data and automatically triggers a graded warning mechanism for key personnel. The entire process makes full use of existing monitoring equipment and significantly improves recognition accuracy and system robustness, especially in complex scenarios such as insufficient light or face occlusion.

[0158] To address situations such as a large number of roll call tasks and limited available resources, the personnel roll call system uses a state machine model for task management, with each state transition following strict business logic, as shown in FIG. 1. Figure 4 When a new roll call task is initiated, the system first performs a resource pre-check: if the current analysis resources are sufficient, it immediately enters the execution state and starts the video stream analysis process; if the resources are insufficient, it enters the waiting queue and automatically activates the task when resources are released through a resource monitoring mechanism. During task execution, dynamic adjustment is supported: when the personnel list needs to be changed, the system can temporarily suspend the task and enter the pause state, maintaining the current progress and releasing computing resources during this period, and then resuming execution after the adjustment is complete.

[0159] Task termination includes two triggering methods: supervisor initiatively terminates or system automatically ends according to preset time. After task termination, it enters the result judgment stage, at this time the system retains all process data but releases the analysis resources, and after the supervisor completes the absence reason analysis and confirms, the task party is formally completed. It is worth noting that the video analysis module is only activated when the task is in the running state, and the video processing resources are immediately released in other states. This design ensures the dynamic optimization allocation of system resources. Each state transition generates detailed operation logs, providing complete traceability basis for subsequent audit.

[0160] Further, the personnel roll call system adopts a regional roll call mode and realizes dynamic personnel management through deep integration with the regional control system. The regional control module records personnel legal access information in real time through an intelligent access head system or video monitoring equipment, and constructs a dynamically updated on-site personnel database. When starting regional roll call, the system automatically obtains the latest personnel whitelist in the region, and links the video monitoring equipment deployed in the region for intelligent verification.

[0161] This mechanism has a double-checking feature: for legal off-site personnel, the regional control system automatically removes them from the current regional list to avoid invalid roll call; for abnormal off-site situations (such as unauthorized off-site), the system maintains the status of the personnel in the list, and the video analysis module confirms their actual absence through continuous face recognition and body detection. This design realizes the business effect of "automatic reduction for legal off-site, and warning for abnormal off-site".

[0162] The personnel roll call system also adopts an event-driven architecture design, and when the access control system detects personnel access, it pushes the list change event to the roll call service in real time. The video analysis module performs targeted identification based on the updated list, greatly improving identification efficiency. For personnel marked as abnormal absence, the system automatically generates disposal suggestions, including associated video playback, last appearance time and location, etc., to assist supervisors in rapid judgment. The whole process realizes the intelligent upgrade from static roll call to dynamic control, which not only guarantees the rigor of supervision, but also avoids unnecessary repeated verification.

[0163] Further, in order to realize the optimization of resource allocation, the personnel roll call system also adopts a hierarchical task scheduling mechanism, which realizes efficient use of resources through hierarchical design of task groups and subtasks. In terms of resource allocation strategy, the system splits each task group into multiple independent subtask units, forming a fine-grained resource management system. When the system starts the roll call task, the resource manager will evaluate the current available computing power in real time, and allocate resources dynamically in the smallest unit of subtasks: subtasks that meet resource requirements immediately enter the execution state, and those that do not enter the standby queue.

[0164] The queuing mechanism employs a two-tiered priority design: first, the priority order of the task groups is followed, and then resources are allocated to subtasks based on this priority order. As shown in the figure, even if the resources required for subtask b in task group B are currently available, if the higher-priority task group A still has subtasks to process, the system will prioritize resources for subtasks in task group A. This design ensures that important tasks are prioritized while fully utilizing resources.

[0165] Task lifecycle management utilizes a centralized control model, with the task group centrally managing the status of all subtasks. The operational status of a task group directly impacts the execution of its subtasks: Subtasks of an inactive task group remain dormant; a running task group activates all subtasks and places them into the resource queue. When a task group is paused or terminated, all associated subtasks are simultaneously paused or have their resources released. Ultimately, the system aggregates the execution results of all subtasks and generates a comprehensive task group roll call report for supervisory review. This architectural design maximizes the efficiency of limited analytical resources while ensuring business integrity.

[0166] In this embodiment, the personnel roll-call system uses multimodal fusion recognition technology to achieve high-precision personnel roll-call, and ensures recognition reliability in different scenarios through a three-level recognition system, such as Figure 5 As shown in the figure, the system first uses facial recognition as its core verification method: the video analysis module extracts facial features from the surveillance footage and performs a 1:N real-time comparison against a pre-built personnel database. A successful match is directly marked as present. To improve recognition rates, the system uses a dynamic quality assessment algorithm, automatically selecting frames with optimal lighting conditions for recognition and enhancing low-quality images.

[0167] When the confidence level of a facial recognition result is low, it is considered difficult to identify a person (e.g., when the face is obscured or facing away from the camera), and the system automatically activates a supplementary recognition solution. The gait recognition module analyzes the spatiotemporal characteristics of a person's walking movements, using a pre-built gait feature library for auxiliary verification. This system utilizes a temporal attention mechanism to effectively capture key posture features during walking, maintaining high recognition accuracy even in crowded scenarios. Simultaneously, the human detection module continuously verifies the presence of a person based on visual features such as clothing color and body shape, cross-verifying this information with the access control system's entry and exit records to prevent missed detections.

[0168] The three recognition technologies complement each other through decision-level fusion: the face recognition result has the highest confidence weight, and when the confidence is insufficient, the system will combine the results of gait recognition and human detection for joint judgment. All recognition processes use real-time stream processing architecture, and data flow between modules is realized through Kafka message queue. The system also designs an exception handling mechanism to automatically trigger a review process for contradictory recognition situations, including calling multi-angle camera data or extending the analysis time, etc., to ensure the accuracy of the roll call results. The final comprehensive report will mark the recognition method and confidence score of each person, providing a comprehensive reference for regulatory decision-making.

[0169] The personnel roll call method provided by the embodiment realizes low-cost deployment by reusing existing cameras in the supervision place, without the need for additional installation of special equipment, and realizes fast roll call and intelligent research and judgment method. The system uses a dynamic fast frame extraction algorithm to intelligently frame the video stream and accurately identify the supervised personnel, achieving the effect of fast roll call. After the roll call is completed, the system automatically generates a structured report, including attendance statistics, details of absent personnel, and recognition confidence scores, supporting visual display and export. For the absence of high-risk personnel, the system automatically assesses the risk level in combination with historical behavior data, triggers a graded alarm mechanism, and correlates and calls related monitoring videos and access control records, providing complete disposal basis for supervisors. This scheme optimizes the utilization rate of existing resources and the intelligent research and judgment process, significantly reducing the system deployment and maintenance cost while ensuring the accuracy of the roll call; the collaborative verification mechanism of face recognition as the main method, gait recognition and human detection as the auxiliary method solves the limitations of single recognition technology in complex scenes (such as occlusion, backlight, low resolution, etc.), and improves the accuracy of roll call. Dynamic quality evaluation and enhancement: automatically select the best face frame and enhance low-quality images to improve recognition success rate. Decision-level fusion strategy: combine the confidence of face, gait, and human detection for comprehensive judgment to ensure high-precision recognition; task group-subtask hierarchical scheduling is used to divide the roll call task into multiple subtasks, dynamically allocate computing resources according to priority, and improve the utilization rate of computing power. Real-time resource preemption and release: resources are released immediately after the completion of subtasks for use by other tasks, avoiding waste of computing power. Task priority-based queuing mechanism: subtasks of high-priority task groups are executed first to ensure that critical roll call tasks are completed on time; through dynamic area roll call management, the system is deeply integrated with the access control system to automatically update the area personnel list and reduce invalid roll call. Automatic early warning for abnormal departure: keep the roll call state for unregistered departing personnel to ensure that there is no oversight in supervision. Multi-camera collaborative analysis: combining monitoring data from different angles can improve recognition coverage and accuracy.

[0170] It should be understood that although the steps in the flowcharts involved in the embodiments described above are shown in sequence according to the arrows, the steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, the execution of the steps is not strictly limited in sequence, and the steps can be executed in other orders. Moreover, at least some of the steps in the flowcharts involved in the embodiments described above can include multiple steps or multiple stages, which are not necessarily executed at the same time but can be executed at different times, and the execution of the steps or stages is not necessarily sequential but can be performed alternately or alternately with at least part of other steps or steps or stages in other steps.

[0171] Based on the same inventive concept, the embodiments of the present application also provide a personnel roll-call system for implementing the personnel roll-call method described above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more personnel roll-call system embodiments provided below can refer to the limitations of the personnel roll-call method described above, which will not be repeated here.

[0172] In one embodiment, as shown in Figure 6 A personnel roll-call system is provided, which includes a task publishing component 100, an image recognition component 300, a personnel recognition component 200, and at least one image acquisition device 400, wherein:

[0173] The task publishing component 100 is configured to store roll-call task information, wherein the roll-call task information includes preset personnel feature information and preset target roll-call information.

[0174] The image recognition component 300 is configured to acquire image acquisition data collected by the at least one image acquisition device 400 in a preset area corresponding to the roll-call task information, and identify the image acquisition data based on a preset image recognition algorithm corresponding to the roll-call task information to obtain personnel feature information, wherein the preset image recognition algorithm includes one or more of a face recognition algorithm, a gait recognition algorithm, and a human body recognition algorithm.

[0175] The personnel recognition component 200 is configured to compare the personnel feature information with the preset personnel feature information to determine recognition personnel information, and match the recognition personnel information with the preset target roll-call information to determine a personnel roll-call result.

[0176] In one embodiment, the personnel roll-call system further includes a message queue component, and the image recognition component 300 is further configured to send the personnel feature information to the message queue component, and the message queue component is configured to send the personnel feature information to the personnel recognition component 200.

[0177] The task publishing component 100 can be directly or indirectly connected in communication with one or more of the personnel identification component 200 and the image identification component 300. In one specific embodiment, the task publishing component 100 can publish a task in a message middleware, and the personnel identification component 200 or the image identification component 300 reads the message middleware. When an executable roll-call task occurs, the personnel identification component 200 and the image identification component 300 negotiate to start a roll-call task process.

[0178] In one embodiment, the task publishing component 100 is further configured to obtain real-time available resource information, determine a target resource requirement corresponding to the roll-call task information based on the roll-call task information, control the image identification component 300 to obtain the image collection data directly or indirectly if the real-time available resource information meets the target resource requirement, and write the roll-call task information into a waiting queue and control the image identification component 300 to obtain the image collection data directly or indirectly when the real-time available resource information meets the target resource requirement if the real-time available resource information does not meet the target resource requirement.

[0179] In one embodiment, the task publishing component 100 is further configured to split the roll-call task based on the roll-call task information to obtain a roll-call task group, the roll-call task group including a plurality of task subunits, and determine a target resource requirement of each of the task subunits to determine whether the real-time available resource information meets the target resource requirement of the task subunit.

[0180] In one embodiment, the task publishing component 100 is further configured to obtain legal personnel access information in a preset area, determine preset target roll-call information in the preset area based on the legal personnel access information, obtain personnel feature information corresponding to the preset target roll-call information from a personnel feature information library as preset personnel feature information, and generate the roll-call task information based on the preset personnel feature information and the preset target roll-call information.

[0181] In one embodiment, the personnel identification component 200 is further configured to match first personnel feature information corresponding to the face recognition algorithm with preset personnel feature information to obtain pending identification personnel information and a confidence level corresponding to the first personnel feature information, take the pending identification personnel information as identification personnel information if the confidence level meets a preset threshold, and verify the pending identification personnel information based on second personnel feature information corresponding to the gait recognition algorithm and third personnel feature information corresponding to the body recognition algorithm to calculate a comprehensive confidence level of the pending identification personnel information if the confidence level does not meet the preset threshold.

[0182] In one of the embodiments, the personnel identification component 200 is further configured to: compare the second personnel feature information corresponding to the gait recognition algorithm with the preset personnel feature information to obtain first verification information; compare the third personnel feature information corresponding to the human body recognition algorithm with the preset personnel feature information to obtain second verification information; verify the to-be-identified personnel information based on the first verification information, the second verification information, and a preset confidence weight combination; and calculate a comprehensive confidence of the to-be-identified personnel information based on the confidence and the verification result.

[0183] The personnel identification component 200 is further configured to: the roll-call task information further includes a roll-call mode; and the preset target roll-call information corresponds to the roll-call mode. The personnel identification component 200 is further configured to: based on the roll-call mode, call a preset image recognition algorithm corresponding to the roll-call mode to identify the image collection data and obtain the personnel feature information.

[0184] In one of the embodiments, the personnel roll-call system further includes a result display component configured to: determine, as an absent personnel information, roll-call information in the preset target roll-call information that cannot be matched with the identified personnel information; and generate change clue information corresponding to the absent personnel information. The change clue information includes historical image collection data, a last appearance time, and a last appearance location corresponding to the absent personnel information.

[0185] The above-mentioned modules in the personnel roll-call system can be all or partially implemented by software, hardware, and a combination thereof. The above-mentioned modules can be embedded in or independent of a processor in a computer device in a hardware form, or can be stored in a memory in a computer device in a software form, so as to be called and executed by a processor to perform operations corresponding to the above-mentioned modules.

[0186] In one embodiment, the terminal 102 can be, for example, a mobile phone, a tablet computer, a wearable device, or the like. Figure 7As shown in one of the computer devices. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium, an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner, and the wireless manner can be achieved through WIFI, mobile cellular network, NFC (near field communication) or other technologies. The computer program is executed by the processor to implement a personnel roll-call method. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0187] Those skilled in the art can understand that, Figure 7 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0188] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0189] Any combination of the technical features of the above embodiments can be made. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combination of the technical features does not exist, it should be considered as the scope of the present application.

[0190] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent of the present application. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of protection of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A personnel roll-call method, characterized in that: The personnel roll call method includes: Based on the roll call task information, image acquisition data acquired by at least one image acquisition device in a preset area corresponding to the roll call task information is acquired; the roll call task information includes preset personnel feature information and preset target roll call information; Based on a preset image recognition algorithm corresponding to the roll-call task information, the image acquisition data is recognized to obtain personnel feature information; the preset image recognition algorithm includes one or more of a face recognition algorithm, a gait recognition algorithm, and a body recognition algorithm; Determine the identification information of the person by comparing the person's characteristic information with the preset person's characteristic information; The identified personnel information is matched with the preset target roll call information to determine the personnel roll call result.

2. The personnel roll-call method according to claim 1, characterized in that: The acquiring, based on the roll-call task information, image acquisition data acquired by at least one image acquisition device in a preset area corresponding to the roll-call task information comprises: Get real-time information on available resources; Based on the roll-call task information, determining target resource requirements corresponding to the roll-call task information; If the real-time available resource information meets the target resource requirement, acquiring the image acquisition data; If the real-time available resource information does not meet the target resource requirement, the roll call task information is written into a waiting queue, and when the real-time available resource information meets the target resource requirement, the image acquisition data is acquired.

3. The personnel roll-call method according to claim 2, characterized in that: The determining, based on the roll-call task information, target resource requirements corresponding to the roll-call task information includes: Performing task splitting based on the roll-call task information to obtain a roll-call task group; the roll-call task group includes a plurality of task sub-units; The target resource requirement of each task subunit is determined respectively to judge whether the real-time available resource information meets the target resource requirement of the task subunit.

4. The personnel roll-call method according to claim 1, characterized in that: Generating the roll call task information includes: Obtain entry and exit information of legal personnel in the preset area; Based on the legal personnel entry and exit information, determining the preset target name information within the preset area; Acquire, from a personnel characteristic information database, personnel characteristic information corresponding to the preset target roll call information as preset personnel characteristic information; The roll call task information is generated based on the preset personnel feature information and the preset target roll call information.

5. The personnel roll-call method according to claim 1, characterized in that: The determining of the identification information of a person according to the comparison of the person characteristic information with the preset person characteristic information includes: Matching the first person feature information corresponding to the face recognition algorithm with the preset person feature information to obtain the pending identified person information and confidence level corresponding to the first person feature information; If the confidence level meets a preset threshold, the pending identification person information is used as the identification person information; If the confidence level does not meet the preset threshold, verifying the pending person information based on the second person feature information corresponding to the gait recognition algorithm and the third person feature information corresponding to the body recognition algorithm, and calculating the comprehensive confidence level of the pending person information; If the comprehensive confidence level meets a preset threshold, the pending identification person information is used as the identification person information.

6. The personnel roll-call method according to claim 5, characterized in that: The verifying the to-be-identified person information based on the second person feature information corresponding to the gait recognition algorithm and the third person feature information corresponding to the human body recognition algorithm and calculating the comprehensive confidence of the to-be-identified person information includes: Comparing the second person characteristic information corresponding to the gait recognition algorithm with the preset person characteristic information to obtain first verification information; Comparing the third person characteristic information corresponding to the human body recognition algorithm with the preset person characteristic information to obtain second verification information; Verifying the to-be-identified person information based on the first verification information, the second verification information, and a preset confidence weight combination; Based on the confidence level and the verification result, the comprehensive confidence level of the to-be-identified person information is calculated.

7. The personnel roll-call method according to claim 1, characterized in that: The roll-call task information also includes a roll-call mode; the preset target roll-call information corresponds to the roll-call mode; The identifying of the image acquisition data based on a preset image recognition algorithm corresponding to the roll call task information to obtain personnel characteristic information includes: based on the roll call pattern, calling a preset image recognition algorithm corresponding to the roll call pattern, identifying the image acquisition data to obtain personnel characteristic information.

8. The personnel roll-call method according to claim 1, characterized in that: The step of matching the identified person information with the preset target roll call information and determining the person roll call result includes: Determining the roll call information in the preset target roll call information that cannot be matched with the identified person information as absent person information; Generate change clue information corresponding to the absent person information; the change clue information includes historical image acquisition data, last appearance time and last appearance location corresponding to the absent person information.

9. A personnel roll call system, characterized in that: The personnel roll-call system includes a task issuing component, an image recognition component, a personnel identification component, and at least one image acquisition device, wherein: The task publishing component is used to store roll call task information; the roll call task information includes preset personnel feature information and preset target roll call information; The image recognition component is configured to obtain image data collected by at least one image acquisition device within a preset area corresponding to the roll call task information; identify the image data based on a preset image recognition algorithm corresponding to the roll call task information to obtain person feature information; the preset image recognition algorithm includes one or more of a face recognition algorithm, a gait recognition algorithm, and a body recognition algorithm; The personnel identification component is used to compare the personnel characteristic information with preset personnel characteristic information to determine the identified personnel information; and match the identified personnel information with preset target roll call information to determine the personnel roll call result.

10. The personnel roll-call system according to claim 9, characterized in that: The personnel roll-call system further includes a message queue component, and the image recognition component is further configured to send the personnel feature information to the message queue component; the message queue component is configured to send the personnel feature information to the personnel recognition component.

Images