Intelligent security management system and method based on image recognition

Abstract

The invention provides a smart security management system and method based on image recognition, and relates to the technical field of smart communities. According to the invention, through setting the public security identification model, the real-time image of each camera in the smart community is preliminarily detected, the suspected image possibly having public security harming behaviors is obtained, and rapid identification of public security harming behaviors is realized. Furthermore, a target identification model matched with the public security type and the environmental parameters is used for carrying out secondary identification on the suspected image and the video data of the suspected image, and an identification result is determined, so that the target identification model can more accurately identify the cause of the event harming the public security behavior and carry out responsibility judgment; accurate determination of public security harming behaviors is realized. Compared with a manual alarm and policeman on-site processing mode, preliminary identification and secondary identification are carried out through an image identification technology, behaviors harming the public security can be automatically, rapidly and accurately identified, and the security management efficiency of the smart community is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of smart communities, and in particular to a smart security management system and method based on image recognition. Background Art

[0002] With the accelerated pace of urbanization and the continuous deepening of smart community construction, public safety prevention and management has become an indispensable core element in community governance. However, the traditional security management model, which relies too much on manual monitoring and patrolling, is not only inefficient, but also has obvious limitations in monitoring coverage, making it difficult to capture and effectively respond to potential public safety risks in a timely manner.

[0003] In current security practice, once a security incident occurs, it often relies on the traditional method of manual alarm and on-site handling by police officers. This handling method not only significantly prolongs the handling time of the incident, but also greatly reduces the handling efficiency, making it difficult to eliminate public safety hazards in a timely and effective manner, and posing a considerable threat to community safety.

[0004] What is more serious is that when the police arrive at the scene, in order to fully understand the situation and make an accurate determination of responsibility, they need to conduct tedious inquiries and mediation, and return to the monitoring center to manually retrieve relevant surveillance videos for detailed manual analysis. This process is not only time-consuming and laborious, but also further prolongs the processing cycle, seriously restricting the improvement of processing efficiency.

[0005] To sum up, the current public security prevention and management faces prominent problems such as low efficiency and long processing cycle, which makes it difficult to meet the urgent needs of modern smart communities for public security prevention and management. Summary of the invention

[0006] The present invention provides an intelligent security management system and method based on image recognition, which can automatically realize the rapid identification and accurate judgment of behaviors endangering public security, and improve the security management efficiency of smart communities.

[0007] In a first aspect, the present invention provides a smart security management method based on image recognition, the method comprising: acquiring real-time images of multiple cameras in a smart community; determining a suspected image and a public security type corresponding to the suspected image based on the real-time image and a preset public security recognition model, the public security recognition model being used to identify suspected images that may involve acts that endanger public security; public security types include fire, criminal, public security and traffic; acquiring video data and environmental parameters corresponding to the suspected image; performing model matching based on the public security type and environmental parameters to determine a target recognition model; determining a recognition result of the suspected image based on the suspected image, the video data, and the target recognition model, the recognition result including an event cause and a responsibility determination result; and performing smart security management based on the recognition result of the suspected image.

[0008] In one possible implementation, based on real-time images and a preset public security identification model, a suspect image and a public security type corresponding to the suspect image are determined, including: based on the real-time image, feature conversion is performed to determine a feature vector of each real-time image; based on the feature vector of each real-time image and a preset public security identification model, an output result corresponding to each real-time image is determined, the output result including the probability that each type of public security type exists in the real-time image; based on the output result corresponding to each real-time image and a preset probability threshold, a suspect image is determined; based on the output result corresponding to each suspect image, a public security type corresponding to each suspect image is determined.

[0009] In a possible implementation, model matching is performed based on the public security type and environmental parameters to determine the target recognition model, including: determining a sub-model related to the public security type from multiple sub-models of the image recognition model based on the public security type and the mapping relationship between the public security type and the model type; performing association analysis based on the public security type and environmental parameters to determine the environmental conditions related to the public security type; the environmental parameters include temperature, humidity, light and pedestrian flow; determining model parameters of the sub-model related to the public security type based on the environmental conditions related to the public security type; and determining the target recognition model based on the sub-model related to the public security type and the model parameters.

[0010] In one possible implementation, the recognition result of the suspect image is determined based on the suspect image, video data, and a target recognition model, including: performing frame processing on the video data to obtain a frame image sequence arranged in chronological order; performing feature extraction based on the suspect image and the frame image sequence to determine a feature vector to be recognized; and inputting the feature vector to be recognized into the target recognition model to obtain the recognition result of the suspect image.

[0011] In one possible implementation, intelligent security management is performed based on the recognition result, including: if the recognition type is criminal or fire, the event data of the suspect image is stored and a first alarm message is generated; the first alarm message is used to instruct the central user to confirm the criminal incident or the fire incident; if the recognition type is public security or traffic, the event data of the suspect image is stored and a second alarm message is generated, the second alarm message is used to instruct the central user to evaluate the impact scope of the public security incident or the traffic incident; the evaluation result input by the central user is received; the evaluation result includes a first-level impact scope or a second-level impact scope; if the evaluation result is a first-level impact scope, an event processing instruction is generated, and the event processing instruction is used to instruct the patrol user to conduct on-site mediation; the event processing instruction is sent to the patrol user's processing terminal.

[0012] In one possible implementation, intelligent security management is performed based on the recognition results, including: responding to the screening operation instructions of the patrol user on the intelligent security terminal, based on the screening operation instructions, screening multiple suspect images stored in the database to obtain one or more alternative images; the screening operation instructions include the query location and query time period selected by the patrol user; displaying one or more alternative images; responding to the patrol user's selection instructions for one or more alternative images; based on the selection instructions, querying the database to obtain event data of the target suspect image, the event data including the target suspect image, video data, public security type, environmental parameters and recognition results; the selection instructions include the target suspect image selected by the patrol user; and displaying the event data of the target suspect image.

[0013] In a possible implementation, the method also includes: receiving a target query instruction from a patrol user, the target query instruction being used to instruct a query on a target person or a target object; the target query instruction including a real-time image of the target person or the target object; based on the target query instruction, querying the target person or the target object to obtain a query report; and displaying the query report.

[0014] In one possible implementation, a target person or a target object is queried based on a target query instruction to obtain a query report, including: extracting features of the target person or the target object based on the target query instruction; performing model matching based on the target query instruction to determine a target query model; determining a query range and multiple cameras within the query range based on a real-time image of the target person or the target object; acquiring real-time images of multiple cameras within the query range; performing traversal matching based on the real-time images of multiple cameras within the query range, features of the target person or the target object, and the target query model to determine a query result, the query result including multiple target real-time images related to the target person or the target object; drawing a trajectory map of the target person or the target object based on camera positions corresponding to the multiple target real-time images and a map of the query range; and generating a query report based on the query result and the trajectory map.

[0015] In a possible implementation, before determining the suspect image and the public security type corresponding to the suspect image based on the real-time image and the preset public security identification model, it also includes: obtaining typical images of various types of public security events that occurred in the historical period, and the public security type corresponding to each typical image; performing feature conversion based on the typical images of the various types of public security events to obtain feature vectors of each typical image; generating multiple first training samples using the feature vectors of each typical image as input and the public security type corresponding to each typical image as output; and performing neural network training based on the multiple first training samples to obtain a public security identification model.

[0016] In a possible implementation, based on the public security type and environmental parameters, model matching is performed, and before the target recognition model is determined, the following is also included: obtaining the processing result data of various events that occurred in the historical period, the processing result data including real-time images, video data, public security types, environmental parameters, event causes and responsibility determination results of various events handled by patrol users during the patrol process; performing cluster analysis based on the public security type and environmental parameters to determine the multiple environmental conditions corresponding to each public security type, wherein any environmental condition is a combination of one or more environmental parameters; based on the public security type and environmental parameters, the processing result data of various events are split to obtain the processing result data corresponding to various environmental conditions under each public security type; for each environmental condition under any public security type, based on the environmental condition under the public security type, Based on the real-time image and video data corresponding to the environmental condition, a plurality of input feature vectors corresponding to the environmental condition under the public security type are generated; based on the event cause and responsibility determination results corresponding to the environmental condition under the public security type, a plurality of output feature vectors corresponding to the environmental condition under the public security type are generated; based on the plurality of input feature vectors and the plurality of output feature vectors, a plurality of second training samples are generated; based on the plurality of second training samples, neural network training is performed to obtain sub-models and model parameters corresponding to the environmental condition under the public security type; based on the processing result data corresponding to various environmental conditions under each public security type, neural network training is performed to obtain sub-models and model parameters corresponding to various environmental conditions under each public security type; based on the sub-models and model parameters corresponding to various environmental conditions under each public security type, an image recognition model is obtained by fusing.

[0017] In a possible implementation, the method also includes: obtaining processing result data of the smart security terminal in the historical period before the current moment, the processing result data including real-time images, video data, public security types, environmental parameters, event causes and responsibility determination results of various events handled by patrol users during the patrol process; based on the processing result data, generating multiple first training samples and multiple second training samples; based on the multiple first training samples, updating the public safety recognition model to obtain an updated public safety recognition model; based on the multiple second training samples, updating each sub-model in the image recognition model to obtain an updated image recognition model; based on the updated public safety recognition model and image recognition model, real-time monitoring of real-time images of the smart community is performed.

[0018] In the second aspect, an embodiment of the present invention provides a smart security management device based on image recognition, the device comprising: a communication module and a processing module, the communication module is used to obtain real-time images of multiple cameras in a smart community. The processing module is used to determine the suspect image and the public security type corresponding to the suspect image based on the real-time image and a preset public security recognition model, the public security recognition model is used to identify suspect images that may have behaviors that endanger public security; public security types include fire, criminal, public security and traffic. The communication module is also used to obtain video data and environmental parameters corresponding to the suspect image; the processing module is also used to perform model matching based on the public security type and environmental parameters to determine the target recognition model; based on the suspect image, video data, and target recognition model, determine the recognition result of the suspect image, the recognition result includes the cause of the event and the responsibility determination result; based on the recognition result of the suspect image, perform smart security management.

[0019] In a third aspect, an embodiment of the present invention provides an intelligent security management system based on image recognition, the system comprising an electronic device, the electronic device comprising a memory and a processor, the memory storing a computer program, the processor being used to call and run the computer program stored in the memory to execute the steps of the method described in the first aspect and any possible implementation method of the first aspect.

[0020] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and wherein when the computer program is executed by a processor, the steps of the method described in the first aspect and any possible implementation method of the first aspect are implemented.

[0021] The present invention provides a smart security management system and method based on image recognition. The present invention sets a public security recognition model to perform preliminary detection on the real-time images of each camera in the smart community, obtains suspected images that may contain behaviors endangering public security, and realizes the rapid identification of behaviors endangering public security. Furthermore, the present invention uses a target recognition model that matches the public security type and environmental parameters to perform secondary recognition on the suspected image and its video data, and determines the recognition result, so that the target recognition model can more accurately identify the cause of the incident endangering public security and make a responsibility determination, thereby realizing an accurate determination of behaviors endangering public security. Compared with manual alarms and on-site handling by police officers, the present invention uses image recognition technology to perform preliminary recognition and secondary recognition, which can automatically, quickly and accurately identify behaviors endangering public security, thereby improving the security management efficiency of smart communities. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0023] Figure 1 It is a flowchart of a smart security management method based on image recognition provided by an embodiment of the present invention; Figure 2 It is a structural schematic diagram of an intelligent security management device based on image recognition provided by an embodiment of the present invention; Figure 3 It is a structural schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0024] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present invention.

[0025] In the description of the present invention, unless otherwise specified, " / " means "or", for example, A / B can mean A or B. "And / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, "at least one" and "plurality" refer to two or more. The words "first", "second", etc. do not limit the quantity and execution order, and the words "first", "second", etc. do not limit them to be different.

[0026] In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a concrete way for easy understanding.

[0027] In addition, the terms "including" and "having" and any variations thereof mentioned in the description of this application are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or modules is not limited to the listed steps or modules, but may optionally include other steps or modules that are not listed, or may optionally include other steps or modules that are inherent to these processes, methods, products or devices.

[0028] In order to make the purpose, technical solutions and advantages of the present invention more clear, the following will be described through specific embodiments in conjunction with the accompanying drawings of the present invention.

[0029] As described in the background technology, the current public security prevention management in smart communities has technical problems such as low processing efficiency and long processing cycle.

[0030] To solve the above technical problems, Figure 1 As shown, an embodiment of the present invention provides a smart security management method based on image recognition. The method includes steps S101-S105.

[0031] S101. Acquire real-time images from multiple cameras in a smart community.

[0032] In some embodiments, the camera can transmit the real-time image to the control center via a wired or wireless network to ensure network stability and avoid image delay or loss. The control center receives the real-time image from each camera and performs preliminary processing, such as format conversion and clarity optimization.

[0033] S102: Based on the real-time image and a preset public security recognition model, determine the suspected image and the public security type corresponding to the suspected image.

[0034] In the embodiment of the present application, the public security recognition model is used to identify suspicious images that may contain behaviors that endanger public security. Public security types include fire protection, criminal, public security and traffic.

[0035] In some embodiments, the public security identification model is obtained by training a neural network based on typical images of various types of public security incidents that occurred in historical periods.

[0036] As a possible implementation manner, step S102 may be specifically implemented as steps S1021 - S1024 .

[0037] S1021. Perform feature conversion based on the real-time image to determine the feature vector of each real-time image.

[0038] Exemplarily, the embodiments of the present invention can pre-process the real-time image, including denoising, grayscale (if the color image is not a necessary feature source), resizing, etc., to ensure the consistency and accuracy of subsequent feature extraction. Use image processing technology (such as SIFT, SURF, HOG, etc.) or deep learning models (such as convolutional neural network CNN) to extract key features in the image. These features may include edges, textures, color distribution, shapes, etc. Encode the extracted features into feature vectors. A feature vector is a high-dimensional array in which each element represents the value of the image on a certain feature dimension.

[0039] S1022: Determine the output result corresponding to each real-time image based on the feature vector of each real-time image and a preset public safety recognition model.

[0040] In some embodiments, the output results include the probability of the presence of various types of police officers in the real-time image.

[0041] Exemplarily, the embodiments of the present invention can load a pre-trained public security identification model. The model is a neural network or machine learning model that receives a feature vector as input and outputs a probability distribution of public security types (such as fire, criminal, public security, and traffic). The feature vectors of each real-time image are input into the public security identification model, and the model calculates the probability that the image belongs to each public security type based on these feature vectors. The model outputs a probability distribution vector, in which each element represents the probability that the image belongs to the corresponding public security type. This vector is the output result corresponding to the real-time image.

[0042] S1023: Determine a suspected image based on the output results corresponding to each real-time image and a preset probability threshold.

[0043] Exemplarily, the embodiments of the present invention may set one or more probability thresholds according to actual application scenarios and security requirements. These thresholds are used to determine whether an image may contain behavior that endangers public safety. For each real-time image output result, check whether the probability that it belongs to each public security type exceeds a preset threshold. If a probability exceeds the threshold, the image is marked as a suspect image. If the model supports multi-label classification (i.e., an image may belong to multiple public security types at the same time), it is necessary to set a threshold for each type separately and filter the suspect images accordingly.

[0044] S1024: Determine the public security type corresponding to each suspect image based on the output results corresponding to each suspect image.

[0045] For example, the embodiment of the present invention can find the police type with the highest probability in the output result for each suspect image, and use the type as the police type corresponding to the suspect image. If a more detailed judgment is required, the probability distribution in the output result of the suspect image can be analyzed, and other high-probability police types and their relative relationships can be considered. When possible, the accuracy of the model judgment can be verified through manual review or additional sensor data.

[0046] S103: Obtain video data and environmental parameters corresponding to the suspected image.

[0047] For example, the embodiment of the present invention can extract video data in the time period corresponding to the suspected image from the camera storage or live stream. The embodiment of the present invention can also collect environmental parameters when the suspected image occurs through sensors (such as temperature sensors, humidity sensors, light detectors, etc.).

[0048] S104: Based on the public security type and environmental parameters, model matching is performed to determine the target recognition model.

[0049] As a possible implementation manner, step S104 may be specifically implemented as steps S1041 - S1044 .

[0050] S1041. Based on the public security type and the mapping relationship between the public security type and the model type, determine a sub-model related to the public security type from multiple sub-models of the image recognition model.

[0051] Exemplarily, the embodiment of the present invention can establish a clear mapping relationship table or database, which records the corresponding relationship between different public security types (such as fire, criminal, public security, and traffic) and each sub-model in the image recognition model. From the mapping relationship table, select one or more sub-models that match the current public security type. These sub-models are specially designed for a certain public security type and can more effectively process image features related to the type.

[0052] S1042. Based on the public security type and environmental parameters, a correlation analysis is performed to determine environmental conditions related to the public security type.

[0053] In some embodiments, environmental parameters include temperature, humidity, light, and human traffic.

[0054] Exemplarily, the embodiments of the present invention can collect environmental parameter data in real time through various sensors installed in the smart community (such as temperature sensors, humidity sensors, light sensors, crowd counters, etc.). Based on historical data and expert knowledge, define association rules between different public security types and environmental conditions. For example, the firefighting type may be associated with environmental conditions such as high temperature and smoke concentration; the public security type may be associated with conditions such as low light and high traffic. Match the currently collected environmental parameters with the association rules to determine the environmental conditions most relevant to the current public security type.

[0055] S1043. Based on the environmental conditions related to the public security type, determine the model parameters of the sub-model related to the public security type.

[0056] Exemplarily, the embodiment of the present invention may establish a correspondence between environmental conditions and model parameters, and different environmental conditions correspond to different model parameters in the same sub-model.

[0057] S1044. Determine a target recognition model based on sub-models related to the public security type and model parameters.

[0058] Exemplarily, the embodiments of the present invention can apply model parameters to the selected sub-model to configure and optimize the sub-model, including the weights of the neural network, parameters of the image processing algorithm, etc.

[0059] S105: Determine a recognition result of the suspect image based on the suspect image, the video data, and the target recognition model.

[0060] In the embodiment of the present application, the identification result includes the cause of the event and the responsibility determination result.

[0061] As a possible implementation manner, step S105 may be specifically implemented as steps S1051 - S1054 .

[0062] S1051, performing frame processing on the video data to obtain a frame image sequence arranged in chronological order.

[0063] Exemplarily, the embodiment of the present invention can decode the video data and convert it from a compressed format (such as MP4, AVI, etc.) into an original image data sequence. This step is usually completed by a video processing library (such as FFmpeg, OpenCV, etc.). In the decoded video data, frame images are extracted at fixed time intervals (such as extracting N frames per second) or according to specific events (such as motion detection triggers). The extracted frame images will be arranged in chronological order to form a frame image sequence. The extracted frame images are stored in memory or written to a temporary file on the disk for subsequent processing. The storage format can be the original image format (such as BMP, PNG, etc.) or a compressed format (such as JPEG) to save storage space.

[0064] S1052: Perform feature extraction based on the suspect image and the frame image sequence to determine the feature vector to be identified.

[0065] Exemplarily, the embodiments of the present invention can find the area corresponding to the suspected image in the frame image sequence. If the suspected image is a frame extracted from the video, the frame can be directly located; if the suspected image is a static image, the best matching frame needs to be found through similarity calculation. In the located frame image area, the same features as the suspected image are extracted. Including low-level features such as color, texture, shape, edge, or higher-level features (such as features extracted by deep learning models).

[0066] The embodiment of the present invention can encode the extracted features into feature vectors. For example, the feature values ​​are quantized into discrete values ​​and combined into a high-dimensional array. For the features extracted by the deep learning model, the feature vector may be the activation value of a certain layer of the network. The extracted feature vector is used as the feature vector to be identified and used for the subsequent target recognition model input.

[0067] S1053: Input the feature vector to be identified into the target recognition model to obtain the identification result of the suspect image.

[0068] Exemplarily, the embodiment of the present invention can load a previously determined target recognition model. Input the feature vector to be recognized into the target recognition model, and the model will calculate the recognition result based on these feature vectors. It includes steps such as forward propagation calculation, feature matching, and classification decision. The recognition result is parsed from the output of the model. It includes specific information such as the cause of the event, responsibility determination, and behavior type. The recognition result is stored in a database or displayed on a user interface for subsequent analysis and processing.

[0069] S106. Perform intelligent security management based on the identification result of the suspicious image.

[0070] As a possible implementation manner, step S105 can be specifically implemented as steps A1-A5.

[0071] A1. If the identification type is criminal or fire, the event data of the suspect image is stored and the first alarm information is generated.

[0072] In some embodiments, the first alarm information is used to instruct the central user to confirm a criminal incident or a fire incident.

[0073] In some embodiments, the control center identifies the type and makes a judgment. If the identified type is criminal or fire, it is considered an emergency; if the identified type is public security or traffic, it is considered a non-emergency but important event.

[0074] For all identified suspect images and their related events, the system needs to store event data. This includes but is not limited to: the suspect image itself, the timestamp of the event, location information (such as camera location), identification type, and possible additional information (such as environmental parameters, other sensor data, etc.). Event data should be stored in a secure and reliable database for subsequent query and analysis.

[0075] For example, when the identification type is criminal or fire, the system immediately generates the first alarm information. The information should include the urgency of the incident, the location of the incident, a brief description (such as "suspected criminal activity found" or "fire source detected"), etc., and clearly instruct the center user to confirm the incident. The first alarm information should be sent to the center user through a high-priority channel (such as SMS, instant messaging software, internal alarm system, etc.).

[0076] A2. If the identification type is public security or traffic, the event data of the suspect image is stored and a second alarm message is generated.

[0077] In some embodiments, the second alarm information is used to instruct the central user to evaluate the impact scope of the public security incident or traffic incident.

[0078] Exemplarily, when the identification type is public security or traffic, the system generates a second alarm message. The message should include the type of event, the location of occurrence, a preliminary estimate of the possible impact range, etc., and instruct the central user to assess the impact range of the event. The second alarm message can be sent to the central user through a lower priority channel so that they can make an assessment while handling the emergency.

[0079] A3. Receive the evaluation results input by the central user.

[0080] In some embodiments, the evaluation results include a primary sphere of influence or a secondary sphere of influence.

[0081] For example, after receiving the alarm information, the central user inputs the evaluation results through the designated interface or system according to the nature and severity of the event. The evaluation results usually include the first-level impact range (i.e., the area or population directly affected by the event) or the second-level impact range (i.e., the area or population that may be indirectly affected by the spread of the event). After receiving the evaluation results, the system decides the next treatment measures according to the level and type of the evaluation.

[0082] A4. If the assessment result is a first-level impact range, an event processing instruction is generated.

[0083] In some embodiments, the event processing instructions are used to instruct patrol users to perform on-site mediation processing.

[0084] A5. Send event processing instructions to the processing terminal of the patrol user.

[0085] For example, if the assessment result is a first-level impact range and the event type is public security or traffic, the system generates an event processing instruction. The instruction includes detailed information of the event, processing requirements (such as on-site mediation, traffic diversion, crowd control, etc.), and possible auxiliary information (such as on-site maps, the location of nearby patrol users, etc.). The system sends the event processing instruction to the processing terminal (such as a smartphone, tablet computer, walkie-talkie, etc.) of the patrol user responsible for the area through the internal communication system or a dedicated mobile application. When sending instructions, the accuracy and timeliness of the information should be ensured so that the patrol user can respond quickly and handle the event.

[0086] As another possible implementation manner, step S105 may be specifically implemented as steps B1-B4.

[0087] B1. In response to the screening operation instruction of the patrol user on the intelligent security terminal, multiple suspicious images stored in the database are screened based on the screening operation instruction to obtain one or more candidate images.

[0088] In some embodiments, the filtering operation instruction includes the query location and query time period selected by the inspection user.

[0089] For example, the patrol user inputs the screening operation instructions through the graphical user interface (GUI) of the intelligent security terminal (such as a mobile application, desktop software or a dedicated hardware terminal). The instructions include the query location (such as a specific street, community or building) and the query time period (such as date and time period) selected by the user. The control center receives these instructions and performs preliminary verification to ensure the validity and format correctness of the input data.

[0090] The control center screens multiple suspect images stored in the database based on the received screening operation instructions. The screening process may involve geographic location matching (ensuring that the image is related to the selected location), timestamp matching (ensuring that the image is captured within the selected time period), and possible additional conditions (such as a specific public security type or environmental parameter range). The screening result may be one or more candidate images that highly match the query conditions of the patrol user.

[0091] B2. Display one or more alternative images.

[0092] Exemplarily, the control center organizes the selected candidate images into a list or grid view, and some basic information (such as capture time, location overview, etc.) may be displayed next to each image. The list or grid view should be easy to navigate, allowing patrol users to quickly browse and select images. The control center displays the list or grid view of candidate images through the GUI of the smart security terminal. The display interface should be clear, intuitive, and contain necessary navigation and selection controls.

[0093] B3. In response to the patrol user's selection instruction for one or more candidate images, query the database based on the selection instruction to obtain event data of the target suspect image.

[0094] In some embodiments, the event data includes a target suspect image, video data, police type, environmental parameters, and recognition results. The selection instruction includes inspecting the target suspect image selected by the user.

[0095] Exemplarily, the patrol user selects one or more target suspect images in the candidate image list or grid view. The selection can be completed by clicking, double-clicking, or dragging. The control center receives the selection instruction and records the target suspect image selected by the user. The control center queries the database for corresponding event data based on the target suspect image selected by the user. The event data includes the target suspect image itself, the video data associated with it (which may be a complete video file or related clips), the public security type (such as criminal, public security, traffic, fire protection, etc.), environmental parameters (such as temperature, humidity, light, etc.), and previous recognition results. The query process should be efficient and accurate to ensure that users can quickly obtain the required information.

[0096] B4. Display event data of the target suspect image.

[0097] Exemplarily, the system organizes the queried event data into a format that is easy to understand and read. This may include image previews, video player interfaces, detailed event descriptions, public security type labels, environmental parameter charts, and text or graphical representations of recognition results. The system displays the event data of the target suspect image through the GUI of the smart security terminal. The display interface should contain all necessary information and be presented in a way that is easy to navigate and read. For video data, the system should provide playback controls (such as play, pause, fast forward, rewind, etc.) to allow patrol users to view video content as needed.

[0098] The present invention provides a smart security management method based on image recognition. By setting a public security recognition model, the real-time images of each camera in the smart community are preliminarily detected to obtain suspected images that may endanger public security, thereby realizing the rapid identification of behaviors endangering public security. Furthermore, the present invention uses a target recognition model that matches the public security type and environmental parameters to perform secondary recognition on the suspected image and its video data, and determines the recognition result, so that the target recognition model can more accurately identify the cause of the incident endangering public security and make a responsibility determination, thereby realizing the accurate determination of behaviors endangering public security. Compared with the method of manual alarm and on-site handling by police officers, the present invention uses image recognition technology to perform preliminary recognition and secondary recognition, which can automatically, quickly and accurately identify behaviors endangering public security, thereby improving the security management efficiency of smart communities.

[0099] Optionally, the image recognition-based intelligent security management method provided by the embodiment of the present invention further includes steps S201-S203.

[0100] S201, receiving a target query instruction from an inspection user.

[0101] In some embodiments, the target query instruction is used to instruct to query a target person or a target object.

[0102] Exemplarily, the target query instruction includes a real-time image of a target person or a target object.

[0103] Exemplarily, the patrol user inputs a target query command through the graphical user interface (GUI) of the intelligent security terminal (such as a mobile application, desktop software or a dedicated hardware terminal). The command is used to query a specific target person or object, such as looking for a missing person, a specific vehicle or object, etc. The command input by the user includes a real-time image of the target person or object, which can be a live image captured by a camera, or a picture selected by the user from an album.

[0104] After receiving the user's input, the system first verifies the data to ensure the integrity and validity of the image. At the same time, the system also checks the user's permissions to ensure that the user has the authority to perform such query operations.

[0105] S202: Based on the target query instruction, query the target person or target object to obtain a query report.

[0106] As a possible implementation manner, step S202 can be specifically implemented as steps S2021-S2027.

[0107] S2021. Extract the features of the target person or object based on the target query instruction.

[0108] For example, when receiving a real-time image in a target query command, the system first uses image processing technology to extract key features in the image. These features may include color, texture, shape, edge, key points (such as corners, spots), etc. For objects with specific structures such as faces and license plates, the system may use more advanced feature extraction methods, such as face recognition algorithms and license plate recognition algorithms based on deep learning.

[0109] S2022. Based on the target query instruction, perform model matching to determine the target query model.

[0110] Exemplarily, the target query model is a sub-model in the image recognition model, and the sub-model realizes the target matching query by comparing the feature similarity of specific areas of two images.

[0111] S2023. Determine a query range and multiple cameras within the query range based on the real-time image of the target person or object.

[0112] In some embodiments, the system determines the query range based on the geographic location information (such as latitude and longitude, address, etc.) in the target query instruction, combined with map data and camera layout information. The range may be a circular area, a rectangular area, or an area of ​​other shapes. The size of the query range may be adjusted according to factors such as the moving speed of the target person or object, the urgency of the query, etc. Within the determined query range, the system screens out all available cameras and obtains their real-time images or video streams. These cameras may come from different monitoring networks, such as urban traffic monitoring, community monitoring, commercial venue monitoring, etc.

[0113] S2024. Obtain real-time images of multiple cameras within the query range.

[0114] Exemplarily, the system can establish communication with selected cameras via a network connection to obtain their images or video streams in real time.

[0115] S2025. Based on the real-time images of multiple cameras within the query range, the characteristics of the target person or object, and the target query model, traversal matching is performed to determine the query result.

[0116] In some embodiments, the query result includes a plurality of target real-time images related to the target person or target object.

[0117] Exemplarily, the system uses a traversal matching method to match the features of the target person or object with the real-time images of each camera within the query range. The matching process may involve steps such as feature comparison and similarity calculation. In order to improve the matching efficiency, the system may adopt some optimization strategies, such as parallel processing and block matching. Based on the matching results, the system selects multiple target real-time images related to the target person or object. These images may come from different cameras, recording the appearance of the target at different times and locations.

[0118] S2026. Based on the camera positions corresponding to the multiple target real-time images and the map of the query range, draw a trajectory map of the target person or object.

[0119] For example, the system draws a trajectory map of the target person or object based on the camera positions corresponding to the real-time images of the multiple targets and the map information of the query range. The trajectory map may represent the moving path of the target in the form of lines, arrows, etc. In order to improve the accuracy and readability of the trajectory map, the system may use some map matching, path smoothing and other algorithms for processing.

[0120] S2027. Generate a query report based on the query results and the trajectory map.

[0121] In some embodiments, the query report includes detailed information such as multiple real-time images of targets related to the target person or object, trajectory diagram, query time, query range, camera information, etc. This information can help users fully understand the movement of the target, the location of the target, etc. The report may also include some additional information, such as the target's identity recognition results (such as name, ID number, etc.), records of related security events, etc.

[0122] S203: Display the query report.

[0123] In this way, the embodiment of the present invention can complete the query instructions of the patrol user through image recognition and matching, provide convenience for the query of personnel and targets, and improve the security management efficiency of the smart community.

[0124] Optionally, the image recognition-based intelligent security management method provided by an embodiment of the present invention further includes steps S301-S304 before step S102.

[0125] S301. Acquire typical images of various types of public security events that occurred in a historical period, and the public security types corresponding to each typical image.

[0126] Exemplarily, the embodiment of the present invention can directly obtain typical images in a historical period and their corresponding public security types from the database of the control center.

[0127] S302: Based on typical images of various types of public security incidents, feature conversion is performed to obtain feature vectors of each typical image.

[0128] S303 , taking the feature vector of each typical image as input and the public security type corresponding to each typical image as output, generating a plurality of first training samples.

[0129] S304: Perform neural network training based on the multiple first training samples to obtain a public safety recognition model.

[0130] Exemplarily, the embodiments of the present invention may select a suitable neural network architecture, such as a convolutional neural network (CNN), a recurrent neural network (RNN), or a deep residual network (ResNet). The parameters of the network, such as the number of layers and the number of nodes, are determined according to the complexity of the problem and the data scale. The training sample is input into the neural network, and the prediction result is calculated by forward propagation. The difference between the prediction result and the true label is measured using a loss function (such as cross entropy loss). The weights and biases of the neural network are updated by a back propagation algorithm to reduce the value of the loss function. The above process is repeated until the value of the loss function converges or reaches a preset number of training rounds. The trained model is evaluated using a validation set or a test set to check the performance of the model. Evaluation indicators include accuracy, recall rate, F1 score, etc. The model is optimized according to the evaluation results, such as adjusting the network structure, increasing training data, using regularization technology, etc. The trained public safety identification model is saved to a disk for subsequent use or deployment.

[0131] In this way, the embodiment of the present invention can deploy a public security recognition model by means of neural network training before image detection of the smart community, thereby facilitating the recognition of suspicious images and public security types.

[0132] Optionally, the image recognition-based intelligent security management method provided by an embodiment of the present invention further includes steps S401-S409 before step S104.

[0133] S401. Obtain processing result data of various events occurring in a historical period.

[0134] In some embodiments, the processing result data includes real-time images, video data, public security types, environmental parameters, causes of events and responsibility determination results of various types of events handled by the patrol user during the patrol process.

[0135] Exemplarily, the embodiment of the present invention can directly obtain the processing result data of various events from the database of the control center.

[0136] S402: Perform cluster analysis based on the police type and environmental parameters to determine the various environmental conditions corresponding to each police type.

[0137] Wherein, any environmental condition is a combination of one or more environmental parameters.

[0138] For example, the embodiment of the present invention can use clustering algorithms such as K-means, hierarchical clustering, DBSCAN, etc. to cluster data according to the police type and environmental parameters. Determine the common environmental condition combination under each police type, such as traffic accidents are more common on rainy days and urban roads at night, while criminal crimes may occur at night and in sparsely populated areas.

[0139] S403: Based on the public security type and environmental parameters, the processing result data of each type of event is split to obtain the processing result data corresponding to various environmental conditions under each public security type.

[0140] Exemplarily, the present invention can split the original data into data subsets under different environmental conditions under different public security types according to the clustering results, and ensure that each subset contains a sufficient number of samples for subsequent model training.

[0141] S404. For each environmental condition under any public security type, based on the real-time image and video data corresponding to the environmental condition under the public security type, generate multiple input feature vectors corresponding to the environmental condition under the public security type.

[0142] Exemplarily, the present invention can extract features such as color, texture, shape, motion trajectory, etc. from real-time image and video data. The extracted features are encoded into vector form for input into a neural network. The most representative features are selected according to importance or relevance to reduce the complexity of the model and improve training efficiency.

[0143] S405. Based on the event cause and responsibility determination result corresponding to the environmental condition under the public security type, generate a plurality of output feature vectors corresponding to the environmental condition under the public security type.

[0144] Exemplarily, the present invention can convert the cause of the event into a quantifiable feature vector, such as using One-Hot Encoding or Word Embedding technology, and convert the responsibility determination result into a feature vector, such as using numerical coding to represent the responsible party and the degree of responsibility.

[0145] S406: Generate multiple second training samples based on the multiple input feature vectors and the multiple output feature vectors.

[0146] Exemplarily, the input feature vector and the output feature vector are combined into a training sample, namely (input feature vector, output feature vector). Ensure that each sample is correctly labeled with the corresponding public security type, environmental conditions, event cause and responsibility determination result.

[0147] S407: Perform neural network training based on the multiple second training samples to obtain a sub-model and model parameters corresponding to the environmental condition under the public security type.

[0148] Exemplarily, the present invention can select a suitable neural network architecture, such as a convolutional neural network (CNN) for image feature extraction, a recurrent neural network (RNN) for processing time series data, or a fully connected neural network (FNN) for classification and regression tasks. The neural network is trained using training samples, the prediction results are calculated by forward propagation, and the network weights and biases are updated by back propagation. The performance of the model is evaluated using a validation set, such as accuracy, recall, F1 score, etc. The network structure, learning rate, regularization parameters, etc. are adjusted according to the evaluation results to improve the model performance.

[0149] S408. Based on the processing result data corresponding to various environmental conditions under various public security types, neural network training is performed to obtain sub-models and model parameters corresponding to various environmental conditions under various public security types.

[0150] S409: Based on the sub-models and model parameters corresponding to various environmental conditions under various public security types, an image recognition model is obtained by integration.

[0151] In this way, the embodiment of the present invention realizes image recognition and prediction under different public security types and environmental conditions by setting up an image recognition model and its sub-models, which facilitates smart security management and improves the security management efficiency of smart communities.

[0152] Optionally, the image recognition-based intelligent security management method provided by the embodiment of the present invention further includes steps S501-S505.

[0153] S501. Obtain processing result data of the intelligent security terminal in the historical period before the current moment.

[0154] In some embodiments, the processing result data includes real-time images, video data, public security types, environmental parameters, causes of events and responsibility determination results of various types of events handled by the patrol user during the patrol process.

[0155] S502: Generate a plurality of first training samples and a plurality of second training samples based on the processing result data.

[0156] Exemplarily, the first training sample can extract real-time image and video data and corresponding public security type information from the processing result data. Feature extraction is performed on the image and video data to generate a feature vector. The feature vector and the public security type information are combined into the first training sample, i.e. (feature vector, public security type).

[0157] Exemplarily, the second training sample can extract real-time image and video data, as well as the corresponding event cause and responsibility determination result from the processing result data. Similarly, feature extraction is performed on the image and video data to generate a feature vector. The feature vector, event cause and responsibility determination result are combined into the second training sample, i.e. (feature vector, event cause, responsibility determination result).

[0158] S503: Based on the multiple first training samples, the public security recognition model is updated to obtain an updated public security recognition model.

[0159] Exemplarily, an existing public safety recognition model is loaded. The model is trained using the first training sample, the prediction result is calculated by forward propagation, and the model weight is updated by back propagation. The performance of the model, such as accuracy and recall, is evaluated using the validation set. Based on the evaluation results, if the model performance is improved, the model is updated to a new public safety recognition model.

[0160] S504: Based on the multiple second training samples, each sub-model in the image recognition model is updated to obtain an updated image recognition model.

[0161] Exemplarily, each sub-model in the existing image recognition model is loaded. For each sub-model (corresponding to different public security types and environmental conditions), the corresponding second training sample is used for training. The performance of each sub-model is evaluated using the validation set. According to the evaluation results, if the performance of the sub-model is improved, the corresponding sub-model is updated. If the image recognition model uses an integrated learning method to fuse multiple sub-models, it is necessary to re-fuse the model to obtain an updated image recognition model.

[0162] S505: Based on the updated public security recognition model and image recognition model, real-time monitoring of the real-time images of the smart community is performed.

[0163] Exemplarily, the present invention can obtain image and video data in real time from cameras or other intelligent security terminals in smart communities. Preprocess the real-time image and video data, such as format conversion, noise removal, etc. Use the updated public security recognition model to recognize the preprocessed image and determine the public security type to which it belongs. For the identified specific public security type events, use the updated image recognition model to further analyze the cause of the event and the responsibility determination results. Output the recognition and analysis results to the terminal device of the monitoring center or relevant personnel so that timely response measures can be taken.

[0164] In this way, the embodiment of the present invention can periodically or in real time update the public security recognition model and the image recognition model to ensure the accuracy of model recognition in the process of smart security management.

[0165] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0166] The following is an embodiment of the device of the present invention. For details not described in detail therein, reference may be made to the corresponding method embodiment described above.

[0167] Figure 2 The structure diagram of a smart security management device based on image recognition provided by an embodiment of the present invention is shown. The smart security management device 600 includes a communication module 601 and a processing module 602 .

[0168] The communication module 601 is used to obtain real-time images from multiple cameras in the smart community.

[0169] Processing module 602 is used to determine the suspect image and the public security type corresponding to the suspect image based on the real-time image and the preset public security recognition model. The public security recognition model is used to identify the suspect image that may have behaviors that endanger public security; the public security types include fire, criminal, public security and traffic.

[0170] The communication module 601 is also used to obtain video data and environmental parameters corresponding to the suspicious image.

[0171] Processing module 602 is also used to perform model matching based on public security type and environmental parameters to determine the target recognition model; determine the recognition result of the suspect image based on the suspect image, video data, and the target recognition model, the recognition result including the cause of the incident and the responsibility determination result; and perform intelligent security management based on the recognition result of the suspect image.

[0172] Figure 3 Schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. Figure 3 As shown, the electronic device 700 includes: a processor 701, a memory 702, and a computer program 703 stored in the memory 702 and executable on the processor 701. When the processor 701 executes the computer program 703, the steps in the above-mentioned method embodiments are implemented, for example Figure 1 Alternatively, when the processor 701 executes the computer program 703, the functions of the modules / units in the above-mentioned device embodiments are realized, for example, Figure 2 The functions of the communication module 601 and the processing module 602 are shown.

[0173] Exemplarily, the computer program 703 may be divided into one or more modules / units, which are stored in the memory 702 and executed by the processor 701 to implement the present invention. The one or more modules / units may be capable of implementing a particular Figure 3 A series of computer program instruction segments with certain functions, which are used to describe the execution process of the computer program 703 in the electronic device 700. For example, the computer program 703 can be divided into Figure 2 A communication module 601 and a processing module 602 are shown.

[0174] The processor 701 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0175] The memory 702 may be an internal storage unit of the electronic device 700, such as a hard disk or memory of the electronic device 700. The memory 702 may also be an external storage device of the electronic device 700, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 700. Further, the memory 702 may also include both an internal storage unit of the electronic device 700 and an external storage device. The memory 702 is used to store the computer program and other programs and data required by the terminal. The memory 702 may also be used to temporarily store data that has been output or is to be output.

[0176] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A smart security management method based on image recognition, characterized in that: include: Obtain real-time images from multiple cameras in a smart community; Based on the real-time image and a preset public safety recognition model, a suspect image and a public security type corresponding to the suspect image are determined, wherein the public safety recognition model is used to identify suspect images that may contain behaviors that endanger public safety; the public security types include fire protection, criminal, public security and traffic; Acquiring video data and environmental parameters corresponding to the suspected image; Based on the public security type and the environmental parameters, model matching is performed to determine a target recognition model; Based on the suspect image, the video data, and the target recognition model, determining a recognition result of the suspect image, wherein the recognition result includes a cause of the event and a responsibility determination result; Based on the recognition result of the suspicious image, intelligent security management is performed.

2. The intelligent security management method based on image recognition according to claim 1 is characterized in that: The determining of the suspected image and the public security type corresponding to the suspected image based on the real-time image and the preset public security recognition model includes: Based on the real-time images, feature conversion is performed to determine feature vectors of each real-time image; Based on the feature vectors of each real-time image and a preset public security identification model, an output result corresponding to each real-time image is determined, wherein the output result includes a probability that each type of public security exists in the real-time image; Determine a suspected image based on the output results corresponding to each of the real-time images and a preset probability threshold; Based on the output results corresponding to the respective suspect images, the public security type corresponding to the respective suspect images is determined.

3. The intelligent security management method based on image recognition according to claim 1 is characterized in that: The performing model matching based on the public security type and the environmental parameters to determine the target recognition model includes: Based on the public security type and the mapping relationship between the public security type and the model type, determining a sub-model related to the public security type from a plurality of sub-models of the image recognition model; Based on the public security type and the environmental parameters, a correlation analysis is performed to determine the environmental conditions related to the public security type; the environmental parameters include temperature, humidity, light and human flow; Determining model parameters of the sub-model related to the public security type based on environmental conditions related to the public security type; The target recognition model is determined based on the sub-model related to the public security type and the model parameters.

4. The intelligent security management method based on image recognition according to claim 1 is characterized in that: The determining the recognition result of the suspect image based on the suspect image, the video data, and the target recognition model includes: Performing frame processing on the video data to obtain a frame image sequence arranged in chronological order; Based on the suspected image and the frame image sequence, feature extraction is performed to determine a feature vector to be identified; The feature vector to be identified is input into the target recognition model to obtain the recognition result of the suspect image.

5. The intelligent security management method based on image recognition according to claim 1 is characterized in that: Based on the recognition result, the intelligent security management is performed, including: If the identification type is criminal or fire, the event data of the suspect image is stored and a first alarm message is generated; the first alarm message is used to instruct the central user to confirm the criminal event or the fire event; If the identification type is public security or traffic, the event data of the suspected image is stored, and a second alarm message is generated, wherein the second alarm message is used to instruct the center user to evaluate the impact scope of the public security event or the traffic event; Receiving an evaluation result input by a central user; the evaluation result includes a primary impact scope or a secondary impact scope; If the assessment result is a first-level impact range, an event processing instruction is generated, and the event processing instruction is used to instruct the patrol user to perform on-site mediation; The event processing instruction is sent to the processing terminal of the patrol user.

6. The intelligent security management method based on image recognition according to claim 1 or 5, characterized in that: Based on the recognition result, the intelligent security management is performed, including: In response to a screening operation instruction of the patrol user on the intelligent security terminal, based on the screening operation instruction, multiple suspicious images stored in the database are screened to obtain one or more candidate images; the screening operation instruction includes a query location and a query time period selected by the patrol user; displaying the one or more candidate images; In response to a selection instruction of the patrol user for the one or more candidate images; based on the selection instruction, querying a database to obtain event data of the target suspect image, the event data including the target suspect image, video data, public security type, environmental parameters and recognition results; the selection instruction includes the target suspect image selected by the patrol user; The event data of the target suspect image is displayed.

7. The intelligent security management method based on image recognition according to any one of claims 1 to 6, characterized in that: The method further comprises: Receive a target query instruction from a patrol user, wherein the target query instruction is used to instruct to query a target person or a target object; the target query instruction includes a real-time image of the target person or the target object; Based on the target query instruction, query the target person or target object to obtain a query report; A report for the query is displayed.

8. The intelligent security management method based on image recognition according to claim 7 is characterized in that: The querying of the target person or the target object based on the target query instruction to obtain a query report includes: Based on the target query instruction, extract the characteristics of the target person or target object; Based on the target query instruction, model matching is performed to determine the target query model; Based on the real-time image of the target person or object, determine a query range and multiple cameras within the query range; Get real-time images of multiple cameras within the query range; Based on the real-time images of multiple cameras within the query range, the characteristics of the target person or target object, and the target query model, traversal matching is performed to determine a query result, wherein the query result includes multiple target real-time images related to the target person or target object; Based on the camera positions corresponding to the multiple target real-time images and the map of the query range, draw a trajectory map of the target person or object; A query report is generated based on the query result and the trajectory map.

9. The intelligent security management method based on image recognition according to any one of claims 1 to 8, characterized in that: The method further comprises: Obtain the processing result data of the intelligent security terminal in the historical period before the current moment, the processing result data includes real-time images, video data, public security types, environmental parameters, event causes and responsibility determination results of various events handled by the patrol user during the patrol process; Based on the processing result data, generating a plurality of first training samples and a plurality of second training samples; Based on the multiple first training samples, the public safety recognition model is updated to obtain an updated public safety recognition model; Based on the multiple second training samples, each sub-model in the image recognition model is updated to obtain an updated image recognition model; Based on the updated public security recognition model and image recognition model, real-time images of the smart community are monitored in real time.

10. An intelligent security management system based on image recognition, characterized in that: The smart security management system includes an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor is used to call and run the computer program stored in the memory to execute the method as described in any one of claims 1 to 9.

Images