Security camera community relation analysis method, device and system based on minimum entropy

Abstract

The invention relates to the technical field of intelligent security and monitoring, and discloses a security camera community relation analysis method, device and system based on minimum entropy, and the method comprises the following steps: collecting multi-dimensional data of each camera, and generating a feature vector; calculating the comprehensive similarity between the cameras, and constructing an association graph in combination with the position, orientation and naming information; performing clustering analysis based on the feature vector and a minimum entropy criterion, and constructing a point location clustering structure; community division based on minimum entropy is executed in combination with the clustering structure, and a community structure reflecting the space and behavior relation between the cameras is generated; the system comprises a data acquisition module, a preprocessing module, a similarity calculation module, a clustering analysis module and a community division module. According to the method, camera clustering and community division are optimized through the minimum entropy criterion, the behavior analysis precision of the security and protection system is improved, and the anomaly detection, resource scheduling and safety early warning capabilities are effectively enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent security and monitoring, and particularly to a method, device, and system for analyzing the community relationship of security cameras based on minimum entropy. Background Art

[0002] With the development of urban intelligent security systems, a large number of cameras play an important role in fields such as public safety, traffic monitoring, and community management. In addition to real-time image acquisition, cameras also generate a large amount of historical data, such as face capture records and personnel file information. These data contain rich location information and can provide a deep analysis basis for the system. However, existing technologies usually focus on real-time monitoring and abnormal alarm, ignoring the comprehensive utilization of historical data, especially the in-depth analysis at the spatial and behavioral levels.

[0003] Existing security systems rely on real-time video monitoring to identify abnormal events, mainly focusing on image recognition and simple behavior detection. This method has problems of data redundancy and low efficiency when dealing with large-scale monitoring data. Especially in complex scenarios or multi-target monitoring, traditional abnormal detection methods are difficult to provide accurate behavior prediction and scenario analysis. In addition, existing technologies often ignore the potential correlation between cameras and are difficult to effectively identify the connection between the spatial layout and behavior patterns of different cameras.

[0004] Although existing methods can perform clustering analysis, most clustering methods only focus on the similarity of cameras and fail to comprehensively consider the weighted fusion of multi-dimensional information such as spatial location, orientation, and naming. This makes the effectiveness and accuracy of traditional clustering results limited in complex deployments and changing environments, and it is difficult to provide flexible and accurate clustering and community division.

[0005] In addition, although existing community discovery methods can group cameras, they lack optimization for security scenarios and fail to effectively combine spatio-temporal characteristics and behavior patterns. These methods are easily affected by noise and redundant information in large-scale data analysis, resulting in unclear division results and difficult to accurately reflect the correlation between cameras and their interaction with the surrounding environment. Summary of the Invention

[0006] The purpose of the present invention is to provide a method, device, and system for analyzing the community relationship of security cameras based on minimum entropy, which solves the problems of insufficient accuracy in camera clustering and community division in the prior art and difficulty in effectively using multi-dimensional data for in-depth spatial and behavioral analysis.

[0007] To achieve the above object, the present invention is realized through the following technical solutions: A method for analyzing the community relationship of security cameras based on minimum entropy, comprising the following steps: Collect multi-dimensional raw data of each camera; Preprocess the collected multi-dimensional raw data to generate feature vectors in a unified format; Calculate the comprehensive similarity between cameras based on the feature vectors. The similarity is weighted and fused by combining position, orientation, and naming information to construct an association graph with cameras as nodes and similarity as edge weights; Perform clustering analysis on each camera position based on the feature vectors and the minimum entropy criterion to construct a position clustering structure; On the association graph of each camera, combine the clustering structure and perform community division based on the minimum entropy to generate a community structure reflecting the spatial and behavioral relationships between cameras.

[0008] Preferably, the step of collecting multi-dimensional raw data of each camera includes: Obtain the captured image data of the camera; Obtain the personnel identity file information associated with the image; Obtain the position information of the camera; Obtain the installation orientation information of the camera; Obtain the device naming information of the camera.

[0009] Preferably, the step of preprocessing the collected multi-dimensional raw data to generate feature vectors in a unified format includes: Extract features from the image data to obtain image feature vectors ; Encode the personnel file information to obtain file feature vectors ; Convert the coordinate of the position information of the camera to obtain a position information vector ; Standardize the installation orientation information of the camera to obtain an orientation vector ; Vectorize the device naming information of the camera to obtain a naming vector ; Concatenate the above vectors into a feature vector in a unified format 。

[0010] Preferably, the step of calculating the comprehensive similarity between cameras based on the feature vectors includes: Calculate the Euclidean distance between the feature vectors of each pair of cameras , and its formula is: ; Among them, and are the feature vectors of camera and camera respectively; and is the corresponding value for each feature dimension; is the dimension of the feature vector; Calculate the similarity of each pair of cameras according to the Euclidean distance , and its formula is: ; Combine the position information, orientation information and naming information between cameras to perform weighted fusion on the similarity to obtain the comprehensive similarity , and its formula is: ; Among them, are the weighting coefficients of the position, orientation and naming information, are the similarity values calculated based on the position, orientation and naming information respectively.

[0011] Preferably, the steps of constructing an association graph with cameras as nodes and similarities as edge weights by performing weighted fusion on the similarity in combination with the position, orientation and naming information include: According to the calculated comprehensive similarity , in combination with the position information similarity and orientation similarity and naming similarity of the cameras, obtain the final similarity value between each pair of cameras through weighted fusion, where the formula is: ; Among them, is the weighting coefficient, controlling the contribution of each similarity information; construct an association graph with cameras as nodes and as the edge weight.

[0012] Preferably, the steps of performing clustering analysis on each camera position based on the feature vector and the minimum entropy criterion to construct a position clustering structure include: Based on the feature vector of the camera, perform clustering analysis using the minimum entropy criterion, and the formula is: ; Among them, is the entropy of the clustering result; is the probability of cluster ; is the number of clusters; By minimizing the entropy , which is used for the optimal clustering assignment of camera positions, obtain the clustering structure of camera positions.

[0013] Preferably, the step of performing community division based on minimum entropy on the association graph of each camera and generating a community structure reflecting the spatial and behavioral relationships between cameras includes: On the association graph constructed based on the camera feature vectors and comprehensive similarity, combined with the clustering structure of the cameras, establish the connection relationships between the cameras; Based on the minimum entropy criterion, calculate the entropy of the community division graph, and optimize the community division of the cameras by minimizing the entropy value; By minimizing the entropy value, optimize the community division between the cameras so that the spatial and behavioral relationships of each community are effectively reflected, and obtain the community structure of the cameras.

[0014] The present invention also provides a security camera community relationship analysis system based on minimum entropy, including: A data acquisition module for acquiring multi-dimensional raw data of each camera; A preprocessing module for preprocessing the acquired multi-dimensional raw data to generate feature vectors in a unified format; A similarity calculation module for calculating the comprehensive similarity between cameras based on the feature vectors, and performing weighted fusion on the similarity by combining position, orientation, and naming information, and constructing an association graph with cameras as nodes and similarity as edge weights; A clustering analysis module for performing clustering analysis on each camera point based on the feature vectors and the minimum entropy criterion, and constructing a point clustering structure; A community division module for performing community division based on the minimum entropy criterion on the basis of the association graph in combination with the clustering structure, and generating a community structure reflecting the spatial and behavioral relationships between cameras.

[0015] The present invention also provides a device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the above method can be implemented.

[0016] In summary, the present invention includes at least one of the following beneficial technical effects: 1. The present invention collects multi-dimensional raw data of each camera, performs weighted fusion by combining the spatial position, orientation, and naming information of the cameras, generates feature vectors in a unified format, and then calculates the comprehensive similarity between the cameras to construct a camera association graph. It achieves a deep analysis of the potential correlations between cameras based on multi-dimensional data, and can effectively capture the similarities between different cameras. Compared with the prior art that only relies on real-time image data for simple abnormal alarm analysis, the present invention solves the deficiencies of ignoring historical data, spatial, and behavioral information, and fully utilizes information such as the installation position and direction of the cameras to comprehensively analyze the relationships between the cameras.

[0017] 2. The present invention performs clustering analysis on camera positions based on the minimum entropy criterion, and further executes community division in combination with the clustering structure to generate a community structure reflecting the spatial and behavioral relationships between cameras. The effects of optimizing camera clustering and community structure are achieved, so that the cameras within each community have high similarity, while there are large differences between different communities. Compared with the clustering methods in the prior art, the present invention introduces the minimum entropy criterion to minimize the entropy value, ensuring the accuracy of clustering and community division, thereby overcoming the problems of low clustering accuracy and insufficient information fusion in traditional methods.

[0018] 3. The present invention adopts the camera clustering structure and the community division result, effectively improving the management and scheduling efficiency of camera resources. Through accurate community structure division, the system can dynamically adjust resource allocation according to the spatial and behavioral associations between cameras to ensure efficient monitoring at critical moments. Compared with the prior art solutions lacking flexible dynamic resource scheduling, the present invention solves the problem that fixed resource allocation in the existing solutions is difficult to meet the real-time security needs, and improves the emergency response ability of the security monitoring system.

[0019] 4. Based on the clustering structure and community division result of camera positions, and combined with historical data and real-time monitoring data, the present invention can efficiently monitor and warn of abnormal behaviors. Through the optimized community structure, the system can identify the occurrence patterns of abnormal behaviors and conduct risk assessment in advance. Compared with the security warning systems in the prior art that mainly rely on image recognition and simple rule judgment, the present invention optimizes clustering and community division by introducing the minimum entropy principle, solves the problem of insufficient recognition of complex behavior patterns in traditional methods, and significantly improves the accurate warning ability and security response efficiency of the security system. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is a schematic flowchart of the method of the present invention; Figure 2 is a schematic diagram of the system architecture of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0021] The following further describes the present invention in detail Figure 1 attached Figure 2 with reference to the attached drawings.

[0022] The embodiment of the present invention provides a method for analyzing the community relationship of security cameras based on minimum entropy, S1. Collect multi-dimensional original data of each camera; S2. Preprocess the collected multi-dimensional original data to generate feature vectors in a unified format; S3. Calculate the comprehensive similarity between cameras based on the feature vectors. The similarity is weighted and fused by combining location, orientation, and naming information to construct an association graph with cameras as nodes and similarity as edge weights. S4. Perform clustering analysis on each camera position based on the feature vectors and the minimum entropy criterion to construct a position clustering structure. S5. Combine the clustering structure on the association graph of each camera and perform community division based on the minimum entropy to generate a community structure reflecting the spatial and behavioral relationships between cameras.

[0023] For step S1, in this embodiment, the "multi-dimensional raw data" is not single-dimensional video or image information, but the aggregation of multi-source heterogeneous information realized through the platform-side data interface based on the deployed camera devices in the security system. Data collection can be configured and deployed in local servers, edge computing nodes, or central platforms to adapt to different scales of security application scenarios.

[0024] Preferably, the data collection step includes the acquisition of the following five categories of information, which are specifically described as follows: In this embodiment, first, capture image data of cameras is obtained.

[0025] The source of the image data is the personnel capture results uploaded in real time by each front-end camera, which are generally generated by an image detection model (such as a face detection algorithm) built-in at the access end or deployed at the edge node. The image data should include at least one clearly recognizable face image, and the image format can be JPEG, BMP, or other structured image formats, along with meta-information such as the capture timestamp and camera number.

[0026] In this embodiment, personnel identity file information associated with the above image data is further obtained.

[0027] The personnel identity information is the file content registered in the platform-side structured database, which can be associated in the background through image comparison. The identity information generally includes information fields such as personnel number, gender, date of birth, and affiliated unit. By introducing the identity information, the image trajectories of the same person under different cameras can be fused, providing semantic layer support for the subsequent construction of a cross-position personnel behavior path model.

[0028] In this embodiment, the location information of the cameras is obtained.

[0029] The location information preferably includes the latitude and longitude coordinate data of each camera position, obtained through recording during deployment or automatic registration by the GIS platform. To further improve the data expression consistency of the location information, in subsequent processing, the system converts its latitude and longitude into three-dimensional Cartesian space coordinates. so as to perform unified processing with the spatial analysis module in subsequent clustering or graph model construction.

[0030] In this embodiment, the installation orientation information of the camera is also obtained.

[0031] The orientation information is used to describe the main direction of the camera shooting angle, and its physical meaning is two parameters: the horizontal pointing angle (Azimuth) and the vertical inclination angle (Elevation) of the camera lens. In terms of the acquisition method, it can be imported through the records during construction or read in real time through sensors. This information is used to construct a discriminant index for perspective consistency in the graph structure and participate in the minimum entropy partition calculation as a correction factor for behavior similarity.

[0032] In this embodiment, the device naming information of the camera is further obtained.

[0033] Device naming refers to the unique identification string when the camera is registered on the platform, which usually contains human-readable elements such as point numbers, names of affiliated regions, building names, and floor marks. Through text analysis methods (such as word segmentation, vector embedding), the present invention converts this naming information into a structured semantic vector, which is used to provide a "semantic neighbor" index in the subsequent feature fusion stage and assist in correcting the abnormal recognition of spatial positions caused by installation oversights.

[0034] In this embodiment, the above-mentioned information collected will be uniformly encapsulated into a structured raw data record, and indexed by the camera number as the primary key, and stored in a temporary data table, which is used as the data source for vector construction and similarity analysis in subsequent steps.

[0035] It should be noted that to adapt to the large-scale urban-level deployment environment, the present invention designs a data reception interface and an asynchronous distribution mechanism based on a message queue in the data acquisition module, which can effectively improve the overall data access throughput capacity of the system. In a multi-threaded or distributed architecture, this acquisition mechanism has good scalability and stability.

[0036] In summary, this embodiment systematically collects five types of multi-dimensional raw data: images, identities, location information, orientation information, and naming information, providing a detailed and complementary information basis for constructing subsequent feature vectors, and effectively supporting the extraction of community structures under the subsequent minimum entropy criterion.

[0037] For step S2, the core objective of this step is to convert the raw information in multi-source heterogeneous formats into a feature vector expression with a unified structure, so as to form a description form of camera points with spatial behavior semantics. The entire processing flow is implemented by a feature encoding module in the system of the present invention, and this module can run on a server or an edge node by calling a preset algorithm model.

[0038] In this embodiment, first, feature extraction is performed on the image data to generate an image feature vector .

[0039] Specifically, the system crops and aligns the face regions appearing in the images captured by the camera, and performs embedding encoding on them using a face recognition network model. Preferably, the deep model adopted can be a convolutional neural network structure pre-trained on a public face recognition dataset, such as the ResNet architecture. It should be noted that face image acquisition, face recognition image acquisition, etc. need to be carried out after being authorized by relevant departments. This model maps the input image to a -dimensional vector: ; wherein, represents the identity semantic position of the image in the feature space and is used to measure the consistency of image capture behaviors between different cameras.

[0040] In this embodiment, the collected personnel file information is further structurally encoded to generate a file feature vector .

[0041] The file information includes several discrete tag items (such as gender, identity category, affiliated unit, etc.). The system converts it into a dense vector representation through one-hot encoding or embedding mapping of multiple types of attributes. If we set as the set of file fields, then there is: ; This vector retains the attribute consistency of the same person under different cameras to a certain extent, which is conducive to aggregating the observation results of the same identity at different positions.

[0042] In this embodiment, coordinate transformation processing is performed on the position information of the camera to generate a position information vector .

[0043] To adapt to the subsequent edge weight model constructed based on geometric distance in the graph space, the system maps the longitude and latitude information of the camera to the three-dimensional Euclidean space coordinates. This mapping preferably uses the spherical projection formula under the large earth model for conversion, specifically as follows: ; ; ; where the variable represents the latitude value of the th camera point, represents its longitude value, which is the radius of the earth or the reference sphere, and the unit can be uniformly set to meters or other units adapted to the coordinate scale.

[0044] Through the above formula calculation, the geographical location of the camera on the spherical surface can be accurately mapped to the Euclidean position coordinates in the three-dimensional space. The finally obtained position information vector is defined as: ; This three-dimensional vector is used to represent the geometric position of the camera in the physical space and constitutes the basis for calculating the spatial similarity in the subsequent composition edge construction and point clustering processes.

[0045] In this embodiment, the installation orientation information of the camera is standardized to generate an orientation vector .

[0046] The original representation of the camera's orientation information is two angular measures, the horizontal angle (Azimuth) and the pitch angle (Elevation), which are respectively used to describe the horizontal direction and the vertical elevation angle of the camera. The system preferably adopts a vectorized expression form to convert the above angle combination into a unit orientation vector for the subsequent consistency analysis of spatial directions in the graph model. Specifically, let the th camera have a pitch angle of , and a horizontal angle of , then its orientation vector can be defined as a three-dimensional unit vector according to the following formula: ; where is the elevation angle of the camera lens relative to the horizontal plane, with the unit of radians; is the azimuth angle of the camera lens in the ground plane (usually counted clockwise from the due north direction), also in radians. This representation can intuitively reflect the shooting direction of the camera and is used to calculate the spatial orientation consistency, which has a significant impact on both graph construction and clustering analysis.

[0047] In this embodiment, the device naming information of the camera is semantically vectorized to generate a naming vector .

[0048] The naming information is usually of string type defined during platform registration and generally includes information elements with semantic tags such as area name, building number, floor location, door direction, functional attribute, etc. These naming texts have obvious spatial indication meanings in human recognition and are also regarded as important auxiliary information sources reflecting the point logic structure in the present invention.

[0049] In actual processing, the system first performs word segmentation on this type of text information to obtain a word-unit sequence. To improve the accuracy of word segmentation, it is preferred to use a natural language processing tool that supports Chinese structure recognition and optimize proper nouns in combination with a security industry terminology dictionary.

[0050] Then, the system calls the pre-trained word embedding model to encode and map the word. The model is preferably a deep embedding model trained on general corpus or industry corpus, such as Word2Vec, GloVe, or the BERT series model based on the Transformer architecture. Mapping to high-dimensional vector , and then perform weighted average or aggregation processing on all word vectors to obtain the naming semantic vector: ; in, Indicates The naming string of the camera; (·) indicates a text vectorization function. It can effectively capture the spatial semantics and point logic contained in the naming. For example, although "Teaching Building 1 East Gate" and "Teaching Building 1 West Gate" are in different locations, they belong to the same structural units and therefore have a high degree of semantic similarity.

[0051] In this embodiment, after completing the vector extraction of image features, identity features, location information, orientation information and naming information, the system combines the five types of features to generate a multi-dimensional point feature vector in a unified format. The definition of this combination is as follows: ; in, is the image content feature vector; is the personnel profile feature vector; is the three-dimensional position vector; is the unit direction vector; is the named semantic vector.

[0052] The resultant vector It is the basic data unit for subsequent calculation of similarity between cameras, construction of camera graph structure, and performance of community division. It fully represents the semantic attributes and spatial behavior status of the camera in multiple dimensions.

[0053] It should be further noted that the above splicing operation does not mechanically combine multiple vectors, but rather performs a fusion on the basis of ensuring information complementarity, dimension unity, and scale comparability. During the implementation process, the system performs a normalization process on each feature component, such as using zero-mean unit-variance normalization (Z-Score) or min-max normalization (Min-Max Scaling), to ensure the comparability of data from different sources in the numerical space.

[0054] In addition, to prevent high-dimensional features from introducing redundancy or information masking, in the present invention, a principal component analysis (PCA) or a feature selection mechanism is preferably introduced to perform dimensionality reduction processing or redundancy elimination on the overall vector space, so as to improve the stability of the fusion representation and the subsequent processing efficiency.

[0055] Through the above method, this embodiment realizes the complete conversion from multi-source heterogeneous security data to structured feature vectors and lays a unified, standardized, and operable data foundation for the association calculation based on the graph model and the community analysis under the minimum entropy criterion.

[0056] For step S3, in this embodiment, the step of calculating the comprehensive similarity between cameras based on the feature vectors, where the similarity is weighted and fused by combining position, orientation, and naming information, and constructing an association graph with cameras as nodes and similarities as edge weights, will be described in detail.

[0057] First, according to the aforementioned method for constructing feature vectors, the system integrates the multi-dimensional features of each camera into a unified feature vector through methods such as weighted average and vector splicing . This feature vector synthesizes data in multiple dimensions such as the image content features, personnel identity features, spatial position information, orientation information, and naming information of the camera. Based on these feature vectors, calculating the similarity between cameras is the core of the next step.

[0058] Specifically, in this embodiment, first calculate the Euclidean distance between the feature vectors of each pair of cameras , and this distance is used to measure the similarity between two cameras. The formula is expressed as: ; where and are the feature vectors of camera and camera respectively; and are the corresponding values of each feature dimension; is the dimension of the feature vector.

[0059] Based on this Euclidean distance, the similarity between two cameras can be further calculated , and its formula is: ; This similarity value The larger it is, the more similar the camera and the camera are in their feature vector spaces.

[0060] After obtaining the preliminary similarity, this embodiment introduces the weighted fusion of position, orientation, and naming information to further improve the accuracy and relevance of similarity calculation. Specifically, the system calculates the comprehensive similarity of the position similarity , orientation similarity , and naming similarity of each pair of cameras based on weighted fusion, and its formula is: ; where are the weighted coefficients of position, orientation, and naming information, and are the similarity values calculated based on position, orientation, and naming information respectively.

[0061] Furthermore, in this embodiment, the constructed camera association graph is extended based on the above comprehensive similarity. In this graph, each camera is a node of the graph, and the edge weight between nodes is determined by the similarity value . The final comprehensive similarity combines the similarities of position, orientation, naming, and the original feature vector, and is calculated through weighted fusion: ; where is the weighted coefficient that controls the contribution of each similarity information; is the contribution coefficient that controls the contribution of the original similarity in the final weighted fusion. Through this weighted fusion strategy, the finally obtained similarity value can more comprehensively reflect the association degree between cameras, and improve the accuracy and robustness of the graph model.

[0062] Based on the calculated comprehensive similarity , the system constructs an association graph with cameras as nodes and similarities as edge weights. In this graph, the edge weight represents the similarity between camera and camera . The construction of this graph provides a basis for subsequent community division and minimum entropy optimization analysis, ensuring the effectiveness and accuracy of the analysis process.

[0063] For step S4, first, according to the technical implementation methods discussed above, the system first vectorizes the multi-dimensional features of the cameras to obtain the feature vectors corresponding to each camera. This feature vector synthesizes multiple aspects such as image content, personnel identity, spatial location, orientation information, and naming information, fully characterizing the comprehensive features of each camera. Based on these feature vectors, the next task is to perform clustering analysis on the cameras to construct a reasonable point clustering structure.

[0064] To this end, the system adopts the minimum entropy criterion to perform clustering analysis on the camera positions. The minimum entropy criterion is a method in information theory used to measure the "purity" or "information content" of the clustering results. Entropy is defined as: ; where, is the entropy of the clustering result; is the probability of cluster ; is the number of clusters; The lower the entropy value, the higher the purity of the clustering result, that is, the stronger the similarity of the camera positions within each cluster.

[0065] In the actual clustering process, the system first divides all camera positions into several clusters according to the feature vectors of the cameras , and each cluster contains camera positions with high similarity. Then, by minimizing the entropy , the system optimizes the clustering assignment of the camera positions. Specifically, the system adjusts the division of the clusters so that the similarity of the camera positions within each cluster is as high as possible, while the similarity between the clusters is as low as possible. In this way, the minimized entropy value indicates the quality of the clustering, and the finally obtained camera position clustering structure can effectively reflect the similarity and relationship between the cameras.

[0066] It should be noted that during the clustering analysis process, the minimization of the entropy value does not solely depend on the spatial distribution of the feature vectors, but comprehensively considers the performance of the camera positions in the multi-dimensional feature space. During the process of minimizing the entropy, the system optimizes the boundaries between different clusters to ensure the maximization of the aggregation degree of the camera positions within each cluster in its feature space.

[0067] Through this clustering analysis method based on the minimum entropy criterion, this embodiment can effectively divide the camera positions reasonably according to their feature similarity and construct an accurate point clustering structure. This structure provides important basic data support for subsequent community division, anomaly detection, correlation analysis, etc., and also provides an effective solution for the optimized management and scheduling of camera positions.

[0068] In practical applications, this clustering method is not only applicable to the spatial distribution of camera positions, but also capable of combining feature information from other dimensions for multi-angle clustering optimization. Ultimately, the point clustering structure constructed through the minimum entropy criterion can effectively improve the analysis efficiency and accuracy of the security monitoring system, providing strong support for intelligent management and decision-making.

[0069] For step S5, in this embodiment, the step of performing community division based on minimum entropy on the association graph of each camera in combination with the clustering structure to generate a community structure reflecting the spatial and behavioral relationships between cameras is described in detail.

[0070] First, in the foregoing steps, the feature vectors and comprehensive similarity of the cameras constructed an association graph between the cameras. In this graph, the cameras are nodes, and the edge weights between the nodes are determined by the similarity values between the cameras, which are calculated through the feature vectors of the cameras. At this time, the connection relationships between the cameras have been quantified by the similarity, reflecting the similarity degrees of the cameras in multiple dimensions such as spatial position, orientation, and naming.

[0071] Next, in combination with the clustering structure obtained from the foregoing clustering steps, the system divides the cameras into several clusters to further optimize the connection relationships between the camera positions. The clustering structure of the cameras is obtained through clustering analysis based on the minimum entropy criterion. The purpose of clustering is to group camera positions with similarity into the same group for further analysis of their spatial and behavioral characteristics. Through this clustering structure, the similarity between the cameras is further quantified and grouped, making the relationships between the cameras clearer and having structured information.

[0072] On this basis, the system performs community division based on the minimum entropy criterion. The minimum entropy criterion is used to optimize the community division of the camera positions, so that the cameras within each community are highly consistent in terms of space and behavior, while there are significant differences between different communities. The goal of community division is to minimize the entropy value of each community, so that the similarity of the cameras within the community is as high as possible, and the differences between the communities are as large as possible.

[0073] Entropy The minimization process is as follows: By minimizing the entropy value, the system will optimize the community division of the cameras, ensuring that the similarity between the cameras within each community is as large as possible, while the differences between the communities are as large as possible. Specifically, the minimized entropy value will indicate the clustering effect. The smaller the entropy value, the higher the similarity between the camera positions within each community, and the more obvious the differences between different communities.

[0074] Through this minimum entropy optimization, the system can ensure that the community structure of camera positions can effectively reflect the spatial and behavioral relationships between cameras. Cameras within each community are similar in terms of spatial distribution, orientation, naming, etc., and have significant differences compared to cameras in other communities. This structure can provide important support for subsequent camera resource management, behavior analysis, and monitoring optimization.

[0075] Finally, the community division optimized based on the minimum entropy criterion provides an accurate quantitative basis for the spatial and behavioral relationships between cameras, thereby supporting the efficient operation of the intelligent security system. Through accurate community division, the system can reasonably schedule and manage cameras in a dynamic environment, improving the efficiency and accuracy of the security monitoring system.

[0076] In addition, based on the optimized camera community structure, the system can combine these clustering and community division results with real-time monitoring data for more accurate analysis and prediction. Specifically, the system can predict the next appearance position of personnel or vehicles based on the community where the camera is located and historical monitoring data, and establish a spatio-temporal dataset to improve the analysis accuracy.

[0077] With this information, the system can achieve automatic alarm and risk assessment of abnormal behaviors, further improving the response speed and accuracy of the security monitoring system. In high-risk areas, the system can intelligently schedule the camera positions in the community to ensure that monitoring resources are reasonably allocated at the most critical moments, thereby improving the overall response ability and security guarantee ability of the system.

[0078] Finally, by integrating real-time monitoring data and the optimized community structure, the system can provide powerful services such as abnormal behavior monitoring, trajectory prediction, and intelligent scheduling, significantly improving the overall effectiveness of the security system and providing comprehensive decision-making support for the security system.

[0079] The present invention also provides a security camera community relationship analysis system based on minimum entropy, including: A data acquisition module for acquiring multi-dimensional raw data of each camera; A preprocessing module for preprocessing the acquired multi-dimensional raw data to generate feature vectors in a unified format; A similarity calculation module for calculating the comprehensive similarity between cameras based on the feature vectors, and weighted fusion of the similarity by combining position, orientation, and naming information to construct an association graph with cameras as nodes and similarity as edge weights; A clustering analysis module for performing clustering analysis on each camera position based on the feature vectors and the minimum entropy criterion to construct a position clustering structure; A community division module, which is used to perform community division based on the minimum entropy criterion by combining a clustering structure on the basis of the association graph, so as to generate a community structure reflecting the spatial and behavioral relationships between cameras.

[0080] The system of this embodiment can be used to execute the above method embodiment, and its principle and technical effect are similar, so they will not be elaborated here.

[0081] The present invention also provides a device, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the computer program is executed by the processor, it executes the above method.

[0082] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for analyzing the community relationship of security cameras based on minimum entropy, characterized in that, It includes the following steps: Collect multi-dimensional raw data of each camera; Preprocess the collected multi-dimensional raw data to generate feature vectors in a unified format; Calculate the comprehensive similarity between cameras based on the feature vectors. The similarity is weighted and fused by combining location, orientation, and naming information to construct an association graph with cameras as nodes and similarity as edge weights; Perform clustering analysis on each camera position based on the feature vectors and the minimum entropy criterion to construct a position clustering structure; On the association graph of each camera, combine the clustering structure and perform community division based on the minimum entropy to generate a community structure reflecting the spatial and behavioral relationships between cameras; 2. The method for analyzing the community relationship of a security camera based on minimum entropy according to claim 1, wherein, The step of collecting multi-dimensional raw data of each camera includes: Obtain the captured image data of the camera; Obtain the personal identity file information associated with the image; Obtain the location information of the camera; Obtain the installation orientation information of the camera; Obtain the device naming information of the camera; 3. The method for analyzing the community relationship of a security camera based on minimum entropy according to claim 1, wherein, The step of preprocessing the collected multi-dimensional raw data to generate feature vectors in a unified format includes: Extract features from the image data to obtain an image feature vector ; Encode the personnel file information to obtain the file feature vector ; Perform coordinate transformation on the position information of the camera to obtain a position information vector ; Standardize the installation orientation information of the camera to obtain an orientation vector ; Vectorize the device naming information of the camera to obtain a naming vector ; Concatenate the above vectors into a feature vector in a unified format .

4. The method for analyzing the community relationship of a security camera based on minimum entropy according to claim 1, wherein, The step of calculating the comprehensive similarity between cameras based on the feature vectors includes: Calculate the Euclidean distance between each pair of camera feature vectors , and its formula is: ； Among them, and are the feature vectors of camera and camera respectively; and are the corresponding values of each feature dimension; is the dimension of the feature vector; Calculate the similarity between each pair of cameras according to the Euclidean distance , and its formula is: ； The similarity is weighted and fused by combining the position information, orientation information, and naming information between cameras to obtain a comprehensive similarity , and its formula is: ； Among them, is the weighted coefficient of the position, orientation, and naming information, are the similarity values calculated based on the position, orientation, and naming information, respectively.

5. The method for analyzing the community relationship of a security camera based on minimum entropy according to claim 1, characterized in that The step of weighting and fusing the similarity by combining location, orientation, and naming information to construct an association graph with cameras as nodes and similarity as edge weights includes: According to the calculated comprehensive similarity , combined with the position information similarity of the cameras , the orientation similarity and the naming similarity , the final similarity value between each pair of cameras is obtained through weighted fusion , and the formula is as follows: ； Among them, is a weighting coefficient that controls the contribution of each similarity information; construct an association graph with cameras as nodes and as edge weights.

6. The method for analyzing the community relationship of a security camera based on minimum entropy according to claim 1, wherein The step of performing clustering analysis on each camera position based on the feature vectors and the minimum entropy criterion to construct a position clustering structure includes: Camera-based feature vector , clustering analysis is performed using the minimum entropy criterion, and the formula is: ； Among them, is the entropy of the clustering result; is the probability of cluster ; is the number of clusters; By minimizing the entropy , for the optimal clustering assignment of camera positions, to obtain the clustering structure of camera positions.

7. The method for analyzing the community relationship of a security camera based on minimum entropy according to claim 1, wherein The step of performing community division based on the minimum entropy on the association graph of each camera by combining the clustering structure to generate a community structure reflecting the spatial and behavioral relationships between cameras includes: On the association graph constructed based on the camera feature vectors and comprehensive similarity, combine the clustering structure of the cameras to establish the connection relationship between cameras; Based on the minimum entropy criterion, calculate the entropy of the community division graph and optimize the community division of the cameras by minimizing the entropy value; By minimizing the entropy value, optimize the community division between cameras so that the spatial and behavioral relationships of each community are effectively reflected to obtain the community structure of the cameras; 8. A security camera community relationship analysis system based on minimum entropy, which is applied to the security camera community relationship analysis method based on minimum entropy as described in any one of claims 1-7, and is characterized in that, It includes: A data collection module for collecting multi-dimensional raw data of each camera; A preprocessing module for preprocessing the collected multi-dimensional raw data to generate feature vectors in a unified format; A similarity calculation module for calculating the comprehensive similarity between cameras based on the feature vectors, and weighting and fusing the similarity by combining location, orientation, and naming information to construct an association graph with cameras as nodes and similarity as edge weights; A clustering analysis module for performing clustering analysis on each camera position based on the feature vectors and the minimum entropy criterion to construct a position clustering structure; A community division module for performing community division based on the minimum entropy criterion on the basis of the association graph by combining the clustering structure to generate a community structure reflecting the spatial and behavioral relationships between cameras; 9. An apparatus, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein, When the processor executes the computer program, it implements the method according to any one of claims 1-7.

Images