Intelligent security and protection monitoring method and device based on big data

Abstract

The invention relates to the technical field of security and protection monitoring, and discloses an intelligent security and protection monitoring method and device based on big data, and the method comprises the steps: dispatching a region monitoring video corresponding to a monitoring region, and extracting a region dynamic foreground and a monitoring region background image in the region monitoring video; single-person foreground primitives and multi-person foreground primitives in the foreground primitives are screened out, and scene knowledge elements corresponding to the background image of the monitoring area are analyzed; identifying a single-person form set corresponding to the single-person knowledge element, performing skeletonization processing on the single-person form set to obtain a form skeleton set, analyzing an individual behavior corresponding to the single-person foreground primitive, and analyzing a group behavior corresponding to the multi-person foreground primitive; and formulating a regional security and protection monitoring scheme corresponding to the monitoring region, and executing security and protection monitoring processing corresponding to the monitoring region by using the regional security and protection monitoring scheme to obtain a monitoring result. The main purpose of the invention is to solve the problem of low monitoring efficiency of intelligent security and protection monitoring.

Description

Technical Field

[0001] The present invention relates to an intelligent security monitoring method and device based on big data, belonging to the technical field of security monitoring. Background Art

[0002] In today's era of rapid technological development, security prevention has become an important concern in all areas of society. Intelligent security monitoring systems play an indispensable role in ensuring public safety, corporate safety, and private property safety. With the acceleration of urbanization and the continuous expansion of the scale of various venues, the scope and complexity of security monitoring are also increasing. Traditional security monitoring methods can no longer meet the growing security needs.

[0003] Traditional security monitoring mainly relies on manual surveillance video. Security personnel stare at the screen for a long time, which is prone to fatigue and negligence, resulting in missed reporting of important security events. Even if there are some automated monitoring systems, their functions are relatively simple, and most of them are based on simple motion detection or threshold setting to trigger alarms. For example, it is common to set a specific area in the monitoring screen. When an object enters the area and exceeds the preset motion amplitude or residence time threshold, an alarm is issued. However, this method is easily disturbed by environmental factors, such as light and shadow changes, small animal activities, etc., resulting in a large number of false alarms, which seriously affects the effectiveness of security monitoring. The existing intelligent security monitoring data processing is achieved through some basic data analysis algorithms. The basic functions and data processing requirements that the monitoring system needs to achieve are determined, and conventional data processing tools or software are selected. The software framework is selected according to the requirements of data volume, processing speed, data compatibility, etc., and the relevant data acquisition equipment, storage devices and processing modules are installed. Data input sources such as cameras and sensors and alarm output devices are connected. Specific data processing languages ​​or tools are used to write data processing programs according to preset security rules. Common ones include simple conditional judgment statements, statistical anomaly detection algorithms, etc. The written programs are deployed to the monitoring system, and relevant parameters such as alarm trigger thresholds and data collection frequency are set according to the requirements of the security scene. However, this method is based on limited sample data and simple algorithm models to write data processing programs, and does not combine big data to analyze the video, which makes it less adaptable to complex and changeable security scenes, and thus leads to low monitoring efficiency of intelligent security monitoring. Summary of the invention

[0004] The present invention provides an intelligent security monitoring method and device based on big data, the main purpose of which is to solve the problem of low monitoring efficiency of intelligent security monitoring.

[0005] To achieve the above object, the present invention provides an intelligent security monitoring method based on big data, comprising:

[0006] Acquire a monitoring area to be processed, schedule a regional monitoring video corresponding to the monitoring area, extract a regional dynamic foreground and a monitoring area background image in the regional monitoring video, and analyze foreground image elements in the regional dynamic foreground;

[0007] Screening out the single-person foreground primitive and the multi-person foreground primitive in the foreground primitive, extracting the single-person knowledge primitive and the multi-person knowledge primitive in the single-person foreground primitive and the multi-person foreground primitive, respectively, and analyzing the scene knowledge elements corresponding to the background image of the monitoring area;

[0008] Identify a single-person morphology set corresponding to the single-person knowledge element, perform skeleton processing on the single-person morphology set to obtain a morphology skeleton set, combine the morphology skeleton set and the scene knowledge elements, analyze the individual behavior corresponding to the single-person foreground image element, locate the geometric position of the image element corresponding to the single-person foreground image element based on the background image of the monitoring area, and analyze the group behavior corresponding to the multi-person foreground image element by combining the single-person individual behavior and the image element geometric position;

[0009] In combination with the individual behavior and the group behavior, a regional security monitoring plan corresponding to the monitoring area is formulated, and the regional security monitoring plan is used to execute the security monitoring processing corresponding to the monitoring area to obtain the monitoring result.

[0010] Optionally, extracting the regional dynamic foreground and monitoring area background image in the regional monitoring video includes:

[0011] Performing frame processing on the regional surveillance video to obtain regional video frames;

[0012] Cleaning the regional video frame to obtain a target video frame;

[0013] Calculating the image optical flow field corresponding to the regional video frame, and setting the optical flow threshold corresponding to the regional video frame;

[0014] Combining the image optical flow field and the optical flow threshold, identifying a target foreground image in the target video frame;

[0015] The regional dynamic foreground in the target foreground image is extracted, and the monitoring area background image in the target video frame is extracted in combination with the regional dynamic foreground.

[0016] Optionally, calculating the image optical flow field corresponding to the regional video frame includes:

[0017] Performing grayscale processing on the regional video frame to obtain a grayscale video frame;

[0018] Performing noise reduction processing on the grayscale video frame to obtain a noise-reduced video frame;

[0019] Calculating the image spatial gradient corresponding to the denoised video frame, and calculating the image partial derivatives between adjacent video frames in the denoised video frame;

[0020] Calculate the optical flow velocity vector corresponding to each pixel in the regional video frame by combining the image spatial gradient and the image partial derivative;

[0021] In combination with the optical flow velocity vector, an image optical flow field corresponding to the regional video frame is constructed.

[0022] Optionally, the extracting the single-person knowledge element and the multi-person knowledge element from the single-person foreground image element and the multi-person foreground image element respectively includes:

[0023] Extracting multi-dimensional features from the single-person foreground primitive, performing semantic analysis on the multi-dimensional features, and obtaining a semantic feature subset;

[0024] Calculating the semantic information entropy corresponding to the semantic feature subset, and filtering the semantic feature subset based on the semantic information entropy to obtain a target feature subset;

[0025] Extracting a single person knowledge element corresponding to the single person foreground element from the target feature subset;

[0026] Based on the target feature subset, analyzing group relationship features between the single person foreground primitives;

[0027] The group relationship feature and the target feature subset are fused to obtain multi-person knowledge elements.

[0028] Optionally, analyzing the group relationship features between the single-person foreground primitives based on the target feature subset includes:

[0029] Identify the semantic identifier corresponding to each feature in the target feature subset, and classify the features in the target feature subset based on the semantic identifier to obtain classified sub-features;

[0030] The classification sub-features are vectorized to obtain sub-feature vectors, and the vector correlation between the sub-feature vectors is calculated by the following formula:

[0031]

[0032] Among them, A represents the vector correlation between sub-feature vectors, B a Represents the vector corresponding to the ath sub-feature in the sub-feature vector, B a+1 represents the vector corresponding to the a+1th sub-feature in the sub-feature vector, a and a+1 both represent the sequence numbers corresponding to the sub-feature vector, and t represents the number of sub-feature vectors;

[0033] Based on the vector association degree, determining a central sub-feature among the classification sub-features;

[0034] The relationship features between the central sub-features are analyzed, and based on the relationship features, group relationship features between the single-person foreground primitives are generated.

[0035] Optionally, the analyzing the scene knowledge elements corresponding to the background image of the monitoring area includes:

[0036] Performing region segmentation processing on the background image of the monitoring area to obtain background segmentation regions;

[0037] Identifying a regional entity in the background segmentation region, and extracting regional visual features in the background segmentation region;

[0038] Combining the regional entity and the regional visual features, analyzing the regional semantic category corresponding to the background segmentation region;

[0039] The category coupling relationship between the semantic categories of the region is analyzed, and knowledge integration processing is performed on the category coupling relationship to obtain scene knowledge elements corresponding to the background image of the monitoring area.

[0040] Optionally, the skeletonization of the single person morphology set to obtain a morphology skeleton set includes:

[0041] Performing smoothing processing on each form in the single-person form set to obtain a smoothed form set;

[0042] Identify a shape boundary corresponding to each shape in the set of smooth shapes;

[0043] Based on the morphological boundary, performing distance change processing on each morphology in the smooth morphological set to obtain a morphological transformation distance;

[0044] Marking the local maximum values ​​in the morphological transformation distance to obtain a marked extreme value set;

[0045] The labeled extreme value set is subjected to skeleton refinement processing to obtain a morphological skeleton set.

[0046] Optionally, combining the morphological skeleton set and the scene knowledge elements to analyze the individual behavior corresponding to the single-person foreground primitive includes:

[0047] Identify the skeleton joint points and joint point timing corresponding to each morphology in the morphology skeleton set;

[0048] In combination with the joint point timing, analyzing the node trajectory corresponding to the skeleton joint point;

[0049] Based on the node trajectory, analyzing the morphology trajectory trend corresponding to each morphology in the morphology skeleton set;

[0050] Associating the morphological trajectory trend with the scene knowledge element to obtain a scene-associated trajectory trend;

[0051] Based on the scene-associated trajectory trend, the individual behavior corresponding to the single-person foreground primitive is analyzed.

[0052] Optionally, the formulating a regional security monitoring plan corresponding to the monitoring area by combining the individual behavior and the group behavior includes:

[0053] Based on the individual behavior and the group behavior, the regional surveillance video is segmented and processed for early warning to obtain an early warning slice video;

[0054] Analyze the video dynamic information corresponding to the warning slice video, and generate a slice video summary of the warning slice video based on the video dynamic information;

[0055] Based on the sliced ​​video summary, a regional security monitoring plan corresponding to the monitoring area is formulated.

[0056] An intelligent security monitoring device based on big data, characterized in that the device comprises:

[0057] A foreground image element analysis module is used to obtain a monitoring area to be processed, schedule a regional monitoring video corresponding to the monitoring area, extract a regional dynamic foreground and a monitoring area background image in the regional monitoring video, and analyze the foreground image elements in the regional dynamic foreground;

[0058] A scene knowledge element analysis module, used to screen out single-person foreground primitives and multi-person foreground primitives in the foreground primitives, extract single-person knowledge elements and multi-person knowledge elements in the single-person foreground primitives and the multi-person foreground primitives, respectively, and analyze the scene knowledge elements corresponding to the background image of the monitoring area;

[0059] A group behavior analysis module is used to identify a single-person morphology set corresponding to the single-person knowledge element, perform skeleton processing on the single-person morphology set to obtain a morphology skeleton set, combine the morphology skeleton set with the scene knowledge elements, analyze the individual behavior corresponding to the single-person foreground image element, locate the geometric position of the image element corresponding to the single-person foreground image element based on the background image of the monitoring area, and analyze the group behavior corresponding to the multi-person foreground image element by combining the single-person individual behavior and the image element geometric position;

[0060] The security monitoring processing module is used to formulate a regional security monitoring plan corresponding to the monitoring area in combination with the individual behavior and the group behavior, and to perform security monitoring processing corresponding to the monitoring area using the regional security monitoring plan to obtain monitoring results.

[0061] Compared with the problems described in the background technology, the present invention extracts the regional dynamic foreground and the monitoring area background image in the regional monitoring video, which helps to more clearly analyze the key dynamic situations in the video, distinguish the foreground activity subject and the background environment, and facilitates the subsequent precise observation and research on the target behavior, and is also conducive to improving the accuracy and efficiency of the interpretation of the monitoring video content. The present invention can obtain valuable information corresponding to the single-person foreground image element and the multi-person foreground image element, such as identity, appearance and gender characteristics, by respectively extracting the single-person knowledge element and the multi-person knowledge element in the single-person foreground image element and the multi-person foreground image element. The present invention uses a preset central axis change algorithm to The skeletonization of the single-person morphology set can simplify the expression of the single-person morphology set, highlight the key structural features, facilitate the subsequent accurate recognition and analysis of human posture, and effectively reduce data redundancy. The present invention combines the individual behavior and the group behavior to formulate a regional security monitoring plan corresponding to the monitoring area, which can accurately predict potential dangers and deploy preventive measures in advance, such as timely deployment of security forces when people gather and behave abnormally, reasonably arrange the layout of monitoring equipment and monitoring priorities according to behavioral hotspots and risk levels, and use the regional security monitoring plan to execute the security monitoring processing corresponding to the monitoring area, greatly improving the monitoring efficiency of security monitoring. Therefore, the intelligent security monitoring method and device based on big data proposed in the present invention can improve the monitoring efficiency of the intelligent security monitoring system. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 A schematic diagram of a process flow of an intelligent security monitoring method based on big data provided by an embodiment of the present invention;

[0063] Figure 2 A functional module diagram for implementing the big data-based intelligent security monitoring device provided in one embodiment of the present invention.

[0064] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0065] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0066] The embodiment of the present application provides an intelligent security monitoring method based on big data. The execution subject of the intelligent security monitoring method based on big data includes but is not limited to at least one of the electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided by the embodiment of the present application. In other words, the intelligent security monitoring method based on big data can be executed by software or hardware installed on a terminal device or a server device. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.

[0067] Embodiment 1:

[0068] Reference Figure 1 FIG. 1 is a flow chart of an intelligent security monitoring method based on big data provided by an embodiment of the present invention. In this embodiment, the intelligent security monitoring method based on big data includes:

[0069] S1. Acquire a monitoring area to be processed, schedule a regional monitoring video corresponding to the monitoring area, extract a regional dynamic foreground and a monitoring area background image in the regional monitoring video, and analyze foreground graphics elements in the regional dynamic foreground.

[0070] The present invention helps to more clearly analyze the key dynamic situations in the video, distinguish the foreground active subject from the background environment, and facilitate subsequent precise observation and research on target behaviors, etc. by extracting the regional dynamic foreground and the monitoring area background image in the regional monitoring video, and is also conducive to improving the accuracy and efficiency of interpreting the monitoring video content. It should be explained that the monitoring area is the destination of the security monitoring, the regional monitoring video is the regional recorded video data corresponding to the monitoring area, the regional dynamic foreground is the active elements such as people and vehicles that are moving and in a relatively forward visual position in the regional monitoring video, and the monitoring area background image is a relatively static fixed environment picture in the regional monitoring video that serves as a backdrop for the active elements; further, the regional monitoring video corresponding to the monitoring area can be obtained by scheduling in the data storage library of the regional monitoring device.

[0071] In detail, the extracting of the regional dynamic foreground and the monitoring area background image in the regional monitoring video includes:

[0072] Performing frame processing on the regional surveillance video to obtain regional video frames;

[0073] Cleaning the regional video frame to obtain a target video frame;

[0074] Calculating the image optical flow field corresponding to the regional video frame, and setting the optical flow threshold corresponding to the regional video frame;

[0075] Combining the image optical flow field and the optical flow threshold, identifying a target foreground image in the target video frame;

[0076] The regional dynamic foreground in the target foreground image is extracted, and the monitoring area background image in the target video frame is extracted in combination with the regional dynamic foreground.

[0077] It should be explained that the regional video frame is a single frame extracted from the regional surveillance video, the target video frame is a related frame that has been specifically screened or specified from the regional video frame, the image optical flow field is a vector field corresponding to the regional video frame that reflects the motion trend and speed of pixels, the optical flow threshold is a numerical standard corresponding to the regional video frame for distinguishing whether a pixel is in a motion state, and the target foreground image is the part of the image containing the moving object that is divided in the target video frame by a relevant algorithm based on the motion situation, etc.

[0078] Furthermore, the regional surveillance video can be segmented by a segmentation tool to obtain regional video frames, such as the PotPlayer tool; the regional video frames can be cleaned by a video deduplication device to obtain target video frames; to set the optical flow threshold corresponding to the regional video frame, firstly, it is necessary to observe the typical moving objects in the regional video frame to understand the optical flow amplitude range corresponding to its normal movement speed, and then, based on actual needs and noise conditions, set an optical flow amplitude that can distinguish between moving objects and background noise as a threshold; combine the image optical flow field and the optical flow threshold to identify the target video frame. Target foreground image, for each pixel in the target video frame, check its optical flow amplitude in the image optical flow field, if the optical flow amplitude is greater than the set optical flow threshold, mark the pixel as a foreground pixel, and the area formed by all foreground pixels is the target foreground image; the regional dynamic foreground in the target foreground image can be extracted by the connected region analysis method; combined with the regional dynamic foreground, extract the monitoring area background image in the target video frame, remove the regional dynamic foreground part from the target video frame, and the remaining part is the monitoring area background image, and this process can be achieved by operations such as image subtraction.

[0079] Further, as an optional embodiment of the present invention, the calculating the image optical flow field corresponding to the regional video frame includes:

[0080] Performing grayscale processing on the regional video frame to obtain a grayscale video frame;

[0081] Performing noise reduction processing on the grayscale video frame to obtain a noise-reduced video frame;

[0082] Calculating the image spatial gradient corresponding to the denoised video frame, and calculating the image partial derivatives between adjacent video frames in the denoised video frame;

[0083] Calculate the optical flow velocity vector corresponding to each pixel in the regional video frame by combining the image spatial gradient and the image partial derivative;

[0084] In combination with the optical flow velocity vector, an image optical flow field corresponding to the regional video frame is constructed.

[0085] It should be explained that the grayscale video frame is a video frame that only contains grayscale information after the regional video frame is grayscaled; the denoised video frame is a video frame after the grayscale video frame is denoised; the image spatial gradient is the information corresponding to the denoised video frame for describing the rate of change of the image grayscale in the horizontal and vertical directions; the image partial derivative is a quantitative representation of the pixel grayscale change between adjacent video frames in the denoised video frame; the optical flow velocity vector is a vector corresponding to each pixel point in the regional video frame for representing the magnitude and direction of its movement speed.

[0086] Furthermore, the grayscale processing of the regional video frame can be achieved by the average method, the average value of the three channels of the color pixel is calculated as the grayscale value, and the grayscale processing is performed using the grayscale value; the denoising processing of the grayscale video frame can be achieved by a low-pass filter; the image spatial gradient corresponding to the denoised video frame can be calculated by a convolution operation method such as the Sobel operator; the grayscale difference between adjacent video frames in the denoised video frame is calculated to obtain the image partial derivative; combining the image spatial gradient and the image partial derivative, the optical flow velocity vector corresponding to each pixel in the regional video frame is calculated, and according to the optical flow constraint equation Ixu+Iyv+It=0 (u and v are the horizontal and vertical optical flow velocity components of the image spatial gradient, respectively), a set of equations is constructed within a set window range according to the least squares principle, and the u and v values ​​corresponding to each pixel are solved by matrix operations to obtain the optical flow velocity vector (u, v) of each pixel; the optical flow velocity vectors are combined together to obtain the image optical flow field corresponding to the regional video frame.

[0087] The present invention can obtain the basic foreground units in the dynamic foreground of the region by analyzing the foreground primitives in the dynamic foreground of the region, thereby facilitating the subsequent related analysis and processing of single-person foreground primitives and multi-person foreground primitives. It should be explained that the foreground primitives are the basic foreground units in the dynamic foreground of the region. Further, the analysis of the foreground primitives in the dynamic foreground of the region can be achieved by the Canny operator.

[0088] S2. Filter out the single-person foreground primitives and the multi-person foreground primitives in the foreground primitives, extract the single-person knowledge primitives and the multi-person knowledge primitives in the single-person foreground primitives and the multi-person foreground primitives respectively, and analyze the scene knowledge elements corresponding to the background image of the monitoring area.

[0089] The present invention can obtain valuable information corresponding to the single-person foreground primitive and the multi-person foreground primitive, such as identity, appearance and gender characteristics, by respectively extracting the single-person knowledge primitive and the multi-person knowledge primitive in the single-person foreground primitive and the multi-person foreground primitive. It should be explained that the single-person foreground primitive is the part of the foreground primitive that represents a single individual, and the multi-person foreground primitive is the part of the foreground primitive that contains multiple individuals; the single-person knowledge primitive is the knowledge element that embodies the relevant characteristics of a single person extracted from the single-person foreground primitive, and the multi-person knowledge primitive is the knowledge element that covers the relationships between multiple people and group characteristics excavated from the multi-person foreground primitive. Furthermore, the screening of the single-person foreground primitive and the multi-person foreground primitive in the foreground primitive can be achieved through a target detection algorithm.

[0090] Specifically, extracting the single-person knowledge element and the multi-person knowledge element from the single-person foreground element and the multi-person foreground element respectively includes:

[0091] Extracting multi-dimensional features from the single-person foreground primitive, performing semantic analysis on the multi-dimensional features, and obtaining a semantic feature subset;

[0092] Calculating the semantic information entropy corresponding to the semantic feature subset, and filtering the semantic feature subset based on the semantic information entropy to obtain a target feature subset;

[0093] Extracting a single person knowledge element corresponding to the single person foreground element from the target feature subset;

[0094] Based on the target feature subset, analyzing group relationship features between the single person foreground primitives;

[0095] The group relationship feature and the target feature subset are fused to obtain multi-person knowledge elements.

[0096] It should be explained that the multi-dimensional features are feature information covering contour, texture, color distribution and other aspects obtained from the single foreground primitive through feature extraction technology; the semantic feature subset is a feature set with actual semantic meanings such as age, gender, clothing and other aspects selected after semantic analysis of the multi-dimensional features; the semantic information entropy represents the metric value corresponding to the semantic feature subset reflecting its uncertainty, amount of information and other conditions; the target feature subset is a key feature subset selected and determined by the semantic feature subset according to certain rules based on the semantic information entropy; the group relationship feature is the feature information between the single foreground primitives that reflects the group relationship such as the relative positions and behavioral interactions of the characters.

[0097] Furthermore, the multi-dimensional features in the single-person foreground primitive can be extracted through a convolutional neural network; the multi-dimensional features can be semantically analyzed through a semantic analysis method to obtain a semantic feature subset; the semantic information entropy corresponding to the semantic feature subset can be calculated through a Shannon entropy algorithm, and based on the numerical value of the semantic information entropy, the semantic feature with the largest semantic information entropy of similar semantics in the semantic feature subset is selected, and the other semantic features are filtered to obtain a target feature subset; the single-person knowledge element corresponding to the single-person foreground primitive is extracted from the target feature subset, and the scattered different semantic features extracted from the image are comprehensively sorted out to finally form a single-person knowledge element that can fully describe the individual key information corresponding to the single-person foreground primitive; the group relationship feature and the target feature subset are fused to obtain a multi-person knowledge element.

[0098] Further, as an optional embodiment of the present invention, the analyzing the group relationship characteristics between the single-person foreground primitives based on the target feature subset includes:

[0099] Identify the semantic identifier corresponding to each feature in the target feature subset, and classify the features in the target feature subset based on the semantic identifier to obtain classified sub-features;

[0100] The classification sub-features are vectorized to obtain sub-feature vectors, and the vector correlation between the sub-feature vectors is calculated by the following formula:

[0101]

[0102] Among them, A represents the vector correlation between sub-feature vectors, B a Represents the vector corresponding to the ath sub-feature in the sub-feature vector, B a+1 represents the vector corresponding to the a+1th sub-feature in the sub-feature vector, a and a+1 both represent the sequence numbers corresponding to the sub-feature vector, and t represents the number of sub-feature vectors;

[0103] Based on the vector association degree, determining a central sub-feature among the classification sub-features;

[0104] The relationship features between the central sub-features are analyzed, and based on the relationship features, group relationship features between the single-person foreground primitives are generated.

[0105] It should be explained that the semantic identifier is a mark corresponding to each feature in the target feature subset, which is used to clarify its semantic meaning; the classification sub-feature is the feature under different categories obtained after the features in the target feature subset are classified according to the semantic identifier; the sub-feature vector is a vector form formed by vectorization of the classification sub-feature for quantitatively representing the classification feature; the vector correlation degree represents the quantitative value of the similarity or correlation degree between the sub-feature vectors; the central sub-feature is a typical feature in the classification sub-feature that represents the overall situation of the corresponding category; the relationship feature is the feature information between the central sub-features that reflects the mutual connection and difference of features of different categories.

[0106] Furthermore, the semantic identifier corresponding to each feature in the target feature subset can be identified through a deep learning model (such as a pre-trained word vector model); the identifier similarity between the semantic identifiers is calculated, and the features in the target feature subset are classified according to the identifier similarity to obtain classification sub-features; the classification sub-features can be vectorized through a feature hashing algorithm to obtain a sub-feature vector; the vector correlation of each sub-feature in the classification sub-features is added together, and the sub-feature with the largest result value is used as the central sub-feature in the classification sub-features; the relationship features between the central sub-features are analyzed, and the similarities and differences between the central sub-features are determined by calculating the vector distance (such as Euclidean distance, cosine distance, etc.) of the central sub-features, and then whether there is an interactive trend between the central sub-features, thereby analyzing the corresponding relationship features, merging the relationship features, and generating group relationship features between the single-person foreground graphics elements.

[0107] By analyzing the scene knowledge elements corresponding to the background image of the monitoring area, the present invention can obtain key information about the scene conditions in the background image of the monitoring area, thereby providing a basis for the subsequent analysis of the individual behavior corresponding to the single-person foreground image element. It should be explained that the scene knowledge elements are key information units in the background image of the monitoring area.

[0108] In detail, the analysis of the scene knowledge elements corresponding to the background image of the monitoring area includes:

[0109] Performing region segmentation processing on the background image of the monitoring area to obtain background segmentation regions;

[0110] Identifying a regional entity in the background segmentation region, and extracting regional visual features in the background segmentation region;

[0111] Combining the regional entity and the regional visual features, analyzing the regional semantic category corresponding to the background segmentation region;

[0112] The category coupling relationship between the semantic categories of the region is analyzed, and knowledge integration processing is performed on the category coupling relationship to obtain scene knowledge elements corresponding to the background image of the monitoring area.

[0113] It should be explained that the background segmentation area is the different sub-areas divided after the background image of the monitoring area is processed by a specific segmentation algorithm; the area entity is a specific object or object in the background segmentation area that has independent physical meaning and can be identified and distinguished; the area visual feature is the characteristic expression in the background segmentation area that can be reflected by visual elements such as color, shape, texture, etc.; the area semantic category is the category affiliation determined by analysis and matching at the semantic level corresponding to the background segmentation area; the category coupling relationship is the internal connection of mutual correlation and mutual influence between the area semantic categories.

[0114] Furthermore, the background image of the monitoring area can be segmented by a region growing method to obtain a background segmentation area; the regional entity in the background segmentation area can be identified by a target detection algorithm (such as the YOLO series, FastR-CNN, etc.); the regional visual features in the background segmentation area can be extracted by a feature extraction algorithm, such as a gray level co-occurrence matrix and an HSV model; the regional semantic category corresponding to the background segmentation area is analyzed by combining the regional entity and the regional visual features, such as if the regional entity is circular and the regional visual features are displayed as red, high saturation, and medium brightness, and the information can be combined to analyze its possible The corresponding regional semantic category is the red light in the traffic light; the category coupling relationship between the regional semantic categories can be analyzed by constructing a semantic category association graph; the category coupling relationship is processed by knowledge integration to obtain the scene knowledge elements corresponding to the background image of the monitoring area. The specific process is: first, the logical context and interdependence pattern in the category coupling relationship are sorted out, the scattered related information is connected in series, and then the key information is refined, the redundant details are removed, and the coupling associations about regional layout, functional coordination, etc. are integrated into one, and finally the scene knowledge elements such as regional distribution pattern and functional interaction logic are formed, so as to comprehensively present the connotation of the background image of the monitoring area.

[0115] S3. Identify the single-person morphology set corresponding to the single-person knowledge element, skeletonize the single-person morphology set to obtain a morphology skeleton set, combine the morphology skeleton set and the scene knowledge elements, analyze the individual behavior corresponding to the single-person foreground primitive, locate the primitive geometric position corresponding to the single-person foreground primitive based on the background image of the monitoring area, and analyze the group behavior corresponding to the multi-person foreground primitive by combining the single-person individual behavior and the primitive geometric position.

[0116] The present invention uses a preset medial axis change algorithm to perform skeletonization processing on the single-person form set, which can simplify the expression of the single-person form set, highlight key structural features, and facilitate subsequent accurate recognition and analysis of human postures. At the same time, it can effectively reduce data redundancy. It should be explained that the single-person form set is a set of human forms corresponding to the single-person knowledge element, and the form skeleton set is a set of skeleton information corresponding to the single-person form set that concisely and accurately presents the structure of various parts of the human body. Furthermore, the recognition of the single-person form set corresponding to the single-person knowledge element can be achieved through a support vector machine method.

[0117] In detail, the skeletonization of the single person morphology set to obtain a morphology skeleton set includes:

[0118] Performing smoothing processing on each form in the single-person form set to obtain a smoothed form set;

[0119] Identify a shape boundary corresponding to each shape in the set of smooth shapes;

[0120] Based on the morphological boundary, performing distance change processing on each morphology in the smooth morphological set to obtain a morphological transformation distance;

[0121] Marking the local maximum values ​​in the morphological transformation distance to obtain a marked extreme value set;

[0122] The labeled extreme value set is subjected to skeleton refinement processing to obtain a morphological skeleton set.

[0123] It should be explained that the smoothed shape set is a shape set in which each shape in the single-person shape set becomes smoother after noise reduction processing, which is conducive to subsequent analysis operations. The shape boundary is the contour boundary corresponding to each shape in the smoothed shape set that distinguishes the shape from the surrounding area (such as the virtual background when the image boundary can be regarded as the background). The shape transformation distance is a set of distance information from each foreground pixel inside the smoothed shape set to the virtual background calculated by distance transformation (using the image boundary as the virtual background, etc.) for each shape in the smoothed shape set. The marked extreme value set is the corresponding set of local maximum values ​​in the shape transformation distance, that is, the set formed by marking those points with the maximum distance value in the local area in the shape transformation distance, which is used for subsequent skeleton extraction and other operations.

[0124] Further, each form in the single-person form set can be smoothed by a median filtering algorithm to obtain a smoothed form set; the form boundary corresponding to each form in the smoothed form set can be identified by an edge detection algorithm, such as the Canny edge detection algorithm mentioned above; the distance value between the pixel point of each form in the smoothed form set and the form boundary can be calculated by a Euclidean distance algorithm, thereby realizing distance change processing and obtaining a form transformation distance; the local maximum in the form transformation distance can be marked by a local maximum search algorithm to obtain a marked extreme value set; the marked extreme value set can be skeletonized by a Zhang-Suen refinement algorithm to obtain a form skeleton set. The processing process of the algorithm is as follows: the Zhang-Suen refinement algorithm starts from the marked extreme value set (preliminary skeleton point), and it determines whether to retain the point by checking the 8-neighborhood pixel situation of each point, such as the connectivity of the pixels around the point, whether it is a boundary point, and other conditions. In the human limbs, it iterates point by point to remove redundant pixels, so that the skeleton is gradually refined to a single pixel width, and finally accurately presents the shape of the bone lines of the human limbs and other parts.

[0125] The present invention combines the morphological skeleton set and the scene knowledge elements to analyze the individual behavior corresponding to the single-person foreground primitive, thereby accurately capturing the human body movement details and environmental associations of the single-person foreground primitive, effectively improving the accuracy of behavior recognition, and better understanding the intention and logic of individual behavior in specific scenes. It should be explained that the individual behavior is the character's action performance, action intention and interaction method with the environment inferred by a comprehensive analysis of the morphological skeleton set corresponding to the single-person foreground primitive combined with the scene knowledge elements.

[0126] In detail, the combining of the morphological skeleton set and the scene knowledge elements to analyze the individual behavior corresponding to the single person foreground primitive includes:

[0127] Identify the skeleton joint points and joint point timing corresponding to each morphology in the morphology skeleton set;

[0128] In combination with the joint point timing, analyzing the node trajectory corresponding to the skeleton joint point;

[0129] Based on the node trajectory, analyzing the morphology trajectory trend corresponding to each morphology in the morphology skeleton set;

[0130] Associating the morphological trajectory trend with the scene knowledge element to obtain a scene-associated trajectory trend;

[0131] Based on the scene-associated trajectory trend, the individual behavior corresponding to the single-person foreground primitive is analyzed.

[0132] It should be explained that the skeleton joints and the joint timing are respectively the connection points of the key parts of the human body corresponding to each morphology in the morphological skeleton set and the sequence of changes of these joints over time; the node trajectory is the trajectory route formed by the position movement of the skeleton joints in a certain time series; the morphological trajectory trend is the overall movement tendency and situation presented by the joint motion trajectory of each morphology in the morphological skeleton set; the scene-associated trajectory trend is the individual behavior movement tendency and development situation combined with the characteristics of a specific scene, which is reflected after the morphological trajectory trend and the scene knowledge elements are processed through association matching.

[0133] Furthermore, the posture estimation model can be used to identify the skeleton joints and joint timing corresponding to each morphology in the morphological skeleton set; combined with the joint timing, the node trajectories corresponding to the skeleton joints are analyzed. For example, the joint timing shows that the human hip joint first translates to the left and then lifts up at consecutive time points. Based on this, it can be analyzed that its node trajectory is a horizontal displacement from the initial position to the left and then forms an arc trajectory upward, thereby clearly depicting the change in the motion path of the skeleton joint within a specific period of time; based on the node trajectory, the morphological trajectory trend corresponding to each morphology in the morphological skeleton set is analyzed. If the node trajectory shows that the leg joint The points move up and down, forward and backward in a regular manner, and the corresponding morphological trajectory trend is the behavioral movement situation; the morphological trajectory trend and the scene knowledge elements are associated with each other to obtain the scene-associated trajectory trend; for example, in a shopping mall scene, the morphological trajectory trend is to walk towards a certain store. Combined with scene knowledge elements such as store type, it can be inferred that its scene-associated trajectory trend is to go into the store to shop or browse goods; based on the scene-associated trajectory trend, the individual behavior corresponding to the single-person foreground element is analyzed. If the scene-associated trajectory trend is towards the restricted area and has a wandering situation, it can be determined that the individual behavior corresponding to the single-person foreground element may be an illegal intrusion or with bad intentions.

[0134] The present invention locates the geometric position of the graphic element corresponding to the single-person foreground graphic element based on the background image of the monitoring area, and can obtain the regional position of the single-person foreground graphic element in the background image of the monitoring area. In combination with the single-person individual behavior and the graphic element geometric position, the group behavior corresponding to the multi-person foreground graphic element is analyzed, and the interaction mode, organizational structure and movement trend of the group can be accurately grasped. It should be explained that the graphic element geometric position is the actual spatial position corresponding to the single-person foreground graphic element, and the group behavior is a description of the interaction mode, organizational structure and movement trend between the multi-person foreground graphic elements. Further, based on the background image of the monitoring area, the graphic element geometric position corresponding to the single-person foreground graphic element can be located by an image difference algorithm; in combination with the single-person individual behavior and the graphic element geometric position, the group behavior corresponding to the multi-person foreground graphic element can be analyzed. For example, in a shopping mall monitoring scene, if the single-person individual behavior has a tendency to gather and the graphic element geometric position shows that multiple people are gradually approaching a specific area, it can be analyzed that the group behavior may be watching or collectively waiting for promotional activities in the area.

[0135] S4. Combining the individual behavior and the group behavior, formulate a regional security monitoring plan corresponding to the monitoring area, and use the regional security monitoring plan to execute security monitoring processing corresponding to the monitoring area to obtain monitoring results.

[0136] The present invention formulates a regional security monitoring plan corresponding to the monitoring area by combining the individual behavior and the group behavior, which can accurately predict potential dangers and deploy preventive measures in advance, such as timely deploying security forces when people gather and behave abnormally, and reasonably arranging the layout of monitoring equipment and monitoring priorities according to behavioral hotspots and risk levels. The regional security monitoring plan is used to execute the security monitoring processing corresponding to the monitoring area, greatly improving the monitoring efficiency of security monitoring. It should be explained that the regional security monitoring plan is a plan corresponding to the monitoring area that includes monitoring resource allocation, security measures planning, emergency response strategies and other aspects based on a comprehensive analysis of individual behavior and group behavior therein.

[0137] In detail, the combining of the individual behavior and the group behavior to formulate a regional security monitoring plan corresponding to the monitoring area includes:

[0138] Based on the individual behavior and the group behavior, the regional surveillance video is segmented and processed for early warning to obtain an early warning slice video;

[0139] Analyze the video dynamic information corresponding to the warning slice video, and generate a slice video summary of the warning slice video based on the video dynamic information;

[0140] Based on the sliced ​​video summary, a regional security monitoring plan corresponding to the monitoring area is formulated.

[0141] It should be explained that the warning slice video is the video clip obtained by segmenting the regional surveillance video according to corresponding standards by identifying the single or multi-person activities with potential safety risks and abnormal performance in the individual behavior and the group behavior; the video dynamic information is the specific content information corresponding to the warning slice video covering the changes over time in many aspects such as character movements, object movement trajectories, scene changes, etc.; the slice video summary is the summary content generated after the video dynamic information is processed through time compression, trajectory simplification, etc., which can concisely present the core content so as to quickly understand the key situations of the video.

[0142] Furthermore, the surveillance video of the area can be segmented and warned through a video management system (VMS) integrated plug-in to obtain a warning slice video; the video dynamic information corresponding to the warning slice video can be analyzed through computer vision technology; based on the slice video summary, a regional security monitoring plan corresponding to the monitoring area is formulated. If the slice video summary shows that a certain area frequently has suspected theft by a single person during a specific period of time, the security monitoring plan may include adding hidden monitoring equipment to the area, increasing the frequency of security patrols, and setting up an intelligent item monitoring system; if the summary shows a group gathering and signs of riots, a crowd evacuation route plan can be planned, anti-riot security forces can be deployed, and regional isolation facilities can be set up to ensure the safety of the monitored area in a targeted manner.

[0143] Compared with the problems described in the background technology, the present invention extracts the regional dynamic foreground and the monitoring area background image in the regional monitoring video, which helps to more clearly analyze the key dynamic situations in the video, distinguish the foreground activity subject and the background environment, and facilitates the subsequent precise observation and research on the target behavior, and is also conducive to improving the accuracy and efficiency of the interpretation of the monitoring video content. The present invention can obtain valuable information corresponding to the single-person foreground image element and the multi-person foreground image element, such as identity, appearance and gender characteristics, by respectively extracting the single-person knowledge element and the multi-person knowledge element in the single-person foreground image element and the multi-person foreground image element. The present invention uses a preset central axis change algorithm to The skeletonization of the single-person morphology set can simplify the expression of the single-person morphology set, highlight the key structural features, facilitate the subsequent accurate recognition and analysis of human posture, and effectively reduce data redundancy. The present invention combines the individual behavior and the group behavior to formulate a regional security monitoring plan corresponding to the monitoring area, which can accurately predict potential dangers and deploy preventive measures in advance, such as timely deployment of security forces when people gather and behave abnormally, reasonably arrange the layout of monitoring equipment and monitoring priorities according to behavioral hotspots and risk levels, and use the regional security monitoring plan to execute the security monitoring processing corresponding to the monitoring area, greatly improving the monitoring efficiency of security monitoring. Therefore, the intelligent security monitoring method based on big data proposed in the present invention improves the monitoring efficiency of the intelligent security monitoring system.

[0144] Embodiment 2:

[0145] like Figure 2 , which is a functional module diagram of an intelligent security monitoring device based on big data provided by one embodiment of the present invention.

[0146] The intelligent security monitoring device 100 based on big data described in the present invention can be installed in an electronic device. According to the functions implemented, the intelligent security monitoring device 100 based on big data can include a foreground element analysis module 101, a scene knowledge element analysis module 102, a group behavior analysis module 103 and a security monitoring processing module 104. The module described in the present invention can also be called a unit, which refers to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, which are stored in the memory of the electronic device.

[0147] In this embodiment, the functions of each module / unit are as follows:

[0148] The foreground image element analysis module 101 is used to obtain a monitoring area to be processed, schedule a regional monitoring video corresponding to the monitoring area, extract a regional dynamic foreground and a monitoring area background image in the regional monitoring video, and analyze the foreground image elements in the regional dynamic foreground;

[0149] The scene knowledge element analysis module 102 is used to screen out the single-person foreground element and the multi-person foreground element in the foreground element, extract the single-person knowledge element and the multi-person knowledge element in the single-person foreground element and the multi-person foreground element, respectively, and analyze the scene knowledge elements corresponding to the background image of the monitoring area;

[0150] The group behavior analysis module 103 is used to identify the single-person morphology set corresponding to the single-person knowledge element, perform skeleton processing on the single-person morphology set to obtain a morphology skeleton set, combine the morphology skeleton set and the scene knowledge elements to analyze the individual behavior corresponding to the single-person foreground primitive, locate the primitive geometric position corresponding to the single-person foreground primitive based on the background image of the monitoring area, and analyze the group behavior corresponding to the multi-person foreground primitive by combining the single-person individual behavior and the primitive geometric position;

[0151] The security monitoring processing module 104 is used to formulate a regional security monitoring plan corresponding to the monitoring area in combination with the individual behavior and the group behavior, and to perform security monitoring processing corresponding to the monitoring area using the regional security monitoring plan to obtain monitoring results.

[0152] In detail, each module described in the intelligent security monitoring device 100 based on big data described in the embodiment of the present application is used in the same manner as described above. Figure 1 The same technical means are used as the intelligent security monitoring method based on big data described in , and can produce the same technical effects, so I will not go into details here.

[0153] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0154] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. An intelligent security monitoring method based on big data, characterized in that: The method comprises: Acquire a monitoring area to be processed, schedule a regional monitoring video corresponding to the monitoring area, extract a regional dynamic foreground and a monitoring area background image in the regional monitoring video, and analyze foreground image elements in the regional dynamic foreground; Screening out the single-person foreground primitive and the multi-person foreground primitive in the foreground primitive, extracting the single-person knowledge primitive and the multi-person knowledge primitive in the single-person foreground primitive and the multi-person foreground primitive, respectively, and analyzing the scene knowledge elements corresponding to the background image of the monitoring area; Identify a single-person morphology set corresponding to the single-person knowledge element, perform skeleton processing on the single-person morphology set to obtain a morphology skeleton set, combine the morphology skeleton set and the scene knowledge elements, analyze the individual behavior corresponding to the single-person foreground image element, locate the geometric position of the image element corresponding to the single-person foreground image element based on the background image of the monitoring area, and analyze the group behavior corresponding to the multi-person foreground image element by combining the single-person individual behavior and the image element geometric position; In combination with the individual behavior and the group behavior, a regional security monitoring plan corresponding to the monitoring area is formulated, and the regional security monitoring plan is used to execute the security monitoring processing corresponding to the monitoring area to obtain the monitoring result.

2. The intelligent security monitoring method based on big data according to claim 1, characterized in that: The extracting of the regional dynamic foreground and the monitoring area background image in the regional monitoring video includes: Performing frame processing on the regional surveillance video to obtain regional video frames; Cleaning the regional video frame to obtain a target video frame; Calculating the image optical flow field corresponding to the regional video frame, and setting the optical flow threshold corresponding to the regional video frame; Combining the image optical flow field and the optical flow threshold, identifying a target foreground image in the target video frame; The regional dynamic foreground in the target foreground image is extracted, and the monitoring area background image in the target video frame is extracted in combination with the regional dynamic foreground.

3. The intelligent security monitoring method based on big data as claimed in claim 2, characterized in that: The calculating the image optical flow field corresponding to the regional video frame includes: Performing grayscale processing on the regional video frame to obtain a grayscale video frame; Performing noise reduction processing on the grayscale video frame to obtain a noise-reduced video frame; Calculating the image spatial gradient corresponding to the denoised video frame, and calculating the image partial derivatives between adjacent video frames in the denoised video frame; Calculate the optical flow velocity vector corresponding to each pixel in the regional video frame by combining the image spatial gradient and the image partial derivative; In combination with the optical flow velocity vector, an image optical flow field corresponding to the regional video frame is constructed.

4. The intelligent security monitoring method based on big data according to claim 1, characterized in that: The extracting of the single-person knowledge element and the multi-person knowledge element from the single-person foreground image element and the multi-person foreground image element respectively includes: Extracting multi-dimensional features from the single-person foreground primitive, performing semantic analysis on the multi-dimensional features, and obtaining a semantic feature subset; Calculating the semantic information entropy corresponding to the semantic feature subset, and filtering the semantic feature subset based on the semantic information entropy to obtain a target feature subset; Extracting a single person knowledge element corresponding to the single person foreground element from the target feature subset; Based on the target feature subset, analyzing group relationship features between the single person foreground primitives; The group relationship feature and the target feature subset are fused to obtain multi-person knowledge elements.

5. The intelligent security monitoring method based on big data as claimed in claim 4, characterized in that: The analyzing the group relationship features between the single person foreground primitives based on the target feature subset includes: Identify the semantic identifier corresponding to each feature in the target feature subset, and classify the features in the target feature subset based on the semantic identifier to obtain classified sub-features; The classification sub-features are vectorized to obtain sub-feature vectors, and the vector correlation between the sub-feature vectors is calculated by the following formula: Among them, A represents the vector correlation between sub-feature vectors, B a represents the vector corresponding to the ath sub-feature in the sub-feature vector, B a+1 represents the vector corresponding to the a+1th sub-feature in the sub-feature vector, a and a+1 both represent the sequence numbers corresponding to the sub-feature vector, and t represents the number of sub-feature vectors; Based on the vector association degree, determining a central sub-feature among the classification sub-features; The relationship features between the central sub-features are analyzed, and based on the relationship features, group relationship features between the single-person foreground primitives are generated.

6. The intelligent security monitoring method based on big data according to claim 1, characterized in that: The analyzing the scene knowledge elements corresponding to the background image of the monitoring area includes: Performing region segmentation processing on the background image of the monitoring area to obtain background segmentation regions; Identifying a regional entity in the background segmentation region, and extracting regional visual features in the background segmentation region; Combining the regional entity and the regional visual features, analyzing the regional semantic category corresponding to the background segmentation region; The category coupling relationship between the semantic categories of the region is analyzed, and knowledge integration processing is performed on the category coupling relationship to obtain scene knowledge elements corresponding to the background image of the monitoring area.

7. The intelligent security monitoring method based on big data according to claim 1, characterized in that: The skeletonization of the single person morphology set to obtain a morphology skeleton set includes: Performing smoothing processing on each form in the single-person form set to obtain a smoothed form set; Identify a shape boundary corresponding to each shape in the set of smooth shapes; Based on the morphological boundary, performing distance change processing on each morphology in the smooth morphological set to obtain a morphological transformation distance; Marking the local maximum values ​​in the morphological transformation distance to obtain a marked extreme value set; The labeled extreme value set is subjected to skeleton refinement processing to obtain a morphological skeleton set.

8. The intelligent security monitoring method based on big data according to claim 1, characterized in that: The combining the morphological skeleton set and the scene knowledge elements to analyze the individual behavior corresponding to the single person foreground primitive includes: Identify the skeleton joint points and joint point timing corresponding to each morphology in the morphology skeleton set; In combination with the joint point timing, analyzing the node trajectory corresponding to the skeleton joint point; Based on the node trajectory, analyzing the morphology trajectory trend corresponding to each morphology in the morphology skeleton set; Associating the morphological trajectory trend with the scene knowledge element to obtain a scene-associated trajectory trend; Based on the scene-associated trajectory trend, the individual behavior corresponding to the single-person foreground primitive is analyzed.

9. The intelligent security monitoring method based on big data according to claim 1, characterized in that: The formulating of a regional security monitoring plan corresponding to the monitoring area in combination with the individual behavior and the group behavior includes: Based on the individual behavior and the group behavior, the regional surveillance video is segmented and processed for early warning to obtain an early warning slice video; Analyze the video dynamic information corresponding to the warning slice video, and generate a slice video summary of the warning slice video based on the video dynamic information; Based on the sliced ​​video summary, a regional security monitoring plan corresponding to the monitoring area is formulated.

10. Intelligent security monitoring device based on big data, characterized in that: The device comprises: A foreground image element analysis module is used to obtain a monitoring area to be processed, schedule a regional monitoring video corresponding to the monitoring area, extract a regional dynamic foreground and a monitoring area background image in the regional monitoring video, and analyze the foreground image elements in the regional dynamic foreground; A scene knowledge element analysis module is used to screen out single-person foreground primitives and multi-person foreground primitives in the foreground primitives, extract single-person knowledge elements and multi-person knowledge elements in the single-person foreground primitives and the multi-person foreground primitives, respectively, and analyze the scene knowledge elements corresponding to the background image of the monitoring area; A group behavior analysis module is used to identify a single-person morphology set corresponding to the single-person knowledge element, perform skeleton processing on the single-person morphology set to obtain a morphology skeleton set, combine the morphology skeleton set with the scene knowledge elements, analyze the individual behavior corresponding to the single-person foreground image element, locate the geometric position of the image element corresponding to the single-person foreground image element based on the background image of the monitoring area, and analyze the group behavior corresponding to the multi-person foreground image element by combining the single-person individual behavior and the image element geometric position; The security monitoring processing module is used to formulate a regional security monitoring plan corresponding to the monitoring area in combination with the individual behavior and the group behavior, and to perform security monitoring processing corresponding to the monitoring area using the regional security monitoring plan to obtain monitoring results.

Images