Event monitoring method and device, electronic equipment and storage medium

Abstract

The invention provides an event monitoring method. The method comprises the following steps: acquiring dynamic data and static data of a target person; performing behavior detection on the dynamic data through a dynamic behavior detection model to obtain a dynamic behavior detection result of the target person; performing behavior analysis on the static data through a static behavior analysis model to obtain a static behavior analysis result of the target person; wherein the dynamic behavior detection model and the static behavior analysis model are obtained by training historical violation case data; and determining whether the target person participates in the violation event based on the dynamic behavior detection result and the static behavior analysis result. According to the method, the problems that hidden behaviors are difficult to identify through single video monitoring, risks are difficult to judge depending on artificial experience, monitoring data are difficult to process manually after an event, and early warning cannot be performed before illegal behaviors occur in an existing method are solved, and potential illegal behaviors and illegal behaviors of personnel can be early warned in advance.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing, and in particular to an event monitoring method and device, electronic equipment and a storage medium. BACKGROUND

[0002] With the continuous progress of science and technology, personnel management and alarm systems have developed rapidly. Now, through various technical means such as face recognition, ID card recognition, camera monitoring, RFID, Bluetooth and other technologies, personnel identification, tracking and analysis of people entering and leaving a place can be achieved, further improving the efficiency and level of safety management. At present, the existing method has the problems of multi-modal data island, video, audio, file, static supervision and other data isolated analysis, lack of fusion correlation means, leading to one-sided risk assessment; relying on manual post-processing of monitoring data, unable to warn before the occurrence of irregular behavior; event assessment mainly relies on personal experience to judge risk, lack of quantitative and standardized risk assessment model; it is difficult to identify hidden behavior through single video monitoring, historical irregular event data has not been systematically mined, and it is difficult to form an iterative optimization prediction ability. Therefore, the existing method has the problems of being difficult to identify hidden behavior through single video monitoring, relying on manual experience to judge risk, manual post-processing of monitoring data, and being unable to warn before the occurrence of irregular behavior. SUMMARY

[0003] The embodiments of the present application provide an event monitoring method, which aims to solve the problems of the existing method that it is difficult to identify hidden behavior through single video monitoring, relies on manual experience to judge risk, manually processes monitoring data, and cannot warn before the occurrence of irregular behavior. The dynamic behavior detection model of the present application detects the behavior of the dynamic data to obtain the dynamic behavior detection result of the target personnel, and analyzes the behavior of the static data through the static behavior analysis model to obtain the static behavior analysis result of the target personnel, and determines whether the target personnel participates in the irregular event according to the dynamic behavior detection result and the static behavior analysis result. The present application solves the problems of the existing method that it is difficult to identify hidden behavior through single video monitoring, relies on manual experience to judge risk, manually processes monitoring data, and cannot warn before the occurrence of irregular behavior, and can early warn potential irregular and illegal behavior of personnel.

[0004] In a first aspect, the embodiments of the present application provide an event monitoring method, which comprises the following steps:

[0005] Obtaining dynamic data and static data of a target personnel;

[0006] performing behavior detection on the dynamic data through a dynamic behavior detection model to obtain a dynamic behavior detection result of the target person, and performing behavior analysis on the static data through a static behavior analysis model to obtain a static behavior analysis result of the target person, wherein the dynamic behavior detection model and the static behavior analysis model are trained according to historical violation case data;

[0007] based on the dynamic behavior detection result and the static behavior analysis result, determining whether the target person participates in a violation event.

[0008] Optionally, the dynamic behavior detection model and the static behavior analysis model are trained according to historical violation case data, comprising:

[0009] obtaining historical violation case information;

[0010] based on the historical violation case information, obtaining a violation event occurrence time point and a violation person;

[0011] obtaining historical dynamic data of a preset time length before the violation event occurrence time point, and obtaining historical static data of the violation person;

[0012] based on the historical dynamic data and the historical static data, constructing a training data set;

[0013] training a to-be-trained dynamic behavior detection model and a to-be-trained static behavior analysis model through the training data set to obtain the trained dynamic behavior detection model and the trained static behavior analysis model.

[0014] Optionally, the historical dynamic data comprises historical video data and historical audio data, the to-be-trained dynamic behavior detection model comprises a to-be-trained video behavior detection model and a to-be-trained audio behavior detection model, the input of the to-be-trained video behavior detection model is multi-dimensional visual behavior features, and the output is a video behavior score value, the input of the to-be-trained audio behavior detection model is multi-dimensional language features and a person relationship graph, and the output is an audio behavior score value, the training data set comprises a video data set and an audio data set, and the constructing of the training data set comprises:

[0015] extracting visual behavior features from the historical video data to obtain multi-dimensional visual behavior features;

[0016] obtaining a behavior score value corresponding to the historical video data, labeling the behavior score value corresponding to the historical video data as a video behavior score value of the multi-dimensional visual behavior features, and constructing a video data set for training the to-be-trained video behavior detection model;

[0017] and language feature extraction is performed on the historical audio data set to obtain multi-dimensional language features.

[0018] An action score value corresponding to the historical audio data is obtained, the action score value corresponding to the historical video data is labeled as a video action score value of the multi-dimensional visual behavior features, and an audio data set for training the to-be-trained audio behavior detection model is constructed.

[0019] Optionally, the visual behavior feature extraction on the historical video data to obtain multi-dimensional visual behavior features comprises:

[0020] Portrait detection and tracking processing are performed on the historical video data to obtain personnel trajectory features corresponding to the historical video data;

[0021] Personnel action recognition processing is performed on the historical video data to obtain personnel action features corresponding to the historical video data;

[0022] Aggregation degree recognition processing is performed on the historical video data to obtain personnel aggregation degree features corresponding to the historical video data;

[0023] The personnel trajectory features, the personnel action features, and the personnel aggregation degree features are spatio-temporally fused and standardized according to classification into long-term features and short-term features to obtain multi-dimensional visual behavior features.

[0024] Optionally, the language feature extraction on the historical audio data set to obtain multi-dimensional language features comprises:

[0025] Text conversion processing is performed on the historical audio data to obtain historical text data corresponding to the historical audio data;

[0026] Basic language feature extraction processing is performed on the text data to obtain basic language features corresponding to the historical text data;

[0027] Deep semantic feature extraction processing is performed on the text data to obtain deep semantic features corresponding to the historical text data;

[0028] Dialogue interaction feature extraction processing is performed on the text data to obtain dialogue interaction features corresponding to the historical text data;

[0029] Based on the basic language features, the deep semantic features, and the dialogue interaction features, multi-dimensional language features are determined.

[0030] Optionally, the historical static data comprises historical structured data and historical event supervision data, and the construction of the training data set comprises:

[0031] construct a personnel relationship graph based on the historical structured data;

[0032] spatially align the historical event supervision data, and perform risk coding to obtain a risk transmission coefficient corresponding to the historical event supervision data;

[0033] construct a risk propagation data set for training the to-be-trained static behavior analysis model based on the personnel relationship graph and the risk transmission coefficient.

[0034] Optionally, the determining whether the target personnel participates in a violation event based on the dynamic behavior detection result and the static behavior analysis result comprises:

[0035] performing weighted calculation on the dynamic behavior detection result and the static behavior analysis result to obtain a total score of the violation event;

[0036] determining whether the target personnel participates in a violation event and a risk level of the violation event based on the total score of the violation event.

[0037] In a second aspect, an embodiment of the present application further provides an event monitoring device, the event monitoring device comprising:

[0038] an acquisition module configured to acquire dynamic data and static data of a target personnel;

[0039] a processing module configured to perform behavior detection on the dynamic data by a dynamic behavior detection model to obtain a dynamic behavior detection result of the target personnel, and perform behavior analysis on the static data by a static behavior analysis model to obtain a static behavior analysis result of the target personnel, wherein the dynamic behavior detection model and the static behavior analysis model are trained according to historical violation case data;

[0040] a determination module configured to determine whether the target personnel participates in a violation event based on the dynamic behavior detection result and the static behavior analysis result.

[0041] In a third aspect, an embodiment of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements steps in the event monitoring method provided by the embodiment of the present application when executing the computer program.

[0042] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program is executable on a processor to implement steps in the event monitoring method provided by the embodiment of the present application.

[0043] In an embodiment of the present invention, dynamic data and static data of a target person are obtained; a dynamic behavior detection model is used to perform behavior detection on the dynamic data to obtain a dynamic behavior detection result of the target person; and a static behavior analysis model is used to perform behavior analysis on the static data to obtain a static behavior analysis result of the target person; wherein, the dynamic behavior detection model and the static behavior analysis model are trained based on historical violation case data; based on the dynamic behavior detection result and the static behavior analysis result, it is determined whether the target person is involved in a violation. The dynamic behavior detection model of the present invention performs behavior detection on the dynamic data to obtain a dynamic behavior detection result of the target person, and a static behavior analysis model is used to perform behavior analysis on the static data to obtain a static behavior analysis result of the target person, and based on the dynamic behavior detection result and the static behavior analysis result, it is determined whether the target person is involved in a violation. The present invention solves the problem that existing methods have difficulty in identifying hidden behaviors through single video surveillance, rely on manual experience to judge risks, and manually process monitoring data afterwards, and cannot provide early warnings before violations occur. It can provide early warnings of potential violations and illegal behaviors of personnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0045] Figure 1 is a flow chart of an event monitoring method provided by an embodiment of the present invention;

[0046] Figure 2 1 is a schematic structural diagram of an event monitoring device provided by an embodiment of the present invention;

[0047] Figure 3 It is a structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0049] like Figure 1 As shown, Figure 1is a flowchart of an event monitoring method provided by an embodiment of the present application, the event monitoring method comprising the steps of:

[0050] 101. Obtain dynamic data and static data of the target person.

[0051] In an embodiment of the present application, the event monitoring method described above can be applied to a personnel management platform, which can be constructed based on a server or a distributed server. The personnel management platform comprises a data interface (for sensors or user uploads), a knowledge database, and a knowledge database construction program. The data interface can be used to obtain dynamic data and static data of the target person. The dynamic data can be real-time monitoring data, and the static data can be personnel violation data, conversation record data, medical record data, etc. The knowledge database construction program can be used to implement knowledge database construction. The knowledge database is dedicated to providing additional associated information for the identified data entities, thereby improving the understanding depth of the content by the data recognition system.

[0052] The target person can be understood as a person who needs to be monitored or analyzed.

[0053] The dynamic data can be understood as data obtained by real-time monitoring through a camera, including behavior data, walking trajectory data, conversation data, etc.

[0054] The static data can be understood as data that does not change with time, such as personnel violation records, conversation records, medical records, etc.

[0055] 102. Perform behavior detection on the dynamic data through a dynamic behavior detection model to obtain a dynamic behavior detection result of the target person, and perform behavior analysis on the static data through a static behavior analysis model to obtain a static behavior analysis result of the target person.

[0056] In an embodiment of the present application, the dynamic behavior detection model and the static behavior analysis model are trained according to historical violation case data. The historical violation case data includes historical video data, historical audio data, structured data, supervision data, etc. The historical video data can be a video record of a violation behavior that occurred in the past, the historical audio data can be an audio record of a violation behavior that occurred in the past, the structured data includes personnel archive information, violation record information, cell environment data, etc., and the supervision data includes personnel violation data, conversation records, medical records, etc. The cell environment data includes personnel density, camera coverage blind area markers, etc.

[0057] The dynamic behavior detection model can identify and predict the violation behavior of the target person. The static behavior detection model can identify and analyze the static behavior characteristics of the target person.

[0058] The behavior detection can be understood as a process of identifying and analyzing the dynamic behavior of the target person in the dynamic data through the dynamic behavior detection model.

[0059] The behavior analysis can be understood as a process of identifying and analyzing the static behavior of the target person in the static data through the static behavior analysis model.

[0060] The dynamic behavior detection result is an evaluation of the dynamic behavior of the target person after analysis by the dynamic behavior detection model. The dynamic behavior detection result includes the behavior pattern, abnormal behavior, and other information of the target person.

[0061] The static behavior analysis result is an evaluation of the static behavior of the target person after analysis by the static behavior analysis model. The static behavior analysis result includes the behavior habit, preference, and other information of the target person.

[0062] It should be noted that the dynamic behavior detection of the dynamic data by the dynamic behavior detection model obtains the dynamic behavior detection result of the target person, and the behavior analysis of the static data by the static behavior analysis model obtains the static behavior analysis result of the target person, which can more comprehensively understand the behavior characteristics of the target person and improve the accuracy of behavior prediction.

[0063] 103、Based on the dynamic behavior detection result and the static behavior analysis result, it is determined whether the target person participates in the irregular event.

[0064] In the embodiment of the application, the dynamic behavior detection result and the static behavior analysis result can be weighted and fused, different weights are given to the dynamic behavior detection result and the static behavior analysis result to calculate the comprehensive influence of the dynamic behavior detection result and the static behavior analysis result. According to the comprehensive influence of the dynamic behavior detection result and the static behavior analysis result, it is determined whether the target person participates in the irregular event.

[0065] It should be noted that if the comprehensive influence is high, it can be considered that the target person has high participation and risk, and more stringent measures need to be taken to prevent or handle.

[0066] In a possible embodiment, the risk value of the target person can be calculated by a risk assessment calculation formula: comprehensive risk value = 0.4 x video analysis score + 0.3 x audio analysis score + 0.2 x static file score (supervision data analysis score) + 0.1 x environmental factor score. The threshold of the comprehensive risk value can be divided into: the risk threshold of low risk is <0.5, which can be recorded by a log; the risk threshold of medium risk is 0.5-0.8, which can be reminded by sound and light to the on-duty personnel; the risk threshold of high risk is >0.8, which can automatically trigger an emergency broadcast and locate the incident area.

[0067] In the embodiments of the present application, the present application increases the proactive prediction of supervision to discover potential risks in advance and improves the accuracy of the prediction of supervision through learning from historical cases.

[0068] In the embodiments of the present application, the dynamic data and the static data of the target person are obtained, the dynamic behavior detection model is used to detect the behavior of the dynamic data to obtain the dynamic behavior detection result of the target person, and the static behavior analysis model is used to analyze the behavior of the static data to obtain the static behavior analysis result of the target person, wherein the dynamic behavior detection model and the static behavior analysis model are trained according to historical violation case data; and whether the target person participates in the violation event is determined based on the dynamic behavior detection result and the static behavior analysis result. Through the dynamic behavior detection model, the behavior of the dynamic data is detected to obtain the dynamic behavior detection result of the target person, and through the static behavior analysis model, the behavior of the static data is analyzed to obtain the static behavior analysis result of the target person, and whether the target person participates in the violation event is determined according to the dynamic behavior detection result and the static behavior analysis result. The present application solves the problem that the existing method cannot identify hidden behavior through single video monitoring, relies on artificial experience to determine risks, and cannot perform early warning before the violation behavior occurs. The present application can early warn potential violation and illegal behavior of the person.

[0069] It can be understood that in the specific embodiments of the present application, data related to dynamic data, static data, user data, etc. are involved. When the embodiments of the present application are applied to specific products or technologies, the user's permission or consent is required, and the collection, use and processing of related data, as well as the training, deployment and calling of algorithm models, need to comply with relevant laws, regulations and standards of the country and region.

[0070] Optionally, in the step of training the dynamic behavior detection model and the static behavior analysis model according to the historical violation case data, historical violation case information can be obtained; based on the historical violation case information, the time point of the violation event and the violation person are obtained; historical dynamic data of a preset time length before the time point of the violation event is obtained, and historical static data of the violation person is obtained; based on the historical dynamic data and the historical static data, a training data set is constructed; and the to-be-trained dynamic behavior detection model and the to-be-trained static behavior analysis model are trained through the training data set to obtain the trained dynamic behavior detection model and the static behavior analysis model.

[0071] In the embodiments of the present application, the above historical violation case information can be understood as related data of past violation events, including video, audio, processing records, etc. of the violation events.

[0072] Specifically, by analyzing historical violation case information, the time point of the past violation event and the personnel involved in the violation can be understood.

[0073] The historical dynamic data described above includes historical monitoring video, in which the annotation data segment before the corresponding violation event occurs is marked.

[0074] The preset time length can be a preset time length of 5-10 minutes, or a preset time length of 6-12 minutes, etc.

[0075] The historical static data described above can be understood as personnel archive information, personnel violation data, conversation records, medical records, etc. The influence of different types of drugs on emotions and behaviors can be analyzed through the medical records of the target personnel.

[0076] The training data set described above is a data sample set used to build and train the model. The role of the training data set is to optimize the model parameters by learning data characteristics and patterns, and to provide a basis for subsequent verification and testing.

[0077] The dynamic behavior detection model to be trained described above can be a dynamic behavior detection model based on deep learning or machine learning, such as a Transformer model, a YOLOv7 model, etc.

[0078] The static behavior analysis model to be trained described above can be a static behavior analysis model based on deep learning or machine learning, such as CNN, RNN, etc.

[0079] The training described above can be supervised training. Supervised training is to use a set of data with known labels to train the model, and to optimize the model parameters so that the model can predict the labels of new data or make decisions based on the characteristics of existing data. During the training process, the parameters of the model can be adjusted using the least loss function to minimize the difference between the output label of the model and the input data. The loss function is used to measure the difference between the model's prediction results and the true results. The purpose is to minimize the loss function value by adjusting the model parameters, so as to improve the prediction accuracy. The loss function described above can be a mean square error loss function, a cross-entropy loss function, etc.

[0080] The trained dynamic behavior detection model described above can identify and predict violation behavior.

[0081] The trained static behavior detection model described above can identify and analyze static behavior characteristics.

[0082] Optionally, the historical dynamic data comprises historical video data and historical audio data, the to-be-trained dynamic behavior detection model comprises a to-be-trained video behavior detection model and a to-be-trained audio behavior detection model, the input of the to-be-trained video behavior detection model is multi-dimensional visual behavior features, the output is a video behavior score value, the input of the to-be-trained audio behavior detection model is multi-dimensional language features and a personnel relationship graph, the output is an audio behavior score value, the training data set comprises a video data set and an audio data set, in the step of constructing the training data set, the historical video data can be subjected to visual behavior feature extraction to obtain multi-dimensional visual behavior features; the behavior score value corresponding to the historical video data is obtained, the behavior score value corresponding to the historical video data is labeled as a video behavior score value of the multi-dimensional visual behavior features, and a video data set for training the to-be-trained video behavior detection model is constructed; and the historical audio data set is subjected to language feature extraction to obtain multi-dimensional language features; the behavior score value corresponding to the historical audio data is obtained, the behavior score value corresponding to the historical video data is labeled as a video behavior score value of the multi-dimensional visual behavior features, and an audio data set for training the to-be-trained audio behavior detection model is constructed.

[0083] In the embodiment of the application, the historical video data can be understood as historical monitoring video, and the historical monitoring video comprises a video segment before a violation event occurs and a video segment when the violation event occurs.

[0084] The historical audio data can be audio file data, and the audio file data comprises text data converted from audio.

[0085] The visual behavior feature extraction can be understood as a process of extracting features capable of describing and distinguishing different behaviors from historical video data.

[0086] The multi-dimensional visual behavior features can be understood as visual behavior features exhibited in different dimensions, such as motion trajectory, speed, acceleration and the like.

[0087] The video behavior score value is an index for evaluating behaviors in a video. The audio behavior score value is an index for evaluating behaviors in an audio.

[0088] The language feature extraction can be understood as a process of analyzing and extracting feature parameters capable of representing the nature of speech from audio data.

[0089] The multi-dimensional language features comprise features such as lexical diversity, syntactic diversity and semantic diversity.

[0090] The personnel relationship graph is used for representing and analyzing various relationships between personnel. The personnel relationship graph displays the relationships between personnel in a graphical manner and can intuitively display a complex relationship network structure.

[0091] The to-be-trained audio behavior detection model can be an audio behavior detection model adopting a Conformer architecture. The Conformer architecture is a hybrid architecture combining a Transformer and a convolutional neural network (CNN), and can support various local dialect speech recognition, special term vocabulary, etc.

[0092] It should be noted that the historical dynamic data includes historical video and audio data. The to-be-trained dynamic behavior detection model includes a video and an audio behavior detection model. The input of the video model is multi-dimensional visual behavior features, and the output is a video behavior score. The input of the audio model is multi-dimensional language features and a personnel relationship graph, and the output is an audio behavior score. First, visual behavior features are extracted from the historical video data to obtain multi-dimensional visual behavior features. Then, the behavior score value corresponding to the historical video data is obtained, the score value corresponding to the historical video data is labeled as the video behavior score value of the multi-dimensional visual behavior features, and a data set for training the video behavior detection model is constructed. Then, language features are extracted from the historical audio data to obtain multi-dimensional language features. Finally, the behavior score value corresponding to the historical audio data is obtained, the score value corresponding to the historical audio data is labeled as the audio behavior score value of the multi-dimensional language features, and a data set for training the audio behavior detection model is constructed.

[0093] Optionally, in the step of extracting visual behavior features from the historical video data to obtain multi-dimensional visual behavior features, the historical video data can be subjected to portrait detection and tracking processing to obtain personnel trajectory features corresponding to the historical video data; the historical video data can be subjected to personnel action recognition processing to obtain personnel action features corresponding to the historical video data; the historical video data can be subjected to gathering degree recognition processing to obtain personnel gathering degree features corresponding to the historical video data; the personnel trajectory features, the personnel action features, and the personnel gathering degree features are subjected to spatio-temporal fusion and standardization processing according to the classification of long-term features and short-term features to obtain multi-dimensional visual behavior features.

[0094] In the embodiments of the present application, the target detection model can be used to detect the human image in the historical video data, output the human target frame, and use the DeepSORT algorithm to track the frames to realize the identity association of the personnel, and record the center point coordinates per second, calculate the moving speed of the personnel (pixel / second→meter / second) to generate the personnel track features corresponding to the historical video data. The target detection model can be a YOLOv7 model, which is a real-time target detection deep learning model that can recognize objects in an image and draw bounding boxes for them. The DeepSORT algorithm is a multi-target tracking algorithm. The DeepSORT algorithm can track the detected targets across frames by analyzing the motion information between consecutive frames, thereby realizing the identity association of the targets. The DeepSORT algorithm can be used to track the motion trajectories of different personnel and associate the identities of different personnel.

[0095] Specifically, after detecting the personnel target frame, the center point coordinates of the target can be extracted, and by calculating the changes of the center point coordinates between adjacent frames, the distance (in pixels) moved by the personnel per second can be calculated. Then, the distance moved by the personnel per second needs to be converted into actual physical units (such as meters), so as to obtain the moving speed of the personnel (in meters / second). Finally, the moving speed of the personnel is recorded to generate the motion trajectory of the personnel.

[0096] The above-mentioned personnel action recognition processing can be understood as a processing process of recognizing and analyzing the human body contour, posture, motion trajectory and other features in the video, thereby obtaining the personnel action features corresponding to the historical video data. The OpenPose model can be used to extract 17 key points of personnel actions, including limb joints and facial feature points, and output 25 sets of key point coordinates per second. Whether there is an item hidden can be detected according to the key parts, and the attention direction can be judged according to the head orientation.

[0097] Further, the joint movement distance and action intensity can be calculated through limb action analysis-action amplitude. The joint movement distance is calculated by the displacement of the elbow joint in 1 second, and the action intensity is calculated by ∑(key point speed2) / key point number. For example, punching, punching is the wrist joint speed>2m / s and the motion trajectory is an arc; hiding and transferring items, hiding and transferring items are two hands meeting time>0.5 seconds and at the edge of the monitoring blind area; self-injury tendency, self-injury tendency is hand staying at neck / wrist>3 seconds and accompanied by tremor (key point displacement variance>15). Specifically, according to the characteristics of hand and head and facial injuries, the transfer of items and the strengthening of group conflict warning can be monitored.

[0098] The above aggregation degree recognition processing can be understood as detecting the density and distribution of the crowd in the video, thereby obtaining the processing process of the personnel aggregation degree feature corresponding to the historical video data. The calculation formula of the aggregation degree is: aggregation degree = minimum enclosing circle radius of personnel / available space radius of the cell. For example, when the distance between two people is <0.5 meters for >3 seconds, it can be marked as suspicious contact; when multiple people form a closed circle, a group event warning can be triggered.

[0099] The above long-term feature can be understood as a behavior feature observed in a minute. For example, the stay time of personnel in each area of the cell in 10 minutes is counted to form an activity manpower diagram; the behavior pattern deviation is formed by comparing the difference between the current behavior and the historical behavior baseline (such as abnormal getting up at night), to detect abnormal behavior.

[0100] The above short-term feature can be understood as a behavior feature observed in a second. For example, whether the acceleration is >3 m / s 2 is detected to identify sudden and violent actions such as fighting or fleeing, and the short-term feature can detect sudden actions; the included angle between the heads of two people <30° for >2 seconds can detect the intersection of the lines of sight of two people, to identify whether the two people are communicating or cooperating.

[0101] The above space-time fusion can be understood as a processing process of combining the time feature and the space feature of the personnel trajectory feature, the personnel action feature, and the personnel aggregation degree feature.

[0102] The above standardization processing can be understood as a processing process of eliminating the dimensional, property, or data set differences in the personnel trajectory feature, the personnel action feature, and the personnel aggregation degree feature, so that the personnel trajectory feature, the personnel action feature, and the personnel aggregation degree feature have consistency. The standardization processing aims to scale different feature values to the same range, such as [0, 1]. Normalization processing can be used to scale each feature value to the [0, 1] interval, and the calculation formula is:

[0103] Standardized value = (original value - historical minimum value) / (historical maximum value - historical minimum value)

[0104] In a possible embodiment, dynamic baseline adjustment can be used. Dynamic baseline adjustment is an adaptive data preprocessing method. Since the distribution of data changes over time, it is necessary to update the extreme value range regularly (for example, every quarter) to more accurately reflect the true distribution of data.

[0105] In another possible embodiment, feature value vectors can be constructed to convert the personnel trajectory feature, the personnel action feature, and the personnel aggregation degree feature into feature vectors, and a set of feature vectors is generated every second, thereby obtaining multi-dimensional visual behavior features.

[0106] In another possible implementation, the trajectory feature, the action feature and the crowd density feature of the personnel can be analyzed through a sliding window analysis by integrating the time sequence features, the window size can be 30 seconds, and the step size can be 5 seconds, and the mean value, the variance and the extreme value of the trajectory feature, the action feature and the crowd density feature of the personnel are obtained, so as to obtain the multi-dimensional visual behavior features.

[0107] Optionally, in the step of extracting the multi-dimensional language features from the historical audio data set, the historical audio data can be subjected to a text conversion processing to obtain historical text data corresponding to the historical audio data; the text data can be subjected to a basic language feature extraction processing to obtain basic language features corresponding to the historical text data; the text data can be subjected to a deep semantic feature extraction processing to obtain deep semantic features corresponding to the historical text data; the text data can be subjected to a dialogue interaction feature extraction processing to obtain dialogue interaction features corresponding to the historical text data; and the multi-dimensional language features are determined based on the basic language features, the deep semantic features and the dialogue interaction features.

[0108] In the embodiments of the present application, the text conversion processing can be understood as a process of converting speech into text for the historical audio data.

[0109] The basic language feature extraction processing can be understood as a process of extracting lexical features, syntactic features and semantic roles from the text data. The lexical feature extraction can be understood as a process of extracting important features that can describe the semantic of the text data, which can be performed by sensitive word TF-IDF weighting. The syntactic feature extraction can be understood as a process of extracting syntactic structure information from the text data, which can be performed by dependency syntax analysis. The semantic role extraction can be understood as a process of extracting semantic relationships between words in a sentence to understand the deep meaning of the sentence, which can be performed by predicate-argument structure analysis.

[0110] The deep semantic feature extraction processing can identify the fine-grained information of the text data, such as emotion and intent. The intent of the text data can be identified by an intent recognition model, which can analyze the input text data and the dialogue history context to identify the user's intent. The emotion of the text data can be identified by an emotion analysis model, which uses a dimensional model to evaluate the emotional state of the user. The dimensional model is a multi-dimensional emotion model, which regards the emotional experience as a multi-dimensional continuous change process, and describes an emotional state by combining different numerical values.

[0111] The dialogue interaction feature extraction processing can be understood as a processing procedure of extracting features reflecting dialogue interaction characteristics from the text data. The dialogue interaction features can reflect the roles of the speakers, the speech rights, etc. The dialogue interaction can be analyzed through the turn-taking pattern, and the dialogue interaction features in the text data can be analyzed, such as, in the dialogue text data, the time interval of the turn-taking can be between 0.5 to 2 seconds, if the time interval of the turn-taking is less than 0.3 seconds, it indicates that someone is talking over; if the long silence is more than 5 seconds, it indicates the premonition of conflict, etc. The dialogue interaction features in the text data can be inferred through the right relationship, such as, by analyzing the frequency of use of the address (such as "big brother") and the imperative sentence in the text data, the power relationship between the dialogue participants can be inferred, which can be understood as that the person who frequently uses "big brother" is in a lower power position, and the person who often uses the imperative sentence is in a higher power position.

[0112] The above-mentioned multi-dimensional language features include lexical diversity, syntactic diversity, and semantic diversity.

[0113] It should be noted that the historical audio data is converted into text data to obtain historical text data corresponding to the historical audio data. The historical text data can be subjected to post-processing, which can be timestamp alignment, slang conversion, context error correction, etc. The above-mentioned timestamp alignment can be to mark the start and end time of each word by a forced alignment algorithm. The above-mentioned slang conversion can be through a special term mapping table established. The above-mentioned context error correction can be to correct recognition errors according to the historical text data.

[0114] Optionally, the historical static data includes historical structured data and historical event supervision data. In the step of constructing the training data set, the personnel relationship graph can be constructed based on the historical structured data; the historical event supervision data is spatio-temporally aligned and risk coded to obtain a risk transmission coefficient corresponding to the historical event supervision data; and the risk propagation data set for training the to-be-trained static behavior analysis model is constructed based on the personnel relationship graph and the risk transmission coefficient.

[0115] In the embodiment of the present application, the above-mentioned historical structured data includes personnel archive information, violation record information, cell environment data, etc. The cell environment data includes personnel density, camera coverage blind area mark, etc. The above-mentioned historical event supervision data includes personnel violation data, dialogue record data, medical treatment data, etc.

[0116] The above-mentioned personnel relationship graph is used to represent and analyze various relationships between personnel. The personnel relationship graph displays the relationship between personnel in a graphical way, which can intuitively display the complex relationship network structure.

[0117] The spatio-temporal alignment can be understood as a process of time synchronization and spatial alignment of historical event monitoring data. The time synchronization can be understood as a process of time synchronization of historical event monitoring data by a time synchronization method, so as to keep the historical event monitoring data synchronized with a standard time source. The spatial alignment can be understood as a process of mapping the time-synchronized historical event monitoring data to the same coordinate reference system, realizing seamless superposition and accurate registration between data.

[0118] The risk coding can be understood as a process of quantifying historical event monitoring data into risk coefficients. Through risk coding, risks can be systematically identified, tracked and processed.

[0119] The risk transmission coefficient can reflect the severity and influence range of the event.

[0120] The risk propagation dataset is constructed by analyzing the personnel relationship graph and the risk transmission coefficient, and contains a data set of risk propagation characteristics. The risk propagation dataset is used to train the to-be-trained static behavior analysis model. The to-be-trained static behavior analysis model can be a static behavior analysis model constructed based on deep learning or machine learning, such as CNN, RNN, etc.

[0121] In a possible embodiment, the risk transmission coefficient can be adjusted by weight, specifically, recent (within 7 days) transmission coefficient x 1.5, night transmission coefficient x 1.2.

[0122] It should be noted that the present application can periodically increase the case to construct the training dataset, adjust the weight parameters of each data feature, and also manually analyze the false positive cases, which can optimize the performance of the model, improve the generalization ability of the model, reduce false positives, and improve the overall effect.

[0123] Optionally, in the step of determining whether the target personnel participates in the irregular event based on the dynamic behavior detection result and the static behavior analysis result, the dynamic behavior detection result and the static behavior analysis result can be weighted and calculated to obtain a total score of the irregular event; and based on the total score of the irregular event, it is determined whether the target personnel participates in the irregular event and the risk level of the irregular event.

[0124] In the embodiment of the present application, the above-mentioned weighted calculation calculates the comprehensive influence of the dynamic behavior detection result and the static behavior analysis result by giving different weights to the dynamic behavior detection result and the static behavior analysis result.

[0125] The total score of the irregular event can reflect the possibility of the target personnel participating in the irregular event and the severity thereof.

[0126] It should be noted that if the total score of the violation event is high, it can be considered that the target personnel has high participation and risk, and stricter measures need to be taken to prevent or handle.

[0127] In a possible embodiment, the risk value of the target personnel can be calculated by a risk assessment calculation formula: comprehensive risk value = 0.4 * video analysis score + 0.3 * audio analysis score + 0.2 * static file score (supervision data analysis score) + 0.1 * environmental factor score. The threshold of the comprehensive risk value can be divided into: the risk threshold of low risk is <0.5, which can be recorded by a log; the risk threshold of medium risk is 0.5-0.8, which can be reminded by sound and light to the on-duty personnel; the risk threshold of high risk is >0.8, which can trigger an emergency broadcast automatically and locate the area where the incident occurs.

[0128] In another possible embodiment, the risk value can be calculated by personnel meeting, meeting risk value = 0.6 * case sensitivity + 0.4 * (meeting duration / historical average duration). The case sensitivity can be a first sensitivity 1.0, a second sensitivity 0.6, a third sensitivity 0.3, etc., the first sensitivity indicating that the case has extremely high importance and attention, the second sensitivity indicating that the case has high importance and attention, and the third sensitivity indicating that the case has relatively low attention and importance. It should be understood that different case sensitivities correspond to different weight values.

[0129] In another possible embodiment, the target personnel can be determined to have a violation event by meeting emotion conduction, for example, the target personnel cries for more than 2 minutes during the meeting, and there is a self-injury risk + 35% in the subsequent n hours; the target personnel has a fierce quarrel during the meeting, and there is a risk of conflict with the cellmates + 25% in the subsequent hours, etc.

[0130] It should be noted that the present application expands the personnel supervision mode, increases active prediction supervision to discover potential risks in advance from the previous single passive supervision, integrates multi-modal data, improves the accuracy of active prediction supervision, and because a scientific quantitative risk assessment system is established, and through learning from historical cases, an evolutionary intelligent prediction and early warning mechanism is realized.

[0131] As shown in Figure 2 The event monitoring device provided by the embodiment of the present application comprises:

[0132] The acquisition module 201 is configured to acquire dynamic data and static data of the target personnel.

[0133] The processing module 202 is configured to perform behavior detection on the dynamic data by a dynamic behavior detection model to obtain a dynamic behavior detection result of the target person, and perform behavior analysis on the static data by a static behavior analysis model to obtain a static behavior analysis result of the target person, wherein the dynamic behavior detection model and the static behavior analysis model are obtained by training according to historical violation case data.

[0134] The determining module 203 is configured to determine whether the target person participates in a violation event based on the dynamic behavior detection result and the static behavior analysis result.

[0135] Optionally, the processing module 202 is further configured to obtain historical violation case information, obtain a violation event occurrence time point and a violation person based on the historical violation case information, obtain historical dynamic data in a preset time period before the violation event occurrence time point, and obtain historical static data of the violation person, construct a training data set based on the historical dynamic data and the historical static data, and train a to-be-trained dynamic behavior detection model and a to-be-trained static behavior analysis model by using the training data set to obtain the trained dynamic behavior detection model and the trained static behavior analysis model.

[0136] Optionally, the historical dynamic data includes historical video data and historical audio data, the to-be-trained dynamic behavior detection model includes a to-be-trained video behavior detection model and a to-be-trained audio behavior detection model, the input of the to-be-trained video behavior detection model is multi-dimensional visual behavior features, and the output is a video behavior score value, the input of the to-be-trained audio behavior detection model is multi-dimensional language features and a person relationship graph, and the output is an audio behavior score value, the training data set includes a video data set and an audio data set, and the processing module 202 is further configured to extract visual behavior features from the historical video data to obtain multi-dimensional visual behavior features, obtain a behavior score value corresponding to the historical video data, mark the behavior score value corresponding to the historical video data as a video behavior score value of the multi-dimensional visual behavior features, and construct a video data set for training the to-be-trained video behavior detection model, and extract language features from the historical audio data set to obtain multi-dimensional language features, obtain a behavior score value corresponding to the historical audio data, mark the behavior score value corresponding to the historical video data as a video behavior score value of the multi-dimensional visual behavior features, and construct an audio data set for training the to-be-trained audio behavior detection model.

[0137] Optionally, the processing module 202 is further configured to perform portrait detection and tracking processing on the historical video data to obtain personnel trajectory features corresponding to the historical video data, perform personnel action recognition processing on the historical video data to obtain personnel action features corresponding to the historical video data, and perform aggregation degree recognition processing on the historical video data to obtain personnel aggregation degree features corresponding to the historical video data; and perform spatio-temporal fusion and standardization processing on the personnel trajectory features, the personnel action features, and the personnel aggregation degree features according to classification of long-term features and short-term features to obtain multi-dimensional visual behavior features.

[0138] Optionally, the processing module 202 is further configured to perform text conversion processing on the historical audio data to obtain historical text data corresponding to the historical audio data, perform basic language feature extraction processing on the text data to obtain basic language features corresponding to the historical text data, perform deep semantic feature extraction processing on the text data to obtain deep semantic features corresponding to the historical text data, and perform dialogue interaction feature extraction processing on the text data to obtain dialogue interaction features corresponding to the historical text data; and determine multi-dimensional language features based on the basic language features, the deep semantic features, and the dialogue interaction features.

[0139] Optionally, the historical static data includes historical structured data and historical event supervision data, and the processing module 202 is further configured to construct a personnel relationship graph based on the historical structured data, perform spatio-temporal alignment on the historical event supervision data, and perform risk coding to obtain a risk transmission coefficient corresponding to the historical event supervision data; and construct a risk propagation data set for training the to-be-trained static behavior analysis model based on the personnel relationship graph and the risk transmission coefficient.

[0140] Optionally, the determining module 203 is further configured to perform weighted calculation on the dynamic behavior detection result and the static behavior analysis result to obtain a total score of a violation event, and determine whether the target personnel participates in the violation event and a risk level of the violation event based on the total score of the violation event.

[0141] As shown in Figure 3 The embodiment of the present application also provides an electronic device, which includes a processor, and the processor can execute any one of the above event monitoring methods.

[0142] Specifically, the electronic device includes a processor 301, a memory 302, and a computer program stored in the memory 302 and capable of running on the processor 301 to execute the event monitoring method, wherein:

[0143] The processor 301 runs the computer program of the event monitoring method stored in the memory 302 to execute the following steps:

[0144] obtaining dynamic data and static data of a target person;

[0145] performing behavior detection on the dynamic data through a dynamic behavior detection model to obtain a dynamic behavior detection result of the target person, and performing behavior analysis on the static data through a static behavior analysis model to obtain a static behavior analysis result of the target person, wherein the dynamic behavior detection model and the static behavior analysis model are obtained by training according to historical violation case data;

[0146] determining whether the target person participates in a violation event based on the dynamic behavior detection result and the static behavior analysis result.

[0147] Optionally, the dynamic behavior detection model and the static behavior analysis model executed by the processor 301 are obtained by training according to historical violation case data, including:

[0148] obtaining historical violation case information;

[0149] obtaining a violation event occurrence time point and a violation person based on the historical violation case information;

[0150] obtaining historical dynamic data of a preset time length before the violation event occurrence time point, and obtaining historical static data of the violation person;

[0151] constructing a training data set based on the historical dynamic data and the historical static data;

[0152] training a to-be-trained dynamic behavior detection model and a to-be-trained static behavior analysis model through the training data set to obtain the trained dynamic behavior detection model and the trained static behavior analysis model.

[0153] Optionally, the historical dynamic data includes historical video data and historical audio data, the to-be-trained dynamic behavior detection model includes a to-be-trained video behavior detection model and a to-be-trained audio behavior detection model, the input of the to-be-trained video behavior detection model is multi-dimensional visual behavior features, and the output is a video behavior score value, the input of the to-be-trained audio behavior detection model is multi-dimensional language features and a person relationship graph, and the output is an audio behavior score value, the training data set includes a video data set and an audio data set, and the processor 301 executes the constructing a training data set, including:

[0154] extracting visual behavior features from the historical video data to obtain multi-dimensional visual behavior features;

[0155] obtaining a behavior score value corresponding to the historical video data, labeling the behavior score value corresponding to the historical video data as a video behavior score value of the multi-dimensional visual behavior feature, and constructing a video data set for training the to-be-trained video behavior detection model;

[0156] and performing language feature extraction on the historical audio data set to obtain multi-dimensional language features;

[0157] obtaining a behavior score value corresponding to the historical audio data, labeling the behavior score value corresponding to the historical video data as a video behavior score value of the multi-dimensional visual behavior feature, and constructing an audio data set for training the to-be-trained audio behavior detection model.

[0158] Optionally, the multi-dimensional visual behavior feature obtained by the processor 301 performing the visual behavior feature extraction on the historical video data comprises:

[0159] performing portrait detection and tracking processing on the historical video data to obtain a personnel trajectory feature corresponding to the historical video data;

[0160] performing personnel action recognition processing on the historical video data to obtain a personnel action feature corresponding to the historical video data;

[0161] performing crowd recognition processing on the historical video data to obtain a personnel crowd feature corresponding to the historical video data;

[0162] performing spatiotemporal fusion and standardization processing on the personnel trajectory feature, the personnel action feature, and the personnel crowd feature according to classification of long-term features and short-term features to obtain the multi-dimensional visual behavior feature.

[0163] Optionally, the multi-dimensional language feature obtained by the processor 301 performing the language feature extraction on the historical audio data set comprises:

[0164] performing text conversion processing on the historical audio data to obtain historical text data corresponding to the historical audio data;

[0165] performing basic language feature extraction processing on the text data to obtain a basic language feature corresponding to the historical text data;

[0166] performing deep semantic feature extraction processing on the text data to obtain a deep semantic feature corresponding to the historical text data;

[0167] performing dialogue interaction feature extraction processing on the text data to obtain a dialogue interaction feature corresponding to the historical text data;

[0168] Based on the basic language features, the deep semantic features and the dialogue interaction features, multi-dimensional language features are determined.

[0169] Optionally, the historical static data includes historical structured data and historical event supervision data, and the processor 301 performs the constructing a training data set, including:

[0170] Based on the historical structured data, a personnel relationship graph is constructed;

[0171] The historical event supervision data is spatio-temporally aligned and risk encoded to obtain a risk transmission coefficient corresponding to the historical event supervision data;

[0172] Based on the personnel relationship graph and the risk transmission coefficient, a risk propagation data set for training the to-be-trained static behavior analysis model is constructed.

[0173] Optionally, the processor 301 performs the determining whether the target personnel participates in a violation event based on the dynamic behavior detection result and the static behavior analysis result, including:

[0174] The dynamic behavior detection result and the static behavior analysis result are weighted and calculated to obtain a total score of the violation event;

[0175] Based on the total score of the violation event, it is determined whether the target personnel participates in a violation event and a risk level of the violation event.

[0176] The embodiment of the application also provides a computer readable storage medium, and the computer readable storage medium stores a computer program.

[0177] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware, and the program can be stored in a computer readable storage medium, and when the program is executed, the processes of the above-mentioned embodiments can be included. The storage medium can be a magnetic disc, an optical disc, a read-only memory (ROM) or a random access memory (RAM).

[0178] The above only describes the preferred embodiments of the application, and of course cannot limit the scope of the application, so equivalent changes made according to the claims of the application are still within the scope of the application.

Claims

1. An event monitoring method, characterized in that: The method comprises the following steps: Obtain dynamic and static data of target personnel; Performing behavior detection on the dynamic data using a dynamic behavior detection model to obtain a dynamic behavior detection result of the target person; and performing behavior analysis on the static data using a static behavior analysis model to obtain a static behavior analysis result of the target person; wherein the dynamic behavior detection model and the static behavior analysis model are trained based on historical violation case data; Based on the dynamic behavior detection result and the static behavior analysis result, it is determined whether the target person is involved in a violation incident.

2. The event monitoring method according to claim 1, wherein: The dynamic behavior detection model and the static behavior analysis model are trained based on historical violation case data and include: Obtain information on historical violation cases; Based on the historical violation case information, obtain the time point of the violation incident and the violator; Obtain historical dynamic data for a preset period of time before the time point of the violation, and obtain historical static data of the violator; Constructing a training data set based on the historical dynamic data and the historical static data; The dynamic behavior detection model to be trained and the static behavior analysis model to be trained are trained using the training data set to obtain the trained dynamic behavior detection model and the trained static behavior analysis model.

3. The event monitoring method according to claim 2, wherein: The historical dynamic data includes historical video data and historical audio data. The dynamic behavior detection model to be trained includes a video behavior detection model to be trained and an audio behavior detection model to be trained. The input of the video behavior detection model to be trained is multi-dimensional visual behavior features, and the output is a video behavior score value. The input of the audio behavior detection model to be trained is multi-dimensional language features and a personnel relationship map, and the output is an audio behavior score value. The training data set includes a video data set and an audio data set. The construction of the training data set includes: Extracting visual behavior features from the historical video data to obtain multi-dimensional visual behavior features; Obtaining behavior score values ​​corresponding to the historical video data, marking the behavior score values ​​corresponding to the historical video data as video behavior score values ​​of the multi-dimensional visual behavior feature, and constructing a video data set for training the video behavior detection model to be trained; and, performing language feature extraction on the historical audio data set to obtain multi-dimensional language features; Obtain the behavior score value corresponding to the historical audio data, mark the behavior score value corresponding to the historical video data as the video behavior score value of the multi-dimensional visual behavior feature, and construct an audio data set for training the audio behavior detection model to be trained.

4. The event monitoring method according to claim 3, wherein: The extracting of visual behavior features from the historical video data to obtain multi-dimensional visual behavior features includes: Performing portrait detection and tracking processing on the historical video data to obtain trajectory features of people corresponding to the historical video data; Performing personnel action recognition processing on the historical video data to obtain personnel action features corresponding to the historical video data; Performing aggregation recognition processing on the historical video data to obtain a personnel aggregation feature corresponding to the historical video data; The personnel trajectory features, the personnel action features and the personnel gathering features are temporally and spatially fused and standardized according to the classification of long-term features and short-term features to obtain multi-dimensional visual behavior features.

5. The event monitoring method according to claim 3, wherein: The language feature extraction of the historical audio data set to obtain multi-dimensional language features includes: Performing text conversion processing on the historical audio data to obtain historical text data corresponding to the historical audio data; Performing basic language feature extraction processing on the text data to obtain basic language features corresponding to the historical text data; Performing deep semantic feature extraction processing on the text data to obtain deep semantic features corresponding to the historical text data; Performing conversation interaction feature extraction processing on the text data to obtain conversation interaction features corresponding to the historical text data; Multi-dimensional language features are determined based on the basic language features, the deep semantic features, and the dialogue interaction features.

6. The event monitoring method according to claim 2, wherein: The historical static data includes historical structured data and historical event supervision data. The construction of the training data set includes: Building a personnel relationship map based on the historical structured data; Performing spatiotemporal alignment on the historical event supervision data and performing risk coding to obtain a risk transmission coefficient corresponding to the historical event supervision data; Based on the personnel relationship map and the risk transmission coefficient, a risk propagation dataset is constructed for training the static behavior analysis model to be trained.

7. The event monitoring method according to any one of claims 1 to 6, characterized in that: Determining whether the target person is involved in a violation based on the dynamic behavior detection result and the static behavior analysis result includes: Performing weighted calculation on the dynamic behavior detection results and the static behavior analysis results to obtain a total score of the violation event; Based on the total score of the violation event, it is determined whether the target person is involved in the violation event and the risk level of the violation event.

8. An event monitoring device, characterized in that: The event monitoring device comprises: Acquisition module, used to obtain dynamic data and static data of target personnel; a processing module, configured to perform behavior detection on the dynamic data using a dynamic behavior detection model to obtain a dynamic behavior detection result of the target person; and to perform behavior analysis on the static data using a static behavior analysis model to obtain a static behavior analysis result of the target person; wherein the dynamic behavior detection model and the static behavior analysis model are trained based on historical violation case data; A determination module is used to determine whether the target person is involved in a violation event based on the dynamic behavior detection result and the static behavior analysis result.

9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the event monitoring method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the event monitoring method according to any one of claims 1 to 7.

Images