An automatic event classification system for outdoor monitoring

Abstract

The present invention discloses an automatic event classification system for outdoor monitoring, which relates to the technical field of intelligent classification and includes a construction module, a classification module, and an output module. The system establishes a first mapping relationship, obtains a level corresponding to the classified event, and outputs the current time classification and the current level to a monitoring center. The present invention establishes a mapping relationship between the event classification preset by the construction module and the historical monitoring video, quickly identifies and classifies the current monitoring event, and grades the event according to the level division logic. The system can analyze the monitoring video in real time, automatically identify abnormal behavior, and trigger corresponding alarms according to the type and severity of the event, so that management personnel can respond quickly. The automated processing reduces the dependence on manual monitoring. Based on AI and deep learning technology, the system can continuously optimize the event classification and level division logic to adapt to different scenarios and needs.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent classification, and in particular to an outdoor monitoring automatic event classification system. Background Art

[0002] In recent years, deep learning technology, particularly convolutional neural networks (CNNs) and the Transformer architecture, has become a core technology for outdoor surveillance event classification. It combines data from multiple sensors (such as cameras and millimeter-wave radar) to improve the accuracy and reliability of event detection. By clustering and analyzing target motion trajectories, it can identify abnormal behavior patterns. Trajectory analysis is used not only for event classification but also for behavior prediction and scene understanding.

[0003] At present, a Chinese invention patent with publication number CN113313060A discloses a method for constructing a classification model of abnormal events based on video surveillance and a detection method. The method clusters the normal event data set through the K-Means clustering algorithm, selects the normal event data located at the abnormal point as the false abnormal event data set, and combines other normal event data sets and abnormal event data sets to form a sample training set; the SVM model is trained based on the above sample training set to obtain a trained SVM model, but the related technology does not analyze the relevant data of different people, things, and events to establish a comprehensive mapping relationship, which is not conducive to the comprehensiveness of abnormal event monitoring, does not classify according to the different abnormal event situations, is not conducive to the clarity of the early warning, and also does not take into account the dynamic development of event types. Summary of the Invention

[0004] The technical problem solved by the present invention is that the relevant technology does not analyze the relevant data of different people, things and events to establish a comprehensive mapping relationship, which is not conducive to the comprehensiveness of abnormal event monitoring; does not classify abnormal events according to their different situations, which is not conducive to the clarity of early warning; and also does not take into account the dynamic development of event types.

[0005] To solve the above technical problems, the present invention provides the following technical solutions: an outdoor monitoring automatic event classification system, comprising a construction module, a division module and an output module;

[0006] The construction module establishes a first mapping relationship based on a preset event classification and historical surveillance videos;

[0007] The classification module classifies the levels of the classified events in the historical surveillance video according to the level classification logic to obtain levels corresponding to the classified events;

[0008] The output module outputs the current time classification and the current level to the monitoring center.

[0009] As a preferred embodiment of the outdoor monitoring automatic event classification system of the present invention, the event classification includes illegal intrusion events and abnormal behavior events, wherein the illegal intrusion events include illegal intrusion by people, illegal entry by vehicles, and illegal intrusion by wild animals, and the abnormal behavior events include fighting and falling;

[0010] The construction module is connected to a security monitoring system, retrieves historical monitoring videos from the security monitoring system, and performs a first preprocessing, a second preprocessing, and a third preprocessing on each historical monitoring video.

[0011] As a preferred solution of the outdoor monitoring automatic event classification system described in the present invention, the first preprocessing includes:

[0012] Select any historical surveillance video, set the first time interval as a segmentation step, and segment the historical surveillance video into N historical sub-videos according to the segmentation step;

[0013] performing a second preprocessing on the sub-videos according to the time sequence of the sub-videos;

[0014] The second preprocessing includes:

[0015] Obtaining the brightness of the sub-video, setting a first value as a brightness threshold, comparing the brightness of the sub-video with the first value, and if the brightness of the sub-video is less than the first value, performing enhancement processing on the sub-video; and if the brightness of the sub-video is greater than or equal to the first value, jumping to the next sub-video and repeating the brightness comparison process until all sub-videos are processed, stopping the second preprocessing;

[0016] The third preprocessing comprises:

[0017] The second preprocessed sub-video is frame-divided and windowed to obtain sub-images of the sub-video, and the sub-images that do not have facial shape features, vehicle shape features, or wild animal shape features are deleted, and the remaining sub-images after deletion are set as the first images corresponding to the sub-video.

[0018] As a preferred solution of the outdoor monitoring automatic event classification system described in the present invention, the facial shape features, vehicle shape features and wild animal shape features are obtained by obtaining facial images, vehicle images and wild animal images from an image database, and then extracting the first shape feature of each image through machine vision. The second shape feature in the sub-image is extracted through machine vision, and the similarity between the first shape feature and the second shape feature is calculated through the cosine similarity formula. The second value is set as the similarity threshold, and the similarity is compared with the second value. When the similarity is greater than or equal to the second value, it is judged that the facial shape feature or the vehicle shape feature or the wild animal shape feature exists in the sub-image. When the similarity is less than the second value, it is judged that the facial shape feature or the vehicle shape feature or the wild animal shape feature does not exist in the sub-image.

[0019] As a preferred solution of the outdoor monitoring automatic event classification system described in the present invention, the construction logic of the first mapping relationship includes:

[0020] Preset event classification, define event classification rules, and establish a first mapping relationship between the defined rules and event classification;

[0021] By inputting rules into the first mapping relationship, event classification is obtained.

[0022] As a preferred solution of the outdoor monitoring automatic event classification system described in the present invention, the event classification rules include:

[0023] Acquire the first image and set up an electronic fence;

[0024] The rules for illegal wildlife intrusion are: there are wild animals distributed on the boundary of the electronic fence or inside the electronic fence;

[0025] The rules for illegal intrusion are: the presence of a face, the face being located on the boundary of the electronic fence, and the person's posture corresponding to the face being in a climbing or leaping posture;

[0026] The rules for illegal vehicle entry are: the vehicle exists, the vehicle is located inside the electronic fence, there is no vehicle entry or exit record, or the recorded license plate number is different from the license plate number in the first image;

[0027] The rules for fighting are: the presence of a human face, the face being inside or outside the electronic fence, and the person holding a weapon or engaging in physical collision.

[0028] The rule for falling is: there is a face, and the person's posture corresponding to the face is lying down;

[0029] When a vehicle is present in the first image and is located within the electronic fence, the license plate number of the vehicle is identified, the monitoring time point at that time is obtained, and the security record database is retrieved to retrieve the vehicle entry and exit records corresponding to the event in the security record database. If no vehicle entry or exit record exists or the recorded license plate number is different from the license plate number in the first image, the event is set as illegal vehicle entry;

[0030] When there are two or more faces in the first image, the postures of the people corresponding to the faces are identified through the first images in time sequence. When the posture is a physical collision, the event classification is set as a fight.

[0031] As a preferred solution of the outdoor monitoring automatic event classification system described in the present invention, the gesture recognition logic includes:

[0032] By retrieving the posture database, the falling posture images, lying posture images, limb collision posture images, climbing posture images and hand-held instrument images are retrieved from the posture database, and the third shape feature quantities of the falling posture images, lying posture images, limb collision posture images, climbing posture images and hand-held instrument images are extracted respectively, and the fourth shape feature quantity of the person corresponding to the face in the first image is extracted. The similarity of the third shape feature quantity and the fourth shape feature quantity is calculated by the cosine similarity formula, and each similarity is compared with the second value. When the similarity is greater than or equal to the second value, the corresponding posture image is set to the posture image of the person in the first image. When the similarity is less than the second value, jump to the next posture image and cyclically compare the similarities. When all posture images are jumped and the similarities are all less than or equal to the second value, the posture in the first image is judged as a normal posture, and subsequent event classification is not performed.

[0033] As a preferred solution of the outdoor monitoring automatic event classification system described in the present invention, the level classification logic includes:

[0034] After the event is classified, the first image of any sub-video is selected, a first number of human faces, wild animals, or vehicles in the corresponding event classification is counted, the first numbers corresponding to the first images in the sub-video are traversed, the growth rates of adjacent first numbers are calculated, and a first average of the growth rates of the sub-videos is calculated;

[0035] The third value and the fourth value are set as growth rate thresholds, wherein the third value is less than the fourth value, and the first average value is compared with the growth rate threshold. When the first average value is less than or equal to the third value, the level is set to the first level. When the first average value is greater than the third value and less than or equal to the fourth value, the level is set to the second level. When the first average value is greater than the fourth value, the level is set to the third level. The severity of the event classification represented by the first level, the second level and the third level are in ascending order.

[0036] As a preferred solution of the outdoor monitoring automatic event classification system described in the present invention, the calculation logic of the growth rate includes:

[0037] In chronological order, a second difference is obtained by subtracting the next first quantity from the previous first quantity, a first ratio of the second difference to the previous first quantity is calculated, and the first ratio is set as the growth rate of the next first quantity relative to the previous first quantity.

[0038] As a preferred solution of the outdoor monitoring automatic event classification system described in the present invention, the output module obtains the current event classification of the current monitoring video according to the first mapping relationship, and obtains the current level corresponding to the current event classification according to the level division logic, and outputs the current time classification and current level to the monitoring center.

[0039] The beneficial effects of the present invention are as follows: by establishing a mapping relationship between the event classification preset by the construction module and the historical monitoring video, the system can quickly identify and classify the current monitoring events, and grade the events according to the hierarchical division logic to ensure that the monitoring center can give priority to high-priority events. It can analyze the monitoring video in real time, automatically identify abnormal behaviors (such as fighting, falling, abnormal parking, etc.), and trigger corresponding alarms according to the type and severity of the event. The classified events and level information are transmitted to the monitoring center in real time to facilitate the management personnel to respond quickly. The automated processing reduces the dependence on manual monitoring and avoids the fatigue and negligence of manual monitoring. Based on AI and deep learning technology, the system can continuously optimize the event classification and level division logic to adapt to different scenarios and needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A schematic diagram of the basic flow of an outdoor monitoring automatic event classification system provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0041] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments.

[0042] Example, see Figure 1 , as an embodiment of the present invention, provides an outdoor monitoring automatic event classification system, including a construction module, a division module and an output module;

[0043] The construction module establishes a first mapping relationship based on a preset event classification and historical surveillance videos;

[0044] The classification module classifies the levels of the classified events in the historical surveillance video according to the level classification logic to obtain levels corresponding to the classified events;

[0045] The output module outputs the current time classification and the current level to the monitoring center.

[0046] The present invention establishes a mapping relationship between the event classification preset by the construction module and the historical monitoring video. The system can quickly identify and classify current monitoring events, and grade the events according to the hierarchical division logic to ensure that the monitoring center can give priority to high-priority events. It can analyze the monitoring video in real time, automatically identify abnormal behaviors (such as fighting, falling, abnormal parking, etc.), and trigger corresponding alarms according to the type and severity of the event. The classified events and level information are transmitted to the monitoring center in real time to facilitate rapid response by management personnel. Automated processing reduces dependence on manual monitoring and avoids fatigue and negligence of manual monitoring. Based on AI and deep learning technology, the system can continuously optimize event classification and level division logic to adapt to different scenarios and needs.

[0047] Event classification includes illegal intrusion events and abnormal behavior events. The illegal intrusion events include illegal intrusion by people, illegal entry by vehicles, and illegal intrusion by wild animals. The abnormal behavior events include fighting and falling.

[0048] The construction module is connected to a security monitoring system, retrieves historical monitoring videos from the security monitoring system, and performs a first preprocessing, a second preprocessing, and a third preprocessing on each historical monitoring video.

[0049] The first preprocessing includes:

[0050] Select any historical surveillance video, set the first time interval as a segmentation step, and segment the historical surveillance video into N historical sub-videos according to the segmentation step;

[0051] performing a second preprocessing on the sub-videos according to the time sequence of the sub-videos;

[0052] The second preprocessing includes:

[0053] Obtaining the brightness of the sub-video, setting a first value as a brightness threshold, comparing the brightness of the sub-video with the first value, and if the brightness of the sub-video is less than the first value, performing enhancement processing on the sub-video; and if the brightness of the sub-video is greater than or equal to the first value, jumping to the next sub-video and repeating the brightness comparison process until all sub-videos are processed, stopping the second preprocessing;

[0054] The third preprocessing comprises:

[0055] The second preprocessed sub-video is frame-divided and windowed to obtain sub-images of the sub-video, and the sub-images that do not have facial shape features, vehicle shape features, or wild animal shape features are deleted, and the remaining sub-images after deletion are set as the first images corresponding to the sub-video.

[0056] In the specific implementation, dividing a long video into multiple shorter sub-videos can reduce the amount of data processed at a single time and improve the processing speed. The division of sub-videos makes the subsequent brightness processing and feature extraction more flexible, and can perform targeted processing according to the characteristics of different sub-videos. The video content in different time periods may vary greatly. After segmentation, it can better adapt to scene changes. Enhancing sub-videos with insufficient brightness can improve image quality and reduce the loss of details caused by insufficient lighting. The enhanced image is more conducive to subsequent feature detection (such as faces, vehicles, etc.). Sub-videos with normal brightness are skipped directly to avoid unnecessary processing, save computing resources, delete irrelevant sub-images, reduce the amount of data for subsequent processing, and improve system efficiency.

[0057] The facial shape features, vehicle shape features and wild animal shape features are obtained by obtaining facial images, vehicle images and wild animal images from an image database, extracting the first shape feature of each image through machine vision, extracting the second shape feature in the sub-image through machine vision, and calculating the similarity between the first shape feature and the second shape feature through a cosine similarity formula, setting the second value as a similarity threshold, and comparing the similarity with the second value. When the similarity is greater than or equal to the second value, it is determined that the facial shape features, vehicle shape features or wild animal shape features exist in the sub-image; when the similarity is less than the second value, it is determined that the facial shape features, vehicle shape features or wild animal shape features do not exist in the sub-image.

[0058] In specific implementation, the shape features of faces, vehicles and wild animals extracted from the image database through machine vision technology can more accurately describe the shape information of the target and adapt to different types of images, including faces, vehicles and wild animals. It has good versatility and can adapt to the recognition needs in different scenarios by adjusting the similarity threshold, thereby improving the adaptability of the system.

[0059] The construction logic of the first mapping relationship includes:

[0060] Preset event classification, define event classification rules, and establish a first mapping relationship between the defined rules and event classification;

[0061] By inputting rules into the first mapping relationship, event classification is obtained.

[0062] In specific implementation, by pre-setting event classification, the event type is clarified, ambiguity and uncertainty are avoided, and the system can quickly identify and respond to different types of events. Through preset rules and mapping relationships, the system can quickly classify monitoring events, reducing the time and complexity of event processing. Through rapid classification, the monitoring center can obtain event type and priority information in a timely manner, thereby quickly responding to and processing events. The classified event data can serve as the basis for intelligent analysis, such as behavior pattern analysis, event trend prediction, etc.

[0063] The event classification rules include:

[0064] Acquire the first image and set up an electronic fence;

[0065] The rules for illegal wildlife intrusion are: there are wild animals distributed on the boundary of the electronic fence or inside the electronic fence;

[0066] The rules for illegal intrusion are: the presence of a face, the face being located on the boundary of the electronic fence, and the person's posture corresponding to the face being in a climbing or leaping posture;

[0067] The rules for illegal vehicle entry are: the vehicle exists, the vehicle is located inside the electronic fence, there is no vehicle entry or exit record, or the recorded license plate number is different from the license plate number in the first image;

[0068] The rules for fighting are: the presence of a human face, the face being inside or outside the electronic fence, and the person holding a weapon or engaging in physical collision.

[0069] The rule for falling is: there is a face, and the person's posture corresponding to the face is lying down;

[0070] When a vehicle is present in the first image and is located within the electronic fence, the license plate number of the vehicle is identified, the monitoring time point at that time is obtained, and the security record database is retrieved to retrieve the vehicle entry and exit records corresponding to the event in the security record database. If no vehicle entry or exit record exists or the recorded license plate number is different from the license plate number in the first image, the event is set as illegal vehicle entry;

[0071] When there are two or more faces in the first image, the postures of the people corresponding to the faces are identified through the first images in time sequence. When the posture is a physical collision, the event classification is set as a fight.

[0072] In the specific implementation, by setting multiple conditions (such as posture, position, license plate number comparison, etc.) to judge the event type, the misjudgment that may be caused by a single condition is reduced, the efficiency of outdoor monitoring is improved, and it is conducive to the comprehensiveness of monitoring.

[0073] The gesture recognition logic includes:

[0074] By retrieving the posture database, the falling posture images, lying posture images, limb collision posture images, climbing posture images and hand-held instrument images are retrieved from the posture database, and the third shape feature quantities of the falling posture images, lying posture images, limb collision posture images, climbing posture images and hand-held instrument images are extracted respectively, and the fourth shape feature quantity of the person corresponding to the face in the first image is extracted. The similarity of the third shape feature quantity and the fourth shape feature quantity is calculated by the cosine similarity formula, and each similarity is compared with the second value. When the similarity is greater than or equal to the second value, the corresponding posture image is set to the posture image of the person in the first image. When the similarity is less than the second value, jump to the next posture image and cyclically compare the similarities. When all posture images are jumped and the similarities are all less than or equal to the second value, the posture in the first image is judged as a normal posture, and subsequent event classification is not performed.

[0075] The grading logic includes:

[0076] After the event is classified, the first image of any sub-video is selected, a first number of human faces, wild animals, or vehicles in the corresponding event classification is counted, the first numbers corresponding to the first images in the sub-video are traversed, the growth rates of adjacent first numbers are calculated, and a first average of the growth rates of the sub-videos is calculated;

[0077] The third value and the fourth value are set as growth rate thresholds, wherein the third value is less than the fourth value, and the first average value is compared with the growth rate threshold. When the first average value is less than or equal to the third value, the level is set to the first level. When the first average value is greater than the third value and less than or equal to the fourth value, the level is set to the second level. When the first average value is greater than the fourth value, the level is set to the third level. The severity of the event classification represented by the first level, the second level and the third level are in ascending order.

[0078] The calculation logic of the growth rate includes:

[0079] In chronological order, a second difference is obtained by subtracting the next first quantity from the previous first quantity, a first ratio of the second difference to the previous first quantity is calculated, and the first ratio is set as the growth rate of the next first quantity relative to the previous first quantity.

[0080] In specific implementation, the first number of human faces, wild animals or vehicles in the statistical event classification provides basic data for evaluating the severity of the event, calculates the growth rate of adjacent first numbers, and calculates the first average of the growth rate of sub-videos, which can reflect the development speed and trend of the event. Through level division, the monitoring center can give priority to events with higher severity and improve response efficiency.

[0081] The output module obtains the current event classification of the current monitoring video according to the first mapping relationship, obtains the current level corresponding to the current event classification according to the level division logic, and outputs the current time classification and the current level to the monitoring center.

[0082] In the specific implementation, the output module obtains the event classification of the current monitoring video according to the first mapping relationship, clarifies the type of event (such as illegal intrusion, abnormal behavior, etc.), so that the monitoring center can quickly understand the nature of the event. According to the level division logic, the output module divides the severity of the event into different levels (such as the first level, the second level, and the third level), providing the monitoring center with more detailed event information, and can output event information in real time to ensure that the monitoring center always grasps the latest monitoring status and takes timely measures.

[0083] The present invention establishes a mapping relationship between the event classification preset by the construction module and the historical monitoring video. The system can quickly identify and classify current monitoring events, and grade the events according to the hierarchical division logic to ensure that the monitoring center can give priority to high-priority events. It can analyze the monitoring video in real time, automatically identify abnormal behaviors (such as fighting, falling, abnormal parking, etc.), and trigger corresponding alarms according to the type and severity of the event. The classified events and level information are transmitted to the monitoring center in real time to facilitate rapid response by management personnel. Automated processing reduces dependence on manual monitoring and avoids fatigue and negligence of manual monitoring. Based on AI and deep learning technology, the system can continuously optimize event classification and level division logic to adapt to different scenarios and needs.

[0084] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium may be implemented by any type of volatile or non-volatile storage device, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0085] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. 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 solutions of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, and all of these should be included in the scope of the claims of the present invention.

Claims

1. An outdoor monitoring automatic event classification system, characterized in that: Includes building module, partitioning module and output module; The construction module establishes a first mapping relationship based on a preset event classification and historical surveillance videos; The classification module classifies the levels of the classified events in the historical surveillance video according to the level classification logic to obtain levels corresponding to the classified events; The output module outputs the current time classification and current level to the monitoring center; The construction logic of the first mapping relationship includes: Preset event classification, define event classification rules, and establish a first mapping relationship between the defined rules and event classification; By inputting rules into the first mapping relationship, an event classification is obtained; The event classification rules include: Acquire the first image and set up an electronic fence; The rules for illegal wildlife intrusion are: there are wild animals distributed on the boundary of the electronic fence or inside the electronic fence; The rules for illegal intrusion are: the presence of a face, the face being located on the boundary of the electronic fence, and the person's posture corresponding to the face being in a climbing or leaping posture; The rules for illegal vehicle entry are: the vehicle exists, the vehicle is located inside the electronic fence, there is no vehicle entry or exit record, or the recorded license plate number is different from the license plate number in the first image; The rules for fighting are: the presence of a human face, the face being inside or outside the electronic fence, and the person holding a weapon or engaging in physical collision. The rule for falling is: there is a face, and the person's posture corresponding to the face is lying down; When a vehicle is present in the first image and is located within the electronic fence, the license plate number of the vehicle is identified, the monitoring time point at that time is obtained, and the security record database is retrieved to retrieve the vehicle entry and exit records corresponding to the event in the security record database. If no vehicle entry or exit record exists or the recorded license plate number is different from the license plate number in the first image, the event is set as illegal vehicle entry; When there are two or more faces in the first image, the postures of the people corresponding to the faces are identified through the first images in time sequence. If the posture is a physical collision, the event classification is set to fighting; The gesture recognition logic includes: By retrieving a posture database, retrieving falling posture images, lying posture images, limb collision posture images, climbing posture images and hand-held instrument images from the posture database, extracting third shape feature quantities of the falling posture images, lying posture images, limb collision posture images, climbing posture images and hand-held instrument images respectively, extracting fourth shape feature quantities of the person corresponding to the face in the first image, calculating the similarity of the third shape feature quantity and the fourth shape feature quantity by using a cosine similarity formula, comparing each similarity with a second value, when the similarity is greater than or equal to the second value, setting the corresponding posture image as the posture image of the person in the first image, when the similarity is less than the second value, jumping to the next posture image, and cyclically comparing the similarities, when all posture images have been jumped and the similarities are all less than or equal to the second value, judging the posture in the first image as a normal posture, and not performing subsequent event classification; The grading logic includes: After the event is classified, the first image of any sub-video is selected, a first number of human faces, wild animals, or vehicles in the corresponding event classification is counted, the first numbers corresponding to the first images in the sub-video are traversed, the growth rates of adjacent first numbers are calculated, and a first average of the growth rates of the sub-videos is calculated; The third value and the fourth value are set as growth rate thresholds, wherein the third value is less than the fourth value, and the first average value is compared with the growth rate threshold. When the first average value is less than or equal to the third value, the level is set to the first level; when the first average value is greater than the third value and less than or equal to the fourth value, the level is set to the second level; when the first average value is greater than the fourth value, the level is set to the third level, and the severity of the event classification represented by the first level, the second level, and the third level are in ascending order; The calculation logic of the growth rate includes: In chronological order, a second difference is obtained by subtracting the next first quantity from the previous first quantity, a first ratio of the second difference to the previous first quantity is calculated, and the first ratio is set as the growth rate of the next first quantity relative to the previous first quantity.

2. The outdoor monitoring automatic event classification system according to claim 1, characterized in that: Event classification includes illegal intrusion events and abnormal behavior events. The illegal intrusion events include illegal intrusion by people, illegal entry by vehicles, and illegal intrusion by wild animals. The abnormal behavior events include fighting and falling. The construction module is connected to a security monitoring system, retrieves historical monitoring videos from the security monitoring system, and performs a first preprocessing, a second preprocessing, and a third preprocessing on each historical monitoring video.

3. The outdoor monitoring automatic event classification system according to claim 2, characterized in that: The first preprocessing includes: Select any historical surveillance video, set the first time interval as a segmentation step, and segment the historical surveillance video into N historical sub-videos according to the segmentation step; performing a second preprocessing on the sub-videos according to the time sequence of the sub-videos; The second preprocessing includes: Obtaining the brightness of the sub-video, setting a first value as a brightness threshold, comparing the brightness of the sub-video with the first value, and if the brightness of the sub-video is less than the first value, performing enhancement processing on the sub-video; and if the brightness of the sub-video is greater than or equal to the first value, jumping to the next sub-video and repeating the brightness comparison process until all sub-videos are processed, stopping the second preprocessing; The third preprocessing comprises: The second preprocessed sub-video is frame-divided and windowed to obtain sub-images of the sub-video, and the sub-images that do not have facial shape features, vehicle shape features, or wild animal shape features are deleted, and the remaining sub-images after deletion are set as the first images corresponding to the sub-video.

4. The outdoor monitoring automatic event classification system according to claim 3, characterized in that: The facial shape features, vehicle shape features and wild animal shape features are obtained by obtaining facial images, vehicle images and wild animal images from an image database, extracting the first shape feature of each image through machine vision, extracting the second shape feature in the sub-image through machine vision, and calculating the similarity between the first shape feature and the second shape feature through a cosine similarity formula, setting the second value as a similarity threshold, and comparing the similarity with the second value. When the similarity is greater than or equal to the second value, it is determined that the facial shape features, vehicle shape features or wild animal shape features exist in the sub-image; when the similarity is less than the second value, it is determined that the facial shape features, vehicle shape features or wild animal shape features do not exist in the sub-image.

5. The outdoor monitoring automatic event classification system according to claim 1, characterized in that: The output module obtains the current event classification of the current monitoring video according to the first mapping relationship, obtains the current level corresponding to the current event classification according to the level division logic, and outputs the current time classification and the current level to the monitoring center.

Images