Campus spoofing online self-identification system

Abstract

The invention provides a campus spoofing online self-recognition system, which comprises a scene marking module for marking the spoofing possibility according to the scene corresponding to the arrangement position of each camera to obtain a scene table; the scene correlation analysis module is used for taking a scene with high possibility of deception in an external input monitoring blind area as a deception high-incidence scene, analyzing the consistency between the deception high-incidence scene and each scene in the scene table, and updating the scene table according to the consistency to obtain an optimization table; the data acquisition module is used for acquiring a current scene image shot by each camera as an image to be analyzed; the suspicious analysis module is used for obtaining a deception suspicious value through analysis of a dynamic analysis strategy, a semantic analysis strategy and a morphological analysis strategy according to the to-be-analyzed image; the decision output module is used for outputting a voice alarm of the camera in the scene and / or pushing an instruction to a guardian according to the comparison of the spoofing suspicious value and a threshold value; the method has the advantage of analyzing whether the spoofing occurs in the current scene according to the possibility of occurrence of the spoofing in the scene.

Description

Technical Field

[0001] The present invention relates to the technical field of educational affairs systems, and more specifically to an online self-identification system for school bullying. Background Art

[0002] There are many forms of school bullying, including but not limited to physical violence, verbal bullying, destruction of personal property, and cyber bullying. Among them, physical violence is the most serious harm to teenagers and may cause long-term psychological damage to the victims. Bullying often occurs on or around campus and is hidden and difficult to detect.

[0003] The current identification method is a traditional monitoring method, which has limitations. Cameras are usually used to monitor various areas of the school. Monitoring personnel need to monitor the images of multiple cameras at the same time, which makes it difficult to achieve comprehensive and detailed monitoring. With the continuous development of artificial intelligence technology, AI-based image recognition and audio monitoring technology provide new solutions for campus bullying monitoring. By deploying high-definition cameras and intelligent audio monitoring equipment in key areas such as teaching buildings, playgrounds, and canteens, real-time monitoring and early warning of abnormal situations on campus can be achieved. For example, the existing invention patent with publication number: CN117237155A discloses an artificial intelligence-based smart campus student behavior analysis system, which mainly obtains audio from suspicious areas, semantically analyzes whether the sentences are bullying sentences, and then analyzes the number of people based on voiceprints. This method is prone to misjudgment due to fuzzy speech or difficulty in judging hidden locations. Summary of the invention

[0004] In view of the shortcomings of the prior art, the purpose of the present invention is to provide an online self-identification system for school bullying, which can further analyze whether bullying occurs in the current scenario based on the possibility of bullying occurring in the scenario, so as to make timely and accurate judgments and warnings.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] An online self-identification system for school bullying, including

[0007] A scene marking module marks the bullying possibility of the corresponding scenes according to the locations of the cameras based on historical data and / or expert experience to obtain a scene table;

[0008] The scene association analysis module takes scenes with high possibility of bullying in the external input monitoring blind area as scenes with high incidence of bullying, analyzes the degree of connection between the scenes with high incidence of bullying and each scene in the scene table, and updates the scene table according to the degree of connection to obtain an optimized table;

[0009] A data acquisition module, which acquires the current scene images captured by each camera in the campus as images to be analyzed according to the scene sorting in the optimization table;

[0010] A suspiciousness analysis module, which obtains a bullying suspicion value by analyzing the image to be analyzed through a dynamic analysis strategy, a semantic analysis strategy, and a morphological analysis strategy, and attaches the bullying suspicion value to an optimization table;

[0011] The decision output module outputs a voice warning of the camera in the scenario and / or pushes instructions to the guardian based on the comparison between the bullying suspicion value and the threshold.

[0012] Furthermore, the dynamic analysis strategy includes a target detection and tracking step and a target behavior analysis step.

[0013] The target detection and tracking step is to obtain a target object through target detection according to the image to be analyzed, and then continuously track the target by extracting target features to obtain a behavior trajectory of the target object;

[0014] The target behavior analysis step obtains the behavior type of the target object through deep learning or database matching based on the behavior trajectory, determines whether there is bullying behavior in the behavior type, and outputs a first scoring instruction or an ignore instruction based on the judgment result.

[0015] Furthermore, the semantic analysis strategy includes a semantic analysis step and a word and sentence matching step.

[0016] The semantic analysis step automatically translates the corresponding audio data in the image to be analyzed into text segments according to the semantic model;

[0017] The phrase matching step is to screen the text segment through bullying text matching or bullying text synonym matching to determine whether there is bullying information in the current scenario, and output a second scoring instruction or an ignore instruction based on the judgment result.

[0018] Furthermore, the morphological analysis strategy includes a face analysis step and a part analysis step.

[0019] The face analysis step includes intercepting the face image of each target object through face recognition according to the image to be analyzed, obtaining the expression type of the current object through deep learning or database matching according to the face image, and outputting a third scoring instruction according to the expression type;

[0020] The part analysis step obtains the head information and / or clothing information of the target object through feature extraction based on the image to be analyzed, and outputs a fourth scoring instruction through deep learning or database matching based on the head information and / or clothing information.

[0021] Furthermore, the decision output module, when the bullying suspicion value is greater than or equal to a threshold, filters the bullied object and the bullying object according to the behavior type of the target object and / or the expression type of the current object and / or the head information and / or the clothing information, connects to an external academic affairs system, indexes the personal profile information of the bullied object and the bullying object in the academic affairs system, and obtains the violence tendency value of the bullying object and the mental health value of the bullied object through weight analysis based on the personal profile information.

[0022] Furthermore, the personal file information includes a learning record column, a family record column and a past evaluation column. The learning record column reflects the student's academic performance over the years, the family record column reflects the student's family members and home visit records, and the past evaluation column reflects the past evaluation records of the student by teachers, parents and classmates.

[0023] Furthermore, the suspicion analysis module screens the time taken by each student to enter and exit high-incidence bullying scenes according to the image to be analyzed, determines the bullying suspicion value according to the time taken, and attaches the bullying suspicion value corresponding to the analysis to the optimization table.

[0024] Furthermore, the phrase matching step determines whether there is a help message in the current scenario through help text matching or help text synonym matching based on the text segment. If so, the help information is transmitted to the decision output module.

[0025] Furthermore, the scene association analysis module obtains a three-dimensional model of the campus in the external academic affairs system, marks the high-incidence bullying scenes in the three-dimensional model to obtain a first marking point, and marks each scene in the scene table in the three-dimensional model to obtain a number of second marking points, and analyzes the lateral distance between the first marking point and each second marking point, wherein the lateral distance reflects the degree of connection.

[0026] Furthermore, the high incidence scenes of bullying include toilets or utility rooms.

[0027] The beneficial effects of the present invention are as follows: by means of scene association analysis, high-incidence scenes of bullying in monitoring blind spots are associated with monitoring scenes, so as to analyze the videos in monitoring scenes and judge the suspicious values ​​of bullying in high-incidence scenes. Specifically, by comprehensively using multiple strategies such as dynamic analysis, semantic analysis and morphological analysis, the system can more accurately identify potential bullying behaviors from images captured by cameras. At the same time, the system can process monitoring data in real time and respond immediately when suspicious behaviors are found, thereby greatly shortening the time interval from the occurrence of an event to its identification. In addition, for some highly vigilant bullying, the morphology of the bullied students can be analyzed to avoid the students from panicking and failing to report to their parents and teachers in time, which leads to reduced mental health of the students.

[0028] In addition, data acquisition can be multi-channel data collection, such as faces, movements, and trajectories obtained from camera video analysis; speech semantic analysis; student health records, etc., comprehensive analysis and decision-making capabilities, including multi-dimensional information integration: scoring and optimizing table mechanisms based on scenarios and relevance; active identification and early warning of early prevention, personalization, and precise intervention. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is the overall flow chart of the present invention;

[0030] Figure 2 It is the optimization table display diagram in the present invention;

[0031] Figure 3 It is a screenshot of the system in the present invention. DETAILED DESCRIPTION

[0032] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0033] The current identification means are traditional monitoring means, which have limitations. Usually, cameras are used to monitor various areas of the school. Monitoring personnel need to monitor the images of multiple cameras at the same time, which makes it difficult to achieve comprehensive and detailed monitoring. Therefore, the present invention designs this school bullying online self-identification system, such as Figure 1-3 Shown, including

[0034] The scene marking module marks the corresponding scenes according to the location of each camera and the degree of bullying possibility based on historical data and / or expert experience to obtain a scene table. By accurately locating high-risk areas, the system can allocate resources more effectively and improve monitoring efficiency. It is assumed that the corridors, top floors, and corners of the top floor stairs are included. According to historical data and expert experience, it can be judged that the top floor is more likely to be bullying. Therefore, the top floor is the first scene in the scene table, and the scenes are sorted in order of bullying possibility.

[0035] Scene association analysis module: Since toilets, changing rooms and other places are private spaces, these places will not be randomly installed with visual cameras for monitoring. Scenes with fewer people, such as utility rooms, will not be randomly installed with visual cameras for monitoring. Therefore, scenes with high possibility of bullying in the external input monitoring blind spots are regarded as high-incidence bullying scenes (wherein high-incidence bullying scenes include toilets, changing rooms or utility rooms, etc.), and the connection between high-incidence bullying scenes and each scene in the scene table is analyzed. The specific connection analysis is: first, obtain the three-dimensional model of the campus in the external academic affairs system, mark the high-incidence bullying scenes in the three-dimensional model to obtain the first marking point, and mark each scene in the scene table in the three-dimensional model to obtain a number of second marking points, and analyze the lateral distance between the first marking point and each second marking point. The lateral distance is the horizontal distance. Since the first marking point and the second marking point may have the smallest longitudinal distance, and the longitudinal distance may reflect the distance between the upper floor and the lower floor, the present invention mainly uses the lateral distance as a reference factor, wherein the lateral distance reflects the degree of contact. Assuming that the toilet is located in the middle of the corridor, that is, the toilet is the first marking point, and there is a visual camera A in the corridor that can capture the junction of the corridor and the toilet, then the camera scene A1 of the corridor is the second marking point, and the lateral distance between the first marking point and the second marking point is the smallest, which reflects the highest degree of contact. There is also a visual camera B at the other corner of the corridor, then the camera scene B1 is the second marking point, and the lateral distance is larger, so the degree of contact is smaller, which corresponds to the fact that A1 is closer to the toilet.

[0036] The data acquisition module obtains the current scene images captured by each camera in the campus as the images to be analyzed according to the scene sorting in the optimization table. The images to be analyzed include several continuous images, which also include matching audio. The images to be analyzed are also attached to the corresponding columns of each scene in the optimization table (in the data acquisition module, the scenes in the campus are sorted according to the optimized scene table, and the real-time image data of the top-ranked scenes are obtained first. The system grabs the currently captured scene images from each camera according to the sorting as the image source to be processed for subsequent analysis. This module ensures key monitoring of high-risk areas for bullying and provides more accurate and timely data support for the analysis module).

[0037] The suspicion analysis module obtains the bullying suspicion value according to the image to be analyzed through dynamic analysis strategy, semantic analysis strategy and morphological analysis strategy. Dynamic analysis is used to capture abnormal movement trajectory, semantic analysis identifies the aggressiveness of speech or expression, and morphological analysis focuses on physical contact or abnormal crowd behavior. Finally, the calculated bullying suspicion value is attached to the optimization table for subsequent risk assessment and early warning response. Specifically, the bullying suspicion value obtained by each analysis strategy can be attached to the optimization table respectively through the above three analysis strategies, or the bullying suspicion values ​​obtained by the three analysis strategies can be obtained by weight calculation to obtain the total bullying suspicion value.

[0038] Specifically, the dynamic analysis strategy includes target detection and tracking steps and target behavior analysis steps.

[0039] In the target detection and tracking step, the input image to be analyzed is preprocessed, including scaling, cropping, and denoising. The target object is then obtained through target detection. The target object is mainly a student (person). All other parts of the image are ignored. The student position is then initialized, that is, the student position is determined in the first frame. In subsequent image frames, the student position is continuously tracked using methods such as feature matching and optical flow, and the student's behavior trajectory is finally obtained.

[0040] In the target behavior analysis step, assuming that bullying behavior occurs under surveillance, key features such as speed, acceleration, trajectory shape, etc. are extracted based on the student's behavior, and the behavior trajectory is identified using a deep learning model (such as a convolutional neural network, a recurrent neural network, etc.) or a database matching method. The behavior trajectory is matched with a predefined behavior type to obtain the behavior type of the target object. For example, if the behavior feature of a student A is raising a fist or kicking a leg, then according to the match, it is obtained that the behavior type contains bullying actions, and the first scoring instruction is output. If at least one student is simultaneously analyzed to have bullying actions within 1 minute, the first scoring score is higher. If no bullying actions are analyzed in this scenario, an ignore instruction is output to exclude the dynamic analysis strategy in this scenario.

[0041] Specifically, the semantic analysis strategy includes a semantic analysis step and a word and sentence matching step.

[0042] In the semantic analysis step, the corresponding audio data in the image to be analyzed is automatically translated into text segments according to the semantic model. Specifically, in audio processing, the corresponding image to be analyzed taken by the camera is matched with the audio data, and the interference sound in the audio is removed by using audio processing technology, and then the audio data is converted into text; the semantic model uses the semantic model in natural language processing technology (such as deep learning model, statistical language model, etc.) to further process the converted text to ensure the accuracy and readability of the text. The semantic model can handle grammatical errors, spelling errors and other problems, and try to understand the semantic meaning in the text; finally, the output result, the text (i.e., text segment) processed by the semantic model will serve as the input of the subsequent word matching step.

[0043] In the phrase matching step, the bullying text segment is screened by bullying text matching or bullying text synonym matching to determine whether there is bullying information in the current scenario, and a second scoring instruction or an ignore instruction is output according to the judgment result; specifically, bullying text matching: the voice corresponding to the image to be analyzed captured by the camera is automatically converted into a text segment by audio translation, and a database containing known bullying words and phrases is established. The converted text segment is matched with this database to check whether there is any direct bullying expression; bullying text synonym matching, because the Chinese word meaning library may have different words or phrases to express similar meanings, it is also necessary to consider synonyms or related expressions of bullying words, and expand the matching range through a synonym library or more complex semantic analysis technology; judgment and output, if bullying information is found in the phrase matching step, the second scoring instruction is output, and if no bullying information is found, the ignore instruction is output;

[0044] The bullying word matching rules include: full word matching or keyword matching. Full word matching is more accurate. For example, if the bullying word "xxx" appears in a word segment, it is considered that there is a possibility of bullying. This method is full word matching. In addition, if the keyword in the bullying word "x1 x2x3" is "x2", and according to the meaning of the bullying word, the keyword "x2" can only be located after the word "x1" or before the word "x3", that is, the order of the keywords is the judgment condition. When this order appears in the word segment, the trigger condition is judged as the possibility of bullying.

[0045] In addition, the same semantic analysis as above is used to analyze the preceding and following paragraphs that are identified as containing bullying words. If the bullying word "xxx" or other bullying words are determined to be, the preceding or following paragraphs "xxxxx1" of the paragraph and the following or following paragraphs "xxxxx2" of the paragraph are analyzed to verify whether the bullying words are continuous bullying paragraphs or whether the bullying words are bullying behaviors in the continuous paragraphs. The purpose is to reduce the error rate in the analysis of bullying words.

[0046] The main purpose of the semantic analysis strategy is to use audio analysis to determine whether there are bullying words in the current scene, and to perform the second scoring based on the frequency and number of bullying words. There is no need to consider who said the bullying words. Compared with the traditional single voice analysis that accurately identifies the speaker of the bullying words through voiceprints, the semantic analysis of the present invention is only to ensure whether there is suspicion of bullying in the current scene, and can be combined with the dynamic analysis strategy to perform a specific bullying possibility analysis. Of course, the present invention can combine voiceprint analysis to accurately locate the speaker of the bullying words, so as to verify the bullying target obtained by the dynamic analysis.

[0047] In addition, the phrase matching step can also filter out the text segments through help text matching or help text synonym matching to determine whether there is a help message in the current scenario. If so, the rescue information is transmitted to the decision output module, where the help text may include "help", "stop fighting", "sorry", etc.

[0048] Specifically, the morphological analysis strategy includes a face analysis step and a part analysis step.

[0049] In the face analysis step, the face image of each target object is intercepted through face recognition according to the image to be analyzed, the expression type of the current object is obtained through deep learning or database matching according to the face image, and the third scoring instruction is output according to the expression type; specifically, first, face recognition and interception, using the existing face recognition technology to identify the facial areas of all target objects (people) from the image to be analyzed, and then intercept these facial images as expression analysis images; secondly, expression recognition, using deep learning models or database matching technology, expression recognition is performed on the intercepted facial images, and subtle changes in facial muscles are analyzed to determine the expression type of the current object, such as happiness, sadness, anger, etc. In general school bullying, the expression of the bullied is usually sad, angry or dull, and the expression of the bully is generally happy or angry, etc.; finally, the third scoring instruction is output. According to the recognized expression type, the system can output the corresponding scoring instruction. When the expression of being bullied appears, the scene will be scored.

[0050] In the part analysis step, the system extracts features from the image to be analyzed, focusing on capturing the head features and / or clothing features of the target object, identifying postures related to bullying behavior (such as abnormal head position, covering the face) or clothing features (such as gang logos, specific ways of wearing uniforms, etc.). When features that match the bullying situation are detected, the fourth scoring instruction is output to add points to the bullying suspicion value of the scene; if there are no relevant features, the analysis results are ignored. This step further improves the system's risk identification ability for the scene by refining the feature analysis, facilitating comprehensive early warning;

[0051] Specifically, the head information and / or clothing information of the target object is obtained through feature extraction based on the image to be analyzed, and the fourth scoring instruction is output through deep learning or database matching based on the head information and / or clothing information; specifically, first, feature extraction, in the part analysis step, the head information and / or clothing information of the target object is obtained from the image to be analyzed by feature extraction technology, and secondly, feature analysis is performed. Assuming that student A and student B appear in scene 1 for the first time, the system will intercept the image that appears for the first time, and record the head information and clothing information. When student A and student B appear in scene 1 again after entering a high-incidence bullying scene, their monitoring images are intercepted. At this time, the hairstyle, clothing position and other features of student A and student B at this time are analyzed and compared with the hairstyle and clothing position of student A and student B who first entered scene 1. It can be preliminarily determined whether bullying occurs, and it can be determined who is the bully and who is the bullied. If the analysis shows that bullying occurs, the fourth scoring instruction is output to score them.

[0052] The above-mentioned morphological analysis strategy mainly solves the problem of monitoring the faces and body parts of both bullies and bullied in monitoring scenarios as bullying scenes or walking out of high-incidence bullying scenes into monitoring scenarios, to prevent bullied persons from not reporting to guardians in time due to threats and intimidation, resulting in long-term living under the shadow of school bullying. In addition, the morphological analysis strategy is combined with the dynamic analysis strategy and the semantic analysis strategy to conduct specific bullying possibility analysis.

[0053] All of the above scores indicate suspicion of bullying.

[0054] In the decision output module, the system compares the suspicious bullying value of each scene with the preset threshold. When the suspicious value exceeds the threshold, the corresponding output mechanism is triggered: a voice warning is played near the scene camera to intervene in potential bullying behavior in real time; or a warning notification and processing instructions are sent to the guardian for quick response. This module realizes automated early warning and graded response, improving the efficiency of monitoring and intervention of bullying risks.

[0055] In addition, when the bullying suspicion value is greater than or equal to the threshold, the bullied object and the bullying object are screened according to the target object's behavior type and / or the current object's expression type and / or head information and / or clothing information, and the external teaching system is connected to index the personal file information of the bullied object and the bullying object in the teaching system, wherein the personal file information includes a learning record column, a family record column and a past evaluation column, wherein the learning record column reflects the student's academic performance over the years, the family record column reflects the student's family members and home visit records, and the past evaluation column reflects the teacher's, parents' and classmates' past evaluation records of the student, and the violent tendency value of the bullying object and the mental health value of the bullied object are obtained through weight analysis based on the personal file information;

[0056] Usually, school bullying often occurs between students with poor academic performance, so the learning record column can be used for preliminary screening. However, it cannot be ruled out that bullying may occur between students with average or above average academic performance. Therefore, screening can also be carried out through family records and past evaluations. For example, if there is parental violence, family conflicts, etc. in the family, this factor can be used as an evaluation weight for the violent tendency value or the mental health value. In addition, in past evaluations, the evaluations of teachers and classmates are also relatively important. The above-mentioned evaluations in the early stage were all recorded in a non-public manner, so the data authenticity is relatively high.

[0057] In addition to the dynamic analysis strategy, semantic analysis strategy, and morphological analysis strategy, the suspiciousness analysis module also needs to analyze the time consumed to enter the high-incidence bullying scene. If several students enter the toilet or utility room at the same time and the time exceeds the preset threshold, abnormal situations may occur, such as smoking, bullying, etc. The time consumed by each student to enter and exit the high-incidence bullying scene is screened according to the image to be analyzed, and the bullying suspicion value is determined based on the time consumed, and the bullying suspicion value corresponding to the analysis is attached to the optimization table;

[0058] Assigning weight logic: Figure 2 As shown, it is assumed that the time when student A and student B first appear in scene 1 (corridor A) is 11:24, and in the surveillance images of scene 1 (corridor A), potential bullying behavior can be indexed in the model based on morphological analysis and assigned a score of F1;

[0059] The verbal interaction between student A and student B is recorded in the surveillance video of scene 1 (corridor A) at 11:24. Then, based on semantic analysis, offensive, threatening or insulting words can be indexed in the database, and the corresponding score in the database is obtained as F2;

[0060] The facial images of students A and B are recorded in the surveillance images of scene 1 (corridor A) at 11:24, and the facial images of students A and B appearing in scene 1 (corridor A) again at 11:39 after entering the high-incidence bullying scene. The facial images are analyzed first to see whether there are any facial expressions (such as anger, fear, pain, etc.). Assuming that the facial expression of student A at 11:24 is fear, the score indexed in the database is f1, and the facial expression of student A at 11:39 is pain, the score indexed in the database is f2. At this time, only student A needs to be scored, and the score is F3 = f1 + f2. At this time, student A can be identified as a suspected victim of bullying.

[0061] In the surveillance image of scene 1 (corridor A) at 11:24, the head information and clothing information of student A and student B are recorded as normal. After entering the high-incidence bullying scene, student A and student B appear in scene 1 (corridor A) again at 11:39. During this period, the two students took 15 minutes to enter the blind spot. According to the abnormal occlusion or covering of the face by the head position of student A, the score f3 is obtained by matching the existing similarity in the matching library. According to whether the clothes of student A are damaged or the neatness of the clothes, the score f4 is obtained by matching the existing similarity in the matching library. At this time, student A can be identified as a suspected victim of bullying, and the total score of the morphological analysis is F4=f3+f4;

[0062] According to the set weight formula configuration:

[0063]

[0064] Among them, S is the suspected bullying value, is the normalization coefficient, which is used to standardize the scores of the four different analyses to the same level, e is the base of the natural logarithm, λ is the attenuation parameter, which controls the shape and sensitivity of the exponential function, and F 1 (t) is the morphological analysis scoring function, which gives scores according to the behavior pattern, F 2 (s) is a semantic analysis scoring function, which gives a score based on the content of the speech. It depends on the frequency and intensity of offensive, threatening or insulting words in the speech. 3 (f, t) is the facial expression analysis scoring function, which gives scores based on the expression type and time changes, where f represents the expression type (such as anger, fear, pain, etc.). The function takes into account the changes and persistence of the expression, F 4 (p,c) is a partial analysis scoring function, which gives scores based on the abnormal head position and the dressing condition. p represents the degree of occlusion or coverage of the abnormal head position, and c represents the degree of damage or neatness of the dressing.

[0065] The value range of S is (0,1], where values ​​close to 1 indicate highly suspicious bullying behavior, while values ​​close to 0 indicate low suspicion of bullying.

[0066] The above are only preferred embodiments of the present invention. The protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. An online self-identification system for school bullying, characterized by: include A scene marking module marks the bullying possibility of the corresponding scenes according to the locations of the cameras based on historical data and / or expert experience to obtain a scene table; The scene association analysis module takes scenes with high possibility of bullying in the external input monitoring blind area as scenes with high incidence of bullying, analyzes the degree of connection between the scenes with high incidence of bullying and each scene in the scene table, and updates the scene table according to the degree of connection to obtain an optimized table; A data acquisition module, which acquires the current scene images captured by each camera in the campus as images to be analyzed according to the scene sorting in the optimization table; A suspiciousness analysis module, which obtains a bullying suspicion value by analyzing the image to be analyzed through a dynamic analysis strategy, a semantic analysis strategy, and a morphological analysis strategy, and attaches the bullying suspicion value to an optimization table; The decision output module outputs a voice warning of the camera in the scenario and / or pushes instructions to the guardian based on the comparison between the bullying suspicion value and the threshold.

2. According to claim 1, a school bullying online self-identification system is characterized by: The dynamic analysis strategy includes a target detection and tracking step and a target behavior analysis step. The target detection and tracking step is to obtain a target object through target detection according to the image to be analyzed, and then continuously track the target by extracting target features to obtain a behavior trajectory of the target object; The target behavior analysis step obtains the behavior type of the target object through deep learning or database matching based on the behavior trajectory, determines whether there is bullying behavior in the behavior type, and outputs a first scoring instruction or an ignore instruction based on the judgment result.

3. According to claim 2, a school bullying online self-identification system is characterized by: The semantic analysis strategy includes a semantic analysis step and a word and sentence matching step. The semantic analysis step automatically translates the corresponding audio data in the image to be analyzed into text segments according to the semantic model; The phrase matching step is to screen the text segment through bullying text matching or bullying text synonym matching to determine whether there is bullying information in the current scenario, and output a second scoring instruction or an ignore instruction based on the judgment result.

4. According to claim 3, the online self-identification system for school bullying is characterized by: The morphological analysis strategy includes a face analysis step and a part analysis step. The face analysis step includes intercepting the face image of each target object through face recognition according to the image to be analyzed, obtaining the expression type of the current object through deep learning or database matching according to the face image, and outputting a third scoring instruction according to the expression type; The part analysis step obtains the head information and / or clothing information of the target object through feature extraction based on the image to be analyzed, and outputs a fourth scoring instruction through deep learning or database matching based on the head information and / or clothing information.

5. According to claim 4, a school bullying online self-identification system is characterized by: The decision output module, when the bullying suspicion value is greater than or equal to a threshold, filters the bullied object and the bullying object according to the behavior type of the target object and / or the expression type of the current object and / or the head information and / or the clothing information, connects to an external educational system, indexes the personal profile information of the bullied object and the bullying object in the educational system, and obtains the violence tendency value of the bullying object and the mental health value of the bullied object through weight analysis based on the personal profile information.

6. According to claim 5, a school bullying online self-identification system is characterized by: The personal profile information includes a learning record column, a family record column and a past evaluation column. The learning record column reflects the student's academic performance over the years, the family record column reflects the student's family members and home visit records, and the past evaluation column reflects the past evaluation records of the student by teachers, parents and classmates.

7. According to claim 1, the online self-identification system for school bullying is characterized by: The suspicion analysis module selects the time taken by each student to enter and exit the high-incidence bullying scene according to the image to be analyzed, determines the bullying suspicion value according to the time taken, and attaches the bullying suspicion value corresponding to the analysis to the optimization table.

8. According to claim 3, the online self-identification system for school bullying is characterized by: The phrase matching step is to screen the text segment through help text matching or help text synonym matching to determine whether there is help information in the current scenario. If so, the help information is transmitted to the decision output module.

9. According to claim 1, a school bullying online self-identification system is characterized by: The scene association analysis module obtains a three-dimensional model of the campus in the external academic affairs system, marks the high-incidence bullying scenes in the three-dimensional model to obtain a first marking point, and marks each scene in the scene table in the three-dimensional model to obtain a number of second marking points, and analyzes the lateral distance between the first marking point and each second marking point, wherein the lateral distance reflects the degree of connection.

10. According to claim 9, a school bullying online self-identification system is characterized by: The high incidence scenes of bullying include toilets or utility rooms.

Images