Human neural perception evaluation and calculation method for campus safety hazard risk and related devices

Abstract

The present invention provides a method and related device for evaluating and calculating human neural perception of campus safety hidden danger risks, and relates to the field of campus security information technology. Video materials of different places are collected from the campus security monitoring video history library, and are organized into video material sets according to the places and annotated; video clips of key time nodes are intercepted from the video materials according to preset rules to generate a test signal set; the EEG-fNIRS device is used to collect the neural activity data of the tested security personnel when watching the test signal, and their subjective scoring data is obtained simultaneously, including test task one and test task two; the LSTM model is trained based on the neural activity data of test task one, and combined with the neural activity data and subjective scoring data of test task two, the overall safety risk value of each place and the risk level of specific hidden danger type are generated. The technical effect of effectively discovering various complex and hidden campus safety hidden dangers is achieved.

Description

Technical Field

[0001] The present application relates to the field of campus security information technology, and in particular to a method and related device for evaluating and calculating human neural perception of campus safety hazards. Background Art

[0002] Campus safety risks are linked to numerous factors, including buildings, locations, facilities, road conditions, crowd activity, and weather changes. Traditional manual inspections are not only time-consuming and labor-intensive, but also difficult to detect in a timely manner, resulting in numerous unexpected safety incidents. With the advancement of security monitoring information technology and big data analytics, campus security risk perception and incident prevention capabilities have been significantly enhanced through machine learning, knowledge graph reasoning, and intelligent video and voice analysis. However, campus safety hazards are closely linked to specific locations, locations, internal and external environments, crowds, and the dynamic nature of weather, making comprehensive perception and prediction of potential risks difficult through these methods.

[0003] Therefore, how to effectively discover various complex and hidden campus safety hazards has become a technical problem that needs to be solved urgently. Summary of the Invention

[0004] In order to effectively discover various complex and hidden campus safety hazards, this application provides a human neural perception evaluation and calculation method for campus safety hazard risks and related devices.

[0005] In the first aspect, the present application provides a method for evaluating and calculating human neural perception of campus safety hazard risks using the following technical solutions:

[0006] A method for evaluating and calculating human neural perception of campus safety hazard risks, including:

[0007] Collect video footage from different locations from the campus security surveillance video history library, including videos of safety accidents and normal scenes, organize them into video material sets by location and annotate them;

[0008] Extracting video clips at key time points from the video material according to preset rules to generate a test signal set;

[0009] Using an EEG-fNIRS device to collect neural activity data of the security personnel under test when viewing the test signal, and simultaneously obtain their subjective scoring data, including test task one and test task two;

[0010] The LSTM model is trained based on the neural activity data of test task one, and combined with the neural activity data and subjective scoring data of test task two to generate the overall safety risk value of each location and the risk level of specific hidden danger types.

[0011] Optionally, the video of the safety incident may include:

[0012] 15 minutes of continuous video footage before the accident;

[0013] Covering no less than 5 types of climate environments and 3 types of activity scenarios;

[0014] If the actual video is insufficient, supplementary video material that meets the requirements can be generated through drills with real people.

[0015] Optionally, the test signal set includes:

[0016] For videos of safety accidents that have occurred, capture key time node segments 15 minutes, 10 minutes, and 5 minutes before the accident;

[0017] For normal scene videos, representative clips are captured in 60-second units according to the scene;

[0018] All clips are marked with location number, time node and scene type.

[0019] Optionally, the test task 1 includes: playing a video clip before the safety accident, and having the tested person score the possibility of the accident according to a preset time interval, and simultaneously collecting a neural activity dataset N1 and a scoring dataset Q;

[0020] Test task 2 includes: playing a video clip of a normal scene, and having the test subjects list the types of potential hidden dangers and risk scores, and simultaneously collecting the neural activity dataset N2 and the binary score dataset J.

[0021] Optionally, the step of training the LSTM model based on the neural activity data of test task one includes:

[0022] Preprocess the data of test task 1 and calculate the safety perception ability score of the test subjects, P = 0.1Q1 + 0.3Q2 + 0.6Q3. Filter the test subjects whose score P is higher than the mean for model training, where Q represents the score dataset, and Q1, Q2, and Q3 represent the specific scores in the score dataset.

[0023] The five frequency band energy parameters of the EEG signal and the mean, variance, and respiratory rate parameters of fNIRS HbO and HbR were extracted from the neural activity data to form a 10-dimensional feature set as LSTM input;

[0024] The LSTM model is trained based on the feature set. The initial parameters include input dimension, hidden layer, number of layers, and output dimension, and are tuned by MAPE value.

[0025] Optionally, the five frequency bands of the EEG signal include: 4-7 Hz, 8-10 Hz, 11-13 Hz, 14-15 Hz and 16-20 Hz.

[0026] Optionally, the step of generating the overall safety risk value and the risk level of specific hidden danger types for each location includes:

[0027] From the binary score dataset J, the types of hidden dangers are summarized by location, and the risk scores of each type of hidden danger are averaged;

[0028] Combined with the overall risk value output by the LSTM model, a safety assessment report is generated that includes the hidden danger type, risk level and priority.

[0029] In a second aspect, the present application provides a human neural perception evaluation and calculation device for campus safety hazard risks, comprising:

[0030] The data acquisition module is used to collect video materials from different locations from the campus security surveillance video history library, including videos of safety accidents and normal scenes, and organize them into video material sets by location and annotate them;

[0031] A test signal set generation module is used to intercept video clips of key time nodes from the video material according to preset rules to generate a test signal set;

[0032] A test task module, for collecting neural activity data of the security personnel under test when viewing the test signal using an EEG-fNIRS device, and simultaneously obtaining their subjective scoring data, including test task one and test task two;

[0033] The output module is used to train the LSTM model based on the neural activity data of test task one, and combine the neural activity data and subjective scoring data of test task two to generate the overall safety risk value of each location and the risk level of specific hidden danger types.

[0034] In a third aspect, the present application provides a computer device, comprising: a memory and a processor, wherein the processor executes the method described above when running computer instructions stored in the memory.

[0035] In a fourth aspect, the present application provides a computer-readable storage medium comprising instructions, which, when executed on a computer, enable the computer to execute the method described above.

[0036] To summarize, this application collects video footage from different locations from a historical library of campus security surveillance videos, organizes them into video material sets by location, and labels them. It then extracts video clips of key time nodes from the video footage according to preset rules to generate a test signal set. It then uses an EEG-fNIRS device to collect neural activity data from the security personnel being tested while they watch the test signals, and simultaneously obtains their subjective rating data, including test task one and test task two. It then trains an LSTM model based on the neural activity data from test task one, and combines the neural activity data and subjective rating data from test task two to generate the overall safety risk value and specific risk level for each location. This approach effectively discovers various complex and hidden campus safety hazards. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a schematic diagram of the computer device structure of the hardware operating environment involved in the embodiment of the present application;

[0038] Figure 2 This is a flow chart of the first embodiment of the human neural perception evaluation and calculation method for campus safety hazard risks in this application;

[0039] Figure 3 This is a structural block diagram of the first embodiment of the human neural perception evaluation and calculation device for campus safety hazard risks in this application. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solutions and advantages of this application more clearly understood, the present application is further described in detail below through the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0041] Reference Figure 1 , Figure 1 This is a schematic diagram of the computer device structure of the hardware operating environment involved in the embodiment of the present application.

[0042] like Figure 1As shown, the computer device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display and an input unit such as a keyboard. Optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a wireless fidelity (Wi-Fi) interface). The memory 1005 may be a high-speed random access memory (RAM) or a stable non-volatile memory (NVM), such as a disk storage device. The memory 1005 may also be a storage device independent of the processor 1001.

[0043] Those skilled in the art will understand that Figure 1 The structure shown in the figure does not constitute a limitation on the computer device, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0044] like Figure 1 As shown, the memory 1005 as a storage medium may include an operating system, a network communication module, a user interface module, and a human neural perception evaluation calculation program for campus safety hazard risks.

[0045] exist Figure 1 In the computer device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in this application can be set in the computer device, and the computer device calls the human neural perception evaluation and calculation program for campus safety hazard risks stored in the memory 1005 through the processor 1001, and executes the human neural perception evaluation and calculation method for campus safety hazard risks provided in the embodiment of this application.

[0046] The present application embodiment provides a method for calculating the human neural perception evaluation of campus safety hazard risk, referring to Figure 2 , Figure 2 This is a flow chart of the first embodiment of the human neural perception evaluation and calculation method for campus safety hazard risks in this application.

[0047] In this embodiment, the human neural perception evaluation and calculation method for campus safety hazard risk includes the following steps:

[0048] Step S10: Collect video materials of different places from the campus security surveillance video history library, including videos of safety accidents and normal scene videos, organize them into video material sets according to the locations and mark them.

[0049] It's important to note that the human brain possesses a rich subconscious safety perception capability. Experienced security personnel often experience subconscious neural activity that triggers anxiety and alarm when observing potentially hazardous situations. However, these subconscious neural activities are difficult to fully perceive and accurately express in a conscious state. Modern neural activity observation technologies provide effective means for assessing these subconscious safety perceptions. For example, electroencephalograms (EEGs) can subconsciously generate beta wave safety warning signals, which have played a crucial role in safety perception in areas such as mining and vehicle driving. It's worth noting that subconscious neural activity typically only reflects the perception of overall safety risk levels and is unable to clearly identify the types and severity of various potential risks. Therefore, it's necessary to combine these perceptions with subjective human evaluations based on cognition to better assess campus safety risks.

[0050] EEG signals have the advantages of fast response and high temporal resolution, but their spatial resolution is very low, making it difficult to accurately locate brain functional areas of neural activity, making it difficult to obtain more comprehensive subconscious safety perception information. Functional near-infrared spectroscopy (fNIRS) has the advantage of high spatial resolution, but its temporal resolution is relatively low.

[0051] It should be noted that the video of the safety accident includes: a continuous video clip of 15 minutes before the accident; covering no less than 5 types of climate environments and 3 types of activity scenes; if the actual video is insufficient, the video material that meets the requirements will be supplemented through real crowd drills.

[0052] In specific implementation, the steps for collecting campus scenario video materials include: collecting video materials from different locations and scenarios from the campus security surveillance video history library, which should include videos of safety accidents that have occurred and representative normal videos of all locations in different climate environments and typical activity scenes, and organizing them into video material sets by location and providing descriptions; among them, videos of safety accidents that have occurred in each location should include no less than 15 minutes of continuous video before the accident in five different climate environments and three different activity scenes; if the safety accident videos that have occurred in the location are insufficient, real-life crowd drills can be used to supplement them;

[0053] Step S20: extracting video clips at key time nodes from the video material according to preset rules to generate a test signal set.

[0054] The test signal set includes: for videos of safety accidents that have occurred, key time node segments are captured 15 minutes, 10 minutes, and 5 minutes before the accident; for normal scene videos, representative segments are captured in units of 60 seconds according to the location; all segments are marked with location number, time node and scene type.

[0055] In a specific implementation, the steps of designing a neural experiment test signal include: from the video materials organized in step S10, selecting the most representative video clips in units of 60 seconds according to the scene to produce a test signal, and annotating them; among them, the video of a safety accident that has occurred should select video clips at time nodes 15 minutes, 10 minutes, and 5 minutes before the accident.

[0056] Step S30: using the EEG-fNIRS device to collect the neural activity data of the security personnel under test when viewing the test signal, and simultaneously obtain their subjective scoring data, including test task one and test task two.

[0057] In a specific implementation, the test task 1 includes: playing a video clip before a safety accident, and the person being tested scores the possibility of the accident according to a preset time interval, and synchronously collecting the neural activity data set N1 and the scoring data set Q. The preset time interval is set according to actual usage requirements, and is set to a time interval of 5 minutes in the task of this embodiment.

[0058] Test task 2 includes: playing a video clip of a normal scene, and having the test subjects list the types of potential hidden dangers and risk scores, and simultaneously collecting the neural activity dataset N2 and the binary score dataset J.

[0059] In a specific implementation, the neural activity observation data collection includes: using experienced campus security personnel as test subjects, playing the test signal produced in step S20 in a dedicated virtual eye mask device according to the location, and collecting neural activity data and personnel score data in the following two test tasks through an EEG-fNIRS portable device;

[0060] Test Task 1: Security Incident Perception Capability Data Collection: This task aims to collect neural activity observation data and security incident perception score data from security personnel in incident scenarios, evaluate their security incident perception capabilities, and select data from those with higher perception capabilities for machine learning.

[0061] Corresponding data collection method: From the neural experiment test signal designed in step S20, select video clips at the time nodes 15 minutes, 10 minutes, and 5 minutes before the safety accident that has occurred according to the location and play them sequentially for testing; after each video clip is played, the test subject scores whether a safety accident is likely to occur on a continuous interval of [0, 5], obtaining scores Q1, Q2, and Q3 for the three videos, and giving the possible safety accident type; collect the neural activity dataset N1, the test subject score dataset Q, and the possible safety accident type set ST in the above test task;

[0062] Test Task 2: Safety Hazard Risk Assessment Data Collection, which aims to collect neural activity observation data of security personnel in normal scenarios and site safety hazard cognitive score data, and conduct a comprehensive assessment and calculation of safety hazard risks in various locations on campus;

[0063] Corresponding data collection method: From the neural experiment test signals designed in step S20, select normal video clips without safety accidents according to the venue for playback test; after each video clip is played, the test subject provides a binary score J(t, r) on the type of safety hazards that may exist in the venue and the risk level of each type of hazard; where t is the type of possible safety hazard, and r is the risk level score of this type of safety hazard, given in the continuous interval [0, 5]. Collect the neural activity dataset N2 in the above test task and the binary score dataset J provided by the test subject;

[0064] Step S40: Train the LSTM model based on the neural activity data of test task one, and combine the neural activity data and subjective scoring data of test task two to generate the overall safety risk value of each location and the risk level of specific hidden danger types.

[0065] In a specific implementation, the steps of training an LSTM model based on the neural activity data from test task one include: preprocessing the data from test task one, calculating the testee's safety perception ability score P = 0.1Q1 + 0.3Q2 + 0.6Q3, screening the testee data with a score P above the mean for model training, where Q represents the score dataset, and Q1, Q2, and Q3 represent specific scores in the score dataset; extracting the five frequency band energy parameters of the EEG signal and the mean, variance, and respiratory rate parameters of HbO and HbR from fNIRS from the neural activity data to form a 10-dimensional feature set as LSTM input; training the LSTM model based on the feature set, with the initial parameters specifically set to input dimension: 10, hidden layers: 20, number of layers: 2, and output dimension: 1, and tuning using the MAPE value. The initial parameter settings include: input dimension, hidden layers, number of layers, and output dimension.

[0066] It should be noted that the five frequency bands of the EEG signal include: 4-7 Hz, 8-10 Hz, 11-13 Hz, 14-15 Hz and 16-20 Hz.

[0067] In a specific implementation, the steps of generating the overall safety risk value of each location and the risk level of a specific hidden danger type include: summarizing the hidden danger types by location from the binary scoring dataset J, and taking the average of the risk scores of each type of hidden danger; combining the overall risk value output by the LSTM model to generate a safety assessment report containing the hidden danger type, risk level and priority.

[0068] It should be noted that the steps of calculating the safety hazard risk assessment include: first, from the safety accident type set ST obtained in the test task one of step S30, the score dataset Q of the tested persons who correctly judged the accident type is selected as the target output, and the neural activity dataset N1 of the above-mentioned persons is used as input, and the LSTM long-short-term neural network model is used to perform machine learning training on a location-by-location basis; then, the neural activity dataset N2 collected in the test task two is used as input, and the trained LSTM model is used to calculate on a location-by-location basis, and its output is used as the overall safety hazard risk assessment value of the location; from the binary score dataset J given by the tested persons, the possible safety hazard types t are listed by location, and the risk level score r of each type of safety hazard is averaged to obtain the specific safety hazard types that may exist in the location and the risk level assessment value of each type of safety hazard;

[0069] The following data preprocessing techniques are used for the test subject scoring dataset Q and neural activity dataset N1 used for LSTM model learning and training: First, the scores Q1, Q2, and Q3 obtained in test task 1 are used to calculate the test subject's safety accident perception and judgment ability score P according to the following formula:

[0070] P=0.1 Q1+0.3 Q2+0.6 Q3

[0071] Then, for each location, only the score dataset Q and neural activity dataset N1 of the subjects whose scores P were greater than the average were selected for LSTM model learning and training. Before the score dataset Q was used for training, linear interpolation was performed according to the values of Q1, Q2, and Q3.

[0072] In addition, the following feature set extraction preprocessing technique was used for the neural activity datasets N1 and N2 used for LSTM model learning and training and calculation after training: first, five parameters of the EEG signal amplitude energy in the frequency ranges of 4-7Hz, 8-10Hz, 11-13Hz, 14-15Hz, and 16-20Hz, as well as the mean and variance of HbO and HbR of the fNIR signal and the five parameters of respiratory rate extracted from them were calculated in units of 3 seconds; then, the above 10 parameters were combined into a feature parameter set as the input of the LSTM model.

[0073] In specific implementation, the core working principle of this embodiment is to combine EEG and fNIRS technology to capture the subconscious neural activity of experienced security personnel when observing campus scenes, and combine these neural activity data with the personnel's subjective scores to achieve a comprehensive assessment of campus safety hazards.

[0074] 1. The EEG-fNIRS device simultaneously collects EEG and functional near-infrared signals, combining the advantages of EEG's high temporal resolution and fNIRS's high spatial resolution to more comprehensively capture subconscious safety perception information.

[0075] 2. By playing videos of past safety incidents, we collect neural activity data and scoring data from security personnel to train the LSTM model. This process allows the model to learn the neural activity characteristics of security personnel when observing potential risk scenarios.

[0076] 3. The LSTM model can process time-series data and is suitable for analyzing continuous neural activity signals. Through training, the model can identify features related to security risk perception from neural activity data.

[0077] 4. Applying the trained model to the neural activity data of normal scenarios can assess the overall security risk level of the scenario, which reflects the subconscious security perception of security personnel.

[0078] 5. At the same time, by collecting the subjective scores of security personnel on the specific types of safety hazards and risk levels, we can supplement the specific hazard information that is difficult to clearly understand through subconscious perception.

[0079] In this way, this embodiment not only utilizes the rich subconscious safety perception ability of human beings, but also combines it with conscious professional judgment, thereby achieving a comprehensive and accurate assessment of campus safety hazards.

[0080] In the specific implementation, the use of the method of this embodiment is explained by taking a scenario as an example.

[0081] 1. Video footage collection: Video footage of the playground was collected from the campus surveillance system, including: a video of a stampede (15 minutes before the accident) and multiple videos of normal playground use at different times and in different weather conditions;

[0082] 2. Test signal design: 60-second segments were captured from the accident video at 15 minutes, 10 minutes, and 5 minutes before the accident, and multiple 60-second segments were captured from normal video.

[0083] 3. Neural Activity Observation Data Collection: Ten experienced campus security personnel were selected and asked to wear EEG-fNIRS devices to watch video clips. Task 1: Play the accident video clips and collect neural activity data N1 and safety incident perception scores Q1, Q2, and Q3. Task 2: Play the normal video clips and collect neural activity data N2 and safety hazard type and risk score J.

[0084] 4. Safety Hazard Risk Assessment Calculation: Use N1 and Q to train an LSTM model. Input N2 into the trained model to obtain the playground's overall safety risk score. Calculate specific safety hazard types (such as crowding, trampling, falls, etc.) and their risk levels based on J.

[0085] Through the above process, we can obtain the overall safety risk score of the playground and the specific types of potential safety hazards (such as crowding and trampling risk 3.2 points, fall risk 2.8 points, etc.). This information can help schools improve playground safety management measures more targetedly.

[0086] In specific implementations, in the application scenario of campus safety hazard risk assessment, this embodiment can also be applied to assessing the safety status of school libraries. Libraries are important places for students to study and research, but they also pose potential safety hazards, such as fire and bookshelf collapse. However, in actual applications, it was found that the special environment of the library interfered with the signal acquisition of the EEG-fNIRS equipment. The metal bookshelves, electronic equipment, and a large number of books in the library will generate electromagnetic interference, affecting the quality of the EEG signal. At the same time, light changes in the library (such as local bright and dark areas) will interfere with the optical measurement of fNIRS, resulting in inaccurate data. The signal interference problem in this specific environment reduces the reliability of neural activity data, thereby affecting the accuracy of safety hazard risk assessment.

[0087] To solve the problem of library environment interfering with EEG-fNIRS device signal acquisition, this embodiment proposes the following optimization solution:

[0088] 1. Signal shielding and filtering enhancement:

[0089] Add an additional electromagnetic shielding layer to the EEG electrodes to reduce environmental electromagnetic interference. Add an adaptive filtering algorithm to the signal processing stage to identify and filter out environmental noise in specific frequency bands in real time.

[0090] 2. fNIRS optical compensation technology:

[0091] An ambient light sensor was added to the fNIRS probe to monitor ambient light intensity in real time. The fNIRS light source intensity was dynamically adjusted based on ambient light intensity to ensure stable signal quality. An optical path compensation algorithm was developed to automatically correct fNIRS data based on ambient light changes.

[0092] 3. Multimodal data fusion:

[0093] An integrated micro-accelerometer monitors the subject's head movements to remove motion artifacts. Combined with video image analysis, it identifies environmental changes (such as movement and lighting) to assist with data correction.

[0094] 4. Environmental adaptability calibration:

[0095] Before the formal test, we conducted a pre-acquisition of environmental adaptability to obtain the background noise characteristics of the library's specific environment. Based on the pre-acquisition data, we automatically adjusted the equipment parameters and signal processing algorithms to optimize data quality.

[0096] 5. Distributed measurement strategy:

[0097] Replace single, long-term measurements with multiple, shorter ones to reduce the impact of environmental changes on the overall data. Develop a data splicing algorithm to intelligently combine multiple short-term measurements to reconstruct complete neural activity signatures.

[0098] These optimization measures significantly improve the signal quality and stability of EEG-fNIRS devices in challenging environments like libraries. These improvements enable them to better adapt to the electromagnetic and optical environments of libraries, effectively reducing interference and providing more reliable neural activity data. This not only enhances the accuracy of safety hazard risk assessments but also expands the method's application in complex environments, providing more comprehensive and precise support for campus safety management.

[0099] This embodiment collects video footage from different locations from a historical library of campus security surveillance videos, organizes them into video material sets by location, and labels them. Video clips at key time points are captured from these video materials according to preset rules to generate a test signal set. An EEG-fNIRS device is used to collect neural activity data from the security personnel being tested while viewing the test signals, and their subjective rating data, including both test tasks one and two, is simultaneously obtained. An LSTM model is trained based on the neural activity data from test task one, and combined with the neural activity data and subjective rating data from test task two, to generate an overall safety risk value for each location and the risk level of specific hidden danger types. This achieves the technical effect of effectively discovering various complex and hidden campus safety hazards.

[0100] In addition, an embodiment of the present application also proposes a computer-readable storage medium, which stores a program for evaluating and calculating human neural perception of campus safety hazard risks. When the program for evaluating and calculating human neural perception of campus safety hazard risks is executed by a processor, the steps of the method for evaluating and calculating human neural perception of campus safety hazard risks as described above are implemented.

[0101] Reference Figure 3 , Figure 3 This is a structural block diagram of the first embodiment of the human neural perception evaluation and calculation device for campus safety hazard risks in this application.

[0102] like Figure 3 As shown, the human neural perception evaluation and calculation device for campus safety hazard risk proposed in the embodiment of the present application includes:

[0103] The data acquisition module 10 is used to collect video materials of different locations from the campus security surveillance video history library, including videos of safety accidents and normal scenes, organize them into video material sets by location and annotate them;

[0104] The test signal set generating module 20 is configured to extract video clips of key time nodes from the video material according to preset rules to generate a test signal set;

[0105] A test task module 30 is used to collect neural activity data of the security personnel under test when viewing the test signal using an EEG-fNIRS device, and simultaneously obtain their subjective scoring data, including test task 1 and test task 2;

[0106] The output module 40 is used to train the LSTM model based on the neural activity data of test task one, and to generate the overall safety risk value of each location and the risk level of specific hidden danger types in combination with the neural activity data and subjective scoring data of test task two.

[0107] It should be understood that the above is only an example and does not constitute any limitation to the technical solution of the present application. In specific applications, technicians in this field can make settings as needed, and the present application does not impose any restrictions on this.

[0108] This embodiment collects video footage from different locations from a historical library of campus security surveillance videos, organizes them into video material sets by location, and labels them. Video clips at key time points are captured from these video materials according to preset rules to generate a test signal set. An EEG-fNIRS device is used to collect neural activity data from the security personnel being tested while viewing the test signals, and their subjective rating data, including both test tasks one and two, is simultaneously obtained. An LSTM model is trained based on the neural activity data from test task one, and combined with the neural activity data and subjective rating data from test task two, to generate an overall safety risk value for each location and the risk level of specific hidden danger types. This achieves the technical effect of effectively discovering various complex and hidden campus safety hazards.

[0109] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of this application. In actual applications, technicians in this field can select part or all of it according to actual needs to achieve the purpose of this embodiment scheme, and no restrictions are imposed here.

[0110] In addition, for technical details not fully described in this embodiment, please refer to the method for human neural perception evaluation and calculation of campus safety hazard risks provided in any embodiment of this application, which will not be repeated here.

[0111] In addition, it should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or system comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or system. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or system comprising the element.

[0112] The serial numbers of the above embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0113] Through the above description of the embodiments, those skilled in the art will clearly understand that the above-mentioned embodiments and methods can be implemented using software plus the necessary general-purpose hardware platform. Of course, hardware can also be used, but in many cases the former is a more preferred embodiment. Based on this understanding, the technical solution of this application, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory (ROM) / RAM, a magnetic disk, or an optical disk) and includes several instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods of each embodiment of this application. The above are only preferred embodiments of this application and do not limit the scope of the patent application. Any equivalent structure or equivalent process transformation made using the contents of this application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the scope of patent protection of this application.

Claims

1. A method for evaluating and calculating human neural perception of campus safety hazards, characterized by: include: Collect video footage from different locations from the campus security surveillance video history library, including videos of safety accidents and normal scenes, organize them into video material sets by location and annotate them; Extracting video clips at key time points from the video material according to preset rules to generate a test signal set; Using an EEG-fNIRS device to collect neural activity data of the security personnel under test when viewing the test signal, and simultaneously obtain their subjective scoring data, including test task one and test task two; The LSTM model is trained based on the neural activity data from test task one, and combined with the neural activity data and subjective scoring data from test task two to generate the overall safety risk value and specific risk level of hidden danger types for each location; The test task 1 includes: playing a video clip before a safety accident, and having the test subjects score the possibility of the accident according to a preset time interval, and simultaneously collecting a neural activity dataset N1 and a scoring dataset Q; Test Task 2 includes: playing a normal scene video clip, asking the test subjects to list the potential hidden danger types and risk scores, and simultaneously collecting the neural activity dataset N2 and the binary score dataset J; The step of training the LSTM model based on the neural activity data of test task 1 includes: Preprocess the data of test task 1 and calculate the safety perception ability score of the test subjects, P = 0.1Q1 + 0.3Q2 + 0.6Q3. Filter the test subjects whose score P is higher than the mean for model training, where Q represents the score dataset, and Q1, Q2, and Q3 represent the specific scores in the score dataset. The five frequency band energy parameters of the EEG signal and the mean, variance, and respiratory rate parameters of fNIRS HbO and HbR were extracted from the neural activity data to form a 10-dimensional feature set as LSTM input; Based on the feature set, the LSTM model is trained. The initial parameters include input dimension, hidden layer, number of layers, and output dimension, and the MAPE value is used for optimization. The step of generating the overall safety risk value and the risk level of specific hidden danger types for each location includes: From the binary score dataset J, the types of hidden dangers are summarized by location, and the risk scores of each type of hidden danger are averaged; Combined with the overall risk value output by the LSTM model, a safety assessment report is generated that includes the hidden danger type, risk level and priority.

2. The method according to claim 1, characterized in that The videos of the safety incidents that have occurred include: 15 minutes of continuous video footage before the accident; Covering no less than 5 types of climate environments and 3 types of activity scenarios; If the actual video is insufficient, supplementary video material that meets the requirements can be generated through drills with real people.

3. The method according to claim 1, characterized in that The test signal set includes: For videos of safety accidents that have occurred, capture key time node segments 15 minutes, 10 minutes, and 5 minutes before the accident; For normal scene videos, representative clips are captured in 60-second units according to the scene; All clips are marked with location number, time node and scene type.

4. The method according to claim 1, wherein The five frequency bands of the EEG signal include: 4-7 Hz, 8-10 Hz, 11-13 Hz, 14-15 Hz and 16-20 Hz.

5. A human neural perception evaluation and calculation device for campus safety hazard risk, characterized by: Executing the method according to claim 1, comprising: The data acquisition module is used to collect video materials from different locations from the campus security surveillance video history library, including videos of safety accidents and normal scenes, and organize them into video material sets by location and annotate them; A test signal set generation module is used to intercept video clips of key time nodes from the video material according to preset rules to generate a test signal set; A test task module, for collecting neural activity data of the security personnel under test when viewing the test signal using an EEG-fNIRS device, and simultaneously obtaining their subjective scoring data, including test task one and test task two; The output module is used to train the LSTM model based on the neural activity data of test task one, and combine the neural activity data and subjective scoring data of test task two to generate the overall safety risk value of each location and the risk level of specific hidden danger types.

6. A computer device, characterized in that: The device comprises: a memory and a processor, wherein the processor executes the method according to any one of claims 1 to 4 when running computer instructions stored in the memory.

7. A computer-readable storage medium, characterized in that The method comprises instructions which, when executed on a computer, cause the computer to execute the method according to any one of claims 1 to 4.

Images