Activity site safety monitoring system and method based on Taylor polynomial gating unit

Abstract

The invention discloses an activity site safety monitoring system and method based on a Taylor polynomial gating unit, and relates to the technical field of safety monitoring, and the system comprises a data center which is used for collecting and storing big data and data information of each subsystem; the supervision center is used for supervising the data information of each subsystem in real time; the user side is used for remotely checking and processing the collected and stored data information; the personnel management subsystem is used for collecting and managing identity information and position information of field workers; and the video monitoring subsystem is used for collecting and identifying image data of the activity site. According to the method, a ConvNeXt architecture is utilized, an original GELU activation function is replaced by a Taylor polynomial gating unit, the abnormal behavior recognition precision can be effectively enhanced, the feature interaction capture capability can be effectively improved, compared with a traditional mode, multi-dimensional interaction features in a complex scene can be captured, a warning threshold value can be adjusted according to scene complexity, the early warning accuracy is effectively improved, and the early warning efficiency is improved. And the false alarm or missing alarm condition is greatly reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of security monitoring, in particular to an activity site security monitoring system and method based on a Taylor polynomial gate control unit. Background Art

[0002] Current event site security monitoring mainly relies on traditional video analysis technology, which has the following shortcomings: 1. The contradiction between real-time and accuracy. When processing high-resolution video streams, the existing monitoring system has a high computational complexity, making it difficult to simultaneously ensure real-time and abnormal behavior recognition accuracy; 2. The ability to capture feature interactions is weak. Traditional convolutional neural networks rely on local perception and static information aggregation, and are unable to capture multi-dimensional interactive features in complex scenes; 3. The early warning accuracy is poor, and alarms are often triggered based on preset rules. There is a lack of dynamic adaptability, which is prone to false alarms or missed alarms.

[0003] Based on this, a system and method for monitoring the safety of an event site based on a Taylor polynomial gating unit are now provided, which can eliminate the drawbacks of existing technical solutions. Summary of the Invention

[0004] The purpose of the present invention is to provide an activity site security monitoring system and method based on Taylor polynomial gating unit to solve the problems in the background technology of being unable to guarantee real-time performance and abnormal behavior recognition accuracy, weak feature interaction capture capability and poor warning accuracy.

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

[0006] The event site security monitoring system based on Taylor polynomial gate control unit includes:

[0007] Data center, used to collect and store big data and data information of various subsystems;

[0008] A monitoring center is used to monitor the data information of each subsystem in real time and realize information sharing with the data center to facilitate the timely issuance of warnings and notifications. The monitoring center includes a controller, a processor, a display panel and a data processing module. The controller is used to receive and output information results from each subsystem and the data center. The processor is used to receive coordination instructions from the controller. The data processing module is used to utilize the ConvNeXt architecture improved by Taylor polynomial gating units and combine it with the spatiotemporal feature pyramid network to enhance the ability to identify abnormal behavior;

[0009] The user end is used to remotely view and process the data information collected and stored in the data center and the supervision center;

[0010] The personnel management subsystem is used to collect and manage the identity and location information of on-site personnel, and to promptly remind personnel to stop and handle abnormal behavior based on the above information characteristics;

[0011] The video surveillance subsystem is used to collect and identify image data at the event site, and to issue warnings for abnormal behavior based on the recognition features of the above image data.

[0012] Preferably, the data center includes an abnormal behavior library, and the abnormal behavior library is used to cooperate with the abnormal behavior identification module to determine whether the behavior is abnormal.

[0013] Preferably, the data processing module includes:

[0014] A preprocessing module is used to perform preprocessing operations on the collected data;

[0015] Feature extraction module, used to extract multi-dimensional spatiotemporal features through a ConvNeXt network enhanced with Taylor polynomial gating units;

[0016] Abnormal behavior recognition module, used to detect abnormal behavior based on spatiotemporal feature pyramid network;

[0017] A fusion module is used to fuse the above information features with the corresponding data information in the data center and the supervision center to obtain the location of the abnormal behavior, risk parameters and identity information of surrounding personnel;

[0018] The early warning and decision-making module is used to generate graded early warnings based on the analysis results, trigger the corresponding emergency response mechanism, and call preset plans based on the type of abnormal behavior in the data center.

[0019] Preferably, the personnel management subsystem includes a face recognition module, a positioning module and an information comparison module. The face recognition module is used to identify and collect the personal identity information of on-site staff. The positioning module is used to collect the basic location information of each staff member at the event site and perform real-time automatic updates of the location data. The information comparison module is used to transmit the identified personal identity information to the supervision center and match it with the location data of abnormal behavior.

[0020] Preferably, the video surveillance subsystem includes several cameras, environmental perception devices and edge computing devices. The cameras are used to capture images and video materials of the event site in real time. The environmental perception devices include several sensor units for real-time perception of various physical quantities. The edge computing devices are used to perform background subtraction and thermal map generation operations on the collected data, and transmit the processed data to the supervision center.

[0021] Preferably, the data processing module also includes a diagnostic component for determining whether a fault occurs in the equipment and lines corresponding to each camera location. The diagnostic component includes an electrically connected fault judgment module and an indicator light. The fault judgment module is used to identify the fault conditions of different equipment and lines, and to indicate whether the equipment and lines meet the fault conditions through the indicator light.

[0022] The method for monitoring the safety of an activity site based on a Taylor polynomial gate control unit specifically includes:

[0023] S1. The video surveillance subsystem's cameras collect multi-viewpoint video and image data from the event site in real time. Environmental sensing devices simultaneously acquire physical quantity data on temperature, sound, and light. Edge computing devices perform dynamic and static background subtraction and image enhancement on the video and image data, and generate a personnel heat map.

[0024] S2, the preprocessing module standardizes, denoises and normalizes the collected data. The data is input into the feature extraction module, and the ConvNeXt network improved by Taylor polynomial gating unit is used to extract features. The original GELU activation function is replaced by Taylor polynomial gating unit, and its expression is: Output multi-dimensional spatiotemporal features;

[0025] S3, the abnormal behavior recognition module combines spatiotemporal feature pyramid network analysis to detect individual abnormal behaviors and identify group and environmental abnormal patterns, and calculate risk parameters to determine whether the risk level exceeds the preset threshold. The fusion module integrates heat maps, environmental data, and personnel location information;

[0026] S4, the early warning and decision-making module generates graded early warning signals based on risk levels, calls the preset plan in the data center, and sends early warning information and processing instructions to the user end and on-site staff through the supervision center;

[0027] S5. The fault judgment module monitors the equipment status in real time, prompts faults through indicator lights, and feeds back the operating status to the data center, records event data, and optimizes the abnormal behavior library and warning thresholds.

[0028] Preferably, the step S1 specifically includes:

[0029] S11. Multiple high-definition cameras cover various areas of the event site, with each camera capturing video and image data of the corresponding area;

[0030] S12. The environmental sensing device collects noise decibel and temperature anomaly data in real time and aligns them with the video and image data timestamps.

[0031] S13. The edge computing device marks the location information and performs background subtraction, noise filtering, and image enhancement on the original video to generate a heat map of the population density distribution in real time. The processed video, heat map, and environmental data are uploaded to the data processing module.

[0032] Preferably, the step S2 specifically includes:

[0033] S21, extracting spatial and temporal dynamic features from video sequences through a ConvNeXt network enhanced by Taylor polynomial gating units;

[0034] S22. Match the extracted features with the behavior templates in the abnormal behavior library, and associate them with the real-time location information provided by the personnel management subsystem;

[0035] S23. The fusion module integrates video features, environmental data, and personnel location information to match on-site staff around the location of abnormal behavior.

[0036] Preferably, the step S3 specifically includes:

[0037] S31, the abnormal behavior recognition module receives the comprehensive features generated by the fusion module and inputs them into the spatiotemporal feature pyramid network for multi-scale analysis to identify abnormal behavior patterns;

[0038] S32. Compare the abnormal behavior database of the data center, match known risk types, and calculate the behavior deviation score;

[0039] S33. Combine the real-time data from environmental sensing devices to perform a weighted assessment of the risk level.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] 1. This event site security monitoring system based on the Taylor polynomial gating unit manages personnel and camera equipment at the event site through the interaction between the data center, the supervision center and various subsystems. It uses a number of cameras to monitor the event site in real time to see if there is any abnormal behavior, and makes subsequent decisions based on the risk level. It also performs early warning operations, allowing nearby staff to quickly rush to the scene of the behavior to stop or deal with the abnormal situation. When the risk level is high, it can alert personnel to evacuate quickly to reduce losses, and quickly understand the cause and process of abnormal behavior, facilitating subsequent analysis and improving management efficiency and quality.

[0042] 2. The present invention uses the ConvNeXt architecture to replace the original GELU activation function with a Taylor polynomial gating unit, and transforms from static features to dynamic feature interaction modeling, so that the accuracy of abnormal behavior recognition can be effectively enhanced and the ability to capture feature interactions can be effectively improved. Compared with traditional methods, it can capture multi-dimensional interactive features in complex scenarios, adjust the warning threshold according to the complexity of the scenario, effectively improve the accuracy of early warning, and greatly reduce the occurrence of false alarms or missed alarms. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a structural schematic diagram of the present invention.

[0044] Figure 2 It is a structural diagram of the supervision center of the present invention.

[0045] Figure 3 Schematic diagram of the structure of the data processing module of the present invention.

[0046] Figure 4 It is a schematic diagram of the process of the present invention.

[0047] Figure 5 Schematic diagram of the process of step S1 of the present invention.

[0048] Figure 6 Schematic diagram of the process of step S2 of the present invention.

[0049] Figure 7 Schematic diagram of the process of step S3 of the present invention.

[0050] Notes on the accompanying figures: data center 100, supervision center 200, controller 210, processor 220, display panel 230, data processing module 240, preprocessing module 241, feature extraction module 242, abnormal behavior recognition module 243, fusion module 244, early warning and decision module 245, fault judgment module 246, indicator light 247, user terminal 300, personnel management subsystem 400, face recognition module 410, positioning module 420, information comparison module 430, video monitoring subsystem 500, camera equipment 510, environmental perception equipment 520, edge computing equipment 530. DETAILED DESCRIPTION

[0051] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0052] In this embodiment, if Figure 1-Figure 7 As shown, the active site safety monitoring system based on the Taylor polynomial gate control unit includes:

[0053] Data center 100, used to collect and store big data and data information of various subsystems;

[0054] The monitoring center 200 is used to monitor the data information of each subsystem in real time and realize information sharing with the data center 100 to facilitate the timely issuance of warnings and notifications. The monitoring center 200 includes a controller 210, a processor 220, a display panel 230 and a data processing module 240. The controller 210 is used to receive and output information results from each subsystem and the data center 100, control personnel and activity site environment information, and output coordination instructions. The processor 220 is used to receive coordination instructions from the controller 210 and publish information according to the instructions. The display panel 230 is used to display the collected data and model data to the user in a visual form. The data processing module 240 is used to utilize the ConvNeXt architecture improved by Taylor polynomial gating units and combine it with the spatiotemporal feature pyramid network to enhance the recognition ability of abnormal behavior;

[0055] The user terminal 300 is used to remotely view and process the data information collected and stored in the data center 100 and the monitoring center 200;

[0056] The personnel management subsystem 400 is used to collect and manage the identity information and location information of on-site personnel, and to promptly remind personnel to stop and handle abnormal behavior based on the above information characteristics;

[0057] The video surveillance subsystem 500 is used to collect and identify image data at the event site and issue warnings for abnormal behavior based on the identification features of the image data;

[0058] Specifically, the monitoring center 200 has good computing power and storage resources to facilitate large-scale data processing and analysis, ensure the accuracy and reliability of the model, facilitate unified management and maintenance of modules, and use 5G / Wi-Fi low-latency communication to ensure real-time data transmission;

[0059] Among them Figure 1 and Figure 2 As shown, the data center 100 includes an abnormal behavior library, which is used to cooperate with the abnormal behavior identification module 243 to determine whether the behavior is abnormal. The abnormal behavior library cooperates with the abnormal behavior identification module 243 to determine whether the behavior is abnormal. The abnormal behaviors in the abnormal behavior library are consistent with the abnormal behavior types defined according to the public security industry standards, including disrupting public order, committing illegal and disciplinary acts such as violence, vulgarity, and pornography, carrying dangerous goods and other prohibited items, and filming indecent actions.

[0060] Among them Figure 2 and Figure 3 As shown, the data processing module 240 includes:

[0061] The preprocessing module 241 is used to perform preprocessing operations on the collected data;

[0062] a feature extraction module 242 for extracting multi-dimensional spatiotemporal features through a ConvNeXt network enhanced with Taylor polynomial gating units;

[0063] Abnormal behavior identification module 243, used to detect abnormal behavior based on spatiotemporal feature pyramid network;

[0064] Fusion module 244 is used to fuse the above information features with the corresponding data information in the data center 100 and the supervision center 200 to obtain the location of the abnormal behavior, risk parameters and identity information of surrounding personnel;

[0065] The warning and decision module 245 is used to generate graded warnings based on the analysis results and trigger the corresponding emergency response mechanism to call preset solutions based on the type of abnormal behavior in the data center 100;

[0066] Specifically, the data preprocessing stage (preprocessing module 241) inputs the original video stream captured by the multi-view camera, and outputs a personnel heat map and a dynamic foreground mask. The feature extraction and interaction enhancement stage (feature extraction module 242) inputs the preprocessed video frame and heat map, extracts features through the ConvNeXt network enhanced by the Taylor polynomial gating unit, and each layer of Taylor polynomial gating unit performs multi-order interaction modeling (such as linear, quadratic, and cubic terms) on the input features, and outputs an enhanced spatiotemporal feature map. The abnormal behavior detection stage (abnormal behavior recognition module 243) inputs the spatiotemporal feature map and heat map, and the spatiotemporal feature pyramid network performs cross-layer fusion of multi-level features to generate a multi-scale feature pyramid. The long short-term memory network performs temporal encoding on the feature sequence of 5 to 10 consecutive frames, outputs a feature vector containing the temporal context, and predicts the category and location of the abnormal behavior. The dynamic warning stage (warning and decision module 245) inputs the abnormal behavior detection results, including confidence, category, and location. According to the dynamic threshold rule, if the parameters exceed the threshold range, a warning signal is generated and pushed to the staff.

[0067] The ConvNeXt network is a pure convolutional neural network architecture that uses a hierarchical feature extraction model. It generates multi-scale feature maps through gradual downsampling and is suitable for tasks such as target detection. It introduces large 7×7 convolution kernels to replace the 3×3 convolution of traditional ResNet, expanding the local receptive field and enhancing the ability to model spatial context. It makes extensive use of depthwise separable convolutions. ConvNeXt simulates long-distance dependencies through large convolution kernels rather than self-attention mechanisms. ConvNeXt supports ImageNet pre-training weight transfer, facilitating rapid fine-tuning in security monitoring scenarios and reducing computational complexity. As a traditional convolutional network, ConvNeXt has the limitation of static feature aggregation. The Taylor Polynomial Gated Unit (TPGU) has the characteristic of dynamic interaction enhancement. Replacing the GELU function with the TPGU can introduce learnable multi-order feature interactions in each convolution block of ConvNeXt, improving the model's ability to model complex patterns such as crowd flow and aggregation.

[0068] ConvNeXt model construction: The preprocessed surveillance video frames are input into the ConvNeXt backbone network. ConvNeXt gradually extracts features through multiple convolutional layers such as the Stem layer and Stage 1-4. Each layer contains operations such as large convolution kernels and depthwise separable convolution. The GELU activation function in ConvNeXt is replaced with TPGU, which outputs multi-scale spatiotemporal feature maps with rich semantics. Abnormal behavior detection and early warning functions are performed through subsequent modules. The initial module structure of ConvNeXt is: Input→Depthwise Conv→LayerNorm→GELU→1A-1Conv→LayerNorm→GELU→1A-1Conv→Output. The module structure after replacement of ConvNeXt is: Input→Depthwise Conv→LayerNorm→TPGU→1A-1Conv→LayerNorm→TPGU→1A-1Conv→Output;

[0069] The TPGU formula is Where N is the polynomial order, which controls the complexity of interaction to enhance the flexibility of feature interaction. i ,b is a learnable parameter, σ(1.702b) is a regulating factor, the first-order term (i=1) can retain linear information, and the high-order term (i≥2) can capture nonlinear associations and complex associations between features, such as the interactive effect of crowd density and movement speed on stampede risk. i Adaptively adjust the weights of interactions of different orders to enhance model flexibility. For example, when detecting pushing behavior, the quadratic term weight is automatically increased to improve sensitivity to acceleration changes.

[0070] ConvNeXt, as a backbone network, does not directly complete the abnormal behavior detection task. It needs to cooperate with the abnormal behavior recognition module 243. The abnormal behavior recognition module 243 has a built-in spatiotemporal feature pyramid network, which integrates the multi-scale features output by ConvNeXt to solve the problem of target scale variation. It performs time series analysis on the feature map sequence of consecutive frames, captures the temporal continuity of the action, and predicts the category, location, and confidence of the abnormal behavior.

[0071] Dynamic threshold rules: Influencing factors include real-time data characteristics, historical statistical laws and scene complexity. Real-time data characteristics include heat map density indicators (people density mean μ density and density variance ), abnormal behavior feature confidence c crowd and the time series change rate (density change rate and the mean velocity μ speed ), historical statistical laws include time period characteristics and prior probability of abnormal events, scene complexity includes camera view coverage and environmental interference factors (light intensity and background complexity), the threshold formula is Threshold(t)=BaseThreshold×α(t)×β(t)×γ(t), where BaseThreshold is the preset basic threshold, α(t) is the real-time data adjustment factor, ranging from 0.8 to 1.2, and the calculation method is λ1,λ2 are weight coefficients, μ normal is the historical normal density mean, β(t) is the time period adjustment factor, and γ(t) is the scene complexity adjustment factor, which is obtained by looking up the table based on the camera data and environmental sensor data. For example, for low-light scenes, γ = 1.05. In the initial stage, μ for each period is calculated based on the historical 7-day data. normal , σ normal ,generate a time period adjustment factor table, label the scene complexity level for each camera, initialize the γ(t) parameter, and refresh the threshold every 5 seconds based on the latest data in the real-time adjustment phase to ensure a rapid response to sudden scenes. Regularly use new data to retrain the model and reinforcement learning parameters to adapt to scene changes;

[0072] The spatiotemporal feature pyramid network includes a spatial pyramid and a temporal pyramid. It generates multi-resolution feature maps through convolutional downsampling at different levels, covering local details and global scenes. It performs hierarchical temporal modeling on the feature sequences of consecutive frames to capture short-term and long-term dynamics. It uses a top-down feature fusion mechanism to upsample high-level semantic features and fuse them layer by layer with low-level high-resolution features. It aligns the number of channels through 1×1 convolution to avoid information loss. It compares the template features in the abnormal behavior library, calculates the behavior deviation score of the current behavior, and adaptively adjusts the threshold based on environmental perception data.

[0073] Among them Figure 1 and Figure 2 As shown, the personnel management subsystem 400 includes a face recognition module 410, a positioning module 420, and an information comparison module 430. The face recognition module 410 is used to identify and collect personal identity information of on-site staff members. The positioning module 420 is used to collect basic location information of each staff member at the event site and automatically update the location data in real time. The information comparison module 430 is used to transmit the identified personal identity information to the supervision center 200 and match it with the location data of abnormal behavior. This facilitates the rapid dispatch of surrounding staff members to the abnormal location when abnormal behavior occurs, thereby reducing the probability of risk occurrence.

[0074] Among them Figure 1 and Figure 2 As shown, the video surveillance subsystem 500 includes several camera devices 510, environmental perception devices 520 and edge computing devices 530. The camera devices 510 are used to capture images and video data of the activity scene in real time. The environmental perception devices 520 include several sensor units for real-time perception of various physical quantities, including but not limited to gas sensors, ultrasonic sensors, laser ranging sensors, temperature and humidity sensors, smoke sensors, light sensors and flame sensors, so as to facilitate the detection of abnormal situations. The edge computing device 530 is used to perform background subtraction and thermal map generation operations on the collected data, and transmit the processed data to the monitoring center 200, so as to facilitate the intuitive understanding of the density and distribution of people based on the thermal map.

[0075] Among them Figure 3 As shown, the data processing module 240 also includes a diagnostic component for determining whether a fault occurs in the equipment and lines corresponding to each camera location. The diagnostic component includes an electrically connected fault judgment module 246 and an indicator light 247. The fault judgment module 246 is used to identify the fault conditions of different equipment and lines, and prompts the equipment and lines whether they meet the fault conditions through the indicator light 247. The red alarm indicates a fault, and the green indicates normal.

[0076] Among them Figure 4-Figure 7 As shown, the method for monitoring the safety of an activity site based on a Taylor polynomial gate control unit specifically includes:

[0077] S1. The camera device 510 of the video surveillance subsystem 500 collects multi-view video and image data of the event site in real time. The environmental perception device 520 simultaneously obtains physical quantity data of temperature, sound, and light. The edge computing device 530 performs dynamic and static background subtraction and image enhancement operations on the video and image data, and generates a personnel heat map.

[0078] S2, the pre-processing module 241 performs standardization, denoising and normalization on the collected data, and the data is input to the feature extraction module 242, which uses the ConvNeXt network improved by the Taylor polynomial gating unit to extract features, replacing the original GELU activation function with the Taylor polynomial gating unit, and its expression is: Output multi-dimensional spatiotemporal features;

[0079] S3, the abnormal behavior identification module 243 combines spatiotemporal feature pyramid network analysis to detect individual abnormal behaviors and identify group and environmental abnormal patterns, and calculates risk parameters to determine whether the risk level exceeds a preset threshold. The fusion module 244 integrates the heat map, environmental data, and personnel location information;

[0080] S4, the warning and decision module 245 generates a graded warning signal according to the risk level, calls the preset plan in the data center 100, and sends warning information and processing instructions to the user terminal 300 and on-site staff through the supervision center 200;

[0081] S5. The fault judgment module 246 monitors the device status in real time, indicates the fault through the indicator light 247, and feeds back the operating status to the data center 100, records event data, and optimizes the abnormal behavior library and warning thresholds;

[0082] Specifically, the early warning and decision module 245 performs a graded response based on the risk level:

[0083] When the risk is low, the abnormal area is marked on the display panel 230 to alert the supervisor. When the risk is medium, the user terminal 300 automatically matches the surrounding staff and issues processing instructions to their terminals. When the risk is high, the power supply of the equipment in the relevant area is cut off, and the evacuation route is broadcast through sound and light alarms and radio. The alarm is also simultaneously notified to the security department for rapid response operations. The abnormal behavior location heat map is updated in real time, the real-time location of the staff is superimposed, the scheduling path is optimized, and the response time and processing results are recorded to the data center 100 for subsequent event tracing and model optimization.

[0084] Among them Figure 4 and Figure 5 As shown, step S1 specifically includes:

[0085] S11. Multiple high-definition camera devices 510 cover various areas of the event site, and each camera device 510 collects video and image data of the corresponding area;

[0086] S12, the environmental sensing device 520 collects noise decibel and temperature anomaly data in real time and aligns them with the video and image data timestamps;

[0087] S13. The edge computing device 530 marks the location information and performs background subtraction, noise filtering, and image enhancement on the original video to generate a heat map of the personnel density distribution in real time. The processed video, heat map, and environmental data are uploaded to the data processing module 240.

[0088] Among them Figure 4 and Figure 6 As shown, the step S2 specifically includes:

[0089] S21, extracting spatial and temporal dynamic features from video sequences through a ConvNeXt network enhanced by Taylor polynomial gating units;

[0090] S22, matching the extracted features with the behavior templates in the abnormal behavior library, and correlating them with the real-time location information provided by the personnel management subsystem 400;

[0091] S23, the fusion module 244 integrates the video features, environmental data and personnel location information to match the on-site staff around the location of the abnormal behavior;

[0092] Among them Figure 4 and Figure 7 As shown, step S3 specifically includes:

[0093] S31, the abnormal behavior recognition module 243 receives the comprehensive features generated by the fusion module 244, inputs them into the spatiotemporal feature pyramid network for multi-scale analysis, and identifies abnormal behavior patterns;

[0094] S32. Compare the abnormal behavior library of the data center 100, match the known risk type, and calculate the behavior deviation score;

[0095] S33. Combining the real-time data from the environment sensing device 520, weightedly assess the risk level;

[0096] Specifically, the identification of abnormal behavior patterns includes violent actions, running in groups, and items left behind. The real-time data of the environmental sensing device 520 includes sudden noise increases, abnormal temperatures, and other situations. The risk levels include low, medium, and high. When the behavior deviation score is ≤30% threshold, it is judged to be low risk. The display panel 230 displays an abnormal behavior reminder on the surface, which only attracts the attention of supervisors and records the behavior log. When 30% < behavior deviation score ≤70%, it is judged to be medium risk. Instructions are pushed to surrounding staff through the user terminal 300, so that staff can quickly stop or deal with abnormal behavior. If the behavior deviation score is >70%, it is judged to be high risk, and the preset plan is called to link the sound and light alarm and access control system for rapid response operations, and the evacuation instruction is pushed to all equipment.

[0097] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. The activity site safety monitoring system based on Taylor polynomial gate control unit is characterized by: include: A data center (100) is used to collect and store big data and data information of each subsystem; A monitoring center (200) is used to monitor data information of each subsystem in real time and realize information sharing with the data center (100) to facilitate timely issuance of warnings and notification announcements. The monitoring center (200) includes a controller (210), a processor (220), a display panel (230) and a data processing module (240). The controller (210) is used to receive and output information results of each subsystem and the data center (100). The processor (220) is used to receive coordination instructions from the controller (210). The data processing module (240) is used to utilize the ConvNeXt architecture improved by the Taylor polynomial gating unit and combine it with the spatiotemporal feature pyramid network to enhance the recognition ability of abnormal behavior. A user terminal (300) is used to remotely view and process data information collected and stored in the data center (100) and the monitoring center (200); The personnel management subsystem (400) is used to collect and manage the identity information and location information of on-site personnel, and to promptly remind personnel to stop and handle abnormal behavior based on the above information characteristics; The video monitoring subsystem (500) is used to collect and identify image data of the activity scene, and to warn of abnormal behavior based on the identification features of the image data.

2. The activity site security monitoring system based on the Taylor polynomial gate control unit according to claim 1 is characterized in that: The data center (100) includes an abnormal behavior library, and the abnormal behavior library is used to cooperate with the abnormal behavior identification module (243) to determine whether the behavior is abnormal.

3. The activity site security monitoring system based on the Taylor polynomial gate control unit according to claim 2 is characterized in that: The data processing module (240) includes: A preprocessing module (241) is used to perform preprocessing operations on the collected data; A feature extraction module (242) for extracting multi-dimensional spatiotemporal features through a ConvNeXt network enhanced by Taylor polynomial gating units; An abnormal behavior recognition module (243), for detecting abnormal behavior based on a spatiotemporal feature pyramid network; A fusion module (244) is used to fuse the above-mentioned information features with corresponding data information in the data center (100) and the supervision center (200) to obtain the location of the abnormal behavior, risk parameters and identity information of surrounding staff; The early warning and decision module (245) is used to generate graded early warnings based on the analysis results, trigger corresponding emergency response mechanisms, and call preset plans based on the abnormal behavior types in the data center (100).

4. The activity site security monitoring system based on the Taylor polynomial gate control unit according to claim 3 is characterized in that: The personnel management subsystem (400) includes a face recognition module (410), a positioning module (420) and an information comparison module (430), wherein the face recognition module (410) is used to identify and collect personal identity information of on-site staff members, the positioning module (420) is used to collect basic location information of each staff member at the activity site and automatically update the location data in real time, and the information comparison module (430) is used to transmit the identified personal identity information to the supervision center (200) and match it with the location data of abnormal behavior.

5. The activity site safety monitoring system based on the Taylor polynomial gate control unit according to claim 4 is characterized in that: The video monitoring subsystem (500) includes a plurality of camera devices (510), an environmental sensing device (520) and an edge computing device (530), wherein the camera devices (510) are used to capture images and video data of an activity scene in real time, the environmental sensing device (520) includes a plurality of sensor units for sensing various physical quantities in real time, and the edge computing device (530) is used to perform background subtraction and heat map generation operations on the collected data, and transmit the processed data to the monitoring center (200).

6. The activity site safety monitoring system based on the Taylor polynomial gate control unit according to claim 5 is characterized in that: The data processing module (240) further includes a diagnostic component for determining whether a fault occurs in the equipment and lines corresponding to each camera location. The diagnostic component includes an electrically connected fault determination module (246) and an indicator light (247). The fault determination module (246) is used to identify the fault conditions of different equipment and lines, and to indicate whether the equipment and lines meet the fault conditions through the indicator light (247).

7. The method for monitoring the safety of an activity site based on a Taylor polynomial gate control unit is characterized in that: The method is used in the activity site safety monitoring system based on the Taylor polynomial gate control unit according to any one of claims 1 to 6, specifically comprising: S1. The video surveillance subsystem (500) uses a camera device (510) to collect multi-view videos and image data of the activity site in real time. The environmental sensing device (520) synchronously obtains physical quantity data of temperature, sound, and light. The edge computing device (530) performs dynamic and static background subtraction and image enhancement operations on the video and image data, and generates a personnel heat map. S2, the pre-processing module (241) performs standardization, denoising and normalization on the collected data, and the data is input to the feature extraction module (242), and the ConvNeXt network improved by the Taylor polynomial gating unit is used to extract features, replacing the original GELU activation function with the Taylor polynomial gating unit, and its expression is: Output multi-dimensional spatiotemporal features; S3, the abnormal behavior identification module (243) combines the spatiotemporal feature pyramid network analysis to detect individual abnormal behavior and identify group and environmental abnormal patterns, and calculates risk parameters to determine whether the risk level exceeds the preset threshold. The fusion module (244) integrates the heat map, environmental data and personnel location information; S4, the warning and decision module (245) generates a graded warning signal according to the risk level, calls the preset plan in the data center (100), and sends warning information and processing instructions to the user terminal (300) and on-site staff through the supervision center (200); S5. The fault judgment module (246) monitors the equipment status in real time, prompts the fault through the indicator light (247), and feeds back the operating status to the data center (100), records event data, and optimizes the abnormal behavior library and warning threshold.

8. The method for using the activity site security monitoring system based on the Taylor polynomial gate control unit according to claim 7, characterized in that: The step S1 specifically includes: S11, multiple high-definition camera devices (510) cover various areas of the event site, and each camera device (510) collects video and image data of the corresponding area; S12, the environmental sensing device (520) collects noise decibel and temperature anomaly data in real time and aligns them with the video and image data timestamps; S13, the edge computing device (530) marks the location information, and performs background subtraction, noise filtering and image enhancement operations on the original video, generates a heat map of the density distribution of people in real time, and uploads the processed video, heat map and environmental data to the data processing module (240).

9. The method for using the activity site security monitoring system based on the Taylor polynomial gate control unit according to claim 7, characterized in that: The step S2 specifically includes: S21, extracting spatial and temporal dynamic features from video sequences through a ConvNeXt network enhanced by Taylor polynomial gating units; S22, matching the extracted features with the behavior templates in the abnormal behavior library, and correlating the real-time location information provided by the personnel management subsystem (400); S23, the fusion module (244) integrates the video features, environmental data and personnel location information to match the on-site staff around the abnormal behavior location.

10. The method for using the activity site security monitoring system based on the Taylor polynomial gate control unit according to claim 7, characterized in that: The step S3 specifically includes: S31, the abnormal behavior identification module (243) receives the comprehensive features generated by the fusion module (244), inputs the comprehensive features into the spatiotemporal feature pyramid network for multi-scale analysis, and identifies abnormal behavior patterns; S32, comparing the abnormal behavior library of the data center (100), matching known risk types, and calculating the behavior deviation score; S33. Combine the real-time data of the environment sensing device (520) and perform a weighted assessment of the risk level.

Images