Intelligent weak current security and protection monitoring method and system fused with computer vision

Abstract

The invention relates to the technical field of intelligent weak current security and protection monitoring methods, in particular to an intelligent weak current security and protection monitoring method and system fused with computer vision, and the method comprises the following steps: deploying a monitoring center, dividing a monitoring region into a plurality of sub-regions, and further dividing each sub-region into a plurality of monitoring region blocks; deploying a camera acquisition module in each monitoring area block for acquiring video data; transmitting the video data to a video processing module through a 5G transmission module; the video processing module preprocesses the received video data; the video processing module executes target trajectory analysis according to the target coordinate information and the tracking data, and judges whether a target behavior is abnormal or not; when an abnormal behavior is detected, the video processing module identifies a target with an abnormal track and performs behavior pre-judgment on a target with a suspicious identification result, so that the identification rate and the response speed of an abnormal event are greatly improved, the false alarm rate is also remarkably reduced, and meanwhile, the system stability and the energy efficiency are optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent weak current security monitoring methods, in particular to an intelligent weak current security monitoring method and system integrating computer vision. Background Art

[0002] With the acceleration of the urbanization process and the increasing demand for security, intelligent security systems play an increasingly important role in modern society. Although traditional weak current security monitoring systems have been continuously developed in the past few decades, their limitations have become increasingly prominent when facing increasingly complex security challenges.

[0003] Currently, the most advanced security monitoring systems usually use high-definition cameras for video acquisition, transmit data through wired or wireless networks, and perform centralized processing and storage in a central control room. These systems have achieved a certain degree of automation and can perform basic motion detection and simple behavior analysis. However, they still have obvious deficiencies in dealing with complex scenarios, identifying potential threats, and predicting abnormal behaviors.

[0004] First of all, there are often bottlenecks in data acquisition and transmission of existing systems. High-definition video streams require a large amount of bandwidth, and traditional network infrastructures are difficult to meet the needs of real-time transmission, resulting in system response delays and even data loss. Secondly, existing video analysis algorithms are mostly limited to simple motion detection and fixed pattern recognition, and it is difficult to cope with complex and changeable real-world scenarios. For example, in a crowded commercial area, the system often has difficulty accurately distinguishing normal behaviors from suspicious activities, resulting in high false alarm rates and missed alarm rates.

[0005] In addition, most existing systems lack environmental perception capabilities and scene adaptability. They cannot automatically adjust the working mode according to environmental factors such as lighting conditions and weather changes, which not only affects the monitoring effect but also causes unnecessary energy waste. At the same time, the scalability and interoperability of the system also face challenges and it is difficult to meet the needs of large and complex environments.

[0006] Finally, existing systems often adopt isolated methods in data processing and analysis and lack the ability to fuse and analyze multi-source data. This results in the system being unable to comprehensively grasp the overall situation of the monitoring environment and it is difficult to achieve active early warning and intelligent decision-making.

[0007] In view of the above problems, there is an urgent need for a new type of security monitoring system that can integrate advanced computer vision technology, high-speed data transmission, intelligent analysis algorithms, and environmental perception capabilities. The system should be able to achieve efficient and accurate abnormal behavior recognition, rapid response, adaptive scene adjustment, and have strong scalability and interoperability. Summary of the Invention

[0008] The intelligent weak-current security monitoring method and system integrating computer vision proposed by the present invention are innovative solutions specifically designed to address these technical challenges. By integrating computer vision, 5G communication, artificial intelligence, and Internet of Things technologies, the present invention aims to revolutionize the working mode and performance level of traditional weak-current security monitoring systems.

[0009] The present invention proposes an intelligent weak-current security monitoring method integrating computer vision, which includes the following steps:

[0010] Deploy a monitoring center including a cloud database, and divide the monitored area into multiple sub-areas, each of which is further divided into multiple monitoring area blocks;

[0011] Deploy a camera acquisition module in each of the monitoring area blocks for collecting video data;

[0012] Transmit the video data to a video processing module through a 5G transmission module;

[0013] The video processing module preprocesses the received video data, including image enhancement and target detection, to obtain target coordinate information and tracking data;

[0014] The video processing module performs target trajectory analysis based on the target coordinate information and the tracking data to determine whether the target behavior is abnormal;

[0015] When an abnormal behavior is detected, the video processing module identifies the target with the abnormal trajectory and performs behavior prediction on the target whose recognition result is suspicious;

[0016] Transmit the abnormal behavior, the recognition result of the suspicious target, and the behavior prediction result to a result output module;

[0017] Deploy a wireless sensor network including temperature sensors, humidity sensors, and air pressure sensors for monitoring the environmental data of the monitored area;

[0018] The video processing module analyzes the environmental data collected by the wireless sensor network and transmits the analysis result to the result output module.

[0019] Preferably, the video processing module includes a data preprocessing model and a video recognition model, wherein:

[0020] The data preprocessing model uses an improved YOLOv5 algorithm for target detection, and this algorithm introduces a dynamic convolution layer, and its definition is as follows:

[0021]

[0022] Among them, is the input feature map, is the number of convolutional kernels, is the weight of the th convolutional kernel, is the

[0023] The video recognition model includes an object tracking model, an object trajectory analysis model, a behavior prediction model, and an object recognition model.

[0024] Preferably, the object recognition model adopts an improved 3D convolutional neural network algorithm, which is defined as follows:

[0025]

[0026] where, is the value of the output feature map at the position, is the activation function, is the convolutional kernel weight, is the value of the input feature map at the (dx,j + dy,k + dz) position, is the bias term, and dx, dy, and dz are learnable dynamic offset parameters.

[0027] Preferably, the training process of the video recognition model includes:

[0028] Collect image datasets of three types of objects: intrusion, fighting, and vehicles;

[0029] Label, process, and enhance the original data;

[0030] Use the improved YOLOv5 algorithm and FPN network model to train the processed data to obtain the weights of various types of objects;

[0031] Optimize the trained model, and adopt an improved attention mechanism during the optimization process, which is defined as follows:

[0032] Attention softmax

[0033] where, represent the query, key, and value matrices respectively, is the dimension of the key, is the mask matrix, used to introduce prior information in space and time, and softmax is the softmax function.

[0034] Preferably, the optimization process further includes:

[0035] Adopt an improved K-means algorithm to cluster the categories and calculate the category centers;

[0036] Calculate the average distance of the class centers and extract the class centers with larger distances;

[0037] Screen and optimize the extracted class centers, where the improved K-means algorithm is defined as follows:

[0038]

[0039] where J is the optimization objective, k is the number of clusters, is the i-th cluster, x is the data point, is the center of the i-th cluster, λ is the regularization coefficient, and R(C) is the regularization term.

[0040] Preferably, the screening process of the class centers includes:

[0041] Calculate the inter-class distance:

[0042] where is the number of samples in the i-th class, represents the i-th class, represents the class center of the i-th class, and x and y represent and the sample points in;

[0043] Calculate the mean of the inter-class distances:

[0044] where k is the total number of classes;

[0045] Calculate the distance threshold:

[0046] where μ is the average of all the means of the inter-class distances, σ is the standard deviation of all the means of the inter-class distances, and α is an adjustable coefficient;

[0047] Delete the data with a mean greater than the threshold T.

[0048] Preferably, the optimization process further includes an optimization process based on multi-kernel learning, which is defined as follows:

[0049]

[0050] where is the final decision function, is the number of kernel functions, is the weight of the -th kernel function, is the -th kernel function, is the input sample, is the support vector.

[0051] Preferably, the process of the video processing module analyzing the environmental information monitored by the wireless sensor network includes:

[0052] Collecting the environmental temperature of the monitored area through the temperature sensor, and dividing the environment into a high-temperature scenario, a low-temperature scenario, and a normal-temperature scenario according to a preset threshold;

[0053] Training the environmental data using an improved neural network, and the loss function of this neural network is defined as follows:

[0054]

[0055] where is the classification loss, is the regression loss, is the auxiliary task loss, and are the weight coefficients.

[0056] Preferably, the video processing module further includes a scene adaptation module for automatically adjusting the working mode according to environmental conditions. This module divides the environment into a daytime scene, a nighttime scene, and a day-night transition scene, and judges the current scene by calculating the solar altitude angle where the calculation formula of is as follows:

[0057]

[0058] where is the latitude of the observation location, is the solar declination, is the solar hour angle.

[0059] An intelligent weak-current security monitoring system integrating computer vision for implementing the method includes:

[0060] A front-end camera acquisition module for acquiring high-definition video data; a 5G transmission module for realizing remote transmission of video data; a video processing module for performing data preprocessing, target detection, trajectory analysis, behavior prediction, and object recognition; a result output module for outputting processing results; a wireless sensor network including a temperature sensor, a humidity sensor, and a barometric pressure sensor for monitoring environmental data of the monitored area; a monitoring center including a local database and a cloud database for storing and managing various types of data generated by the system.

[0061] The beneficial effects of the present invention are mainly reflected in the following aspects:

[0062] Through innovative system architecture and algorithm design, the present invention effectively solves multiple key problems existing in the prior art. First, by adopting 5G technology and optimized data transmission strategies, the bottleneck of real-time high-definition video transmission is overcome, significantly enhancing the system's response speed and reliability. Second, by introducing the improved YOLOv5 algorithm and 3D convolutional neural network, combined with attention mechanism and multi-core learning optimization, the system's anomaly detection and behavior analysis capabilities in complex scenarios are significantly improved.

[0063] More importantly, the present invention innovatively combines environmental perception and video analysis. Environmental data such as temperature, humidity, and air pressure are collected through a wireless sensor network, and intelligent algorithms are used for multi-source data fusion analysis. This method not only improves the system's situation awareness but also enables the system to adaptively adjust its working mode according to environmental changes, enhancing both the monitoring effect and optimizing energy usage.

[0064] Another important innovation point of the present invention lies in its high scalability and modular design. By dividing the monitoring area into multi-level sub-regions and monitoring blocks, the system can flexibly adapt to application scenarios of different scales and complexities, from small offices to large commercial complexes.

[0065] At the algorithm level, the present invention cleverly solves the contradiction between accuracy and real-time performance. By introducing dynamic convolutional layers and learnable dynamic offset parameters, the system can maintain high recognition accuracy while significantly reducing computational complexity, achieving true real-time analysis. In addition, the combination of the improved attention mechanism and multi-core learning method enables the system to perform well in processing long time series and multi-modal data, effectively enhancing the prediction ability of abnormal behaviors.

[0066] Generally speaking, through the synergistic effect of multiple technological innovations, the present invention achieves a qualitative leap in the performance of the security monitoring system. It not only significantly improves the recognition rate and response speed of abnormal events but also significantly reduces the false alarm rate, while optimizing system stability and energy efficiency. This all-round performance improvement enables the system to provide more intelligent, reliable, and efficient security solutions for various complex environments, bringing a revolutionary change to urban security and social governance. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 It is the overall system logic block diagram of the present invention.

[0068] Figure 2 It is the detailed logic block diagram of the video processing module of the present invention.

[0069] Figure 3 It is the environmental data processing logic block diagram of the present invention.

[0070] Figure 4 It is a logic block diagram of the scene adaptation module of the present invention.

[0071] Figure 5 It is the video recognition model training process of the present invention. Detailed implementation manners

[0072] In order to further elaborate on the technical means and their effects adopted by the present invention to achieve the predetermined invention purpose, the following combines the accompanying drawings and their preferred embodiments, and details their specific implementation manners, structures, features and their effects as follows. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0073] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs.

[0074] Embodiment 1

[0075] Refer to Figures 1-5 , the present invention relates to an intelligent weak-current security monitoring method and system integrating computer vision, belonging to the field of security technology. This method and system achieve efficient and intelligent security monitoring by integrating computer vision technology, 5G communication and Internet of Things sensors.

[0076] First, the intelligent weak-current security monitoring method integrating computer vision proposed by the present invention includes the following steps: deploying a monitoring center including a cloud database, dividing the monitoring area into multiple sub-areas, and further dividing each sub-area into multiple monitoring area blocks. Then, deploying a camera acquisition module 1 in each monitoring area block for collecting video data. Next, transmitting the video data to a video processing module 3 through a 5G transmission module 2.

[0077] The video processing module 3 preprocesses the received video data, including image enhancement and target detection, to obtain target coordinate information and tracking data. Subsequently, the video processing module 3 performs target trajectory analysis based on the target coordinate information and tracking data to determine whether the target behavior is abnormal. When an abnormal behavior is detected, the video processing module 3 identifies the target with the abnormal trajectory and performs behavior prediction on the target with a suspicious recognition result.

[0078] After that, the system transmits the abnormal behavior, the suspicious target recognition result and the behavior prediction result to a result output module 4. At the same time, the present invention also deploys a wireless sensor network 8 including a temperature sensor 5, a humidity sensor 6 and a pressure sensor 7 for monitoring the environmental data of the monitored area. The video processing module 3 analyzes the environmental data collected by the wireless sensor network 8 and transmits the analysis result to the result output module 4.

[0079] For example, in the application scenario of a large commercial complex, the entire building can be divided into sub-regions such as a shopping mall area, an office area, and a parking lot area, and each sub-region can be further divided into several monitoring area blocks. This hierarchical division method helps to improve the scalability and management efficiency of the system.

[0080] Secondly, the video processing module 3 of the present invention includes a data preprocessing model and a video recognition model. The data preprocessing model uses an improved YOLOv5 algorithm for object detection, and this algorithm introduces a dynamic convolutional layer, which is defined as follows:

[0081]

[0082] Among them, is the input feature map, is the number of convolutional kernels (in an embodiment of the present invention, N is set to 4), is the weight of the th convolutional kernel, is the

[0083] th convolutional operation. This dynamic convolutional layer can adaptively adjust the convolutional operation, improving the accuracy and robustness of object detection.

[0084] Again, the object recognition model of the present invention uses an improved 3D convolutional neural network algorithm, which is defined as follows:

[0085]

[0086] Among them, is the value of the output feature map at the position, is the activation function, is the convolutional kernel weight, is the value of the input feature map at the (dx,j + dy,k + dz) position, is the bias term, and dx, dy, dz are learnable dynamic offset parameters. This improved 3D convolutional neural network can better capture spatio-temporal features and improve the accuracy of object recognition.

[0087] For example, in the shopping mall area, if the system detects that a person has lingered near the jewelry counter multiple times but has not carried out normal shopping behavior, the object recognition model may identify them as a potential pickpocket, thus triggering the alert of the security personnel.

[0088] Then, the training process of the video recognition model of the present invention includes collecting image datasets of three types of objects: intrusion, fighting, and vehicles, annotating, processing, and enhancing the original data, and then using the improved YOLOv5 algorithm and FPN network model for training. During the training process, an improved attention mechanism is adopted, which is defined as follows:

[0089] Attention softmax

[0090] Among them, respectively represent the query, key, and value matrices, is the dimension of the key (set to 128 in the present invention), is the mask matrix, used to introduce prior information in space and time, and softmax is the softmax function. This improved attention mechanism can better capture long-term dependencies in the video sequence and improve the performance of the model.

[0091] Preferably, in an embodiment of the present invention, in order to further optimize the model, an improved K-means algorithm is adopted to cluster the categories and calculate the category centers. The definition of this algorithm is as follows:

[0092]

[0093] Among them, J is the optimization objective, k is the number of clusters (set to 5 in the present invention, corresponding to five states: normal, suspicious, dangerous, urgent, and unknown), is the i-th cluster, x is the data point, is the center of the i-th cluster, λ is the regularization coefficient (taking the value of 0.05), and R(C) is the regularization term. This improved K-means algorithm can better balance the size of the clusters and improve the stability of clustering.

[0094] Next, in the process of screening the category centers of the present invention, a series of steps are adopted to ensure the accuracy of screening. First, calculate the inter-class distance:

[0095]

[0096] Among them, is the number of samples in the i-th category, represents the i-th category, represents the category center of the i-th category, and x and y respectively represent and sample points in. Then, calculate the mean of the inter-class distances:

[0097]

[0098] Among them, k is the total number of categories. Finally, calculate the distance threshold:

[0099]

[0100] Among them, μ is the average of the means of all inter-class distances, σ is the standard deviation of the means of all inter-class distances, and α is an adjustable coefficient (taking the value of 2.0 in the present invention). The system will delete the data with a mean greater than the threshold T to improve the stability and generalization ability of the model.

[0101] In addition, the present invention also introduces an optimization process based on multi-kernel learning, which is defined as follows:

[0102]

[0103] Among them, is the final decision function, is the number of kernel functions (selecting 4 kernel functions in the present invention), is the weight of the th kernel function, is the th kernel function, is the input sample, is the support vector. This multi-kernel learning method can better adapt to different feature spaces and improve the generalization ability of the model.

[0104] In terms of environmental information analysis, the video processing module 3 of the present invention adopts an improved neural network, and its loss function is defined as follows:

[0105]

[0106] Among them, is the classification loss, is the regression loss, is the auxiliary task loss, and are weight coefficients (in the present invention takes 0.7, takes 0.2). This multi-task learning method can optimize multiple objectives simultaneously and improve the overall performance of the model.

[0107] For example, in the parking lot area, the system can analyze the driving trajectory of the vehicle (classification task), predict the parking position of the vehicle (regression task), and determine whether there is any abnormal behavior of the vehicle (auxiliary task) simultaneously.

[0108] Finally, the present invention also includes a scene adaptation module 9 for automatically adjusting the working mode according to environmental conditions. This module judges the current scene by calculating the solar altitude angle h, and its calculation formula is as follows:

[0109]

[0110] Among them, is the latitude of the observation location, is the solar declination, is the solar hour angle. Through this method, the system can accurately determine whether it is daytime, night, or the period of day-night alternation, and thus select the most suitable working mode. For example, in the night mode, the system will rely more on the data of the infrared camera and adjust the algorithm parameters to adapt to the low-light conditions.

[0111] Generally speaking, the intelligent low-voltage security monitoring system integrating computer vision of the present invention includes a front-end camera acquisition module 1, a 5G transmission module 2, a video processing module 3, a result output module 4, a wireless sensor network 8 (including a temperature sensor 5, a humidity sensor 6, and a barometric pressure sensor 7), and a monitoring center 10. This system architecture realizes the full-process intelligence from data acquisition, transmission, processing to result output, greatly improving the efficiency and accuracy of security monitoring.

[0112] Through the above detailed description, it can be seen that the present invention has significant innovations and advantages in the field of intelligent low-voltage security monitoring. It not only integrates a number of advanced technologies, but also makes innovations in algorithm optimization, system architecture, and scene adaptability. This method of integrated innovation transforms the low-voltage security system from a simple monitoring and recording tool into a comprehensive security solution with active perception, intelligent analysis, and prediction and early warning capabilities, and has broad application prospects in various scenarios such as commercial complexes, smart cities, and industrial parks.

[0113] Embodiment 1: In a large commercial complex, we deployed the intelligent low-voltage security monitoring system integrating computer vision of the present invention. The complex includes a shopping center, an office area, and an underground parking lot, with a total area of about 100,000 square meters. We divided the entire area into 15 sub-areas, and each sub-area was further divided into 10 - 20 monitoring area blocks. In each monitoring area block, we installed high-definition cameras and far-infrared cameras, and at the same time deployed temperature, humidity, and barometric pressure sensors.

[0114] The system uses a 5G network for data transmission to ensure the real-time transmission of a large amount of high-definition video and sensor data. The video processing module uses our improved YOLOv5 algorithm and 3D convolutional neural network, combined with attention mechanism and multi-core learning optimization. The scene adaptation module automatically adjusts the working mode according to time and lighting conditions.

[0115] Comparative Example 1: In another commercial complex of similar scale, we deployed a traditional weak current security monitoring system. This system uses ordinary high-definition cameras and transmits data through a wired network. Video processing adopts a conventional motion detection algorithm and does not have intelligent analysis functions. The system does not include environmental sensors and has no scene adaptation ability.

[0116] The test methods and standards are as follows:

[0117] We conducted a 30-day test in these two commercial complexes. The main test indicators include: abnormal event recognition rate, false alarm rate, response time, system stability, and energy efficiency. The specific test methods are as follows:

[0118] 1. Abnormal event recognition rate: We simulated 100 different types of abnormal events (such as theft, fighting, suspicious behavior, etc.) and recorded the number of times the system correctly recognized them.

[0119] 2. False alarm rate: Record the ratio of the number of times the system gives false alarms to the total number of alarms.

[0120] 3. Response time: The average time from the occurrence of an abnormal event to the system issuing an alarm.

[0121] 4. System stability: Record the number and duration of system failures or interruptions.

[0122] 5. Energy efficiency: Record the average daily power consumption of the entire system.

[0123] The test results are as follows:

[0124] Test indicators Example 1 Comparative Example 1 Abnormal event recognition rate 95% 68% False alarm rate 3% 15% Average response time 2.5 seconds 12 seconds System stability (number of failures within 30 days) 1 time 5 times Daily average energy consumption 180 kWh 220 kWh

[0125] The test results clearly demonstrate the superiority of the system of the present invention. First of all, in terms of the abnormal event recognition rate, the system of the present invention reaches a high recognition rate of 95%, far exceeding the 68% of the traditional system. This benefits from our improved YOLOv5 algorithm and 3D convolutional neural network, which can capture abnormal behaviors in complex scenes more accurately.

[0126] Secondly, the false alarm rate of the system of the present invention is only 3%, much lower than the 15% of the traditional system. This significant improvement is attributed to the multi-core learning optimization and attention mechanism we introduced, which improve the judgment accuracy of the system and effectively reduce false alarms.

[0127] In terms of response time, the system of the present invention only needs an average of 2.5 seconds to identify and report abnormal events, while the traditional system requires 12 seconds. This fast response ability stems from our 5G data transmission and real-time video processing technology, providing more reaction time for security personnel.

[0128] The improvement in system stability is also very significant. During the 30-day test period, the system of the present invention had only 1 minor failure, while the traditional system had 5 interruptions. This reflects the reliability and robustness of our system architecture.

[0129] Finally, although the system of the present invention integrates more advanced functions, its daily average energy consumption is 18% lower than that of the traditional system. This is mainly due to our scenario adaptation module and intelligent management. The system can adjust its working mode according to actual needs, avoiding unnecessary energy waste.

[0130] In summary, the intelligent weak-current security monitoring system integrating computer vision of the present invention is significantly superior to the traditional system in all key indicators. It not only improves the security efficiency and accuracy but also reduces the operation cost. These advantages make this system particularly suitable for application in large and complex environments such as commercial complexes, smart cities, and industrial parks. With the acceleration of urbanization and the improvement of security requirements, we believe that the present invention will play an increasingly important role in the field of intelligent security.

[0131] It should be noted that the above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An intelligent weak current security monitoring method integrating computer vision, characterized in that It includes the following steps: Deploy a monitoring center containing a cloud database, divide the monitoring area into multiple sub-areas, and further divide each sub-area into multiple monitoring area blocks; Deploy a camera acquisition module in each of the monitoring area blocks for acquiring video data; Transmit the video data to a video processing module through a 5G transmission module; The video processing module preprocesses the received video data, including image enhancement and target detection, to obtain target coordinate information and tracking data; The video processing module performs target trajectory analysis based on the target coordinate information and the tracking data to determine whether the target behavior is abnormal; When an abnormal behavior is detected, the video processing module identifies the target with the abnormal trajectory and makes a behavior prediction for the target with a suspicious recognition result; Transmit the abnormal behavior, the suspicious target recognition result, and the behavior prediction result to a result output module; Deploy a wireless sensor network including temperature sensors, humidity sensors, and barometric pressure sensors for monitoring the environmental data of the monitored area; The video processing module analyzes the environmental data collected by the wireless sensor network and transmits the analysis result to the result output module.

2. The intelligent weak current security monitoring method integrating computer vision according to claim 1, characterized in that, The video processing module includes a data preprocessing model and a video recognition model, where: The data preprocessing model uses an improved YOLOv5 algorithm for target detection. This algorithm introduces a dynamic convolutional layer, and its definition is as follows: Among them, is the input feature map, is the number of convolutional kernels, is the weight of the th convolutional kernel, is the th convolution operation; The video recognition model includes a target tracking model, a target trajectory analysis model, a behavior prediction model, and an object recognition model.

3. The intelligent weak current security monitoring method integrating computer vision according to claim 2, wherein, The object recognition model uses an improved 3D convolutional neural network algorithm, and its definition is as follows: Among them, is the value of the output feature map at position, is the activation function, is the convolutional kernel weight, is the value of the input feature map at (dx,j + dy,k + dz) position, is the bias term, and dx, dy, and dz are learnable dynamic offset parameters.

4. The intelligent weak current security monitoring method integrating computer vision according to claim 2, characterized in that, The training process of the video recognition model includes: Collect image data sets of three types of objects: intrusion, fighting, and vehicles; Label, process, and enhance the original data; Use the improved YOLOv5 algorithm and the FPN network model to train the processed data to obtain the weights of various types of objects; Optimize the trained model. During the optimization process, an improved attention mechanism is adopted, and its definition is as follows: Attention softmax Among them, respectively represent the query, key, and value matrices, is the dimension of the key, is the mask matrix, used to introduce prior information in space and time, and softmax is the softmax function.

5. The intelligent weak-current security monitoring method integrating computer vision according to claim 4, characterized in that, The optimization process also includes: Use an improved K-means algorithm to cluster the categories and calculate the category centers; Calculate the average distance of the category centers, and extract the category centers with larger distances; Screen and optimize the extracted category centers, where the improved K-means algorithm is defined as follows: where J is the optimization objective, k is the number of clusters, is the i-th cluster, x is the data point, is the center of the i-th cluster, λ is the regularization coefficient, and R(C) is the regularization term.

6. The intelligent weak current security monitoring method integrating computer vision according to claim 5, characterized in that The screening process of the category centers includes: Calculate the distance between classes: Among them, is the number of samples in the i-th category, represents the i-th category, represents the category center of the i-th category, where x and y respectively represent and sample points in Calculate the mean inter-class distance: where k is the total number of categories; Calculate distance threshold: where μ is the average of all inter-class distance means, σ is the standard deviation of all inter-class distance means, and α is an adjustable coefficient; Delete the data with a mean greater than the threshold T.

7. The intelligent weak-current security monitoring method integrating computer vision according to claim 5, characterized in that, The optimization process also includes an optimization process based on multi-core learning, and its definition is as follows: in, is the final decision function, is the number of kernel functions, For the The weight of the kernel function, For the Kernel function, is the input sample, is the support vector.

8. The intelligent weak current security monitoring method integrating computer vision according to claim 1, wherein, The process by which the video processing module analyzes the environmental information monitored by the wireless sensor network includes: Collect the environmental temperature of the monitoring area through the temperature sensor, and divide the environment into a high-temperature scenario, a low-temperature scenario, and a normal-temperature scenario according to a preset threshold; Use an improved neural network to train the environmental data, and the loss function of this neural network is defined as follows: Among them, is the classification loss, is the regression loss, is the auxiliary task loss, and are weight coefficients.

9. The intelligent weak-current security monitoring method integrating computer vision according to claim 1, wherein The video processing module further includes a scene adaptation module for automatically adjusting the working mode according to environmental conditions. This module divides the environment into a daytime scene, a nighttime scene, and a day-night transition scene, and determines the current scene by calculating the solar altitude angle wherein the calculation formula is as follows: wherein, is the latitude of the observation location, is the solar declination, is the solar hour angle.

10. An intelligent weak-current security monitoring system integrating computer vision for implementing the method according to any one of claims 1-9, characterized in that, It includes: The front-end camera acquisition module is used to acquire high-definition video data; The 5G transmission module is used to realize the remote transmission of video data; The video processing module is used to perform data preprocessing, target detection, trajectory analysis, behavior prediction, and object recognition; The result output module is used to output the processing results; The wireless sensor network, including temperature sensors, humidity sensors, and barometric pressure sensors, is used to monitor the environmental data of the monitored area; the monitoring center, including a local database and a cloud database, is used to store and manage various types of data generated by the system.

Images