Intelligent monitoring platform and method based on AI vision

Abstract

The invention discloses an intelligent monitoring platform and method based on AI vision, and relates to the technical field of security monitoring, the intelligent monitoring platform comprises a video acquisition module, an intelligent calculation module, an AI vision analysis module, an alarm module and a comprehensive management platform, the video acquisition module is used for acquiring video data of a monitoring area in real time; the intelligent calculation module comprises intelligent equipment deployed in a monitoring area, and the intelligent equipment is responsible for performing feature extraction and preliminary analysis on the video data; the AI visual analysis module is used for using a cloud server and is responsible for further performing feature extraction and analysis classification on the video data processed by the intelligent calculation module, and the alarm module is used for triggering an alarm when the intelligent calculation module and the AI visual analysis module cooperatively detect abnormal behaviors of the video data. According to the invention, the real-time performance and the accuracy of monitoring are improved, and the false alarm rate and the missing report rate are reduced; and the specific environment is monitored in a customized manner, so that the pertinence and effectiveness of monitoring are improved.

Description

Technical Field

[0001] The present invention mainly relates to the field of security monitoring technology, specifically to an intelligent monitoring platform and method based on AI vision. Background Art

[0002] Security monitoring technology refers to a monitoring and management system that uses cameras and other related hardware devices, as well as software systems, to record real-time video of the monitored area. It mainly includes multiple subsystems such as video security monitoring systems, intrusion alarm systems, and access control systems. These systems can be used individually, in combination, or integrated to achieve comprehensive security protection for the monitored area.

[0003] Traditional monitoring platforms have long been the core tools in various security monitoring scenarios. These platforms mainly rely on manual inspections and post-analysis to process monitoring data. However, with the expansion of monitoring scope and the increase in monitoring needs, the limitations of traditional monitoring platforms have gradually emerged. First, there are obvious efficiency issues in the manual inspection method. Due to the large amount of monitoring data and the wide monitoring scope, manual inspections often consume a lot of time and manpower costs, and are prone to missed detection or false detection due to human factors. At the same time, the manual inspection method cannot achieve real-time monitoring and early warning, and can often only be analyzed post-event after the incident occurs, which greatly limits the response speed and emergency response capabilities of the monitoring platform. Secondly, traditional monitoring platforms also have shortcomings in data processing and recognition. Due to technical limitations, traditional monitoring platforms can often only perform simple image processing and recognition, and are difficult to handle tasks such as target detection and behavior recognition in complex scenarios. This not only leads to low accuracy of monitoring results, but also easily causes problems such as false alarms and missed alarms, which brings additional troubles to monitoring work.

[0004] On the one hand, data processing delay is one of the main problems faced by intelligent monitoring platforms. Due to the huge amount of monitoring data and the need for real-time processing and analysis, higher requirements are placed on the computing power and data transmission speed of the platform. However, most current intelligent monitoring platforms still have certain delays when processing large amounts of data, which will affect the real-time and accuracy of the monitoring results. On the other hand, insufficient recognition accuracy is also a problem that intelligent monitoring platforms need to solve. Although AI technology has made great progress in image processing and recognition, recognition accuracy will still be affected to a certain extent in complex scenes and under conditions of target occlusion, lighting changes, etc. This will lead to problems such as false alarms and missed alarms, reducing the reliability and stability of the monitoring system.

[0005] To sum up, the technical problems of data processing delay and insufficient recognition accuracy urgently need to be solved by technical personnel in related fields. Summary of the invention

[0006] Based on this, the purpose of the present invention is to provide an intelligent monitoring platform and method based on AI vision to solve the technical problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions: The AI ​​vision-based intelligent monitoring platform includes a video acquisition module, an intelligent computing module, an AI vision analysis module, an alarm module, and an integrated management platform connected to the alarm module by telecommunications. Specifically: The video acquisition module is used to collect video data of the monitoring area in real time based on a high-definition camera; The intelligent computing module includes intelligent devices deployed in the monitoring area, which are responsible for feature extraction and preliminary analysis of video data; and for identifying and tracking people, vehicles, and objects; The AI ​​visual analysis module is used to use a cloud server to further extract features and analyze and classify the video data processed by the intelligent computing module for abnormal behavior detection; The alarm module is used to trigger an alarm when the intelligent computing module and the AI ​​visual analysis module jointly detect abnormal behavior in the video data, and push the alarm information to the integrated management platform; The integrated management platform is used to realize the centralized management, analysis and display of video data, and supports users to define monitoring rules and alarm thresholds.

[0008] Preferably, the smart device is responsible for preliminary feature extraction and simple preliminary classification by using a lightweight learning model, and the cloud server is responsible for feature extraction and analysis classification by using a deep learning model; the lightweight learning model and the deep learning model together constitute a collaborative model.

[0009] Preferably, the lightweight learning model is expressed as: ; in, Represents a lightweight learning model function; : Represented as the parameters of the lightweight learning model; : represents the original video frame of the video data collected by the high-definition camera in the video acquisition module; The feature extraction process is expressed as: ,in, Represents a feature vector extracted from a raw video frame of video data.

[0010] Preferably, the deep learning model is expressed as: ; in, : represents the deep learning model function; : represents the parameters of the deep learning model; : represents the feature vector extracted from the original video frame of the video data; The deep learning model output is represented as: ,in, Represents the output result of the deep learning model.

[0011] Preferably, the collaborative model is used to balance recognition accuracy and computational complexity; the recognition accuracy is expressed as: ; in, represents the recognition accuracy function; Reference data indicating unusual behavior; The computational complexity is expressed as: ,in, Represents the computational complexity function.

[0012] Preferably, the collaborative model is expressed as: ; in, : represents the original video frame of the video data, which is the input item; : Indicates the balance coefficient.

[0013] Preferably, the integrated management platform sends reminder notifications by sending text messages and emails.

[0014] Preferably, the comprehensive management platform can be integrated and linked with the access control system and the alarm system to achieve sharing and collaborative processing of video data.

[0015] The intelligent monitoring method based on AI vision includes the following steps: Step 1: Video data collection: Deploy high-definition cameras at the main entrances and exits of the industrial park, production lines, and key areas of warehouses. The cameras collect video data of the monitored areas in real time and transmit the video data to the intelligent computing module. Step 2: Preprocessing: After receiving the video data, the intelligent computing module performs preprocessing operations, including video format conversion, frame rate adjustment, and noise removal. The intelligent computing module uses the built-in lightweight learning model to perform preliminary analysis on the video data. Step 3: Intelligent analysis: upload the pre-processed video data to the AI ​​visual analysis module, which uses a deep learning model to perform intelligent analysis on the video data; Step 4: Alarm triggering and information push: When the AI ​​visual analysis module detects abnormal behavior, an alarm is triggered immediately. The alarm information includes the type of abnormal behavior, time and location of occurrence, and related video clips or screenshots; Step 5. Multi-system integration and linkage: When abnormal behavior occurs, the integrated management platform can automatically trigger the response mechanism of the relevant security system to achieve sharing and collaborative processing of security information.

[0016] In summary, the present invention mainly has the following beneficial effects: The present invention improves the real-time and accuracy of monitoring and reduces the false alarm rate and missed alarm rate. With the help of artificial intelligence technologies such as deep learning, the present invention can continuously optimize the monitoring model, realize the accurate identification and distinction of specific environmental characteristics, effectively distinguish normal activities from potential threats, and significantly reduce the false alarms and missed alarms caused by environmental interference or algorithm limitations. This not only improves the reliability of the monitoring system, but also reduces the unnecessary troubles caused to managers by frequent false alarms.

[0017] Customized monitoring of specific environments is achieved, which improves the pertinence and effectiveness of monitoring. In response to the specific needs of different application scenarios, the present invention provides a highly configurable monitoring solution. Users can adjust monitoring parameters, set alarm thresholds, and even customize exclusive monitoring logic according to actual needs. This flexibility ensures that the monitoring system can closely meet the monitoring needs of specific environments and achieve more accurate and effective monitoring management.

[0018] It supports multi-system integration and linkage, improving the overall efficiency of the security system. It reduces the frequency and cost of manual inspections and improves monitoring efficiency. The present invention is designed with an open and powerful interface that can seamlessly access various existing security systems (such as access control, fire protection, video surveillance, etc.) to achieve cross-system information sharing and linkage control. This multi-system integration capability not only promotes the effective integration of resources, but also improves the collaborative efficiency of security response, building a more comprehensive and intelligent security protection network. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a framework diagram of the overall process of intelligent monitoring of the present invention.

[0020] Description of the drawings: 10. Video acquisition module; 20. Intelligent computing module; 30. AI visual analysis module; 40. Alarm module; 50. Integrated management platform. DETAILED DESCRIPTION

[0021] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0022] Example like Figure 1 As shown, the AI ​​vision-based intelligent monitoring platform includes a video acquisition module 10, an intelligent computing module 20, an AI vision analysis module 30, an alarm module 40, and an integrated management platform 50 connected to the alarm module 40 by telecommunications. The video acquisition module 10 is used to collect video data of the monitoring area in real time based on a high-definition camera; The intelligent computing module 20 includes intelligent devices deployed in the monitoring area, which are responsible for feature extraction and preliminary analysis of video data; and for identifying and tracking people, vehicles, and objects; The AI ​​visual analysis module 30 is used to use the cloud server to further extract features and analyze and classify the video data processed by the intelligent computing module 20 for abnormal behavior detection; The alarm module 40 is used to trigger an alarm when the intelligent computing module 20 and the AI ​​visual analysis module 30 cooperate to detect abnormal behavior in the video data, and push the alarm information to the integrated management platform 50; The integrated management platform 50 is used to realize the centralized management, analysis and display of video data, and supports users to define monitoring rules and alarm thresholds.

[0023] Smart devices are responsible for preliminary feature extraction and simple preliminary classification by using lightweight learning models, while cloud servers use deep learning models for feature extraction and analysis and classification; The lightweight learning model and the deep learning model together form a collaborative model.

[0024] The lightweight learning model is expressed as: ; in, Represents a lightweight learning model function; : Represented as the parameters of the lightweight learning model; : represents the original video frame of the video data collected by the high-definition camera in the video acquisition module 10; The feature extraction process is expressed as: ,in, Represents a feature vector extracted from a raw video frame of video data.

[0025] The deep learning model is represented as: ; in, : represents the deep learning model function; : represents the parameters of the deep learning model; The deep learning model output is represented as: ,in, Represents the output result of the deep learning model.

[0026] The collaborative model is used to balance recognition accuracy and computational complexity; recognition accuracy is expressed as: ; in, represents the recognition accuracy function; Reference data indicating unusual behavior; The computational complexity is expressed as: ,in, Represents the computational complexity function.

[0027] The collaborative model is expressed as: ; in, : represents the original video frame of the video data, which is the input item; : Indicates the balance coefficient.

[0028] The integrated management platform 50 sends reminder notifications by sending text messages and emails.

[0029] The comprehensive management platform can be integrated and linked with the access control system and alarm system to achieve sharing and collaborative processing of video data.

[0030] It should be noted that in this embodiment, the deep learning framework: a distributed deep learning framework is designed, which can perform some deep learning calculations on smart devices and work with cloud servers at the same time. Smart devices are responsible for preliminary feature extraction and simple classification tasks, while complex model training and large-scale data processing are completed by cloud servers.

[0031] Deploy lightweight deep learning models such as MobileNet and ShuffleNet on smart devices. These models have a small number of parameters and computational complexity, and are suitable for running on resource-constrained smart devices. The lightweight model is responsible for extracting key features from raw video data and sending these features to the cloud for further analysis.

[0032] Therefore, we design a collaborative optimization algorithm that can dynamically adjust the size and complexity of the model according to the computing power and network bandwidth of the smart device. When the computing power of the smart device is strong or the network bandwidth is high, a more complex model can be deployed to improve the recognition accuracy; otherwise, a simpler model is used to ensure real-time performance.

[0033] In this example, we design the algorithm and select the model: MobileNetV2 is used as a lightweight feature extraction model on smart devices. MobileNetV2 is an efficient convolutional neural network suitable for running on resource-constrained devices. Input and output: The input is the original video frame and the output is the feature vector.

[0034] Training and optimization: MobileNetV2 is pre-trained using a large-scale dataset in the cloud, then migrated to smart devices and fine-tuned based on the computing power and dataset of the smart devices.

[0035] Model selection: ResNet50 is used as the advanced classification model in the cloud. ResNet50 has a deeper network structure and stronger feature extraction capabilities, which is suitable for classifying complex scenarios. Input and output: The input is the feature vector extracted by the smart device, and the output is the classification result. Training and optimization: ResNet50 is trained in the cloud using the feature vector and the corresponding label to optimize the classification performance.

[0036] The collaborative optimization objective function is designed to balance the recognition accuracy and computational complexity of the model. The objective function can be expressed as: ; in, and They are the parameters of MobileNetV2 and ResNet50 respectively. and are the computational complexity of the two models, is the original video frame, which is the input of the model; To identify the accuracy function, it is used to measure the accuracy of the model output; is the reference data of abnormal behavior, used to calculate the recognition accuracy. is the balance coefficient.

[0037] Optimization algorithms such as gradient descent are used to minimize the objective function to find the optimal model parameters.

[0038] Experimental verification: Dataset preparation,Training set: Use public image classification datasets for pre-training, and collect datasets in specific surveillance scenarios for fine-tuning. Test set: Collect test datasets containing various surveillance scenarios to evaluate the performance of the model.

[0039] Experimental setup: Hardware environment: Deploy MobileNetV2 on smart devices and ResNet50 on cloud servers. Software environment: Use deep learning framework for model training and reasoning. Evaluation indicators: Use indicators such as accuracy and computational complexity to evaluate the performance of the model.

[0040] Experimental results and analysis: Accuracy: Evaluate the recognition performance of the model by comparing the accuracy of the model on the test set. Computational complexity: Evaluate the computational complexity of the model by measuring the inference time of the model on smart devices and cloud servers. Collaborative optimization effect: By adjusting the balance coefficient , observe the trade-off between model recognition accuracy and computational complexity, and verify the effectiveness of the collaborative optimization algorithm.

[0041] According to the experimental results, we can draw the following conclusions: The combination of lightweight feature extraction models on smart devices and advanced classification models in the cloud can reduce computational complexity while ensuring recognition accuracy. The collaborative optimization algorithm can find a balance between recognition accuracy and computational complexity to improve the overall performance of the model. Future work can further explore more efficient feature extraction models, more accurate classification algorithms, and smarter collaborative optimization strategies to further improve the performance and feasibility of intelligent monitoring systems.

[0042] Specific application examples: Taking the industrial park as an example, high-definition cameras should be deployed at the main entrances and exits of the industrial park, including the main gate, side gate and emergency exit. These cameras should have night vision function to ensure all-weather monitoring effect.

[0043] Production lines: High-definition cameras are deployed at key locations of the production line, such as machine operation areas and material storage areas, to monitor safe operations and material management during the production process.

[0044] Warehouses deploy high-definition cameras at warehouse entrances and exits, shelf areas, and cargo loading and unloading areas to ensure the safe management of warehouse materials.

[0045] High-speed network transmission technology is used to ensure real-time and stable transmission of video data. At the same time, in order to cope with emergencies such as network failures, the system also has a data cache function to ensure data integrity. Data storage: Distributed storage technology is used to store video data on multiple nodes to ensure data security and reliability. At the same time, the system also has data backup and recovery functions to prevent data loss.

[0046] Intelligent computing modules are deployed at the monitoring site to pre-process and perform preliminary analysis on video data, such as personnel counting. Through image recognition technology, the number of people entering and leaving the park is counted in real time to provide data support for park security management. Vehicle identification uses license plate recognition technology to identify and record vehicles entering and leaving the park to ensure compliance management of park vehicles.

[0047] The preprocessed video data is uploaded to the AI ​​visual analysis module, and deep learning algorithms are used for intelligent analysis, such as behavior recognition and anomaly detection.

[0048] When abnormal behavior is detected, such as people entering a restricted area, by setting the restricted area range, the system can detect and trigger an alarm in real time when someone enters the restricted area without permission. Speeding vehicles are set at key locations on the park roads. When a vehicle exceeds the speed limit, the system can detect and trigger an alarm in real time.

[0049] Other abnormal behaviors: such as lost items, fighting, etc., can also be detected and identified by the system through the behavior recognition algorithm. When the AI ​​visual analysis module detects abnormal behavior, the system can immediately trigger an alarm and push the alarm information to the integrated management platform. The alarm information includes the type of abnormal behavior, location, time, and related video screenshots or recordings.

[0050] After receiving the alarm information, the integrated management platform will send the alarm information to relevant personnel via SMS, email, etc. to ensure that they can know and handle abnormal behavior in a timely manner.

[0051] Activate emergency plan: Depending on the type and severity of abnormal behavior, the system can automatically or manually activate the corresponding emergency plan, such as dispatching security personnel to the scene, initiating emergency evacuation, etc.

[0052] The working principle of the present invention is: Detailed description of the variables involved: : The original video frame, which is the input of the model.

[0053] : Feature vector extracted from the original video frame.

[0054] : The final output of the model, that is, the classification result.

[0055] : Reference data of abnormal behavior, i.e., true labels, are used to calculate recognition accuracy.

[0056] : Parameters of the lightweight model.

[0057] : Parameters of the cloud model.

[0058] : Lightweight deep learning model for feature extraction on smart devices.

[0059] : Cloud-based deep learning model for further analysis and classification of feature vectors.

[0060] : Recognition accuracy, used to measure the accuracy of model output.

[0061] : Computational complexity, used to measure the computational complexity and resource consumption of the model.

[0062] : Balance coefficient, used to adjust the weight between recognition accuracy and computational complexity.

[0063] Lightweight deep learning model for feature extraction, assuming the model is ,in is the parameter of the lightweight model. The feature extraction process can be expressed as: .

[0064] In the cloud, we use more complex deep learning models For the feature vector Further analysis and classification are performed. The final output is: .

[0065] Design a collaborative optimization objective function that aims to balance the recognition accuracy and computational complexity of the model. Assume that the recognition accuracy is in is the true label; the computational complexity is .

[0066] The collaborative optimization objective function can be expressed as: .

[0067] The above embodiments are only for illustrating the technical idea of ​​the present invention, and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the present invention.

Claims

1. An AI vision-based intelligent monitoring platform, comprising a video acquisition module (10), an intelligent computing module (20), an AI vision analysis module (30), an alarm module (40), and an integrated management platform (50) connected to the alarm module (40) by telecommunications, characterized in that: Specifically: The video acquisition module (10) is used to acquire video data of the monitoring area in real time based on a high-definition camera; The intelligent computing module (20) includes an intelligent device deployed in the monitoring area, and the intelligent device is responsible for extracting features and performing preliminary analysis on the video data; realizing the identification and tracking of people, vehicles, and objects; The AI ​​visual analysis module (30) is used to use a cloud server to further extract features and analyze and classify the video data processed by the intelligent computing module (20) for the detection of abnormal behavior; The alarm module (40) is used to trigger an alarm when the intelligent computing module (20) and the AI ​​visual analysis module (30) detect abnormal behavior in the video data in collaboration, and push the alarm information to the integrated management platform (50); The comprehensive management platform (50) is used to realize the centralized management, analysis and display of video data, and supports users to define monitoring rules and alarm thresholds.

2. The AI ​​vision-based intelligent monitoring platform according to claim 1, characterized in that: The smart device is responsible for preliminary feature extraction and simple preliminary classification by using a lightweight learning model, and the cloud server is responsible for feature extraction and analysis classification by using a deep learning model; the lightweight learning model and the deep learning model together constitute a collaborative model.

3. The AI ​​vision-based intelligent monitoring platform according to claim 2 is characterized in that: The lightweight learning model is expressed as: ; in, Represents a lightweight learning model function; : Represented as the parameters of the lightweight learning model; : represents the original video frame of the video data collected by the high-definition camera in the video acquisition module (10); The feature extraction process is expressed as: ,in, Represents a feature vector extracted from a raw video frame of video data.

4. The AI ​​vision-based intelligent monitoring platform according to claim 3 is characterized in that: The deep learning model is expressed as: ; in, : represents the deep learning model function; : Represents the parameters of the deep learning model; : represents the feature vector extracted from the original video frame of the video data; The deep learning model output is represented as: ,in, Represents the output result of the deep learning model.

5. The AI ​​vision-based intelligent monitoring platform according to claim 4 is characterized in that: The collaborative model is used to balance recognition accuracy and computational complexity; the recognition accuracy is expressed as: ; in, represents the recognition accuracy function; Reference data indicating unusual behavior; The computational complexity is expressed as: ,in, Represents the computational complexity function.

6. The AI ​​vision-based intelligent monitoring platform according to claim 5, characterized in that: The collaborative model is expressed as: ； in, : represents the original video frame of the video data, which is the input item; : Indicates the balance coefficient.

7. The AI ​​vision-based intelligent monitoring platform according to claim 1, characterized in that: The integrated management platform (50) sends reminder notifications by sending text messages and emails.

8. The AI ​​vision-based intelligent monitoring platform according to claim 1, characterized in that: The integrated management platform (50) can be integrated and linked with the access control system and the alarm system to achieve sharing and collaborative processing of video data.

9. An AI vision-based intelligent monitoring method, applied to an AI vision-based intelligent monitoring platform according to any one of claims 1 to 8, characterized in that: The following steps are involved: Step 1: Video data collection: HD cameras are deployed at the main entrances and exits of the industrial park, production lines, and key warehouse areas. The HD cameras collect video data of the monitored areas in real time and transmit the video data to the intelligent computing module (20); Step 2: Preprocessing: After receiving the video data, the intelligent computing module (20) performs preprocessing operations, including video format conversion, frame rate adjustment, and noise removal; the intelligent computing module (20) uses a built-in lightweight learning model to perform preliminary analysis on the video data; Step 3: Intelligent analysis: uploading the pre-processed video data to the AI ​​visual analysis module (30), which uses a deep learning model to perform intelligent analysis on the video data; Step 4: Alarm triggering and information push: When the AI ​​visual analysis module (30) detects abnormal behavior, an alarm is immediately triggered. The alarm information includes the type of abnormal behavior, the time and location of occurrence, and related video clips or screenshots; Step 5. Multi-system integration and linkage: When abnormal behavior occurs, the integrated management platform can automatically trigger the response mechanism of the relevant security system to achieve sharing and collaborative processing of security information.

Images