Storage security protection and fire-fighting early warning method based on edge AI mode

Abstract

The invention discloses a storage security protection and fire-fighting early warning method based on an edge AI mode, and belongs to the technical field of safety monitoring, and the early warning method employs an edge AI artificial intelligence video data processing analysis method to carry out intelligent security protection behavior identification early warning and intelligent fire-fighting behavior identification early warning; the edge AI artificial intelligence video data processing analysis method comprises the steps that a data center pushes a trained model to an edge AI server, and the edge AI server analyzes streaming data collected by edge Internet of Things equipment in real time and carries out AI reasoning; the edge AI server uploads the received new condition related data to the cloud to train the model, and the trained model is pushed to the edge AI server again; according to the invention, security and protection and fire-fighting applications are deeply fused, an information island is thoroughly broken, data sharing and interconnection are realized, unified output of early warning signals is ensured, and the emergency response speed and accuracy are improved.

Description

Technical Field

[0001] The present invention belongs to the field of security monitoring technology, and in particular relates to a warehouse security and fire warning method based on an edge AI mode. Background Art

[0002] In the current warehouse security and fire protection system, key security information such as video surveillance, intelligent access control, perimeter protection, attendance management and fire warning sensors are isolated in different systems, lacking the necessary integration and linkage mechanism, resulting in information islands, which seriously weakens the overall security effectiveness. Managers need to frequently shuttle between multiple systems and manually track various security data, making it difficult to achieve comprehensive and real-time monitoring of warehouse security information, making it difficult to immediately understand and prevent potential business operation risks, and there is an information island phenomenon; in addition, the warehouse security and fire protection system has insufficient hardware facilities. Due to the early construction or limited funds, some warehouses have aging and serious damage to fire protection facilities, which cannot meet the needs of modern fire safety, which increases the risk of fire and the severity of consequences; the existing warehouse security and fire protection system relies more on human factors and over-reliance on the traditional mode of patrol and monitoring by security personnel, which is not only inefficient and prone to errors, but also prone to weak safety awareness and illegal operations. In the face of sudden security incidents, the lag in manual response makes it difficult to respond quickly and effectively, which buries hidden dangers for warehouse safety.

[0003] Therefore, developing an efficient warehouse security and fire warning method to avoid information island phenomenon, reduce the impact of human factors, improve the intelligence of system operation, and reduce the false alarm rate of warning is of great significance for warehouse security and fire safety. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention discloses a warehouse security and fire warning method based on edge AI mode.

[0005] The present invention adopts the following technical solution:

[0006] A warehouse security and fire warning method based on edge AI mode, wherein the warning method uses edge AI artificial intelligence video data processing and analysis method to perform intelligent security behavior recognition and warning and intelligent fire behavior recognition and warning;

[0007] The edge AI artificial intelligence video data processing and analysis method includes: the data center pushes the trained model to the edge AI server, the edge AI server analyzes the streaming data collected by the edge IoT device in real time and performs AI reasoning; the edge AI server uploads the received new situation related data to the cloud to train the model, and the trained model is pushed to the edge AI server again;

[0008] The intelligent security behavior recognition and warning include: regional intrusion monitoring, using deep learning algorithms to identify unauthorized entry into the selected area; sleeping or leaving the post monitoring, analyzing the activities of people in the work area and judging whether the employees are in working state; electric bicycle prohibition elevator monitoring, monitoring whether there are electric bicycles trying to enter the elevator; headcount and over-limit monitoring, real-time monitoring of the flow of people in each area of ​​the warehouse, setting thresholds for the flow of people, and triggering alarms when the flow of people exceeds the threshold; personal protective equipment monitoring, the personal protective equipment monitoring includes the verification and identification of the wearing of safety helmets, work clothes and reflective clothing; dangerous behavior identification, the dangerous behavior includes behaviors that can cause safety accidents, the dangerous behavior includes smoking or using mobile phones; tripwire counting and fall monitoring, used to set virtual boundary lines to count the number of people entering and leaving, and immediately notify relevant personnel when someone falls; mask wearing monitoring, used to monitor whether all personnel entering the work area wear masks correctly;

[0009] The intelligent fire behavior recognition and warning include: flame monitoring and smoke monitoring, using image processing technology to quickly locate the source of fire and assess the size of the fire; emergency passage occupancy monitoring, monitoring and identifying the occupancy of emergency passages to ensure smooth evacuation routes; obstacle monitoring, issuing warnings when obstacles that block the escape route are identified; object monitoring and tracking, tracking and recording changes in the location of important materials; and abnormal environmental parameter monitoring.

[0010] Furthermore, the edge AI artificial intelligence video data processing and analysis method includes:

[0011] Hardware configuration: The hardware configuration includes a high-definition digital surveillance camera; the surveillance camera has the function of pulling network video streams;

[0012] Edge computing server: The edge computing server includes a high-performance small server or edge gateway deployed in the warehouse to perform algorithm analysis tasks;

[0013] Sensor network: The sensor network includes environmental monitoring devices, including temperature sensors and smoke detectors, which are connected to edge computing nodes to provide additional data input;

[0014] Warehouse inspection robot: The robot obtains real-time inspection data through inspection. The inspection data includes: fire, loss of fire extinguisher, blocked fire passage, illegal personnel or temperature and humidity. The real-time inspection data is connected to the edge computing node to provide dynamic additional data input.

[0015] Furthermore, the edge AI artificial intelligence video data processing and analysis method includes:

[0016] Software architecture, the software architecture includes: lightweight operating system, the lightweight operating system uses an optimized embedded operating system; containerization technology, the containerization technology uses the Docker containerization solution; message queue, the message queue uses MQTT or other lightweight message protocols; edge AI framework, the edge AI framework uses an artificial intelligence framework suitable for edge computing scenarios;

[0017] Data preprocessing and feature extraction, the data preprocessing and feature extraction include: after the video stream enters the edge computing node, first remove noise and adjust the format through a preprocessing step; apply computer vision technology to perform preliminary analysis on the image, the preliminary analysis includes background subtraction or motion detection; extract key feature vectors from the video frame as the basis for the next step of recognition.

[0018] Furthermore, the edge AI artificial intelligence video data processing and analysis method includes:

[0019] Intelligent security algorithm, the intelligent security algorithm includes: identifying regional intrusion, sleeping on duty or leaving the post through target tracking and behavior analysis algorithms; prohibiting electric bicycles from entering elevators and counting the number of people in the area through object classification and counting technology; monitoring personal protective equipment through image segmentation and semantic understanding, the personal protective equipment includes helmet monitoring and reflective clothing recognition; judging dangerous behaviors of personnel through posture estimation models, the dangerous behaviors include smoking and making phone calls;

[0020] An intelligent fire fighting algorithm includes: realizing open flame and smoke monitoring through a flame recognition model and a method based on color space conversion; evaluating the occupancy of fire passages through a static obstacle monitoring algorithm; and tracking lost items through comparative analysis of historical trajectory information and current scenes.

[0021] Furthermore, the edge AI artificial intelligence video data processing and analysis method includes:

[0022] Abnormal alarm mechanism, which includes: when the edge computing node detects any violation of safety rules, it immediately triggers the alarm process; records the timestamps, location coordinates, and related evidence pictures or video clips of all violations;

[0023] The cloud-edge collaborative working mode includes: for difficult tasks that are highly complex or require long-term data accumulation to complete, the difficult tasks are uploaded to the cloud for processing;

[0024] Security measures include: ensuring that all transmitted data is encrypted; regularly updating software versions and patches on edge devices; and implementing strict access control policies.

[0025] Furthermore, the intelligent security behavior recognition and early warning includes:

[0026] Image acquisition and preprocessing, which includes: using a high-definition digital camera to acquire real-time video data and transmitting the real-time video data to an edge AI server; performing preprocessing operations on the original video frames of the real-time video data, which includes denoising and contrast enhancement to improve the effect of subsequent analysis;

[0027] Target monitoring and tracking, which includes: applying deep learning target monitoring algorithms to identify objects and people in the video; using multi-target tracking technology to continuously track the movement trajectory of identified objects between different frames.

[0028] Furthermore, the intelligent security behavior recognition and early warning includes:

[0029] The regional intrusion monitoring includes: setting a virtual boundary line for a specific area, and triggering an alarm when an unauthorized person enters the area;

[0030] The sleeping or leaving the post monitoring includes: monitoring the posture changes and activity frequency of human bodies in the designated area to determine whether there are workers who are stationary or leaving their posts for a long time;

[0031] The electric bicycle prohibition elevator monitoring includes: analyzing the type of vehicles entering and exiting the elevator entrance, and issuing an alarm if an electric bicycle attempts to enter;

[0032] The headcount counting and over-limit monitoring includes: monitoring and counting the current number of people in the selected area, and comparing it with the preset maximum capacity threshold, and triggering an alarm when the maximum capacity threshold is exceeded;

[0033] The tripwire counting and fall monitoring includes: setting a virtual tripwire in the detection channel to monitor whether a person falls.

[0034] Furthermore, the intelligent security behavior recognition and early warning includes:

[0035] Scene understanding and environmental perception, which includes parsing the surveillance images in combination with semantic segmentation technology to distinguish different background elements and foreground objects; using 3D reconstruction technology to construct the spatial layout inside the warehouse and locate the specific location where the event occurred;

[0036] Algorithm optimization, including: adopting a lightweight neural network architecture; using model compression technology to reduce computational complexity; using a multi-task learning framework to enable a single model to handle multiple types of monitoring tasks simultaneously;

[0037] The alarm mechanism and response strategy include: automatically generating alarm information when any intelligent security rule is violated, and the alarm information includes timestamp, location description, on-site screenshot or video clip; providing a visual interface to display the real-time status and historical records of all cameras in the warehouse.

[0038] Furthermore, the intelligent firefighting behavior recognition and early warning includes:

[0039] Image and environmental data acquisition, including video monitoring and sensor integration;

[0040] The video monitoring includes: a high-definition digital camera continuously captures real-time images in the warehouse to obtain video data, and transmits the video data to an edge AI server for processing;

[0041] The sensor integration includes: the sensor includes a physical quantity sensing device, and the physical quantity sensing device includes a visual sensor, a temperature sensor or a smoke detector;

[0042] Flame monitoring and smoke monitoring include: using deep learning algorithms to analyze video frames and identify open flame or smoke features; the flame monitoring includes: using color space conversion combined with morphological operations to enhance the expressiveness of the flame area, and using dynamic threshold segmentation technology to separate the flame part from the background; the smoke monitoring includes: extracting texture features on grayscale images, and judging whether there is abnormal diffusion by comparing the changes between consecutive frames;

[0043] Emergency passage occupancy monitoring includes: using target tracking technology to monitor all safety exits, and triggering an alarm if an obstacle is found blocking the escape route; identifying a specific area by setting a virtual boundary line or using a semantic segmentation model to continuously monitor whether there are vehicles or other objects parked in the area; when there is an object target in the emergency passage, if the size of the object target is larger than the preset size and the time it stays in the passage exceeds the specified time, an alarm is triggered;

[0044] Item monitoring and tracking includes: combining historical data to establish standard placement information for important materials in the warehouse, and comparing the differences between the current layout and the standard placement information; when the location of important materials is detected to have changed, the event is automatically recorded and an alarm is triggered;

[0045] Abnormal environmental parameter monitoring includes: integrating data from temperature sensors and smoke detectors, and setting thresholds to distinguish between normal ranges and potentially dangerous conditions.

[0046] Furthermore, the intelligent firefighting behavior recognition and early warning includes:

[0047] Multimodal fusion decision-making, the multimodal fusion decision-making includes: improving judgment accuracy by fusing cross-modal information when a single sensor may have a false alarm;

[0048] Emergency communication and command and dispatch, including: sending an alarm notification to designated personnel when an emergency is confirmed, including the location, time and initial assessment of the severity of the accident; providing an interactive map interface to display the specific location of the affected area and the layout of surrounding facilities;

[0049] Data analysis and post-event review, the data analysis and post-event review include: recording the relevant alarm data information of each alarm triggering, the alarm data information includes: video clips or sensor value curve graphs.

[0050] Beneficial effects:

[0051] The method of the present invention deeply integrates security and fire protection applications, completely breaks down information silos, and realizes data sharing and interconnection; all security and fire protection data are centrally managed and analyzed on the same highly integrated platform, ensuring unified output of early warning signals and improving emergency response speed and accuracy.

[0052] By utilizing the widely deployed surveillance cameras in the warehouse and directly accessing them through smart terminals, duplicate construction is avoided. Intelligent analysis technology automatically analyzes abnormal behaviors in images and outputs instant alarm signals, effectively reducing the cost of adding monitoring points and achieving efficient use of resources.

[0053] This method enables 24-hour uninterrupted monitoring, instantly captures and handles multiple potential security threats, has higher monitoring efficiency, and greatly reduces the probability of security accidents.

[0054] The method of the present invention has an automatic alarm function, which can instantly trigger an alarm or push alarm information to a unified platform, ensuring that security personnel or management personnel can receive notifications in the first place and take countermeasures quickly, effectively shortening the response time and avoiding the deterioration of the situation.

[0055] The data storage and analysis functions integrated in this method can comprehensively record and analyze historical events and monitoring data, provide solid data support for the formulation and optimization of security management strategies, and help make decisions more scientific and accurate.

[0056] This method supports remote monitoring functions, allowing managers to easily monitor multiple storage equipment and locations at the same time, and fully understand the security status no matter where they are, greatly improving the flexibility and efficiency of management.

[0057] This method is not limited to monitoring and alarming, but also supports advanced behavior analysis technology, which can intelligently identify and link relevant personnel information. This function enables the rapid mobilization of relevant resources when responding to security incidents, achieving more intelligent and efficient emergency response. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0059] Figure 1 It is a schematic diagram of the framework of a warehouse security and fire warning method based on edge AI mode of the present invention;

[0060] Figure 2 It is a schematic diagram of target monitoring and tracking in a warehouse security and fire warning method based on an edge AI mode of the present invention. DETAILED DESCRIPTION

[0061] In order to better understand the technical solution of the present invention, the embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0062] It should be clear that the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0063] Example 1

[0064] like Figure 1 As shown, Figure 1 It is a schematic diagram of the framework of a warehouse security and fire warning method based on edge AI mode of the present invention; Figure 2 It is a schematic diagram of target monitoring and tracking in a warehouse security and fire warning method based on an edge AI mode of the present invention.

[0065] A warehouse security and fire warning method based on edge AI mode, wherein the warning method uses edge AI artificial intelligence video data processing and analysis method to perform intelligent security behavior recognition and warning and intelligent fire behavior recognition and warning;

[0066] The edge AI artificial intelligence video data processing and analysis method includes: the data center pushes the trained model to the edge AI server, the edge AI server analyzes the streaming data collected by the edge IoT device in real time and performs AI reasoning; the edge AI server uploads the received new situation related data to the cloud to train the model, and the trained model is pushed to the edge AI server again;

[0067] The intelligent security behavior recognition and warning include: regional intrusion monitoring, using deep learning algorithms to identify unauthorized entry into the selected area; sleeping or leaving the post monitoring, analyzing the activities of people in the work area and judging whether the employees are in working state; electric bicycle prohibition elevator monitoring, monitoring whether there are electric bicycles trying to enter the elevator; headcount and over-limit monitoring, real-time monitoring of the flow of people in each area of ​​the warehouse, setting thresholds for the flow of people, and triggering alarms when the flow of people exceeds the threshold; personal protective equipment monitoring, the personal protective equipment monitoring includes the verification and identification of the wearing of safety helmets, work clothes and reflective clothing; dangerous behavior identification, the dangerous behavior includes behaviors that can cause safety accidents; tripwire counting and fall monitoring, used to set virtual boundary lines to count the number of people entering and leaving, and immediately notify relevant personnel when someone falls; mask wearing monitoring, used to monitor whether all people entering the work area wear masks correctly;

[0068] The intelligent fire behavior recognition and warning include: flame monitoring and smoke monitoring, using image processing technology to quickly locate the source of fire and assess the size of the fire; emergency passage occupancy monitoring, monitoring and identifying the occupancy of emergency passages to ensure smooth evacuation routes; obstacle monitoring, issuing warnings when obstacles that block the escape route are identified; object monitoring and tracking, tracking and recording changes in the location of important materials; and abnormal environmental parameter monitoring.

[0069] Furthermore, the edge AI artificial intelligence video data processing and analysis method includes:

[0070] Hardware configuration: The hardware configuration includes a high-definition digital surveillance camera; the surveillance camera has the function of pulling network video streams;

[0071] Edge computing server: The edge computing server includes a high-performance small server or edge gateway deployed in the warehouse to perform algorithm analysis tasks;

[0072] Sensor network: The sensor network includes environmental monitoring devices, including temperature sensors and smoke detectors, which are connected to edge computing nodes to provide additional data input;

[0073] Warehouse inspection robot: The robot obtains real-time inspection data through inspection. The inspection data includes: fire, loss of fire extinguisher, blocked fire passage, illegal personnel or temperature and humidity. The real-time inspection data is connected to the edge computing node to provide dynamic additional data input.

[0074] Furthermore, the edge AI artificial intelligence video data processing and analysis method includes:

[0075] Software architecture, the software architecture includes: lightweight operating system, the lightweight operating system uses an optimized embedded operating system; containerization technology, the containerization technology uses the Docker containerization solution; message queue, the message queue uses MQTT or other lightweight message protocols; edge AI framework, the edge AI framework uses an artificial intelligence framework suitable for edge computing scenarios;

[0076] Data preprocessing and feature extraction, the data preprocessing and feature extraction include: after the video stream enters the edge computing node, first remove noise and adjust the format through a preprocessing step; apply computer vision technology to perform preliminary analysis on the image, the preliminary analysis includes background subtraction or motion detection; extract key feature vectors from the video frame as the basis for the next step of recognition.

[0077] Furthermore, the edge AI artificial intelligence video data processing and analysis method includes:

[0078] Intelligent security algorithm, the intelligent security algorithm includes: identifying regional intrusion, sleeping on duty or leaving the post through target tracking and behavior analysis algorithms; prohibiting electric bicycles from entering elevators and counting the number of people in the area through object classification and counting technology; monitoring personal protective equipment through image segmentation and semantic understanding, the personal protective equipment includes helmet monitoring and reflective clothing recognition; judging dangerous behaviors of personnel through posture estimation models, the dangerous behaviors include smoking and making phone calls;

[0079] An intelligent fire fighting algorithm includes: realizing open flame and smoke monitoring through a flame recognition model and a method based on color space conversion; evaluating the occupancy of fire passages through a static obstacle monitoring algorithm; and tracking lost items through comparative analysis of historical trajectory information and current scenes.

[0080] Furthermore, the edge AI artificial intelligence video data processing and analysis method includes:

[0081] Abnormal alarm mechanism, which includes: when the edge computing node detects any violation of safety rules, it immediately triggers the alarm process; records the timestamps, location coordinates, and related evidence pictures or video clips of all violations;

[0082] The cloud-edge collaborative working mode includes: for difficult tasks that are highly complex or require long-term data accumulation to complete, the difficult tasks are uploaded to the cloud for processing;

[0083] Security measures include: ensuring that all transmitted data is encrypted; regularly updating software versions and patches on edge devices; and implementing strict access control policies.

[0084] Furthermore, the intelligent security behavior recognition and early warning includes:

[0085] Image acquisition and preprocessing, the image acquisition and preprocessing including: using a high-definition digital camera to acquire real-time video data, and transmitting the real-time video data to an edge AI server; performing preprocessing operations on the original video frames of the real-time video data, the preprocessing operations including denoising and contrast enhancement;

[0086] Target monitoring and tracking, which includes: applying deep learning target monitoring algorithms to identify objects and people in the video; using multi-target tracking technology to continuously track the movement trajectory of identified objects between different frames.

[0087] Furthermore, the intelligent security behavior recognition and early warning includes:

[0088] Area intrusion monitoring includes: setting a virtual boundary line for a specific area, triggering an alarm when an unauthorized person enters the area;

[0089] Sleeping or leaving the post monitoring includes: monitoring the posture changes and activity frequency of human bodies in the designated area to determine whether there are workers who are stationary or away from their posts for a long time;

[0090] The electric bicycle prohibition elevator monitoring includes: analyzing the type of vehicles entering and exiting the elevator entrance, and issuing an alarm if an electric bicycle attempts to enter;

[0091] Crowd counting and over-limit monitoring include: monitoring and counting the current number of people in the selected area, and comparing it with the preset maximum capacity threshold, triggering an alarm when the maximum capacity threshold is exceeded;

[0092] Tripwire counting and fall monitoring include: setting up virtual tripwires in the detected channel to monitor whether there are people falling.

[0093] Furthermore, the intelligent security behavior recognition and early warning includes:

[0094] Scene understanding and environmental perception, which includes parsing the surveillance images in combination with semantic segmentation technology to distinguish different background elements and foreground objects; using 3D reconstruction technology to construct the spatial layout inside the warehouse and locate the specific location where the event occurred;

[0095] Algorithm optimization, including: adopting a lightweight neural network architecture; using model compression technology to reduce computational complexity; using a multi-task learning framework to enable a single model to handle multiple types of monitoring tasks simultaneously;

[0096] The alarm mechanism and response strategy include: automatically generating alarm information when any intelligent security rule is violated, and the alarm information includes timestamp, location description, on-site screenshot or video clip; providing a visual interface to display the real-time status and historical records of all cameras in the warehouse.

[0097] Furthermore, the intelligent firefighting behavior recognition and early warning includes:

[0098] Image and environmental data acquisition, including video monitoring and sensor integration;

[0099] The video monitoring includes: a high-definition digital camera continuously captures real-time images in the warehouse to obtain video data, and transmits the video data to an edge AI server for processing;

[0100] The sensor integration includes: the sensor includes a physical quantity sensing device, and the physical quantity sensing device includes a visual sensor, a temperature sensor or a smoke detector;

[0101] Flame monitoring and smoke monitoring include using deep learning algorithms to analyze video frames and identify open flame or smoke features; the flame monitoring includes: using color space conversion combined with morphological operations to enhance the expressiveness of the flame area, and using dynamic threshold segmentation technology to separate the flame part from the background; the smoke monitoring includes: extracting texture features on grayscale images, and judging whether there is abnormal diffusion by comparing the changes between consecutive frames;

[0102] Emergency passage occupancy monitoring includes: using target tracking technology to monitor all safety exits, and triggering an alarm if an obstacle is found blocking the escape route; identifying a specific area by setting a virtual boundary line or using a semantic segmentation model to continuously monitor whether there are vehicles or other objects parked in the area; when there is an object target in the emergency passage, if the size of the object target is larger than the preset size and the time it stays in the passage exceeds the specified time, an alarm is triggered;

[0103] Item monitoring and tracking includes: combining historical data to establish standard placement information for important materials in the warehouse, and comparing the differences between the current layout and the standard placement information; when the location of important materials is detected to have changed, the event is automatically recorded and an alarm is triggered;

[0104] The abnormal environmental parameter monitoring includes: integrating the data of the temperature sensor and the smoke detector, and setting a threshold value to distinguish between a normal range and a potential dangerous condition.

[0105] Furthermore, the intelligent firefighting behavior recognition and early warning includes:

[0106] Multimodal fusion decision-making, the multimodal fusion decision-making includes: improving judgment accuracy by fusing cross-modal information when a single sensor may have a false alarm;

[0107] Emergency communication and command and dispatch, including: sending an alarm notification to designated personnel when an emergency is confirmed, including the location, time and initial assessment of the severity of the accident; providing an interactive map interface to display the specific location of the affected area and the layout of surrounding facilities;

[0108] Data analysis and post-event review, the data analysis and post-event review include: recording the relevant alarm data information of each alarm triggering, the alarm data information includes: video clips or sensor value curve graphs.

[0109] Example 2

[0110] 1. Edge AI performs video data processing and analysis, such as Figure 1 As shown in the figure: the data center pushes the trained model to the edge AI server, which analyzes the streaming data collected by edge IoT devices in real time and performs AI reasoning; when AI encounters a problem, the edge AI server uploads the difficult data to the cloud to further train the original AI model, and the trained AI model is pushed to the edge AI server again. This feedback loop plays an important role in improving model performance, and the edge AI model will only become smarter after deployment.

[0111] 2. Intelligent security behavior recognition and warning: Regional intrusion monitoring, using deep learning algorithms to identify unauthorized entry into specific areas. Sleeping / leaving monitoring, analyzing activities in the work area and automatically determining whether employees are in normal working conditions. Electric bicycles are prohibited from entering the elevator management, monitoring whether there are electric bicycles or non-designated items trying to enter the elevator. People counting and over-limit monitoring, real-time monitoring of the flow of people in each area of ​​the warehouse, and setting thresholds to trigger alarms. Personal protective equipment inspection, including but not limited to automatic verification of the wearing of safety helmets, work clothes, and reflective clothing. Identification of dangerous behaviors, such as smoking and using mobile phones, which may cause safety accidents. Tripwire counting and fall detection, setting virtual boundaries to count the number of people entering and leaving or notifying relevant personnel immediately when someone falls. Mask wearing monitoring, if necessary, ensure that all entrants wear masks correctly.

[0112] 3. Intelligent fire behavior recognition and early warning: Flame and smoke perception, using image processing technology to quickly locate the source of the fire and assess the size of the fire. Emergency passage occupancy monitoring to ensure that the evacuation route is unobstructed. Obstacle removal reminder, when any object that blocks the escape path is found, a warning is issued. Asset loss tracking, tracking and recording the location changes of important materials to prevent theft or misplacement.

[0113] Example 3

[0114] Edge AI artificial intelligence video data processing and analysis implementation method:

[0115] 1. Hardware configuration: High-definition digital surveillance camera, with network video stream pulling function, supporting rtsp protocol.

[0116] 2. Edge server: Deploy high-performance small servers or edge gateways in the warehouse to perform complex algorithm analysis tasks. These devices usually have sufficient computing resources to support the operation of deep learning models.

[0117] 3. Sensor network: including temperature sensors, smoke detectors and other types of environmental monitoring devices, connected to edge computing nodes to provide additional data input.

[0118] 4. Warehouse inspection robot: Real-time inspection data during the robot inspection process, including: fire, loss of fire extinguishers, blocked fire passages, illegal personnel, temperature and humidity environment monitoring, connected to the edge computing node to provide dynamic additional data input.

[0119] 5. Software architecture: Lightweight operating system: Use specially optimized embedded operating systems (such as YoctoProject, Ubuntu Core) to ensure that the system can run efficiently in a resource-constrained environment. Containerization technology: Use Docker containerization solutions to facilitate the rapid deployment and update of different application services. Message queue: Use MQTT or other lightweight message protocols to ensure the communication efficiency between different components and the scalability of the system. Edge AI framework: Select an artificial intelligence framework suitable for edge computing scenarios (such as TensorFlow Lite, PyTorch Mobile) to facilitate the porting of trained models to edge devices for reasoning.

[0120] The perception layer accesses the warehouse monitoring video stream and aggregates the video stream to the data processing unit. The data processing layer uses edge computing to process and analyze the collected data in real time, supporting the establishment of intelligent early warning and prediction models. The application service layer realizes intelligent analysis and recognition of video images, and provides intelligent early warning and prediction models. The management platform: equipment status monitoring, alarm recording, and unified configuration. The display layer supports client access and data linkage to the digital twin system and warehouse system.

[0121] 6. Data preprocessing and feature extraction: After the video stream enters the edge computing node, it first removes noise and adjusts the format through preprocessing steps to meet the needs of subsequent processing. Apply computer vision technology to perform preliminary analysis of the image, such as background subtraction and motion detection, to reduce unnecessary computing overhead. Extract key feature vectors from the video frame as the basis for the next step of recognition.

[0122] 7. Intelligent algorithm deployment:

[0123] Intelligent security algorithm: Area intrusion, sleeping on duty, and leaving the post can be identified through target tracking and behavior analysis algorithms. The prohibition of electric bicycles from entering elevators and the regional population counting function rely on object classification and counting technology. Safety helmet monitoring, reflective clothing recognition, and personal protective equipment inspection mainly rely on image segmentation and semantic understanding capabilities. Specific actions such as smoking and making phone calls can be judged by posture estimation models.

[0124] Intelligent fire fighting algorithm: Open fire and smoke monitoring can be achieved with the help of flame recognition models and color space conversion-based methods. The occupancy of fire passages is evaluated through static obstacle monitoring algorithms. Lost item tracking requires a comparative analysis of historical trajectory information and the current scene.

[0125] 8. Abnormal alarm mechanism: When the edge computing node detects any violation of safety rules, the alarm process is triggered immediately. Depending on the type of event, you can choose a variety of notification methods (SMS, email, sound and light alarm) to inform relevant personnel in a timely manner. Record the timestamps, location coordinates and related evidence pictures / video clips of all violations to provide a basis for subsequent investigations.

[0126] 9. Cloud-edge collaborative working mode: For some tasks that are highly complex or require long-term accumulation of large amounts of data to complete (such as large-scale data analysis and model retraining), relevant tasks can be uploaded to the cloud for processing. Through reasonable task scheduling strategies, the powerful computing power support of cloud computing can be fully utilized while ensuring real-time performance.

[0127] 10. Security measures: Ensure that all transmitted data is encrypted to prevent sensitive information leakage. Regularly update the software version and patches on edge devices to prevent potential security vulnerabilities. Implement strict access control policies to restrict access by unauthorized users.

[0128] Example 4

[0129] Intelligent security identification and early warning implementation method:

[0130] 1. Image acquisition and preprocessing: Use a high-definition digital camera to collect real-time video streams and transmit video data to the edge AI server. Perform preprocessing operations on the original video frames, including denoising and contrast enhancement, to improve the effect of subsequent analysis.

[0131] 2. Target monitoring and tracking: Use deep learning target monitoring algorithms (such as YOLO, SSD or Faster R-CNN) to identify objects and people in the video. Use multi-target tracking technology (such as Deep SORT) to continuously track the movement trajectory of the identified objects between different frames, such as Figure 2 shown.

[0132] 3. Behavior analysis and abnormality monitoring:

[0133] Area Intrusion: Set a virtual boundary line for a specific area and trigger an alarm when an unauthorized person enters the area.

[0134] Sleeping or leaving the post: By monitoring the changes in human posture and activity frequency in the designated area, it is determined whether there are workers who are stationary for a long time or away from their posts.

[0135] Electric bicycles are prohibited from entering elevators: Analyze the types of vehicles entering and exiting the elevator entrance, and issue a warning immediately if an electric bicycle attempts to enter.

[0136] People counting and over-limit monitoring: Count the current number of people in each area and compare it with the preset maximum capacity threshold, and notify the management staff when the limit is exceeded.

[0137] Personal protective equipment inspection: Use semantic segmentation models to identify whether workers are wearing necessary safety equipment (such as helmets and reflective clothing).

[0138] Identification of dangerous behaviors: including smoking and using mobile phones, by analyzing the characteristics of human body movements.

[0139] Tripwire counting and fall detection: Set up virtual tripwires in certain important passages and monitor whether anyone falls.

[0140] Mask wearing monitoring: Ensure that all personnel wear masks correctly when necessary.

[0141] 4. Scene understanding and environmental perception: Combine semantic segmentation technology to analyze the monitoring images and distinguish different background elements (such as walls, floors, shelves) and foreground objects. Use 3D reconstruction technology to build a spatial layout model inside the warehouse, which helps to more accurately locate the specific location of the event.

[0142] 5. Algorithm optimization: In view of the limited edge computing resources, a lightweight neural network architecture (such as MobileNet, EfficientDet) is adopted, and the number of model parameters is reduced as much as possible while ensuring performance. Model compression technology (such as pruning and quantization) is implemented to further reduce computational complexity. The multi-task learning framework allows a single model to handle multiple types of monitoring tasks at the same time, thereby saving computing resources.

[0143] 6. Alarm mechanism and response strategy: When any smart security rule is violated, an alarm message including timestamp, location description and on-site screenshots / video clips will be automatically generated. According to the different urgency, select the appropriate communication channel (SMS, email, APP push) to send alarms to relevant personnel. Provide a visual interface to display the real-time status and historical records of all cameras in the warehouse, so as to quickly find the root cause of the problem and take corresponding measures.

[0144] 7. Integration and management: Integrate the above functional modules into a unified operating platform to support remote configuration, upgrade and maintenance functions. Provide API interface to seamlessly connect with other management systems (such as smart warehouse, access control, lighting adjustment) to form a comprehensive smart warehousing solution.

[0145] Example 5

[0146] Intelligent fire identification and early warning implementation method:

[0147] 1. Image and environmental data collection:

[0148] Video surveillance: High-definition digital cameras continuously capture real-time images inside the warehouse and transmit video streams to edge AI servers for processing.

[0149] Sensor integration: In addition to visual information, temperature sensors and smoke detectors are also required to provide real-time environmental change data.

[0150] 2. Flame monitoring and smoke monitoring: Use deep learning algorithms (such as convolutional neural networks (CNNs)) to analyze video frames and identify possible open flames or smoke features. For flame monitoring, the color space conversion combined with morphological operations can be used to enhance the expressiveness of the flame area; at the same time, dynamic threshold segmentation technology can be used to separate the flame part from the background. Smoke monitoring focuses on the extraction of texture features on grayscale images, and determines whether there is abnormal diffusion by comparing the changes between consecutive frames.

[0151] 3. Emergency passage occupancy monitoring: Use target tracking technology to ensure that all emergency exits remain unobstructed, and immediately trigger an alarm if an obstacle is found blocking the escape route. Specific areas (such as fire truck passages) are identified by setting virtual boundaries or using semantic segmentation models, and continuous monitoring is performed to see if there are vehicles or other objects illegally parked in the area.

[0152] If there is an object in the channel, the size of the object is larger than the preset size, and the object stays for more than a certain time, the alarm target will be reported.

[0153] 4. Item monitoring and tracking: Combine historical data to establish a standard placement model for important materials in the warehouse, and regularly compare the differences between the current layout and the standard model. When a change in the location of a key asset is detected, the event is automatically recorded and reported to management personnel so that timely measures can be taken to prevent loss or misplacement.

[0154] 5. Abnormal environmental parameter alarm: Integrate the data of temperature sensors and smoke detectors into a unified platform, set reasonable thresholds to distinguish between normal ranges and potential dangerous conditions. If the readings at one or more points exceed the preset safety limit, the emergency plan will be immediately activated, including but not limited to sound and light warnings and activation of automatic sprinkler fire extinguishing devices.

[0155] 6. Multimodal fusion decision-making: In the case of a single sensor that may have a false alarm, the judgment accuracy can be improved through cross-modal information fusion (for example, combining video analysis results with temperature / smoke data). Based on the comprehensive evaluation results, it is decided whether further manual intervention is needed or whether to directly execute an automated response action.

[0156] 7. Emergency communication and command and dispatch: Once a fire emergency is confirmed, a detailed alarm notification will be quickly sent to designated personnel, including but not limited to the location, time, and initial assessment of the severity of the accident. Support multiple communication methods (SMS, phone, email) to ensure that information can be quickly delivered to relevant departments and individuals. Provide an interactive map interface to display the specific location of the affected area and the layout of surrounding facilities to help rescue teams develop the most effective action plan.

[0157] 8. Data analysis and post-event review: Record the process of each alarm triggering and related evidence (such as video clips, sensor value curves) to provide support for subsequent investigation and analysis. Regularly conduct statistical analysis of accumulated historical data to identify areas or time periods where hidden dangers may exist, thereby guiding the improvement of existing safety management strategies.

[0158] The embodiments of the present invention are described in detail above. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A warehouse security and fire warning method based on edge AI mode, characterized in that: The early warning method uses edge AI artificial intelligence video data processing and analysis methods to perform intelligent security behavior recognition and early warning and intelligent fire behavior recognition and early warning; The edge AI artificial intelligence video data processing and analysis method includes: the data center pushes the trained model to the edge AI server, the edge AI server analyzes the streaming data collected by the edge IoT device in real time and performs AI reasoning; the edge AI server uploads the received new situation related data to the cloud to train the model, and the trained model is pushed to the edge AI server again; The intelligent security behavior recognition and warning includes: area intrusion monitoring; sleeping or leaving the post monitoring; electric bicycle prohibition from entering the elevator monitoring; headcount and over-limit monitoring; personal protective equipment monitoring; dangerous behavior identification; trip wire counting and fall monitoring, which is used to set a virtual boundary line to count the number of people entering and leaving, and immediately notify relevant personnel when someone falls; mask wearing monitoring; The intelligent fire behavior recognition and early warning include: flame monitoring and smoke monitoring; emergency passage occupancy monitoring; obstacle monitoring; object monitoring and tracking; and environmental parameter abnormality monitoring.

2. The early warning method according to claim 1, characterized in that: The edge AI artificial intelligence video data processing and analysis method includes: Hardware configuration: The hardware configuration includes a high-definition digital surveillance camera; the surveillance camera has the function of pulling network video streams; Edge computing server: The edge computing server includes a high-performance small server or edge gateway deployed in the warehouse to perform algorithm analysis tasks; Sensor network: The sensor network includes environmental monitoring devices, including temperature sensors and smoke detectors, which are connected to edge computing nodes to provide additional data input; Warehouse inspection robot: The robot obtains real-time inspection data through inspection. The inspection data includes: fire, loss of fire extinguisher, blocked fire passage, illegal personnel or temperature and humidity. The real-time inspection data is connected to the edge computing node to provide dynamic additional data input.

3. The early warning method according to claim 2, characterized in that: The edge AI artificial intelligence video data processing and analysis method includes: Software architecture, the software architecture includes: lightweight operating system, the lightweight operating system uses an optimized embedded operating system; containerization technology; message queue; edge AI framework, the edge AI framework uses an artificial intelligence framework suitable for edge computing scenarios; Data preprocessing and feature extraction, the data preprocessing and feature extraction include: after the video stream enters the edge computing node, first remove noise and adjust the format through a preprocessing step; apply computer vision technology to perform preliminary analysis on the image, the preliminary analysis includes background subtraction or motion detection; extract key feature vectors from the video frame as the basis for the next step of recognition.

4. The early warning method according to claim 3, characterized in that: The edge AI artificial intelligence video data processing and analysis method includes: Intelligent security algorithm, the intelligent security algorithm includes: identifying regional intrusion, sleeping on duty or leaving the post through target tracking and behavior analysis algorithms; prohibiting electric bicycles from entering elevators and counting the number of people in the area through object classification and counting technology; monitoring personal protective equipment through image segmentation and semantic understanding, the personal protective equipment includes helmet monitoring and reflective clothing recognition; judging dangerous behaviors of personnel through posture estimation models, the dangerous behaviors include smoking and making phone calls; An intelligent fire fighting algorithm includes: realizing open flame and smoke monitoring through a flame recognition model and a method based on color space conversion; evaluating the occupancy of fire passages through a static obstacle monitoring algorithm; and tracking lost items through comparative analysis of historical trajectory information and current scenes.

5. The early warning method according to claim 4, characterized in that: The edge AI artificial intelligence video data processing and analysis method includes: Abnormal alarm mechanism, which includes: when the edge computing node detects any violation of safety rules, it immediately triggers the alarm process; records the timestamps, location coordinates, and related evidence pictures or video clips of all violations; The cloud-edge collaborative working mode includes: for difficult tasks that are highly complex or require data accumulation to complete, uploading the difficult tasks to the cloud for processing; Security measures include: ensuring that all transmitted data is encrypted; regularly updating software versions and patches on edge devices; and implementing strict access control policies.

6. The early warning method according to claim 1, characterized in that: The intelligent security behavior recognition and warning includes: Image acquisition and preprocessing, the image acquisition and preprocessing including: using a high-definition digital camera to acquire real-time video data, and transmitting the real-time video data to an edge AI server; performing preprocessing operations on the original video frames of the real-time video data, the preprocessing operations including denoising and contrast enhancement; Target monitoring and tracking, which includes: applying deep learning target monitoring algorithms to identify objects and people in the video; using multi-target tracking technology to continuously track the movement trajectory of identified objects between different frames.

7. The early warning method according to claim 6, characterized in that: The intelligent security behavior recognition and warning includes: The regional intrusion monitoring includes: setting a virtual boundary line for a specific area, and triggering an alarm when an unauthorized person enters the area; The sleeping or leaving the post monitoring includes: monitoring the posture changes and activity frequency of human bodies in the designated area to determine whether there are workers who are stationary or leaving their posts for a long time; The electric bicycle prohibition elevator monitoring includes: analyzing the type of vehicles entering and exiting the elevator entrance, and issuing an alarm if an electric bicycle attempts to enter; The headcount counting and over-limit monitoring includes: monitoring and counting the current number of people in the selected area, and comparing it with the preset maximum capacity threshold, and triggering an alarm when the maximum capacity threshold is exceeded; The tripwire counting and fall monitoring includes: setting a virtual tripwire in the detection channel to monitor whether a person falls.

8. The early warning method according to claim 7, characterized in that: The intelligent security behavior recognition and warning includes: Scene understanding and environmental perception, which includes parsing the surveillance images in combination with semantic segmentation technology to distinguish different background elements and foreground objects; using 3D reconstruction technology to construct the spatial layout inside the warehouse and locate the specific location where the event occurred; Algorithm optimization, including: adopting a lightweight neural network architecture; using model compression technology to reduce computational complexity; using a multi-task learning framework to enable a single model to handle multiple types of monitoring tasks simultaneously; The alarm mechanism and response strategy include: automatically generating alarm information when any intelligent security rule is violated, and the alarm information includes timestamp, location description, on-site screenshot or video clip; providing a visual interface to display the real-time status and historical records of all cameras in the warehouse.

9. The early warning method according to claim 1, characterized in that: The intelligent fire behavior recognition and warning includes: Image and environment data acquisition, the image and environment data acquisition includes video monitoring and sensor integration; the video monitoring includes: a high-definition digital camera continuously captures real-time images in the warehouse to obtain video data, and transmits the video data to the edge AI server for processing; the sensor integration includes: the sensor includes a physical quantity sensing device, and the physical quantity sensing device includes a visual sensor, a temperature sensor or a smoke detector; The flame monitoring and smoke monitoring include: using deep learning algorithms to analyze video frames and identify open flame or smoke features; the flame monitoring includes: using color space conversion combined with morphological operations to enhance the expressiveness of the flame area, and using dynamic threshold segmentation technology to separate the flame part from the background; the smoke monitoring includes: extracting texture features on grayscale images, and judging whether there is abnormal diffusion by comparing the changes between consecutive frames; The emergency passage occupancy monitoring includes: using target tracking technology to monitor all safety exits, and triggering an alarm if an obstacle is found to be blocking the escape route; continuously monitoring whether there are vehicles or other objects parked in a specific area by setting a virtual boundary line or using a semantic segmentation model; when there is an object target in the emergency passage, if the size of the object target is larger than a preset size and the time it stays in the passage exceeds a specified time, an alarm is activated; The item monitoring and tracking includes: establishing standard placement information of important materials in the warehouse in combination with historical data, comparing the difference between the current layout and the standard placement information; when the location of important materials is detected to have changed, automatically recording the event and initiating an alarm; The abnormal environmental parameter monitoring includes: integrating the data of the temperature sensor and the smoke detector, and setting a threshold value to distinguish between a normal range and a potential dangerous condition.

10. The early warning method according to claim 9, characterized in that: The intelligent fire behavior recognition and early warning includes: Multimodal fusion decision-making, the multimodal fusion decision-making includes: improving judgment accuracy by fusing cross-modal information when a single sensor may have a false alarm; Emergency communication and command and dispatch, including: sending an alarm notification to designated personnel when an emergency is confirmed, including the location, time and initial assessment of the severity of the accident; providing an interactive map interface to display the specific location of the affected area and the layout of surrounding facilities; Data analysis and post-event review, the data analysis and post-event review include: recording the relevant alarm data information of each alarm triggering, the alarm data information includes: video clips or sensor value curve graphs.

Images