Multi-view multi-dimensional area intelligent monitoring method

Abstract

The invention relates to the field of area monitoring, in particular to a multi-view multi-dimensional area intelligent monitoring method, which comprises the following steps of: deploying various types of sensors in an area; acquiring RGB images and infrared images at different angles and angles; collecting the concentrations of the gas sensors at different positions; system alarm and decision support; in the using process, it is found that a high-temperature source also relates to vehicles in an area, and in order to shield the high-temperature source, a YOLO model is adopted for vehicle detection; a multi-dimensional monitoring method is used for detection, and alarm strategies of different degrees are configured at the same time. According to the invention, multiple types of sensors are deployed, and regional intelligent monitoring is carried out based on a multi-dimensional fusion detection network, so that the robustness of the problem of environment variability is improved.

Description

Technical Field

[0001] The present invention relates to the field of regional monitoring technology, and in particular to a multi-view and multi-dimensional regional intelligent monitoring method. Background Art

[0002] With the development of society and the acceleration of urbanization, people's demand for security is increasing. As an important means to ensure public safety, regional monitoring has received more and more attention and attention. Due to the wide variety of usage scenarios, traditional standardized video surveillance products have been difficult to meet market demand. Users are increasingly inclined to choose products that can be tailored to meet their actual needs. The application scenarios of the method provided in the present invention include monitoring methods of different dimensions, such as: heat sources, combustible gases, etc.

[0003] Most of the current monitoring methods in the above scenarios usually use fixed visible light cameras, which have the following main problems: single shooting angle, blind spots, and no other monitoring methods. With the rapid development of intelligent manufacturing, the demand for automation, real-time, intelligence, and efficiency is getting higher and higher, and the current monitoring methods are increasingly unable to meet the development requirements of modern industrial production. Summary of the invention

[0004] The present invention provides a multi-view and multi-dimensional regional intelligent monitoring method and system, deploys multiple types of sensors, and performs regional intelligent monitoring based on a multi-dimensional fusion detection network, thereby improving the robustness of environmental variability problems.

[0005] In order to solve the above technical problems, the present invention provides a multi-view and multi-dimensional regional intelligent monitoring method, comprising the following steps:

[0006] S1. Deploy multiple types of sensors in the area;

[0007] S2, collect RGB images and infrared images at different angles and perspectives;

[0008] S3, collecting gas sensor concentrations at different locations;

[0009] S4, system alarm and decision support;

[0010] S5. During use, it is found that the high temperature source also involves vehicles in the area. In order to shield this high temperature source, the present invention uses the YOLO model to detect vehicles;

[0011] S6. A multi-dimensional monitoring method is used for detection, and alarm strategies of different degrees are configured.

[0012] Furthermore, the various types of sensors in S1 include visual sensors, combustible gas detection alarms, on-site wiring boxes, main control cabinets, display equipment and alarm equipment. The visual sensors are deployed in factory buildings, packaging rooms, production workshops and other areas.

[0013] Furthermore, S2 includes data collection and preparation, image data annotation, image preprocessing, data enhancement, visible light image fireworks target detection model and infrared temperature detection.

[0014] Furthermore, the data collection and preparation includes obtaining image and video data containing smoke, flames, and non-pyrotechnic interferences from public data sets to enhance the generalization ability of the algorithm;

[0015] The image data annotation includes annotating fireworks images through the constructed image annotation platform to generate XML and JSON files containing smoke and flame position and category information;

[0016] The image preprocessing includes preprocessing the image data, image scaling, cropping, normalization, and frame extraction and key frame selection of the video data;

[0017] The data enhancement includes methods such as random rotation, flipping, scaling, translation, filtering, changing color space components, random pixel perturbation, singular value decomposition, etc.

[0018] Furthermore, the visible light image fireworks target detection model adopts a ResNet architecture and includes multiple convolution groups, each convolution group includes multiple residual blocks, and outputs classification results through a global average pooling layer and a softmax activation function.

[0019] Furthermore, the infrared temperature detection includes acquiring infrared temperature images at different positions and angles, and obtaining a comprehensive regional temperature distribution information map after splicing the images.

[0020] Furthermore, in S3, the concentration values ​​of gas sensors at different positions are collected to form a regional concentration field, and detection processing is performed by two methods: whether the concentration reaches a threshold value and the concentration trend;

[0021] The sensor generates corresponding electrical signals by sensing the change of gas concentration, and the processing unit collects these signals; the collected signals contain noise or interference, and are filtered and amplified; the concentration of the gas to be detected is calculated based on the output signal of the sensor and the calibration data, and the calculated concentration value is displayed on the display screen. At the same time, the system will obtain this information and trigger the alarm mechanism according to the concentration threshold set in advance;

[0022] When there is a rapid change in concentration within a certain period of time, the system will identify this state and issue an alarm notification;

[0023] Where (C_1) and (C_2) are the concentration values ​​measured at time (t_1) and (t_2).

[0024] Furthermore, the system alarm and decision support includes monitoring alarm, real-time data and custom control.

[0025] Furthermore, in S5, the YOLO model is used to detect vehicles, and the specific real-time method includes: data collection, data preprocessing, data labeling, model deployment and application.

[0026] Furthermore, S6 includes smoke and infrared fusion analysis, combustible gas detection and visual sensor coordinated detection and alarm strategies.

[0027] By means of the above technical solution, the present invention provides a multi-view and multi-dimensional regional intelligent monitoring method, which has at least the following beneficial effects:

[0028] By acquiring image information from different collection locations, feature extraction is performed on the image information respectively to obtain multiple groups of image features; by acquiring the combustible gas concentration information at different locations and combining it with the regional distribution map, a concentration distribution map is obtained; at the same time, based on the detection results of the above two, the system configures different alarm strategies and decision support according to the detection results. In the present invention, multiple types of sensors are deployed, and regional intelligent monitoring is performed based on a multi-dimensional fusion detection network, which improves the robustness of environmental variability problems. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0030] Figure 1 It is a schematic diagram of a multi-view and multi-dimensional regional intelligent monitoring method;

[0031] Figure 2 Schematic diagram of sensor deployment for multi-view and multi-dimensional regional intelligent monitoring method. DETAILED DESCRIPTION

[0032] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, 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.

[0033] like Figure 1 and Figure 2 As shown, the multi-view and multi-dimensional regional intelligent monitoring method provided by the present invention includes the following steps:

[0034] S1. Deploy multiple types of sensors in the area;

[0035] As market competition intensifies, regional monitoring equipment manufacturers will pay more attention to product customization and differentiated services. According to the actual needs and usage scenarios of users, customized monitoring solutions and personalized service experience are provided. The deployment of various sensors in the present invention is as follows: Figure 2 shown.

[0036] Step 1.1: The length, width and height of the factory building are 72m*27m*10.5m. Shelves: 5.475m high, 1.1m deep, 9m long, 2 rows of shelves per span. The shelf spacing is about 5.5m. The area of ​​each span is 8.5m*11m, 9 spans in total, 6m*10m for the packaging workshop and 21m*27m for the production workshop. In addition to arranging a set of visual sensors in each span, a set of visual sensors is arranged in the packaging room and the production workshop, a total of 11 sets of visual sensors are arranged to patrol the industrial hazardous waste warehouse 24 hours a day.

[0037] Step 1.2: The deployment method of combustible gas detection alarm is: one is arranged crosswise at 0.6 meters above the ground on the column surface in the aisle and 5 meters from the center of the beam, and one is arranged in each span at 0.6 meters above the ground on the steel structure on both sides of the outer wall. There is no wall shielding in the production workshop, the gas is easy to diffuse and the accumulation is mainly piled on both sides of the production equipment area. 4 combustible gas detection alarm devices are installed, totaling 25.

[0038] Step 1.3: Arrange three on-site junction boxes at a height of about 1.5m from the inner steel structure.

[0039] Step 1.4: A main control cabinet and a 4K large screen display are arranged in the central control room on the first floor of the office.

[0040] Step 1.5: A terminal display is arranged in the rotary hearth furnace control room.

[0041] Step 1.6: Arrange three sound and light alarm indicator lights on site, and arrange one voice alarm system on the first floor of the fly ash control room and one voice alarm system in the rotary hearth furnace control room.

[0042] S2, collect RGB images and infrared images at different angles and perspectives;

[0043] Step 2.1: Data collection and preparation;

[0044] Obtaining image and video data containing smoke, flames, and non-pyrotechnic interference (obtained through practical field) and from public data sets (such as the Fire Smoke Database, Kaggle related competition data sets, etc.) helps to enhance the generalization ability of the algorithm because the data comes from different scenarios and conditions.

[0045] Step 2.2: Image data annotation;

[0046] The fireworks images are annotated through the built image annotation platform to generate XML and JSON files containing smoke, flame location (bounding box) and category information.

[0047] Step 2.3: Image preprocessing;

[0048] Perform preprocessing of image data, image scaling, cropping, normalization, frame extraction and key frame selection of video data, etc.

[0049] Normalization:

[0050] Among them, x and y are the values ​​before and after the conversion, MaxValue is the maximum value of the sample, and MinValue is the minimum value of the sample.

[0051] The image contrast is improved by histogram equalization, gamma correction and other methods to make the characteristics of smoke and flame more obvious.

[0052] Step 2.4: Data enhancement, in order to further increase the diversity and quantity of the dataset, data enhancement techniques are used;

[0053] (1) Randomly rotate the image by a certain angle; (2) Flip the image horizontally or vertically; (3) Enlarge or reduce the image according to a certain ratio; (4) Translate the image in a certain way on the image plane to change the position of the image content; (5) Filter the image using a specified scale factor to change the size or blur of the image content; (6) Change the saturation (S) and brightness component (V) in the HSV color space of the image, keep the hue (H) unchanged, and increase the illumination change; (7) Randomly perturb each pixel of the image; (8) Perform singular value decomposition on the image, and then restore the image using the eigenvectors corresponding to a certain number of larger eigenvalues; (9) Generate a new training image by randomly selecting the processed training images and performing weighted averaging.

[0054] Use affine transformation to scale the image:

[0055] Among them, (x, y) is the point on the original graph, and (x', y') is the corresponding point on the scaled graph. a, b, c, d, e, f are the coefficients of affine transformation. a and d are used to control whether to zoom in or out, and different values ​​are used to achieve the purpose of data enhancement.

[0056] Step 2.5: Visible light image fireworks target detection model;

[0057] The smoke and fire recognition algorithm architecture used in the present invention is mainly based on the adjustment of the CNN architecture of ResNet. ResNet mainly introduces a residual network, which can build a deeper neural network and improve the performance of the model to adapt to the specific task of smoke and fire recognition. The model includes multiple convolutional layers, pooling layers, activation layers, batch normalization layers and fully connected layers. The loss function adopts the cross entropy loss function and is optimized for multi-classification problems (smoke, fire, non-fireworks).

[0058] The input layer resizes the original image to 224x224x3 (height x width x number of color channels). Before passing the image to the network, the above preprocessing, normalization and other processes will be performed.

[0059] In the network, the first convolutional layer uses a larger convolution kernel (7x7) and a larger stride (3) to reduce the spatial dimension of the feature map. This is followed by batch normalization and ReLU activation function. This layer also includes a maximum pooling layer to further reduce the spatial dimension of the feature map.

[0060] In this embodiment, the ResNet used contains four convolution groups (Stage 2, Stage 3, Stage 4, and Stage 5), each of which contains multiple residual blocks. And at the end of each convolution group, a downsampling operation is performed to reduce the spatial dimension of the feature map. Downsampling is implemented with a convolution layer with a stride of 2, and a maximum pooling layer is used in a specific convolution group. As the convolution group goes deeper, the number of channels of the feature map gradually increases to capture richer feature information.

[0061] Stage 2: Contains 3 Bottleneck residual blocks, each of which contains three convolutional layers (1x1, 3x3, 1x1). The first 1x1 convolutional layer is mainly used to reduce the number of channels of the input feature map (i.e., dimensionality reduction) to reduce computational complexity; the second 3x3 convolutional layer is the main feature extraction layer, used to learn the feature representation of the input data; the third 1x1 convolutional layer is used to restore the number of channels of the feature map to match the number of input channels of the residual block for subsequent addition operations. After the convolutional layer is completed, there is a global average pooling layer for dimensionality reduction and feature extraction.

[0062] Stage 3: To further extract high-level features of the image, four Bottleneck residual blocks are used in this convolutional unit, each of which contains three convolutional layers, which is consistent with Stage 2. After the convolutional layer is completed, there is a global average pooling layer for dimensionality reduction and feature extraction.

[0063] Stage 4: To further extract high-level features of the image, six Bottleneck residual blocks are used in this convolutional unit, each of which contains three convolutional layers, which is consistent with Stage 2. After the convolutional layer is completed, there is a global average pooling layer for dimensionality reduction and feature extraction.

[0064] Stage 5: As the last convolution group in NewResNet, Stage 5 uses 3 Bottleneck residual blocks. These residual blocks further extract and abstract features in the final stage of the network in preparation for classification or other tasks. The global average pooling layer operates on the feature map output by the last convolution group. The feature map of each channel is globally averaged to obtain a feature vector with a length of the number of channels. This feature vector is used as the input of the softmax activation function, and the softmax activation function is used to output the probability of each category (a total of 5 categories in this embodiment). The training process is to divide the labeled training set into multiple batches and use an optimized learning algorithm. The stochastic gradient descent optimization algorithm is used for parameter update, and the learning rate can be adjusted according to the training progress.

[0065] Finally, the trained model is deployed to the server in this embodiment, and the real-time video stream is subjected to model detection processing. At the same time, the recognition result is subjected to non-maximum suppression processing to remove overlapping boundary boxes and improve detection accuracy.

[0066] Step 2.6: Infrared temperature detection;

[0067] Infrared temperature images at different positions and angles are obtained, and after the images are spliced, a comprehensive regional temperature distribution information map is obtained. In this embodiment, temperature detection is mainly divided into two functions.

[0068] One is to conduct real-time detection and alarm through temperature detection combined with threshold adaptive setting. However, due to factors such as the on-site environment and weather, the threshold cannot use a unified value. Therefore, by obtaining the temperature field distribution law at different times and locations, and then combining the big data regression model to establish the relationship between the temperature field and the threshold, the adaptive threshold is determined. At the same time, this threshold will change according to the regression model, so as to alarm suspicious temperatures. All alarm data will be saved and counted to facilitate subsequent investigation and tracking.

[0069] Among them, is the temperature field, which is a two-dimensional continuous function that describes the change of temperature with position in the entire two-dimensional plane. The binary function f is a function that accepts two inputs (x and y in this case) and returns an output (temperature T in this case). It is a function fitted according to the actual data on site, which defines the mapping relationship between temperature and coordinates.

[0070] The second is to issue an alarm based on the temperature trend. Although the temperature has not reached the adaptive threshold, the temperature is rising rapidly within a certain period of time. This state is considered an abnormal state. Therefore, in this embodiment, an alarm needs to be issued in such a situation.

[0071] Wherein, ΔT represents the change in temperature, Δt represents the change in time (1 minute is used in this embodiment), and k is the slope of the linear regression, which represents the rate of temperature increase per unit time.

[0072] S3, collecting gas sensor concentrations at different locations;

[0073] The concentration values ​​of gas sensors at different positions are collected to form a regional concentration field. In this embodiment, detection and processing are mainly performed in two ways: whether the concentration reaches a threshold value and the concentration trend.

[0074] First, the sensor generates corresponding electrical signals by sensing changes in gas concentration, and the processing unit collects these signals; the collected signals contain noise or interference, which are filtered and amplified; the concentration of the gas to be detected is calculated based on the output signal of the sensor and the calibration data. The calculated concentration value is displayed on the display screen and the system will obtain this information and trigger the alarm mechanism based on the concentration threshold set in advance.

[0075] The second is the concentration trend. When there is a rapid change in concentration within a certain period of time, the system will identify this state and issue an alarm.

[0076] Where (C_1) and (C_2) are the concentration values ​​measured at time (t_1) and (t_2).

[0077] S4, system alarm and decision support;

[0078] 4.1. Monitoring alarm: In the independent thread ThreadPoolExecutor, Modbus, Socket and other technologies are continuously used to check various types of sensors to obtain their status, detection values ​​and other information; then the system automatically makes decisions and makes corresponding alarm processing according to certain rules; and uses Asyncio, Websocket and other technical modules to achieve ideal alarm timeliness and system resource usage.

[0079] Among them, in this embodiment, ThreadPoolExecutor is a thread pool of 4 threads to handle related transactions; at the same time, all data transmission and communication are carried out in Asyncio (asynchronous) mode, ensuring the smooth operation of the main process.

[0080] 4.2. Real-time data: The system’s built-in data repository will automatically record historical monitoring data in order to perform some data analysis, model optimization and other operations, and the real-time monitoring curve can be seen in the visual interface to control the system’s monitoring status.

[0081] 4.3. Customized control: The system is highly configurable. Monitoring switches, time, methods, etc. can all be configured on the visualization page, and Asyncio technology can be used to make the configuration take effect in real time.

[0082] S5. During the use of the embodiment, it is found that the high temperature source also involves vehicles in the area. In order to shield the high temperature source, the present invention uses the YOLO model to detect vehicles. The specific real-time method is as follows:

[0083] 5.1. Data Collection: Collect datasets containing vehicle images and videos. Most of these datasets are collected at different angles and lighting conditions on site, which increases the generalization ability of the model.

[0084] 5.2. Data preprocessing: Preprocess the collected images, including scaling, cropping, grayscale, etc., to meet the input requirements of the model. Perform image enhancement, such as translation, rotation, scaling, etc., to increase the diversity of the data set and improve the robustness of the model.

[0085] 5.3. Data Annotation: Annotate the vehicles in the image, and mark the location (bounding box) and category of the vehicle. The annotation information includes the coordinates of the bounding box (upper left corner and lower right corner coordinates) and the category label.

[0086] 5.4. Model training: Select the YOLOv5 model version suitable for the current vehicle detection for model training, optimize the model weights and biases through multiple iterations, use different loss functions to evaluate the model performance, and adjust the model hyperparameters (such as learning rate, batch size, number of iterations, etc.).

[0087] 5.5. Model deployment and application: When the system detects a high temperature source (infrared device), it determines whether there is a vehicle in the image (visible light device) and the intersection and union ratio of the high temperature point and the vehicle, and uses this as the basis for subsequent processing.

[0088] Among them, result is the output of the model, (x, y) is the coordinate of the center point of the bounding box, the offset value relative to the grid boundary, normalized to between 0 and 1; (w, h) is the width and height of the bounding box, the ratio relative to the width and height of the entire image, also normalized to between 0 and 1; confidence is the IOU (intersection-over-union) of the predicted box and the true box, indicating the accuracy of the predicted box.

[0089] S6. In actual application, the present invention uses a multi-dimensional monitoring method for detection, and configures alarm strategies of different degrees;

[0090] 6.1: Fireworks and infrared fusion analysis;

[0091] The method of pairing and analyzing the results of fireworks image detection with the infrared temperature monitoring results is adopted to effectively improve the judgment accuracy and sensitivity of the system. First of all, in terms of data acquisition, an infrared and visible light integrated machine is used to realize the synchronous acquisition of data; then, the fireworks image is pre-processed by denoising, contrast enhancement and other operations to improve the image quality, facilitate the subsequent feature extraction, and the infrared thermal image is corrected and normalized to eliminate the influence of equipment differences and environmental factors on temperature measurement; finally, the extracted fireworks image features are matched with the infrared temperature features to find the corresponding relationship between the two, and the fireworks image and infrared temperature information are integrated through the fusion algorithm (in this embodiment, the IOU (intersection over union) and threshold are mainly used for algorithm fusion) to form a more comprehensive fire monitoring result.

[0092] 6.2: Combustible gas detection and visual sensor detection;

[0093] In this embodiment, when the hardware is deployed, all combustible gas detectors are placed within the range of the visual sensor, thus providing the possibility of collaboration. For the installation location of the combustible gas detector, the visual sensor pre-sets the cruise point of this location. When a combustible gas detector shows an abnormal concentration or concentration trend, the system will notify the nearby visual sensors to focus their field of view on the abnormal combustible gas detector. At the same time, on this basis, an alarm is issued to provide corresponding decision support for relevant personnel on site.

[0094] 6.3 Alarm strategy;

[0095] In this embodiment, there are three types of alarms as follows: At the same time, different alarm strategies are provided according to different detection results.

[0096] Alarm form 1: Visual interface display: Display alarm information and content (such as temperature, fireworks, combustible gas concentration, etc.).

[0097] Alarm form 2: Sound and light alarm: The sound and light alarm deployed on site is triggered by the light-on / light-off signal of the remote I / O module, and is equipped with a variety of ways to turn off the alarm.

[0098] Alarm form three: SMS reminder: The alarm information and content will be sent to the responsible person via SMS through the pre-configured mobile phone number (on-site person in charge, on-duty personnel, etc.), and a solution that the system considers reasonable will be provided, such as: having relevant personnel go to the site for reconfirmation, triggering fire linkage, etc.

[0099] Alarm form four: Voice reminder: Through the pre-configured mobile phone number (on-site person in charge, on-duty personnel, etc.), the alarm information and content will be notified to the responsible person by voice, and a solution that the system considers reasonable will be provided, such as: having relevant personnel go to the site for reconfirmation, triggering fire linkage, etc.

[0100] (1) Alarm strategy 1: The highest level alarm. When fireworks and infrared fusion analysis are abnormal, this level of alarm will be triggered and all actions of the above alarm forms 1, 2, 3 and 4 will be executed.

[0101] (2) Alarm strategy 2: When an abnormality is detected in the analysis of the combustible gas detector and there is no heat source, fire point or smoke in the area, an intermediate alarm will be triggered and all actions of the above alarm forms 1, 3 and 4 will be executed.

[0102] (3) Alarm strategy three: When some system alarms occur (such as a visual sensor being offline, a combustible gas detector being offline, etc.), an intermediate alarm will be triggered and the actions of the above alarm forms one and three will be executed.

[0103] The beneficial effects of the present invention are:

[0104] 1. Early warning and rapid response: The fire warning system can monitor the smoke / fire, heat, gas and other fire precursors in the environment in real time. The system can capture the initial signs of fire and ensure timely alarm at the early stage of fire, greatly shortening the time interval from the occurrence of fire to its discovery.

[0105] 2. All-weather monitoring and unmanned operation: The system can achieve 24-hour uninterrupted operation, and can stably and reliably perform monitoring tasks day and night, reducing the pressure of manpower monitoring, especially at night and during unmanned periods, and can still ensure effective fire protection.

[0106] 3. Multi-channel alarm and emergency notification: When the intelligent early warning system detects an abnormality, it can not only use on-site sound and light alarms, but also quickly transmit the alarm information to relevant responsible persons and management departments through mobile phone text messages / voice.

[0107] 4. Accident records and post-event tracing: Even if an accident does occur, the system can record detailed information about the fire, including time, location, snapshots before and after the alarm, video recordings, etc., providing a strong basis for post-event investigation and analysis, responsibility determination and improvement of preventive measures.

[0108] 5. Save lives and reduce losses: Most importantly, the fire warning system can provide decision-making support at the initial stage of a fire, notify people to evacuate the scene according to the situation, reduce the risk of casualties, and reduce property losses through early intervention, which is of great significance to public safety and social stability.

[0109] It should be noted that the terms "comprises", "includes" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or apparatus that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or apparatus.

[0110] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-view and multi-dimensional regional intelligent monitoring method, characterized in that: The following steps are involved: S1. Deploy multiple types of sensors in the area; S2, collect RGB images and infrared images at different angles and perspectives; S3, collecting gas sensor concentrations at different locations; S4, system alarm and decision support; S5. During use, it is found that the high temperature source also involves vehicles in the area. In order to shield this high temperature source, the YOLO model is used to detect vehicles; S6. A multi-dimensional monitoring method is used for detection, and alarm strategies of different degrees are configured.

2. The multi-view and multi-dimensional regional intelligent monitoring method according to claim 1 is characterized in that: The various types of sensors in S1 include visual sensors, combustible gas detection alarms, on-site wiring boxes, main control cabinets, display equipment and alarm equipment. The visual sensors are deployed in factory buildings, packaging rooms and production workshops.

3. The multi-view and multi-dimensional regional intelligent monitoring method according to claim 1 is characterized in that: S2 includes data collection and preparation, image data annotation, image preprocessing, data enhancement, visible light image fireworks target detection model and infrared temperature detection.

4. The multi-view and multi-dimensional regional intelligent monitoring method according to claim 3 is characterized in that: The data collection and preparation includes obtaining image and video data containing smoke, flames, and non-pyrotechnic interferences from public data sets to enhance the generalization ability of the algorithm; The image data annotation includes annotating fireworks images through the constructed image annotation platform to generate XML and JSON files containing smoke and flame position and category information; The image preprocessing includes preprocessing the image data, image scaling, cropping, normalization, and frame extraction and key frame selection of the video data; The data enhancement includes random rotation, flipping, scaling, translation, filtering, changing color space components, random pixel perturbation and singular value decomposition.

5. The multi-view and multi-dimensional regional intelligent monitoring method according to claim 4 is characterized in that: The visible light image fireworks target detection model adopts the ResNet architecture and includes multiple convolution groups, each of which includes multiple residual blocks, and outputs classification results through a global average pooling layer and a softmax activation function.

6. The multi-view and multi-dimensional regional intelligent monitoring method according to claim 5 is characterized in that: The infrared temperature detection includes acquiring infrared temperature images at different positions and angles, and splicing the images to obtain a comprehensive regional temperature distribution information map.

7. The multi-view and multi-dimensional regional intelligent monitoring method according to claim 1 is characterized in that: In S3, the concentration values ​​of gas sensors at different locations are collected to form a regional concentration field, and detection and processing are performed by two methods: whether the concentration reaches the threshold value and the concentration trend; The sensor generates corresponding electrical signals by sensing changes in gas concentration, and the processing unit collects these signals; the collected signals contain noise or interference; Perform filtering and amplification processing; calculate the concentration of the gas to be detected based on the output signal of the sensor and the calibration data. The calculated concentration value is displayed on the display screen. At the same time, the system will obtain this information and trigger the alarm mechanism based on the concentration threshold set in advance; When there is a rapid change in concentration within a certain period of time, the system will identify this state and issue an alarm notification; Where (C_1) and (C_2) are the concentration values ​​measured at time (t_1) and (t_2).

8. The multi-view and multi-dimensional regional intelligent monitoring method according to claim 1 is characterized by: The system alarm and decision support includes monitoring alarm, real-time data and custom control.

9. The multi-view and multi-dimensional regional intelligent monitoring method according to claim 1, characterized in that: In S5, the YOLO model is used to detect vehicles. The specific real-time methods include: data collection, data preprocessing, data labeling, model deployment and application.

10. The multi-view and multi-dimensional regional intelligent monitoring method according to claim 1, characterized in that: S6 includes smoke and infrared fusion analysis, combustible gas detection and visual sensor coordinated detection and alarm strategies.

Images