A method and system for monitoring abnormal objects in a substation based on image AI technology

Abstract

The application provides a substation abnormal object monitoring method and system based on image AI technology, comprising: acquiring a monitoring image; pre-processing the monitoring image to obtain a pre-processed image; determining whether the pre-processed image changes based on the pre-processed image and a reference image, and if yes, acquiring a sub-monitoring image of a changed area; inputting the sub-monitoring image into a detection model, and outputting an alarm type from the model; using segmentation and detection technology of images to monitor the station in addition to patrol, improving the utilization rate of high-definition video equipment, and alarming when an abnormality is monitored, so that operation and maintenance personnel can discover hidden dangers in time and deal with them, thereby ensuring the safety of the equipment and lines in the station.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to a substation abnormal object monitoring method and system based on image AI technology. BACKGROUND

[0002] With the wide application of high-definition video in power inspection, the target detection technology based on deep learning and machine vision of the station inspection system can identify and analyze the defects in the substation. In the case of 2 inspections per day, there is still a large amount of idle time, and the target detection technology based on video highly occupies computing resources during identification, which cannot achieve multi-scene simultaneous and continuous identification.

[0003] Therefore, the present application provides a substation abnormal object monitoring method and system based on image AI technology to improve the utilization rate of high-definition video equipment in the substation and the detection efficiency and accuracy of abnormal conditions. SUMMARY

[0004] The present application aims to provide a substation abnormal object monitoring method based on image AI technology, which comprises obtaining a monitoring image; preprocessing the monitoring image to obtain a preprocessed image; determining whether the preprocessed image has changed based on the preprocessed image and a reference image, and if so, obtaining a sub-monitoring image of the changed area; inputting the sub-monitoring image into a detection model, and outputting an alarm type from the model.

[0005] Further, the sub-monitoring image of the changed area is obtained by inputting the preprocessed image and the reference image into a segmentation model, and the model outputs a segmentation image; the segmentation image is an image of the area with differences in the preprocessed image and the reference image; based on the segmentation image, a camera device is controlled to obtain a sub-monitoring image including the segmentation image.

[0006] Further, the segmentation model is obtained by training, which comprises obtaining a first training sample group, the first training sample group comprising a sample reference image and a sample monitoring image; obtaining a first label of the first training sample group, the first label being an image of the area with differences in the sample reference image and the sample monitoring image; inputting the first training sample group into an initial segmentation model, and iteratively updating the parameters of the initial segmentation model based on the output of the model and the first label to obtain a trained segmentation model.

[0007] Further, the initial segmentation model is a change detection network of a twin neural network and a FCN.

[0008] Further, it further comprises issuing a warning based on the alarm type, and outputting the merged segmentation image.

[0009] Further, the preprocessing includes grayscale processing, noise judgment and removal, and / or binary processing.

[0010] Further, the detection model is obtained by training, including: obtaining a second training sample; the second training sample includes an image of an abnormal situation; obtaining a second label of the second training sample, the second label being an alarm type corresponding to the image of the abnormal situation; inputting the second training sample into an initial detection model, and iteratively updating parameters of the initial detection model based on an output of the model and the second label to obtain a trained detection model.

[0011] Further, the initial detection model is YOLOV4.

[0012] Further, the abnormal situation in the transformer substation is analyzed during a non-patrol task of the monitoring device.

[0013] The purpose of the present application is to provide a transformer abnormal object monitoring system based on image AI technology, including an acquisition module, a preprocessing module, a sub-monitoring image acquisition module and a determination module; the acquisition module is used to acquire a monitoring image; the preprocessing module is used to preprocess the monitoring image to obtain a preprocessed image; the sub-monitoring image acquisition module is used to determine whether the preprocessed image changes based on the preprocessed image and a reference image, and if so, a sub-monitoring image of a changed area is acquired; and the determination module is used to input the sub-monitoring image into a detection model, and the model outputs an alarm type.

[0014] The technical scheme of the embodiment of the present application has at least the following advantages and beneficial effects:

[0015] Some embodiments in the present specification use image segmentation and detection technology through a silent monitoring method, and monitor large oil-filled equipment such as transformers (reactors) and main entrances and inspection channels in addition to patrol, effectively improving the utilization rate of high-definition video equipment. When an abnormality is monitored, an alarm is given, and maintenance personnel can discover hidden dangers in time and deal with them, ensuring the safety of equipment and lines.

[0016] Some embodiments in the present specification use the segmentation technology of the change detection network of the twin neural network and FCN and the target detection technology based on the YOLO V4 network to monitor the transformer substation, which can improve the utilization rate of video equipment in the transformer substation and improve the accuracy of abnormal situation detection.

[0017] Some embodiments in the present specification first segment a monitoring image to obtain a segmented image of a difference area, then detect the segmented image to obtain an alarm type, which can improve the monitoring efficiency and accuracy of abnormal situations and avoid processing too much irrelevant data. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 An exemplary flowchart of a method for substation abnormal object monitoring based on image AI technology is provided for some embodiments of the present application.

[0019] Figure 2 An exemplary schematic diagram of training a segmentation model is provided for some embodiments of the present application.

[0020] Figure 3 An exemplary schematic diagram of training a detection model is provided for some embodiments of the present application.

[0021] Figure 4 An exemplary module diagram of a system for substation abnormal object monitoring based on image AI technology is provided for some embodiments of the present application. DETAILED DESCRIPTION

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations.

[0023] Figure 1 An exemplary flowchart of a method for substation abnormal object monitoring based on image AI technology is provided for some embodiments of the present application. In some embodiments, the flow 100 can be executed by the system 400. As shown in FIG. 1, the flow 100 includes the following steps: Figure 1

[0024] Step 110, obtaining a monitoring image. In some embodiments, step 110 can be executed by the obtaining module 410.

[0025] The monitoring image can refer to an image obtained by monitoring a target region. The target region can refer to a region that needs to be monitored. For example, the target region can include one or more of a region where a large oil-filled device such as a transformer is located, a main entrance and exit, a patrol channel, etc. In some embodiments, the monitoring image can be obtained by various image acquisition devices. For example, the image acquisition device can monitor the running state of the device in the target region, the substation operating environment, and the personnel behavior at the entrance and exit at a frequency of not more than 2 minutes each time during the non-patrol task execution, and alarm for abnormal situations.

[0026] Step 120, pre-processing the monitoring image to obtain a pre-processed image. In some embodiments, step 120 can be executed by the pre-processing module 420.

[0027] ​In some embodiments, the preprocessing can include, but is not limited to, one or more of grayscale processing, noise judgment and removal, binary processing, and the like.

[0028] The grayscale processing can refer to grayscale processing of a color image, weighted average processing of R, G, and B three colors, and thus obtaining a grayscale image. For example, the coordinates of a picture are (X, Y), and X e (0, H1) and Y e (0, W1). The horizontal coordinate is intercepted as The vertical coordinate is intercepted as and the image matrix is normalized to obtain a grayscale picture matrix Mat2Gray[], the width of which is W2 and the height of which is H2.

[0029] The noise judgment and removal can refer to a method of filtering, using a Canny algorithm to suppress noise and accurately determine the position of the image edge. In some embodiments, a double-threshold method is used to calculate a threshold (a low threshold T L and a high threshold T H ). If the gradient value of the edge pixel point is greater than the high threshold, it is considered to be a strong edge point. If the edge gradient value is less than the high threshold and greater than the low threshold, it is marked as a weak edge point. Points less than the low threshold are suppressed.

[0030] T H = β * T L

[0031] wherein β represents a proportion parameter of the high threshold and the low threshold, and the high and low thresholds are usually in a ratio of two to one or three to one.

[0032] In some embodiments, in order to obtain the most effective threshold, β = 2.25 is obtained by comparison in different color spaces.

[0033] In some embodiments, based on the low threshold T L and the high threshold T h , a strong edge map and a weak edge map are obtained, and a long edge is obtained using connectivity analysis to form a final edge image, so as to achieve the purpose of image smoothing and noise removal.

[0034] The binary processing can refer to allowing only two colors to appear in the image. For example, the grayscale value of each pixel in the pixel matrix in the picture is 0 or 255, wherein 0 is black and 255 is white. The image after processing only presents black and white effects. In some embodiments, the image after binary processing only includes the device to be detected and the background.

[0035] Step 130, based on the preprocessed image and the reference image, determining whether the preprocessed image has changed, if yes, obtaining a sub-monitoring image of the changed area; if no, continuing to monitor.

[0036] The reference image can refer to an image in which the target region is in a normal state. In some embodiments, the reference image can be obtained by a monitoring device capturing the target region in a normal state. The monitoring device can be various camera devices, such as a camera, a video camera, etc.

[0037] In some embodiments, each camera device can be associated with multiple other camera devices. After detecting a changed region, the master camera device can control the slave camera devices to obtain sub-monitoring images containing the changed region. The sub-monitoring image can refer to an image captured by the camera device for the changed region.

[0038] In some embodiments, one master camera device can be associated with multiple slave camera devices. The master camera device and the slave camera devices can obtain monitoring images. When it is determined that the difference between the monitoring images and the reference image is greater than a preset threshold, a region with a difference is determined. The master camera device can mark the position, size, etc. of the region with a difference, and then control a suitable slave camera device to obtain an image of the region. In some embodiments, the suitable slave camera device can refer to the camera device closest to the region with a difference.

[0039] In some embodiments, the pre-processed image and the reference image can be input into a segmentation model, and the model outputs a segmentation image. The segmentation image is an image of a region with a difference in the pre-processed image and the reference image. Based on the segmentation image, the camera device is controlled to obtain a sub-monitoring image including the segmentation image.

[0040] The segmentation image is an image of a region with a difference in the pre-processed image and the reference image. In some embodiments, there can be multiple differences between the pre-processed image and the reference image. The pre-processed image can be segmented along multiple difference parts to obtain multiple segmentation images. The master camera device can control multiple slave camera devices to obtain images of regions where the multiple segmentation images are located. For more information about the segmentation model, see Figure 2 and related descriptions.

[0041] In step 140, the sub-monitoring image is input into a detection model, and the model outputs an alarm type. In some embodiments, step 140 can be performed by the determination module 440.

[0042] The alarm type can refer to the type to which an abnormal situation belongs. The alarm type can include a behavior class (such as not wearing a safety helmet correctly, crossing a line, not wearing a long-sleeved work uniform, etc.), an environment class (such as site fireworks, water accumulation, foreign matter, small animals, etc.), a device class (such as device fireworks, oil leakage, device deformation, device fracture, device tilt, foreign matter intrusion, etc.). For more information about the detection model, see Figure 3 and related descriptions.

[0043] In some embodiments, the plurality of camera devices can jointly analyze and process the sub-monitoring image to obtain the alarm type. For example, when the sub-monitoring image is still in a busy state after being acquired from the camera device (e.g., other monitoring images still need to be acquired), the camera device can send the acquired sub-monitoring image to the master camera device, and the master camera device can send the sub-monitoring image to the slave camera device in an idle state based on the idle / busy state of the slave camera device for analysis and processing to obtain the alarm type.

[0044] In some embodiments, a warning is also issued based on the alarm type, and the segmented images are merged and output. For example, different alarm types can have different warning methods (e.g., pop-up windows, buzzers, etc.), and the staff can be prompted to handle the abnormal situation through different alarm methods. In some embodiments, the positions of the plurality of segmented images in the monitoring image can be identified, and then the images at other positions in the monitoring image are normalized to output a difference image that only retains the difference information. The normalization can mean deleting or taking a single color for the images at other positions.

[0045] Some embodiments in the specification group camera devices so that each group of camera devices can detect sub-monitoring images, improving detection efficiency, and other idle camera devices associated with the camera device that acquires the sub-monitoring image can be used to detect the segmented image during the working period of the camera device (e.g., acquiring monitoring images), so that the workload of the camera device can be transferred to other camera devices with lighter workload, balancing the working pressure of the camera devices.

[0046] Figure 2 An exemplary schematic diagram for training a segmentation model is provided for some embodiments of the present application. In some embodiments, Figure 2 The flow 200 shown can be performed by the segmentation module 430. As Figure 2 As shown, the flow 200 includes the following:

[0047] The pre-processed image and the reference image are input into the segmentation model, and the model outputs a segmented image.

[0048] In some embodiments, the segmentation model can be a trained ChangeNet of a twin neural network and a FCN.

[0049] The network structure of ChangeNet uses ResNet to extract features of the pre-processed image and the reference image, and combines convolution to output change positioning information at different levels. Then the same network is used to identify the detected changes and output object-level labeled change detection results. The change detection result can be a detection positioning and classification map of the change region in the pre-processed image.

[0050] In some embodiments, the ChangeNet can include a siamese network and a FCN. The siamese network can include a plurality of ResNet residual blocks to extract features of the pre-processed image and the reference image. The FCN can include a plurality of full convolution layers with a kernel size of 1x1 to integrate the extracted features. In some embodiments, the ChangeNet can further include a combination layer and a classification layer to add the integrated features and then classify them.

[0051] In some embodiments, for the output of the combination layer, a 1x1 convolution kernel can be used to reduce the dimension to N, and then a softmax classifier can be used for classification. A segmentation image with a size of w x h x N is obtained, which is the difference between the pre-processed image and the reference image. Wherein w and h are the length and width of the style image, and N is the number of segmentation images. For more information about the pre-processed image, the reference image and the segmentation image, see Figure 1 and related descriptions.

[0052] Some embodiments in this specification can capture coarse information and detailed information of an object by combining the outputs of different levels of convolution layers.

[0053] In some embodiments, the siamese network can be a parallel weight sharing network with the same number of parameters and values.

[0054] Some embodiments in this specification can improve the accuracy of the segmentation image by using a parallel weight sharing network for feature extraction, so that the same features can be learned from the pre-processed image and the reference image.

[0055] Obtain a first training sample set.

[0056] The first training sample set can refer to a combination of images used to train the initial segmentation model. The first training sample set can include a sample reference image and a sample monitoring image. The sample reference image can refer to an image used to train the initial segmentation model, and the sample monitoring image can refer to a monitoring image corresponding to the sample reference image. For example, the sample monitoring image can refer to a monitoring image that has been segmented and obtained a segmentation image. In some embodiments, historical monitoring data can be processed to obtain a sample reference image. The method of obtaining the sample reference image is consistent with the method of obtaining the reference image. For more information about obtaining the sample reference image, see Figure 1 and related descriptions.

[0057] Obtain a first label for the first training sample set. The first label can refer to an image of the region that exists in the difference between the pre-processed image of the sample monitoring image and the sample reference image. In some embodiments, the first label can be manually extracted.

[0058] input the first training sample set into the initial segmentation model, update the parameters of the initial segmentation model based on the output of the model and the first label iteratively, and obtain the trained segmentation model.

[0059] In some embodiments, the parameters of the initial segmentation model can be obtained through transfer learning, and the initial segmentation model can use ResNet50. The residual block of the initial segmentation model mainly consists of a convolution layer, batch normalization (BN) and a ReLU activation function. When training, a loss function can be constructed based on the first label and the output of the initial segmentation model, and the parameters of the initial segmentation model are updated iteratively based on the loss function until the loss function reaches a preset condition, and the initial segmentation model is used as the segmentation model. The preset condition can mean that the loss function converges or the number of iterations reaches a threshold. In some embodiments, the loss function can be constructed based on the edge coordinates of the segmentation image output by the initial segmentation model and the edge coordinates of the first label.

[0060] Figure 3 An exemplary schematic diagram for training a detection model is provided for some embodiments of the present application. In some embodiments, Figure 3 The flow 300 shown can be performed by the determination module 440. As Figure 3 As shown, the flow 300 includes the following:

[0061] input the segmentation image into the detection model, and the model outputs the alarm type.

[0062] In some embodiments, the detection model can be YOLO V4, which obtains the alarm type by performing target detection on the segmentation image.

[0063] Obtain a second training sample.

[0064] The second training sample can refer to an image used for training the detection model. The second training sample includes images of abnormal situations. In some embodiments, the second training sample can be obtained by collecting images of each type of abnormal situation in the substation.

[0065] Obtain a second label of the second training sample.

[0066] The second label can be the alarm type corresponding to the image of the abnormal situation. In some embodiments, the labels of the alarm types can be labeled for the images of each type of abnormal situation in the substation through manual labeling.

[0067] input the second training sample into the initial detection model, update the parameters of the initial detection model based on the output of the model and the second label iteratively, and obtain the trained detection model.

[0068] In some embodiments, the BackBone backbone network can be constructed based on the YOLO V4 algorithm to obtain an initial detection model. The BackBone backbone network includes a CSPDarknet53 module, a Mish activation function and a Dropblock module. During training, a loss function can be constructed based on the output of the initial detection model and the second label, and the parameters of the initial detection model are updated iteratively based on the loss function. When the loss function reaches a preset condition, the initial detection model is used as the detection model. The preset condition can be that the loss function converges or the number of iterations reaches a threshold.

[0069] Figure 4 An exemplary block diagram of a system for substation abnormal object monitoring based on image AI technology is provided for some embodiments of the present application. As shown in Figure 4 , the system 400 includes an acquisition module 410, a preprocessing module 420, a sub-monitoring image acquisition module 430 and a determination module 440.

[0070] The acquisition module 410 is configured to acquire a monitoring image. For more information about the acquisition module 410, see Figure 1 and related descriptions.

[0071] The preprocessing module 420 is configured to preprocess the monitoring image to obtain a preprocessed image. For more information about the preprocessing module 420, see Figure 1 and related descriptions.

[0072] The sub-monitoring image acquisition module 430 is configured to determine whether the preprocessed image has changed based on the preprocessed image and a reference image, and if so, acquire a sub-monitoring image of the changed area. For more information about the sub-monitoring image acquisition module 430, see Figure 1 and related descriptions.

[0073] The determination module 440 is configured to input the segmented image into a detection model, and the model outputs an alarm type. For more information about the determination module 440, see Figure 1 and related descriptions.

[0074] In some embodiments, the system 400 can further include an alarm module configured to issue a warning based on the alarm type and output the merged segmented image. For more information about the alarm module, see Figure 1 and related descriptions.

[0075] The above is only a preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for monitoring substation anomalies based on image AI technology, characterized in that, include Acquire monitoring images; The monitoring image is preprocessed to obtain a preprocessed image; Based on the preprocessed image and the reference image, determine whether the preprocessed image has changed; if so, obtain a sub-monitoring image of the changed area. The sub-monitored image is input into the detection model, and the model outputs the alarm type. Multiple camera devices jointly analyze and process the sub-monitoring image to obtain alarm types, including: when the sub-monitoring image is acquired by a camera device but it is still in a busy state, the sub-camera device sends the acquired sub-monitoring image to the master camera device. The master camera device, based on the idle / busy state of the sub-camera device, sends the sub-monitoring image to the idle sub-camera device for analysis and processing to obtain alarm types. The acquisition of sub-monitoring images of the changed areas includes, The preprocessed image and the reference image are input into the segmentation model, and the model outputs a segmented image. The segmented image is an image of the region where there are differences between the preprocessed image and the reference image; the preprocessed image has multiple differences from the reference image, and the preprocessed image is segmented along multiple difference regions to obtain multiple segmented images; the main camera device controls multiple cameras to acquire images of the regions where the multiple segmented images are located; Based on the segmented image, the camera device is controlled to acquire a sub-monitoring image including the segmented image; each camera device is associated with multiple other camera devices, and after a change area is detected, the master camera device controls the slave camera devices to acquire a sub-monitoring image containing the change area; the sub-monitoring image refers to the image captured by the camera device for the change area; one master camera device is associated with multiple slave camera devices, and the master camera device and slave camera devices acquire monitoring images. When it is determined that the difference between the monitoring image and the reference image is greater than a preset threshold, the area with the difference is determined, the master camera device marks the location and size information of the area with the difference, and controls the slave camera device closest to the area with the difference to acquire the image of the area.

2. The method for monitoring substation anomalies based on image AI technology according to claim 1, characterized in that, The segmentation model is obtained through training and includes, Obtain a first training sample group, which includes a sample reference image and a sample monitoring image; Obtain the first label of the first training sample group, where the first label is an image of a region that differs between the sample reference image and the sample monitoring image; The first training sample group is input into the initial segmentation model, and the parameters of the initial segmentation model are iteratively updated based on the model's output and the first label to obtain the trained segmentation model.

3. The method for monitoring substation anomalies based on image AI technology according to claim 2, characterized in that, The initial segmentation model is a twin neural network and an FCN change detection network; The change detection network uses ResNet to extract features from the preprocessed image and the reference image, and combines convolution to output change localization information at different levels; the same network is used to identify the detected changes and output labeled change detection results; the change detection result is a detection localization and classification map of the changed regions in the preprocessed image; The Siamese neural network is a parallel weight-sharing network with the same number and values ​​of parameters, including multiple ResNet residual blocks, which extract features from the preprocessed image and the reference image; the FCN includes multiple fully convolutional layers with a kernel size of 1×1, which integrate the extracted features; The change detection network also includes a combination layer and a classification layer. The integrated features are added together and then classified. For the output of the combination layer, a 1×1 convolutional kernel is used to reduce the dimension to N, and then a softmax classifier is used for classification. A segmented image of size w×h×N is obtained. This segmented image is the difference between the preprocessed image and the reference image. Here, w and h are the length and width of the style image, and N is the number of segmented images. A loss function is constructed based on the edge coordinates of the segmented image output by the initial segmentation model and the edge coordinates of the first label.

4. The method for monitoring substation anomalies based on image AI technology according to claim 1, characterized in that, It also includes issuing a warning based on the alarm type and merging the segmented images before outputting them.

5. The method for monitoring substation anomalies based on image AI technology according to claim 1, characterized in that, The preprocessing includes grayscale processing, noise detection and removal, and / or binarization processing; Grayscale processing includes: the image coordinates are (X, Y), and , ; and These are the height and width of the image, respectively; the horizontal axis of the image is cropped to... The vertical axis is truncated as The image matrix is ​​then normalized to obtain a grayscale image matrix, with a width of W2 and a height of H2. Noise detection includes: calculating thresholds using a dual threshold method; edge pixels with gradient values ​​greater than the high threshold are considered strong edge pixels; edge pixels with gradient values ​​less than the high threshold but greater than the low threshold are marked as weak edge pixels; pixels with gradient values ​​less than the low threshold are suppressed. ; in, Represents a high threshold; This represents a low threshold. A parameter representing the ratio of high threshold to low threshold; Binarization processing includes: making the image contain only two colors, and the binarized image consists only of the device to be detected and the background.

6. The method for monitoring substation anomalies based on image AI technology according to claim 1, characterized in that, The detection model is obtained through training and includes, Obtain a second training sample; the second training sample includes images of abnormal situations; Obtain the second label of the second training sample, where the second label is the alarm type corresponding to the image of the abnormal situation; The second training sample is input into the initial detection model, and the parameters of the initial detection model are iteratively updated based on the model's output and the second label to obtain the trained detection model.

7. The method for monitoring substation anomalies based on image AI technology according to claim 6, characterized in that, The initial detection model is YOLOv4; the BackBone backbone network is constructed based on the YOLOv4 algorithm to obtain the initial detection model; the BackBone backbone network includes the CSPDarknet53 module, the Mish activation function, and the Dropblock module.

8. The method for monitoring substation anomalies based on image AI technology according to any one of claims 1-7, characterized in that, Analyze abnormal situations within the substation during non-inspection periods of the monitoring equipment.

9. A system for monitoring substation anomalies based on image AI technology, characterized in that, It includes an acquisition module, a preprocessing module, a sub-monitoring image acquisition module, and a determination module; The acquisition module is used to acquire monitoring images; The preprocessing module is used to preprocess the monitoring image to obtain a preprocessed image; The sub-monitoring image acquisition module is used to determine whether the pre-processed image has changed based on the pre-processed image and the reference image; if so, it acquires the sub-monitoring image of the changed area. The determining module is used to input the sub-monitoring image into the detection model, and the model outputs the alarm type; Multiple camera devices jointly analyze and process the sub-monitoring image to obtain alarm types, including: when the sub-monitoring image is acquired by a camera device but it is still in a busy state, the sub-camera device sends the acquired sub-monitoring image to the master camera device. The master camera device, based on the idle / busy state of the sub-camera device, sends the sub-monitoring image to the idle sub-camera device for analysis and processing to obtain alarm types. The acquisition of sub-monitoring images of the changed areas includes, The preprocessed image and the reference image are input into the segmentation model, and the model outputs a segmented image. The segmented image is an image of the region where there are differences between the preprocessed image and the reference image; the preprocessed image has multiple differences from the reference image, and the preprocessed image is segmented along multiple difference regions to obtain multiple segmented images; the main camera device controls multiple cameras to acquire images of the regions where the multiple segmented images are located; Based on the segmented image, the camera device is controlled to acquire a sub-monitoring image including the segmented image; each camera device is associated with multiple other camera devices, and after a change area is detected, the master camera device controls the slave camera devices to acquire a sub-monitoring image containing the change area; the sub-monitoring image refers to the image captured by the camera device for the change area; one master camera device is associated with multiple slave camera devices, and the master camera device and slave camera devices acquire monitoring images. When it is determined that the difference between the monitoring image and the reference image is greater than a preset threshold, the area with the difference is determined, the master camera device marks the location and size information of the area with the difference, and controls the slave camera device closest to the area with the difference to acquire the image of the area.

Images