Smart city video monitoring adjustment method and system based on 5G network
By deploying high-resolution cameras and edge computing devices in smart cities, combined with 5G networks and efficient video processing algorithms, the bandwidth limitation problem of video surveillance systems has been solved, enabling real-time transmission and efficient processing of high-definition video, and improving the real-time performance and target recognition capabilities of the surveillance system.
Patent Information
- Application Number
- CN202411545282.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-01
- Publication Date
- 2024-11-29
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing video surveillance systems struggle to support high-definition video and multiple video streams due to bandwidth limitations, resulting in blurry or delayed images that affect real-time monitoring and rapid response capabilities. Furthermore, they lack advanced features such as automated event detection and behavior analysis.
By deploying high-resolution surveillance cameras and edge computing devices using Geographic Information System (GIS) technology, video streams are captured in real time using the H.265 compression format. Video data is processed by combining adaptive dynamic Gaussian filters and the YOLO algorithm, a convolutional neural network is constructed for target detection, and data summarization and latency adjustment are performed through edge computing devices to achieve efficient transmission and processing of video data.
It improves the real-time performance and response speed of the monitoring system, reduces data transmission latency, enhances image detail and target recognition accuracy, optimizes storage and transmission efficiency, and ensures the efficient operation of the monitoring system under different conditions.
Smart Images

Figure CN119052398A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of smart cities, and in particular to a smart city video surveillance adjustment method and system based on a 5G network. Background Art
[0002] With the rapid development of 5G networks, new breakthroughs have been made in solving urban safety issues, providing strong support for coping with increasingly complex security challenges in the process of urbanization. The high bandwidth, low latency and wide connectivity of 5G technology have brought urban safety management into a new stage. In critical moments such as traffic accident handling and public safety incident response, it can greatly improve reaction speed and processing efficiency, ensuring timely measures to maintain safety. Due to the wide connectivity of 5G technology, the popularization and integration of smart devices have been promoted. With the support of 5G networks, various smart sensors, cameras and other security devices in the city can exchange and collaborate data more efficiently.
[0003] The existing technology has the following deficiencies: Bandwidth limitations of existing video surveillance systems are a major problem. As the demand for high-definition video and multi-channel video streams increases, traditional systems often have difficulty providing sufficient bandwidth support. This lack of bandwidth limits the capabilities of the system, causing the surveillance footage to become blurry or delayed during peak hours or in high-traffic environments. This not only affects the clarity of the image, but also limits the ability to monitor in real time and respond quickly. In an emergency, this delay may affect the timeliness of event processing, resulting in the inability to quickly take necessary measures. As the size of the city expands and the demand for surveillance increases, the original system may not be able to support the access of more cameras, and only perform basic video recording and playback, but lack advanced data analysis and processing capabilities, making it difficult to perform advanced functions such as automated event detection and behavior analysis.
[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention
[0005] The purpose of the present invention is to provide a smart city video surveillance adjustment method based on 5G network to solve the problems in the above-mentioned background technology.
[0006] In order to achieve the above object, the present invention provides the following technical solutions: Step 1: Draw a city map using geographic information system technology, deploy high-resolution surveillance cameras supported by 5G networks and edge computing devices, and use built-in image sensors to capture real-time video streams in H.265 compression format; Step 2: Select a sliding window for the video data and calculate the standard deviation of the frame area in the window as a filtering parameter. Apply an adaptive dynamic Gaussian filter based on the local noise level and filtering parameters of the sliding window. Grayscale each small block in the frame area of the video data separately, and reconstruct the frame area of the new video data through the cumulative distribution function. Step 3: Segment the region of interest of the video data and the background region of the video data, and determine whether to add adjacent pixels to the region of interest by defining a growth criterion; Step 4: Connect the high-resolution surveillance camera to the edge computing device, use the YOLO algorithm to detect the target in the segmented frame area of the area of interest, and construct the behavior target of the convolutional neural network output video data; Step 5: Use edge computing devices to summarize data, construct network indicator characteristics and dynamically adjust delays. Use regression coefficients to concentrate the mean characteristics of load resources and the slope characteristics to predict the linear elements of load resources. Dynamically adjust resources based on the predicted load, and transmit video data related content to the monitoring center for feedback through edge computing devices.
[0007] Preferably, a city map of the monitoring area is drawn using geographic information system technology, including city squares, major traffic arteries, commercial areas, and schools, high-resolution surveillance cameras supporting 5G networks are deployed, and built-in image sensors are used to capture video streams in H.265 compression format in real time, and edge computing devices are deployed near the high-resolution surveillance cameras.
[0008] Preferably, a sliding window is selected for the video data and the standard deviation of the frame area within the window is calculated as a filtering parameter, a circular filtering window of size 5*5 is selected, and for each pixel in the frame area of the current video data, an adaptive dynamic Gaussian filter is applied based on the local noise level and filtering parameters of the sliding window in which it is located.
[0009] Preferably, the frame area of the video data is divided into a plurality of small blocks, each small block is individually grayed, and the number of pixels at each gray level is counted to calculate the cumulative distribution function value, and the specific formula is:
[0010] in, Represents the cumulative distribution function value of the current gray level i, represents the probability of a pixel at gray level j, Represents the parameters that control nonlinear enhancement and adjust the sensitivity to high-frequency details. Represents the parameters that control the gray level effect, enhancing the contrast of low gray levels, Indicates the total number of pixels in each small block. The grayscale index is represented, a contrast limit parameter is introduced, small blocks are reorganized to construct a new frame area of video data, and the boundaries between small blocks are smoothed using bilinear interpolation.
[0011] Preferably, the region of interest of the video data and the background region of the video data are segmented, a pixel point located at the inner center point of the region of interest is selected as a seed point, and according to the pixel point characteristics of the seed point, a growth criterion is defined to determine whether to add adjacent pixel points to the region of interest, and the specific steps of the determination are: S1.1. Using the grayscale value between pixels as the growth criterion, when the difference between the grayscale value of the pixel to be processed and the grayscale value of the seed point is less than the set threshold, the pixel is added to the region of interest.
[0012] S1.2. Using the color difference between pixels as a growth criterion, if the difference between the color value of the pixel to be processed and the color value of the seed point is less than a set difference threshold, the pixel is added to the region of interest.
[0013] S1.3, starting from the seed point, traverse its neighborhood pixels. When the pixel point meets the growth criteria and has not yet been added to the region of interest, it will be added to the region of interest and used as a new seed point to continue growing. The growth is iterated continuously until there are no pixels that meet the growth criteria. The segmented frame area representing the region of interest and the background area is obtained through iterative generation.
[0014] Preferably, the VideoCapture class of OpenCV is used to connect the video data to the edge computing device through a high-resolution surveillance camera, and the YOLO algorithm is used to perform target detection on the segmented frame area of the area of interest, and the confidence level is selected as 0.5 and the bounding box coordinates of the segmented frame area are obtained. By tracking the center point of the segmented frame area, the crowd count in the monitoring area is counted, and the center point coordinates of the person bounding box of each target detection in the segmented frame area are obtained.
[0015] Preferably, a convolutional neural network is constructed and video data is received through an input layer, a convolutional layer is added, multiple convolution kernels are used to perform sliding local perception on the video data, and a dot product operation is performed on the segmented frame area to generate a behavior target feature map, an activation layer is added, ReLU is used as an activation function and nonlinearity is introduced, and the specific formula of the dot product operation is:
[0016] in An element representing the behavior target feature in the video data, The pixel value representing the target feature of the behavior in the video data, represents the weight value of the behavioral target feature in the video data, Respectively represent the row and column indices inside the convolution kernel, They represent the row and column indices of the behavior target features in the video data respectively, downsample the behavior target features in the video data output by the convolution layer through the pooling layer, repeat the iterative convolution and pooling operations, connect all neurons of the convolution layer, activation layer and pooling layer through each neuron of the fully connected layer, flatten the behavior target features into a one-dimensional vector, convert them into the probability distribution of the behavior target through Softmax, map the behavior target through the weight matrix and output the behavior target feature map of the video data.
[0017] Preferably, edge computing equipment is used to perform data summarization on the segmented frame area of the area of interest, and network indicator characteristics are constructed and delay is dynamically adjusted according to the real-time network and computing load. The specific formula is:
[0018] in, represents dynamic delay, represents the basic delay, Indicates the degree of network congestion. represents the network delay factor, Indicates the available bandwidth, Represents the dynamic processing time, and the closed-form solution formula of ridge regression is used to calculate the regression coefficient vector. The specific formula is:
[0019] in, represents the regression coefficient vector, represents the design matrix, represents the regularization parameter in ridge regression, represents the identity matrix, Represents load resources. The linear elements of load resources are predicted by the mean characteristics and slope characteristics of the load resources through the regression coefficient. The resources are dynamically adjusted according to the predicted load. The specific formula is:
[0020] in, Represents a dynamically loaded resource, Indicates the basic load resource. represents the linear element of the forecast resource, Indicates the current load resources. Indicates the resource adjustment factor.
[0021] Preferably, edge computing devices are used to transmit relevant content of video data to the monitoring center through a secure communication channel, and the event type, the event that occurred, and the specific portraits of the segmented frames are fed back to the relevant personnel, and a visual instrument is established to view the feedback information in real time.
[0022] A smart city video surveillance adjustment system based on a 5G network, comprising a device deployment module, a data processing module, a region segmentation module, an edge processing module and an edge optimization module; Equipment deployment module: Draw city maps of the monitoring area through geographic information system technology, deploy high-resolution surveillance cameras and edge computing devices that support 5G networks, and use built-in image sensors to capture video streams in H.265 compression format in real time; Data processing module: select a sliding window for video data and calculate the standard deviation of the frame area in the window as a filtering parameter, apply an adaptive dynamic Gaussian filter based on the local noise level and filtering parameters of the sliding window, grayscale each small block in the frame area of the video data separately, and reconstruct the frame area of the new video data through the cumulative distribution function; Region segmentation module: segment the region of interest of video data and the background region of video data, and determine whether to add adjacent pixels to the region of interest by defining growth criteria; Edge processing module: Connect high-resolution surveillance cameras to edge computing devices, use the YOLO algorithm to detect targets in segmented frame areas of interest, and construct a convolutional neural network to output behavioral targets for video data; Edge optimization module: constructs network indicator characteristics and dynamically adjusts delays, predicts linear elements of load resources through regression coefficients based on the mean characteristics and slope characteristics of load resources, dynamically adjusts resources based on the predicted load, and transmits video data related content to the monitoring center for feedback through edge computing devices.
[0023] In the above technical solution, the technical effects and advantages provided by the present invention are: GIS technology is used to provide detailed city maps to ensure the reasonable layout of surveillance cameras and maximize coverage. High-resolution cameras combined with high-speed transmission of 5G networks are used to ensure the real-time nature of surveillance data, quickly respond to emergencies, and realize real-time processing of video data and sensor data, reduce data transmission delays, and improve response speed. The H.265 compression format is used to ensure video quality while significantly reducing bandwidth occupancy and optimizing storage and transmission efficiency. The standard deviation of the frame area within the sliding window is calculated to effectively reflect the local noise level of the current video data. A 5x5 circular filter window is selected to avoid excessive blurring of the center of the frame area and reduce the impact on the edges. The grayscale level of each small block is calculated. Perform detailed analysis, enhance the contrast of low gray levels, improve the details of the overall image, use the seed point growth algorithm to segment the region of interest and the background area in the video data, improve the recognition accuracy and processing efficiency of the target, access the high-resolution surveillance camera through the OpenCV VideoCapture class, and combine the YOLO algorithm to detect the target in the region of interest, which can identify and track the flow of people in real time and improve the monitoring efficiency. Convolutional neural network is constructed to perform local perception of video data through sliding convolution kernels, and edge computing devices are used to summarize the data of the segmented frame area. The processing delay can be dynamically adjusted according to the real-time network and computing load to ensure the efficient operation of the monitoring system under different conditions. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.
[0025] Figure 1 This is a method flow chart of a smart city video surveillance adjustment method based on 5G network in the present invention.
[0026] Figure 2 This is a module schematic diagram of a smart city video surveillance and adjustment system based on 5G network in the present invention. DETAILED DESCRIPTION
[0027] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.
[0028] The present invention provides Figure 1A smart city video surveillance adjustment method based on a 5G network is shown, comprising the following steps: Step 1: Draw a city map using geographic information system technology, deploy high-resolution surveillance cameras supported by 5G networks and edge computing devices, and use built-in image sensors to capture real-time video streams in H.265 compression format; Use geographic information system technology to draw urban maps of monitoring areas, including city squares, major traffic arteries, commercial areas and schools, deploy high-resolution surveillance cameras that support 5G networks, and use built-in image sensors to capture video streams in H.265 compression format in real time. Deploy edge computing devices near high-resolution surveillance cameras to process video data and sensor data in real time.
[0029] Step 2: Select a sliding window for the video data and calculate the standard deviation of the frame area in the window as a filtering parameter. Apply an adaptive dynamic Gaussian filter based on the local noise level and filtering parameters of the sliding window. Grayscale each small block in the frame area of the video data separately, and reconstruct the frame area of the new video data through the cumulative distribution function. A sliding window is selected for the video data and the standard deviation of the frame area within the window is calculated as a filtering parameter. The standard deviation is used to reflect the local noise level of the current video data. A circular filtering window of size 5*5 is selected to avoid excessive blurring of the center of the frame area of the video data and reduce the impact on the edges. For each pixel in the frame area of the current video data, an adaptive dynamic Gaussian filter is applied based on the local noise level and filtering parameters of the sliding window in which it is located.
[0030] The frame area of the video data is divided into multiple small blocks, each small block is grayed separately, and the number of pixels at each gray level is counted to calculate the cumulative distribution function value. The specific formula is:
[0031] in, Represents the cumulative distribution function value of the current gray level i, represents the probability of a pixel at gray level j, Represents the parameters that control nonlinear enhancement and adjust the sensitivity to high-frequency details. Represents the parameters that control the gray level effect, enhancing the contrast of low gray levels, Indicates the total number of pixels in each small block. Represents the index of the grayscale level, which is used to traverse all grayscale levels and introduce a contrast limit parameter. When the number of pixels at a certain grayscale level exceeds 500, the small blocks are cropped and the excess pixels are evenly distributed to other grayscale levels. The small blocks are reorganized to construct a new frame area of the video data, and the boundaries between the small blocks are smoothed using bilinear interpolation to avoid obvious seams.
[0032] Step 3: Segment the region of interest of the video data and the background region of the video data, and determine whether to add adjacent pixels to the region of interest by defining a growth criterion; Segment the region of interest of the video data and the background region of the video data, where the region of interest includes people and vehicles, select the pixel point located at the inner center point of the region of interest as the seed point, and determine whether to add the adjacent pixel points to the region of interest by defining the growth criterion according to the pixel point characteristics of the seed point. The specific steps of the determination are as follows: S1.1. Use the grayscale value between pixels as the growth criterion. When the difference between the grayscale value of the pixel to be processed and the grayscale value of the seed point is less than the set threshold, the pixel is added to the region of interest, where the threshold is set to 20.
[0033] S1.2. Use the color difference between pixels as the growth criterion. If the difference between the color value of the pixel to be processed and the color value of the seed point is less than the set difference threshold, the pixel is added to the region of interest, where the difference threshold is set to 5.
[0034] S1.3, starting from the seed point, traverse its neighborhood pixels. When the pixel point meets the growth criteria and has not yet been added to the region of interest, it will be added to the region of interest and used as a new seed point to continue growing. The growth is iterated continuously until there are no pixels that meet the growth criteria. The segmented frame area representing the region of interest and the background area is obtained through iterative generation.
[0035] Step 4: Connect the high-resolution surveillance camera to the edge computing device, use the YOLO algorithm to detect the target in the segmented frame area of the area of interest, and construct the behavior target of the convolutional neural network output video data; Use OpenCV's VideoCapture class to connect video data to edge computing devices through high-resolution surveillance cameras, use the YOLO algorithm to detect targets in the segmented frame area of the area of interest, select a confidence level of 0.5 and obtain the bounding box coordinates of the segmented frame area, track the center point of the segmented frame area, count the number of people in the monitored area, obtain the center point coordinates of the bounding box of each target detected person in the segmented frame area, set the position of y=500 in the segmented frame area as the entry and exit line, and update the crowd count when the center point coordinates cross the entry and exit line.
[0036] Construct a convolutional neural network and receive video data through the input layer, add a convolutional layer, use multiple convolution kernels to perform sliding local perception on the video data, and perform dot product operations on the segmented frame area to generate a behavioral target feature map, add an activation layer, use ReLU as the activation function and introduce nonlinearity. The specific formula of the dot product operation is:
[0037] in An element representing the behavior target feature in the video data, The pixel value representing the target feature of the behavior in the video data, represents the weight value of the behavioral target feature in the video data, Respectively represent the row and column indices inside the convolution kernel, They represent the row and column indices of the behavior target features in the video data respectively, downsample the behavior target features in the video data output by the convolution layer through the pooling layer, repeat the iterative convolution and pooling operations, connect all neurons of the convolution layer, activation layer and pooling layer through each neuron of the fully connected layer, flatten the behavior target features into a one-dimensional vector, convert them into the probability distribution of the behavior target through Softmax, map the behavior target through the weight matrix and output the behavior target feature map of the video data.
[0038] Step 5: Use edge computing devices to perform data summarization, construct network indicator characteristics and dynamically adjust delays. Use regression coefficients to concentrate the mean characteristics of load resources and the slope characteristics to predict the linear elements of load resources. Dynamically adjust resources according to the predicted load, and transmit video data related content to the monitoring center for feedback through edge computing devices. The edge computing device is used to perform data summarization on the segmented frame area of the area of interest. According to the real-time network and computing load, the network indicator characteristics are constructed and the delay is dynamically adjusted. The specific formula is:
[0039] in, represents dynamic delay, represents the basic delay, Indicates the degree of network congestion. represents the network delay factor, Indicates the available bandwidth, Represents the dynamic processing time, and the closed-form solution formula of ridge regression is used to calculate the regression coefficient vector. The specific formula is:
[0040] in, represents the regression coefficient vector, represents the design matrix, represents the regularization parameter in ridge regression, represents the identity matrix, Represents load resources. The linear elements of load resources are predicted by the mean characteristics and slope characteristics of the load resources through the regression coefficient. The resources are dynamically adjusted according to the predicted load. The specific formula is:
[0041] in, Represents a dynamically loaded resource, Indicates the basic load resource. represents the linear element of the forecast resource, Indicates the current load resources. Indicates the resource adjustment factor.
[0042] Edge computing devices are used to transmit video data-related content to the monitoring center through secure communication channels, and the event type, event, and specific portraits of segmented frames are fed back to relevant personnel, and a visual instrument is established to view feedback information in real time.
[0043] GIS technology is used to provide detailed city maps to ensure the reasonable layout of surveillance cameras and maximize coverage. High-resolution cameras combined with high-speed transmission of 5G networks are used to ensure the real-time nature of surveillance data, quickly respond to emergencies, and realize real-time processing of video data and sensor data, reduce data transmission delays, and improve response speed. The H.265 compression format is used to ensure video quality while significantly reducing bandwidth occupancy and optimizing storage and transmission efficiency. The standard deviation of the frame area within the sliding window is calculated to effectively reflect the local noise level of the current video data. A 5x5 circular filter window is selected to avoid excessive blurring of the center of the frame area and reduce the impact on the edges. The grayscale level of each small block is calculated. Perform detailed analysis, enhance the contrast of low gray levels, improve the details of the overall image, use the seed point growth algorithm to segment the region of interest and the background area in the video data, improve the recognition accuracy and processing efficiency of the target, access the high-resolution surveillance camera through the OpenCV VideoCapture class, and combine the YOLO algorithm to detect the target in the region of interest, which can identify and track the flow of people in real time and improve the monitoring efficiency. Convolutional neural network is constructed to perform local perception of video data through sliding convolution kernels, and edge computing devices are used to summarize the data of the segmented frame area. The processing delay can be dynamically adjusted according to the real-time network and computing load to ensure the efficient operation of the monitoring system under different conditions.
[0044] The present invention provides Figure 2 A smart city video surveillance and adjustment system based on a 5G network is shown, comprising a device deployment module, a data processing module, a region segmentation module, an edge processing module, and an edge optimization module; Equipment deployment module: Draw city maps of the monitoring area through geographic information system technology, deploy high-resolution surveillance cameras and edge computing devices that support 5G networks, and use built-in image sensors to capture video streams in H.265 compression format in real time; Data processing module: select a sliding window for video data and calculate the standard deviation of the frame area in the window as a filtering parameter, apply an adaptive dynamic Gaussian filter based on the local noise level and filtering parameters of the sliding window, grayscale each small block in the frame area of the video data separately, and reconstruct the frame area of the new video data through the cumulative distribution function; Region segmentation module: segment the region of interest of video data and the background region of video data, and determine whether to add adjacent pixels to the region of interest by defining growth criteria; Edge processing module: Connect high-resolution surveillance cameras to edge computing devices, use the YOLO algorithm to detect targets in segmented frame areas of interest, and construct a convolutional neural network to output behavioral targets for video data; Edge optimization module: constructs network indicator characteristics and dynamically adjusts delays, predicts linear elements of load resources through regression coefficients based on the mean characteristics and slope characteristics of load resources, dynamically adjusts resources based on the predicted load, and transmits video data related content to the monitoring center for feedback through edge computing devices.
[0045] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.
Claims
1. A smart city video surveillance adjustment method based on 5G network, comprising the following steps: Features: Step 1: Draw a city map using geographic information system technology, deploy high-resolution surveillance cameras supported by 5G networks and edge computing devices, and use built-in image sensors to capture real-time video streams in H.265 compression format; Step 2: Select a sliding window for the video data and calculate the standard deviation of the frame area in the window as a filtering parameter. Apply an adaptive dynamic Gaussian filter based on the local noise level and filtering parameters of the sliding window. Grayscale each small block in the frame area of the video data separately, and reconstruct the frame area of the new video data through the cumulative distribution function. Step 3: Segment the region of interest of the video data and the background region of the video data, and determine whether to add adjacent pixels to the region of interest by defining a growth criterion; Step 4: Connect the high-resolution surveillance camera to the edge computing device, use the YOLO algorithm to detect the target in the segmented frame area of the area of interest, and construct the behavior target of the convolutional neural network output video data; Step 5: Use edge computing devices to summarize data, construct network indicator characteristics and dynamically adjust delays. Use regression coefficients to concentrate the mean characteristics of load resources and the slope characteristics to predict the linear elements of load resources. Dynamically adjust resources based on the predicted load, and transmit video data related content to the monitoring center for feedback through edge computing devices.
2. According to the 5G network-based smart city video surveillance adjustment method described in claim 1, Features: In step 1, a city map of the monitoring area, including city squares, major traffic arteries, commercial areas and schools, is drawn using geographic information system technology, high-resolution surveillance cameras supporting 5G networks are deployed, and built-in image sensors are used to capture video streams in H.265 compression format in real time, and edge computing devices are deployed near the high-resolution surveillance cameras.
3. According to the 5G network-based smart city video surveillance adjustment method of claim 1, Features: In the step 2, a sliding window is selected for the video data and the standard deviation of the frame area in the window is calculated as a filtering parameter, a circular filtering window of size 5*5 is selected, and for each pixel in the frame area of the current video data, an adaptive dynamic Gaussian filter is applied based on the local noise level and filtering parameters of the sliding window in which it is located, the frame area of the video data is divided into multiple small blocks, each small block is grayed separately, and the number of pixels at each gray level is counted, the cumulative distribution function value is calculated, a contrast limit parameter is introduced, the small blocks are reorganized to construct a new frame area of the video data, and the boundaries between the small blocks are smoothed using bilinear interpolation.
4. According to the 5G network-based smart city video surveillance adjustment method of claim 3, Features: The specific formula for calculating the cumulative distribution function value is: ; in, Represents the cumulative distribution function value of the current gray level i, represents the probability of a pixel at gray level j, Represents the parameters that control nonlinear enhancement and adjust the sensitivity to high-frequency details. Represents the parameters that control the gray level effect, enhancing the contrast of low gray levels, Indicates the total number of pixels in each small block. An index representing the grayscale level.
5. According to the 5G network-based smart city video surveillance adjustment method of claim 1, Features: In the step three, the region of interest of the video data and the background region of the video data are segmented, and a pixel point located at the inner center point of the region of interest is selected as a seed point. Based on the pixel point characteristics of the seed point, a growth criterion is defined to determine whether to add adjacent pixels to the region of interest.
6. According to the 5G network-based smart city video surveillance adjustment method of claim 1, Features: In the step 4, the video data is connected to the edge computing device through the high-resolution surveillance camera using the VideoCapture class of OpenCV, the segmented frame area of the area of interest is detected by the YOLO algorithm, the confidence level is selected as 0.5 and the bounding box coordinates of the segmented frame area are obtained, the center point of the segmented frame area is tracked, the flow of people in the monitoring area is counted, and the center point coordinates of the person bounding box in the segmented frame area of each target detection are obtained, a convolutional neural network is constructed and the video data is received through the input layer, a convolutional layer is added, and the video data is slidingly perceived locally using multiple convolution kernels, and a dot product operation is performed on the segmented frame area to generate a behavior target feature map, an activation layer is added, ReLU is used as an activation function and nonlinearity is introduced, the behavior target features in the video data output by the convolutional layer are downsampled through the pooling layer, the convolution and pooling operations are repeated, and all neurons of the convolutional layer, the activation layer and the pooling layer are connected through each neuron of the fully connected layer, and the behavior target features are flattened into a one-dimensional vector, converted into a probability distribution of the behavior target through Softmax, and the behavior target is mapped through the weight matrix and the behavior target feature map of the video data is output.
7. According to claim 6, a smart city video surveillance adjustment method based on 5G network, Features: The specific formula of the dot product operation is: ; in An element representing the behavior target feature in the video data, The pixel value representing the target feature of the behavior in the video data, represents the weight value of the behavioral target feature in the video data, Respectively represent the row and column indices inside the convolution kernel, The indices of the rows and columns representing the target features in the video data, respectively.
8. According to the 5G network-based smart city video surveillance adjustment method of claim 1, Features: In the step five, edge computing equipment is used to perform data summary on the segmented frame area of the area of interest, network indicator characteristics are constructed and delay is dynamically adjusted based on the real-time network and computing load, the regression coefficient vector is calculated using the closed-form solution formula of ridge regression, the linear elements of the load resources are predicted through the mean characteristics and slope characteristics of the load resources in the regression coefficient concentration, and the resources are dynamically adjusted based on the predicted load.
9. According to the method for adjusting smart city video surveillance based on 5G network as claimed in claim 8, Features: The specific formula for dynamically adjusting the delay is: ; in, represents dynamic delay, represents the basic delay, Indicates the degree of network congestion. represents the network delay factor, Indicates the available bandwidth, Represents the dynamic processing time, and the specific formula for calculating the regression coefficient vector is: ; in, represents the regression coefficient vector, represents the design matrix, represents the regularization parameter in ridge regression, represents the identity matrix, represents the load resource, and the specific formula of the load dynamic adjustment resource is: ; in, Represents a dynamic load resource, Indicates the basic load resource. represents the linear element of the forecast resource, Indicates the current load resources. Indicates the resource adjustment factor.
10. A smart city video surveillance adjustment system based on a 5G network, used to implement a smart city video surveillance adjustment method based on a 5G network as described in any one of claims 1 to 9 above, It is characterized in that It includes a device deployment module, a data processing module, a region segmentation module, an edge processing module, and an edge optimization module; Equipment deployment module: Draw city maps of the monitoring area through geographic information system technology, deploy high-resolution surveillance cameras and edge computing devices that support 5G networks, and use built-in image sensors to capture video streams in H.265 compression format in real time; Data processing module: select a sliding window for video data and calculate the standard deviation of the frame area in the window as a filtering parameter, apply an adaptive dynamic Gaussian filter based on the local noise level and filtering parameters of the sliding window, grayscale each small block in the frame area of the video data separately, and reconstruct the frame area of the new video data through the cumulative distribution function; Region segmentation module: segment the region of interest of the video data and the background region of the video data, and determine whether to add adjacent pixels to the region of interest by defining a growth criterion; Edge processing module: Connect high-resolution surveillance cameras to edge computing devices, use the YOLO algorithm to detect targets in segmented frame areas of interest, and construct a convolutional neural network to output behavioral targets for video data; Edge optimization module: constructs network indicator characteristics and dynamically adjusts delays, predicts linear elements of load resources through regression coefficients based on the mean characteristics and slope characteristics of load resources, dynamically adjusts resources based on the predicted load, and transmits video data related content to the monitoring center for feedback through edge computing devices.
Citation Information
Patent Citations
Intelligent video monitoring method and system thereof
CN101635835A
Moving object detection method based on improved mixing gauss and image cutting
CN103077530A
Lens jitter prevention video movement target detection method for intelligent mobile terminal
CN105374051A
Video monitoring target detection method based on W4 algorithm and frame difference
CN105678803A
Improved region growing method for optic cup image segmentation
CN107346545A
Cited By
Tunnel safety monitoring system and method based on edge calculation
CN120451581A
Urban grid management method, system and equipment based on machine vision and medium
CN120525694A
DeepStream-based multi-channel heterogeneous video detection method and system
CN121239900A
Deepstream-based multi-path heterogeneous video detection method and system
CN121239900B
