A cloud edge collaboration based scene adaptive urban management platform

Abstract

This invention relates to the field of cloud-edge collaboration technology, specifically disclosing a scene-adaptive urban management platform based on cloud-edge collaboration, comprising: an image acquisition module: commanding a drone to fly along a preset route and capture images along the way, preprocessing and recognizing the video stream data, and extracting keyframes; a keyframe cropping module: acquiring the three primary color values ​​of pixels, marking target pixels based on the three primary color channels, constructing a planar coordinate system, and cropping frames based on the maximum and minimum values ​​of the horizontal and vertical coordinates; and an analysis and processing module: dividing the cropped frames into groups, acquiring the maximum proportion of cropped frames in the same group, determining the target event corresponding to the urban management problem, and distributing the data to various drone terminals. By lightweighting the data, the bandwidth pressure during upload is reduced, transmission resources are saved, the efficiency from problem discovery to handling is improved, management resources are targeted, and the transformation of urban management from passive response to proactive prediction is supported.

Description

Technical Field

[0001] This invention relates to the field of cloud-edge collaboration technology, and more specifically to a scene-adaptive urban management platform based on cloud-edge collaboration. Background Technology

[0002] In recent years, drone technology has demonstrated enormous application potential in urban management due to its unique aerial mobility and wide field of view. Traditional urban patrol and management mainly rely on manual inspections or fixed-point monitoring, which suffers from inherent drawbacks such as low efficiency, limited coverage, high risk, and slow response. Drones are seen as an effective tool for solving these problems; however, existing drone-based patrol solutions still have a series of significant technical shortcomings, making it difficult to meet the needs of modern urban management.

[0003] In existing technologies, drones lack real-time on-site analysis capabilities. Most drones merely act as aerial cameras, transmitting raw video data back to the user. This places enormous bandwidth pressure on communication links and relies on manual interpretation, resulting in low efficiency and a high risk of missing crucial information. The lack of real-time AI analysis capabilities at the flight terminal or near-terminal level prevents the immediate identification of specific targets such as road cracks, illegal road occupancy, and water accumulation, hindering the qualitative leap from simply "seeing" to "understanding." Summary of the Invention

[0004] The purpose of this invention is to provide a scene-adaptive urban management platform based on cloud-edge collaboration to solve the above-mentioned technical problems.

[0005] The objective of this invention can be achieved through the following technical solutions: A scenario-adaptive urban management platform based on cloud-edge collaboration includes the following steps: S1: Command the drone to fly along a preset route, and simultaneously activate the onboard camera equipment to preprocess the video stream data collected in real time during the flight based on the onboard AI; The processed video stream is input into a preset algorithm model to identify target events appearing in the video stream. If the identification is successful, the frame sequence corresponding to the target event in the video stream is extracted and recorded as a key frame. The target events represent road cracks, illegal road occupation, and water accumulation. S2: Obtain the three primary color values ​​of the pixels in the keyframe. The three primary color channels include the red channel value R, the green channel value G, and the blue channel value B. Mark the target pixels based on the three primary color channels. Construct a planar coordinate system with the preset pixel as the origin, obtain the coordinates (x, y) of the target pixel, and filter out the maximum value x of the target pixel's x-coordinate. max With minimum value x min The maximum value of the y-axis. max With minimum value ymin The cropped frames are obtained by cropping keyframes based on the maximum and minimum values ​​of the horizontal and vertical coordinates. S3: Upload the cropped frames to the cloud and obtain the upload time and the location of the cropped frames. Then, divide the cropped frames belonging to the same location into the same group of cropped frames. Obtain the number N of the i-th type of target events corresponding to the cropped frames in the same group. i Calculate the proportion B i =N i / N, where N represents the number of cropped frames contained in the same cropped frame group; Get the maximum percentage B max The urban management problems that occurred at the locations corresponding to the cropped frames in the same group accounted for the largest proportion, B. max The corresponding target event will send the above results to each drone terminal, and the drone will then stop uploading keyframes for the locations corresponding to the cropped frames in the same group.

[0006] As a further aspect of the present invention: the method for preprocessing real-time video stream data acquired during flight based on airborne AI specifically includes: Obtain the total number M of target events detected by the UAV while flying along a preset route, and preset the original detection period T and reference value M. s Calculate the detection period t = T × M s / M; The video stream obtained after the drone flies along a preset route is acquired. Frame sequences are extracted at each detection period t and recorded as valid frames. Valid frames are retained and invalid frames are discarded.

[0007] As a further aspect of the present invention: a method for identifying target events appearing in a video stream includes: Get the number of times A of the target event appears in the video stream. If A≠0, the recognition is successful; if A=0, the recognition fails and subsequent steps are omitted.

[0008] As a further aspect of the present invention: the method for marking target pixels based on the three primary color channels includes: Calculate the color value of the pixel SC=2 16 ×R+2 8 ×G+B, retrieves the maximum color value (SC) of pixels corresponding to the same type of target event. max and minimum value SC min Generate color value range [SC] min SC max If the color value SC is within the color value range [SC], min SC max In the image, the corresponding pixel is marked as the target pixel.

[0009] As a further aspect of the present invention: obtain the number of bytes B used to display the three primary color channels; if B ≠ 3, modify the color value calculation formula to SC = (2... B / 3×16 ×R)+(2 B / 3×8 ×G)+B.

[0010] As a further aspect of the present invention, a method for cropping keyframes based on the maximum and minimum values ​​of the horizontal and vertical coordinates to obtain cropped frames includes: The coordinates (X, Y) of the cropped frame pixels are determined according to constraints. Crop the keyframes.

[0011] As a further aspect of the present invention: if the same group of cropped frames corresponds to only one target event, then the subsequent steps are stopped, and the urban management problem occurring at the location corresponding to the same group of cropped frames is directly determined as the corresponding target event.

[0012] As a further aspect of the present invention: if the number of target events contained in the cropped frame is not unique, then when calculating their respective proportions, the number N of the i-th type of target events is counted independently. i .

[0013] The beneficial effects of this invention are as follows: This solution, through the coordinated operation of three modules—image acquisition, keyframe cropping, and analysis processing—forms an intelligent identification and management system for urban road problems. Its beneficial effects are reflected in multiple aspects: Firstly, in the image acquisition stage, the drone automatically flies along a preset route and simultaneously activates the camera equipment. Combined with onboard AI, the real-time video stream is preprocessed, and then a preset algorithm model is used to accurately identify three types of target events: road cracks, illegal road occupation, and water accumulation. This process replaces traditional manual inspection, which not only significantly improves the coverage and acquisition efficiency of urban areas but also avoids the safety risks of manual operations. At the same time, by preprocessing and identifying and filtering out irrelevant images, only keyframes are extracted to retain valid information, avoiding interference from invalid data and providing a focused foundation for subsequent analysis.

[0014] Secondly, the keyframe cropping module marks target pixels with the three primary color values ​​and combines the coordinate system to filter extreme values ​​for cropping. This can accurately select the problem area and remove irrelevant background. This not only makes the cropped frame more focused on the problem itself and improves the pertinence of subsequent analysis, but also reduces the bandwidth usage and cloud storage pressure during upload by making the data lightweight, thus saving transmission and storage resources.

[0015] Finally, the analysis and processing module groups the cropped frames by location, calculates the proportion of various target events in each group, and ultimately identifies the core issue corresponding to the largest proportion, clarifying the key directions of urban management and avoiding the ambiguity of priorities caused by multiple issues being mixed together. At the same time, the results are sent to the drone terminal and repeated uploads are stopped, which not only avoids redundant data collection at the same location, further saving resources, but also enables collaborative operation of multiple drones, improving the overall coordination and coverage of monitoring. In addition, the complete traceability of data recorded in the cloud supports the management closed loop.

[0016] In summary, this solution, with intelligence and automation at its core, promotes the transformation of urban management from the traditional model of inefficient manual labor and wasteful resources to a new model of intelligent efficiency and resource optimization. It improves the efficiency of problem detection and handling, enables targeted allocation of management resources, and effectively supports the transformation of urban management from passive response to proactive prediction. Attached Figure Description

[0017] The invention will now be further described with reference to the accompanying drawings.

[0018] Figure 1 This is a schematic diagram of the structure of a scene-adaptive urban management platform based on cloud-edge collaboration according to the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1 As shown, this invention is a scene-adaptive urban management platform based on cloud-edge collaboration, comprising: Image acquisition module: The drone flies along a preset route, and the onboard camera equipment is activated simultaneously. Based on the onboard AI, the video stream data acquired in real time during the flight is preprocessed. The processed video stream is input into a preset algorithm model to identify target events appearing in the video stream. If the identification is successful, the frame sequence corresponding to the target event in the video stream is extracted and recorded as a key frame. The target events represent road cracks, illegal road occupation, and water accumulation. Keyframe cropping module: Obtains the three primary color values ​​of pixels in the keyframe, wherein the three primary color channels include red channel value R, green channel value G and blue channel value B, and marks the target pixel based on the three primary color channels; Construct a planar coordinate system with the preset pixel as the origin, obtain the coordinates (x, y) of the target pixel, and filter out the maximum value x of the target pixel's x-coordinate. max With minimum value x min The maximum value of the y-axis. max With minimum value y min The cropped frames are obtained by cropping keyframes based on the maximum and minimum values ​​of the horizontal and vertical coordinates. Analysis and processing module: Uploads the cropped frames to the cloud, obtains the upload time and the location of the cropped frames, and divides the cropped frames belonging to the same location into the same group of cropped frames; Obtain the number N of the i-th type of target events corresponding to the cropped frames in the same group. i Calculate the proportion B i =N i / N, where N represents the number of cropped frames contained in the same cropped frame group; Get the maximum percentage B max The urban management problems that occurred at the locations corresponding to the cropped frames in the same group accounted for the largest proportion, B. max The corresponding target event will send the above results to each drone terminal, and the drone will then stop uploading keyframes for the locations corresponding to the cropped frames in the same group.

[0021] It should be noted that this solution constructs an intelligent identification and management system for urban road problems. This system mainly relies on three modules: image acquisition, keyframe cropping, and analysis and processing. In the image acquisition stage, drone technology is used as the primary data collection tool. The drone flies autonomously along a pre-set route. During flight, the camera equipment on the drone simultaneously starts capturing real-time footage of urban roads. Meanwhile, the onboard artificial intelligence system preprocesses the acquired real-time video stream, thereby reducing the workload of video processing. The drone identifies three representative target events: road cracks, illegal road occupation, and water accumulation. It should be noted that the determination of target events is based on the actual situation, and this invention does not impose rigid requirements.

[0022] This image acquisition process overcomes the shortcomings of traditional manual inspections, which are inefficient and limited by manpower and time costs, preventing comprehensive coverage of the entire city. Leveraging the powerful mobility and efficiency of drones, roads in every corner of the city can be included in the monitoring range, significantly improving the breadth and depth of urban coverage. Furthermore, since no personnel are required to directly participate in hazardous on-site operations, it effectively avoids various safety risks that may be encountered during manual inspections, such as traffic accidents and hazards in inclement weather.

[0023] Furthermore, through preprocessing, the system can filter out a large amount of irrelevant image information, extracting only frames containing key information as valid data for storage. This targeted data filtering mechanism greatly reduces interference from invalid data, allowing subsequent analysis and processing to focus more on the core issues.

[0024] Secondly, the keyframe cropping module plays a crucial role in the entire intelligent recognition and management system, acting as a bridge between different components. Specifically, it uses the unique visual characteristics of the three primary colors (red, green, and blue) to mark target pixels. In the field of digital imaging, each pixel is composed of a mixture of red, green, and blue primary colors in different proportions, and different objects or scenes often exhibit specific numerical distribution patterns in these color channels. Based on this, the system can accurately identify pixels related to road problems, such as abnormal color changes at road cracks, the distinctive color range of illegally parked vehicles, and the unique reflective hues of waterlogged areas.

[0025] After marking the target pixels, the module further uses a coordinate system to filter extreme values. This coordinate system is a two-dimensional Cartesian coordinate system built upon the image. By setting reasonable mathematical models and algorithmic rules, the module analyzes and calculates the positions of the marked target pixels within this coordinate system. The system automatically searches for extreme points located at boundaries or with significant features; these extreme points act as key nodes outlining the problem area. Based on the positional information of these extreme points, the module can accurately determine a bounding box, also known as a "cropping box," completely and precisely selecting the actual problem area.

[0026] In this process, irrelevant background information is effectively removed. Whether it's fixed structures like trees and lampposts along the road, or dynamic elements such as pedestrians and other unrelated vehicles, anything not related to the current road issue is excluded. The cropped frame image is highly focused on the problem itself, eliminating all factors that might distract or interfere with the analysis. This highly targeted image processing method greatly improves the accuracy and efficiency of subsequent analysis. Analysts no longer need to spend a lot of time and effort distinguishing useful information from useless noise, and can directly conduct in-depth research on the core issue. Because a large amount of irrelevant background is removed, the amount of image data after cropping is significantly reduced, achieving data lightweighting. In actual network transmission, bandwidth is a valuable resource, especially in large-scale urban road monitoring scenarios where massive amounts of data need to be uploaded to the cloud for processing in a timely manner. The reduction in data volume directly leads to a significant reduction in bandwidth usage during upload, avoiding transmission delays and packet loss caused by data congestion. Simultaneously, in terms of cloud storage, lightweight data also means lower storage costs and higher storage efficiency. Servers that would have previously required a large amount of storage space to hold the original image data can now handle a large number of cropped images more easily, saving valuable transmission and storage resources and providing a strong guarantee for the long-term stable operation of the entire system.

[0027] This keyframe cropping technology, based on primary color value marking and coordinate system filtering, also boasts excellent scalability and compatibility. As urban road environments continuously change and new problems emerge, adjustments to the primary color value judgment thresholds and coordinate system filtering algorithms are all that's needed to adapt to new demands, ensuring consistently accurate and efficient keyframe cropping and providing high-quality data support for subsequent analysis and processing. Furthermore, this technology can seamlessly integrate with other image processing algorithms and data analysis tools, further enhancing the overall performance and intelligence level of the entire intelligent recognition and management system.

[0028] Finally, based on geographical location, all the cropped frames were meticulously grouped. Each location was an independent data unit, aggregating road condition information collected at different times in that area.

[0029] For data within each location group, the analysis and processing module employs advanced statistical analysis algorithms to calculate the proportion of each type of target event within the group. This quantitative analysis method transforms previously scattered and disordered data into a clear and easily interpretable whole. By comparing the proportions of different categories of target events, the system can quickly pinpoint the dominant core issue at a specific location. If, in a certain road section, the proportion of road cracks is significantly higher than the other two types of issues, then the primary hidden danger for that section is clearly road surface damage, requiring priority for repair resources. In this way, urban management departments can clearly identify the key management directions for each area, avoiding the previous situation where work priorities were unclear due to a mix of multiple issues. Managers no longer need to hesitate among numerous problems but can instead develop targeted solutions and rationally allocate human, material, and financial resources based on the system's analysis results.

[0030] Meanwhile, to further improve system efficiency and resource utilization, the analysis and processing module also features an intelligent feedback mechanism. Once the core issue at a location is identified, the system immediately sends the analysis results to the drone terminals responsible for that area. Upon receiving the instruction, the drones automatically adjust their operating mode and cease repeatedly uploading data to that location. This measure has dual significance: firstly, it effectively avoids redundant data collection at the same location, preventing a large influx of invalid data into the system, thereby further saving valuable network bandwidth and cloud storage space; secondly, it creates conditions for collaborative operations among multiple drones. Multiple drones can conduct monitoring work in different areas in an orderly manner according to their respective task assignments, without interfering with each other while cooperating, greatly improving the overall monitoring synergy and coverage.

[0031] It's worth noting that the cloud, acting as the "brain" of the entire system, comprehensively records all collected data, as well as the analysis and processing procedures and results. These data resources not only provide strong support for current management decisions but also possess powerful traceability capabilities. Whether reviewing historical data to understand the development trend of a problem or evaluating the effectiveness of handling specific events, both can be easily achieved through the cloud-based data query system.

[0032] In summary, this solution, driven by intelligence and automation, overturns the inefficiencies and resource waste inherent in traditional urban management models. By introducing advanced technologies and algorithms, this solution achieves an intelligent upgrade of urban management. It can accurately and in real-time identify problems on urban roads and respond quickly; guide the targeted allocation of management resources to ensure maximum effectiveness of every resource; and transform urban management from passively responding to emergencies to proactively anticipating potential risks and taking preventative measures, significantly improving the safety and stability of urban operations. This innovative management model not only enhances the overall efficiency of problem detection and resolution but also injects new vitality into the sustainable development of cities, effectively supporting the improvement of urban management and the advancement of modernization.

[0033] In another preferred embodiment of the present invention, the method for preprocessing video stream data acquired in real time during flight based on airborne AI specifically includes: Obtain the total number M of target events detected by the UAV while flying along a preset route, and preset the original detection period T and reference value M. s Calculate the detection period t = T × M s / M; The video stream obtained after the drone flies along a preset route is acquired. Frame sequences are extracted at each detection period t and recorded as valid frames. Valid frames are retained and invalid frames are discarded.

[0034] It is worth noting that this video stream preprocessing method based on airborne AI achieves accurate sampling by dynamically calculating the detection cycle, intelligently matching the data collection rhythm with the actual road conditions. When target events occur frequently, the sampling frequency is automatically increased to capture details, while the interval is extended in sparse areas to reduce redundancy, thus avoiding the omission of key information and eliminating the accumulation of invalid data. The effective frame sequences extracted according to the optimized cycle constitute a high-purity dataset, significantly reducing transmission bandwidth consumption and cloud storage pressure, while maintaining spatiotemporal continuity to provide structured support for subsequent analysis. Through a quantitative model, resources are allocated in a tiered manner, and an interpretable decision-making closed loop is constructed under the premise of ensuring the integrity of monitoring. Ultimately, an efficient link from data refinement to intelligent judgment is formed, providing an economical and reliable technical foundation for the accurate identification and targeted management of urban road problems.

[0035] In another preferred embodiment of the present invention, the method for identifying target events appearing in a video stream includes: Get the number of times A of the target event appears in the video stream. If A≠0, the recognition is successful; if A=0, the recognition fails and subsequent steps are omitted.

[0036] Understandably, the recognition result is determined by counting the occurrences of the target event in the video stream. If at least one target event is detected, the recognition is considered successful; if no target event is detected, the recognition is considered a failure and the subsequent process is terminated. This solution uses counting as the core basis to achieve rapid filtering and decision-making, simplifying the processing logic in complex scenarios.

[0037] In another preferred embodiment of the present invention, the method for marking target pixels based on the three primary color channels includes: Calculate the color value of the pixel SC=2 16 ×R+2 8 ×G+B, retrieves the maximum color value (SC) of pixels corresponding to the same type of target event. max and minimum value SC min Generate color value range [SC] min SC max If the color value SC is within the color value range [SC], min SC max In the image, the corresponding pixel is marked as the target pixel.

[0038] It is important to note that target pixel recognition is achieved by constructing a quantization model based on the three primary colors. First, the comprehensive color value of each pixel is calculated according to a specific weighting formula. Then, the maximum / minimum values ​​of similar target events are extracted to form a dynamic threshold range [SC]. min SC max The system ultimately identifies pixels falling within this range as target points. This effectively separates target features from background interference, providing a quantifiable standard for accurate positioning.

[0039] In a preferred embodiment, the number of bytes B used to display the three primary color channels is obtained. If B ≠ 3, the color value calculation formula is modified to SC = (2... B / 3×16 ×R)+(2 B / 3×8 ×G)+B.

[0040] It should be noted that when the number of bytes in a channel is not 3, a modified weighted algorithm is used to accurately extract color features from images with different bit depths, ensuring compatibility and consistency of calculations across different data formats.

[0041] In another preferred embodiment of the present invention, the method for cropping keyframes based on the maximum and minimum values ​​of the horizontal and vertical coordinates to obtain cropped frames includes: The coordinates (X, Y) of the cropped frame pixels are determined according to constraints. Crop the keyframes.

[0042] Understandably, using coordinate extreme values ​​to accurately locate the target area and eliminate irrelevant background can achieve data lightweighting and focus enhancement, thereby improving the efficiency of subsequent analysis.

[0043] In another preferred embodiment of the present invention, if the same group of cropped frames corresponds to only one target event, then the subsequent steps are stopped, and the urban management problem occurring at the location corresponding to the same group of cropped frames is directly determined as the corresponding target event.

[0044] It is worth noting that when the cropped frames in the same group are associated with only a single target event, the process is terminated immediately and the nature of the event is accurately determined, avoiding redundant analysis that consumes computing resources; the processing efficiency is improved through a rapid convergence mechanism to ensure the timeliness of problem location; interference from multiple events is eliminated, making the judgment results more certain; the decision-making chain is simplified, making it easier for the management system to directly generate targeted handling plans; at the same time, the workload of manual review is reduced, realizing automated closed-loop management and effectively improving the response speed and resource utilization rate of urban governance.

[0045] In another preferred embodiment of the present invention, if the number of target events contained in the cropped frame is not unique, the number N of the i-th type of target events is counted independently when calculating their respective proportions. i .

[0046] It is worth noting that refined statistics are implemented for scenarios with multiple events stored together. When the cropped frame contains multiple target events, N is counted independently for each type. i The proportions are calculated separately to ensure fine data granularity. This approach avoids category confusion while retaining complete information, providing a quantitative basis for priority assessment under complex road conditions and improving the scientific nature of management decisions.

[0047] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the present invention should still fall within the scope of the present invention.

Claims

1. A cloud-edge collaboration based scene adaptive urban management platform, characterized in that, include: Image acquisition module: The drone flies along a preset route, and the onboard camera equipment is activated simultaneously. Based on the onboard AI, the video stream data acquired in real time during the flight is preprocessed. The processed video stream is input into a preset algorithm model to identify target events appearing in the video stream. If the identification is successful, the frame sequence corresponding to the target event in the video stream is extracted and recorded as a key frame. The target events represent road cracks, illegal road occupation, and water accumulation. Keyframe cropping module: Obtains the three primary color values ​​of pixels in the keyframe, wherein the three primary color channels include red channel value R, green channel value G and blue channel value B, and marks the target pixel based on the three primary color channels; Construct a planar coordinate system with the preset pixel as the origin, obtain the coordinates (x, y) of the target pixel, and filter out the maximum value x of the target pixel's x-coordinate. max With minimum value x min The maximum value of the y-axis max With minimum value y min The cropped frames are obtained by cropping keyframes based on the maximum and minimum values ​​of the horizontal and vertical coordinates. Analysis and processing module: Uploads the cropped frames to the cloud, obtains the upload time and the location of the cropped frames, and divides the cropped frames belonging to the same location into the same group of cropped frames; Obtain the number N of the i-th type of target events corresponding to the cropped frames in the same group. i Calculate the proportion B i =N i / N, where N represents the number of cropped frames contained in the same cropped frame group; Get the maximum percentage B max The urban management problems that occurred at the locations corresponding to the cropped frames in the same group accounted for the largest proportion, B. max The corresponding target event will send the above results to each drone terminal, and the drone will then stop uploading keyframes for the locations corresponding to the cropped frames in the same group. 2.The cloud-edge collaborative scene adaptive urban management platform according to claim 1, wherein, The specific methods for preprocessing real-time video stream data acquired during flight based on airborne AI include: Obtain the total number M of target events detected by the UAV while flying along a preset route, and preset the original detection period T and reference value M. s Calculate the detection period t = T × M s / M; The video stream obtained after the drone flies along a preset route is acquired. Frame sequences are extracted at each detection period t and recorded as valid frames. Valid frames are retained and invalid frames are discarded.

3. The scene-adaptive urban management platform based on cloud-edge collaboration according to claim 1, characterized in that, Methods for identifying target events in video streams include: Get the number of times A of the target event appears in the video stream. If A≠0, the recognition is successful; if A=0, the recognition fails and subsequent steps are omitted.

4. The scene-adaptive urban management platform based on cloud-edge collaboration according to claim 1, characterized in that, Methods for marking target pixels based on the three primary color channels include: Calculate the color value of the pixel SC=2 16 ×R+2 8 ×G+B, retrieves the maximum color value (SC) of pixels corresponding to the same type of target event. max and minimum value SC min Generate color value range [SC] min SC max If the color value SC is within the color value range [SC], min SC max In the image, the corresponding pixel is marked as the target pixel.

5. The cloud-edge collaborative scene-adaptive urban management platform according to claim 4, characterized in that, The number of bytes B for displaying three primary color channels is obtained, and if B≠3, the formula for calculating the color value is modified to SC=(2 B / 3×16 ×R)+(2 B / 3×8 ×G)+B. 6.The cloud-edge collaborative scene adaptive urban management platform of claim 1, wherein, Methods for obtaining cropped frames by cropping keyframes based on the maximum and minimum values ​​of the horizontal and vertical coordinates include: Crop the frame pixel point coordinate (X, Y) according to the constraint condition Crop the key frame.

7. The cloud-edge collaborative scene-adaptive urban management platform according to claim 1, wherein, If a cropped frame in the same group corresponds to only one target event, then stop the subsequent steps and directly determine the urban management problem that occurs at the location corresponding to the cropped frame in the same group as the corresponding target event. 8.The cloud-edge collaborative scene adaptive urban management platform of claim 1, wherein, If the number of target events contained in the cropped frame is not unique, the number N of the i-th target event is independently counted when calculating the respective proportion i .

Images