Real-time quality detection method and system based on 5G edge calculation

Abstract

The invention relates to a real-time quality detection method and system based on 5G edge calculation. The method comprises the steps that dynamic video stream data of real-time video monitoring are acquired, video frame low-delay feature extraction is carried out on the dynamic video stream data, video frame feature extraction data are obtained, and the video frame feature extraction data comprise definition indexes, color distribution features and brightness fluctuation features; generating a corresponding video frame quality feature matrix according to the definition index, the color distribution feature and the brightness fluctuation feature, and constructing a dynamic quality feature sequence based on timestamp information; and performing real-time difference analysis on the quality feature sequences of the adjacent video frames through the edge nodes to obtain a quality correlation coefficient. According to the invention, through cooperation of the 5G network and edge calculation, video quality low-delay detection and anomaly positioning are realized. The edge nodes extract and compress features such as definition, color, brightness and the like of video frames, thereby reducing bandwidth occupation and reducing delay.

Description

Technical Field

[0001] The present application relates to the technical field of data processing, and particularly to a real-time quality detection method and system based on 5G edge computing. Background Art

[0002] In the current trend of digital production and intelligent monitoring, the real-time quality detection method and system based on 5G edge computing have emerged as the key technical means to ensure product quality and improve production efficiency. This technology combines the excellent characteristics of 5G communication and the local processing advantages of edge computing, aiming to achieve real-time and accurate quality detection for various production processes and video monitoring scenarios.

[0003] Existing quality detection methods mostly independently evaluate based on single-frame pixel features (such as clarity index), lacking the ability to model the temporal changes of video quality. When facing gradual anomalies such as slow lens fouling and periodic light source flickering, single-frame analysis cannot capture them, and the miss detection rate will increase at this time.

[0004] However, the above-mentioned coordination control scheme determination models provided in the related technologies all have limitations, resulting in poor regulation effects of the determined coordination control schemes and having room for further improvement. Summary of the Invention

[0005] The present application provides a real-time quality detection method and system based on 5G edge computing to at least solve the problems in the related technologies.

[0006] In a first aspect, the present application provides a real-time quality detection method based on 5G edge computing, and the method includes: Obtain dynamic video stream data of real-time video monitoring, and perform video frame low-latency feature extraction on the dynamic video stream data to obtain video frame feature extraction data, where the video frame feature extraction data includes a clarity index, a color distribution feature, and a brightness fluctuation feature; Generate a corresponding video frame quality feature matrix according to the clarity index, the color distribution feature, and the brightness fluctuation feature, and construct a dynamic quality feature sequence based on timestamp information; Perform real-time difference analysis on the quality feature sequences of adjacent video frames through an edge node to obtain a quality correlation coefficient, and obtain an influence coefficient of adjacent video frames based on the quality correlation coefficient; Construct a real-time quality association graph based on the influence coefficient, and obtain video frame data to be evaluated according to the real-time quality association graph, where the video frame data to be evaluated includes a frame marker sequence and video frame type information; Obtain a video frame evaluation coefficient according to the frame marker sequence and the video frame type information, and determine whether the abnormal video frame evaluation coefficient is less than a preset threshold; If it is less than or equal to, it is determined that the video frame data to be evaluated is normal; If it is greater than, it is determined that the video frame data to be evaluated is abnormal.

[0007] Optionally, the step of performing video frame low-latency feature extraction on the dynamic video stream data to obtain video frame feature extraction data includes: Obtain the resolution, frame rate, and contrast data of consecutive frames according to the real-time video stream, and obtain the calculated clarity index based on the resolution, frame rate, and contrast data; Perform RGB to HSV color space conversion according to the color distribution characteristics of the real-time video stream, and obtain the color temperature deviation value and the saturation fluctuation range; Generate a color distribution feature vector based on the color temperature deviation value and the saturation fluctuation range; Obtain the average brightness value and the maximum brightness value of the video frame according to the real-time video stream, and obtain the brightness fluctuation characteristics according to the average brightness value and the maximum brightness value.

[0008] Optionally, the step of generating a corresponding video frame quality feature matrix based on the clarity index, color distribution characteristics, and brightness fluctuation characteristics, and constructing a dynamic quality feature sequence based on the timestamp information includes: Assign a unique timestamp to each video frame, and sort the quality feature matrix in timestamp order; Obtain the time interval between adjacent frames, and perform a sliding window process on the quality feature matrix based on the time interval to generate a local quality feature subsequence; Calculate the mean, variance, and change rate for each local quality feature subsequence to obtain a dynamic quality feature vector; Combine the dynamic quality feature vector with the original quality feature matrix to generate a dynamic quality feature sequence; Perform a discrete Fourier transform on the dynamic quality feature sequence to extract periodic quality fluctuation characteristics; Fuse the periodic quality fluctuation characteristics with the dynamic quality feature sequence to form a complete dynamic quality feature sequence.

[0009] Optionally, the step of performing real-time difference analysis on the quality feature sequences of adjacent video frames through an edge node to obtain a quality correlation coefficient, and obtaining an influence coefficient of adjacent video frames based on the quality correlation coefficient includes: Perform a pre-trained difference analysis on adjacent video frames in the edge node to obtain initialized model parameters; Perform a frame-by-frame difference operation on the quality feature sequences of adjacent video frames to generate a feature difference matrix; Obtain the relative change rate of each feature item based on the feature difference matrix, and obtain the probability distribution of each feature item; Obtain the characteristic abnormal entropy value according to the probability distribution; Generate a quality correlation coefficient matrix according to the initialized model parameters and the characteristic abnormal entropy value; Perform normalization processing on the quality correlation coefficient matrix to obtain the influence coefficients of adjacent video frames.

[0010] Optionally, the step of constructing a real-time quality correlation graph based on the influence coefficients includes: Represent each video frame as a node, and represent the influence coefficients of adjacent video frames as edge weights; Construct an initial correlation graph based on the edge weights, and mark the positions of abnormal video frame sequences of each edge; Obtain the correlation path of each abnormal video frame by according to the positions of the abnormal video frame sequences; Obtain the total delay sum for each correlation path, obtain the path with the minimum propagation delay, and use it as the critical path; Generate a real-time quality correlation graph through topological structure analysis according to the critical path.

[0011] Optionally, the step of obtaining the video frame data to be evaluated according to the real-time quality correlation graph includes: Obtain the video frames of the critical path according to the real-time quality correlation graph, and record them as candidate frames to be evaluated; Parse the original annotation data of the candidate frames to be evaluated to obtain a frame label sequence; Obtain the shooting scene, device type, and content features of the video frame, and generate video frame type information according to the shooting scene, device type, and content features of the video frame; Perform local sliding window processing on the dynamic quality feature sequence of the candidate frames to be evaluated, and extract time correlation features; Combine the confidence values of the frame label sequence to generate a label confidence feature vector; Fuse the time correlation features and the label confidence feature vector to construct a dynamic feature correlation matrix; Obtain the video frame data to be evaluated according to the video frame type information and the dynamic feature correlation matrix.

[0012] Optionally, the step of obtaining the video frame evaluation coefficient according to the frame label sequence and the video frame type information includes: Obtain the label information from the frame label sequence, where the label information includes normal labels, suspicious labels, and abnormal labels; Classify and encode the video frame type information to generate a type feature vector; Construct an evaluation input matrix based on the label information and the type feature vector; Generate a preliminary evaluation coefficient according to the evaluation input matrix; Obtain the final video frame evaluation coefficient according to the preliminary evaluation coefficient.

[0013] In a second aspect, the present application provides a real-time quality detection system based on 5G edge computing, and the system includes: A video processing module, configured to obtain dynamic video stream data of real-time video monitoring, and perform video frame low-latency feature extraction on the dynamic video stream data to obtain video frame feature extraction data, where the video frame feature extraction data includes a clarity index, a color distribution feature, and a brightness fluctuation feature; A sequence construction module, configured to generate a corresponding video frame quality feature matrix according to the clarity index, the color distribution feature, and the brightness fluctuation feature, and construct a dynamic quality feature sequence based on timestamp information; A difference analysis module, configured to perform real-time difference analysis on the quality feature sequences of adjacent video frames through an edge node to obtain a quality correlation coefficient, and obtain an influence coefficient of adjacent video frames based on the quality correlation coefficient; An association graph construction module, configured to construct a real-time quality association graph based on the influence coefficient, and obtain video frame data to be evaluated according to the real-time quality association graph, where the video frame data to be evaluated includes a frame marker sequence and video frame type information; An evaluation decision module, configured to obtain a video frame evaluation coefficient according to the frame marker sequence and the video frame type information, and determine whether the abnormal video frame evaluation coefficient is less than a preset threshold; If it is less than or equal to, it is determined that the video frame data to be evaluated is normal; If it is greater than, it is determined that the video frame data to be evaluated is abnormal.

[0014] Preferably, the sequence construction module includes: A clarity extraction unit, configured to obtain resolution, frame rate, and contrast data of consecutive frames according to a real-time video stream, and obtain a calculated clarity index based on the resolution, frame rate, and contrast data; A color analysis unit, configured to perform RGB to HSV color space conversion according to the color distribution feature of the real-time video stream, and obtain a color temperature deviation value and a saturation fluctuation range; A feature generation unit, configured to generate a color distribution feature vector based on the color temperature deviation value and the saturation fluctuation range; A brightness analysis unit, configured to obtain an average brightness value and a maximum brightness value of a video frame according to the real-time video stream, and obtain a brightness fluctuation feature according to the average brightness value and the maximum brightness value.

[0015] In a third aspect, the present application provides a computer-readable storage medium, on which a program is stored. When the program is executed by a processor, the real-time quality detection method based on 5G edge computing provided in the first aspect above is implemented.

[0016] Compared with the related art, the present application provides at least the following technical effects: Through the collaboration of 5G network and edge computing, low-latency detection and anomaly localization of video quality are realized. The edge node extracts and compresses features such as clarity, color, and brightness of video frames, reducing bandwidth occupancy and latency. By using multi-dimensional fusion such as dynamic quality feature sequences and real-time quality correlation graphs, the accuracy of anomaly determination is improved. With the help of graph theory to construct a correlation graph and combined with critical path analysis, the anomaly source localization time is shortened. Through the video frame type information, scene adaptive evaluation is realized, and the threshold is dynamically adjusted to adapt to different environments.

[0017] Details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more concise and understandable. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present application and form a part of the present application. The schematic embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation of the present application. In the drawings: Figure 1 It is a schematic flowchart of the method according to an embodiment of the present invention.

[0019] Figure 2 It is a schematic structural diagram of the system according to an embodiment of the present invention.

[0020] Figure 3 It is a schematic internal structure diagram of a computer device according to an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] In order to make the objectives, technical solutions, and advantages of the present application clearer, the present application will be described and explained below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application. Based on the embodiments provided in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present application.

[0022] Obviously, the accompanying drawings in the following description are only some examples or embodiments of the present application. For those of ordinary skill in the art, without creative efforts, the present application can also be applied to other similar scenarios based on these drawings. In addition, it can also be understood that although the efforts made in such a development process may be complex and lengthy, for those of ordinary skill in the art related to the content disclosed in the present application, some design, manufacturing, or production changes based on the technical content disclosed in the present application are only conventional technical means and should not be understood as the content disclosed in the present application being insufficient.

[0023] When the term "embodiment" is mentioned in the present application, it means that the specific features, structures, or characteristics described in connection with the embodiment can be included in at least one embodiment of the present application. The occurrence of this phrase at various positions in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. It is explicitly and implicitly understood by those of ordinary skill in the art that the embodiments described in the present application can be combined with other embodiments without conflict.

[0024] Unless otherwise defined, the technical terms or scientific terms involved in the present application should have the ordinary meaning understood by those of ordinary skill in the technical field to which the present application belongs. The terms "a", "an", "one kind", "the", and other similar words involved in the present application do not represent a quantity limitation and can represent a singular or plural number. The terms "including", "comprising", "having", and any variations thereof involved in the present application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include unlisted steps or units, or may also include other steps or units inherent to these processes, methods, products, or devices. The terms "connected", "coupled", and other similar words involved in the present application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The term "plurality" involved in the present application refers to two or more. "And / or" describes the association relationship of associated objects and indicates that three relationships may exist. For example, "A and / or B" may represent: A exists alone, A and B exist simultaneously, and B exists alone. The character " / " generally represents an "or" relationship between the associated objects before and after. The terms "first", "second", "third", etc. involved in the present application are only used to distinguish similar objects and do not represent a specific order for the objects.

[0025] Embodiment 1 The embodiment of the present invention provides a real-time quality detection method based on 5G edge computing. Figure 1 It is a flowchart shown according to an exemplary embodiment. It includes: S1. Obtain the dynamic video stream data of real-time video monitoring, and perform low-latency feature extraction on the dynamic video stream data to obtain video frame feature extraction data, where the video frame feature extraction data includes clarity index, color distribution feature, and brightness fluctuation feature; S2. Generate a corresponding video frame quality feature matrix according to the clarity index, color distribution feature, and brightness fluctuation feature, and construct a dynamic quality feature sequence based on the timestamp information; S3. Perform real-time difference analysis on the quality feature sequences of adjacent video frames through edge nodes to obtain quality correlation coefficients, and obtain influence coefficients of adjacent video frames based on the quality correlation coefficients; S4. Construct a real-time quality association graph based on the influence coefficients, and obtain video frame data to be evaluated according to the real-time quality association graph, where the video frame data to be evaluated includes a frame marker sequence and video frame type information; S5. Obtain a video frame evaluation coefficient according to the frame marker sequence and video frame type information, and determine whether the abnormal video frame evaluation coefficient is less than a preset threshold; If it is less than or equal to, it is determined that the video frame data to be evaluated is normal; If it is greater than, it is determined that the video frame data to be evaluated is abnormal.

[0026] As described in S1 - S5 above, the present invention obtains dynamic video stream data in real time through a 5G network, and uses a low-latency algorithm to extract features of video frames. Edge nodes quickly analyze parameters such as the resolution, frame rate, and contrast of video frames through a lightweight model (such as a convolutional neural network), and calculate the clarity index by combining frequency domain analysis; at the same time, extract color distribution features (such as color temperature deviation, saturation fluctuation) through the conversion between RGB and HSV color spaces, and calculate the brightness fluctuation feature by the brightness mean and maximum value. This method reduces the feature dimension through normalization processing and principal component analysis (PCA) to generate a standardized quality feature matrix. Compared with traditional cloud centralized processing, the local feature extraction of edge nodes significantly reduces the transmission delay and ensures the real-time performance of video quality evaluation. For example, in an industrial monitoring scenario, when the camera lens is damaged and the picture is blurred, the sudden drop in the clarity index can be captured in real time to avoid misjudgment in subsequent defect detection.

[0027] Then, a dynamic quality feature sequence is constructed through timestamp information to capture the temporal change trend of the video stream. Specifically, the edge node assigns millisecond-level timestamps to each frame, and based on the sliding window technique, local analysis is performed on the feature matrices of adjacent frames to calculate the mean, variance, and change rate, generating dynamic feature vectors. The periodic fluctuation features are further extracted through the discrete Fourier transform (DFT). Finally, the static features and dynamic features are fused to form a complete dynamic quality feature sequence. This solves the problem that traditional static feature analysis cannot reflect the quality change trend, and is especially suitable for detecting brightness fluctuations caused by environmental light changes in night monitoring. For example, when the light source on the factory production line flickers periodically due to unstable voltage, the dynamic feature sequence can accurately capture the periodic pattern of the brightness fluctuation, providing a reliable basis for subsequent anomaly determination.

[0028] After that, real-time difference analysis is performed on the dynamic quality feature sequences of adjacent frames by the edge node to quantify the quality degradation relationship between frames. In the specific implementation, frame-by-frame difference operation is used to generate a feature difference matrix, and the feature anomaly entropy value is calculated through probability distribution. Combining the time interval, a quality correlation coefficient is generated. The network bandwidth utilization rate and transmission delay duration are further introduced, and an influence coefficient is generated through weighted fusion. This method reduces the dependence on the cloud through local difference analysis. For example, in a 5G edge server, the single-frame processing time can be less than 5 ms, significantly improving the real-time performance.

[0029] Next, a real-time quality association graph is constructed based on the influence coefficient, and high-risk video frames are screened. Specifically, each frame is represented as a node, and the influence coefficient is used as the edge weight. The abnormal propagation path is identified through breadth-first search. Combining the label sequence (normal / suspicious / abnormal) and video frame type information (such as industrial monitoring frames, live broadcast frames), metadata of the frame to be evaluated is generated. High-risk frames are screened through the dynamic feature association matrix, and the abnormal confidence score is calculated. This method accurately identifies the abnormal propagation chain through graph theory optimization.

[0030] Finally, by fusing the label sequence, type information, and dynamic feature vectors, a random forest model is used to calculate the video frame evaluation coefficient, and it is compared with a preset threshold to determine anomalies. In the specific implementation, the dynamic threshold mechanism calculates the rolling mean and standard deviation based on historical data, and adaptively adjusts the threshold range to avoid false alarms. For example, the threshold is lower when the daylight is stable during the day, and automatically increases when the fluctuations are larger at night.

[0031] In one embodiment, the step of performing low-latency feature extraction on the dynamic video stream data to obtain video frame feature extraction data includes: S101. Obtain the resolution, frame rate, and contrast data of consecutive frames according to the real-time video stream, and obtain a calculated clarity index based on the resolution, frame rate, and contrast data; S102. Perform RGB to HSV color space conversion according to the color distribution characteristics of the real-time video stream, and obtain the color temperature deviation value and the saturation fluctuation range; S103. Generate a color distribution feature vector based on the color temperature deviation value and the saturation fluctuation range; S104. Obtain the average brightness value and the maximum brightness value of the video frame according to the real-time video stream, and obtain the brightness fluctuation feature according to the average brightness value and the maximum brightness value.

[0032] In summary, the present invention realizes the quantitative analysis of the clarity of video frames by extracting the resolution, frame rate and contrast data of the video stream. Obtain basic parameters such as resolution and frame rate from the video stream metadata, use the Canny edge detection operator to process the grayscale matrix of the video frame, obtain the edge intensity by calculating the square root of the sum of the squares of the gradients in the x and y directions, and then count the edge pixel density per unit area to generate a clarity index. Characterize the image clarity by calculating the edge density on the edge side locally, which can effectively identify fuzzy abnormalities such as lens contamination and out-of-focus.

[0033] After that, to achieve accurate identification of video color anomalies, convert the RGB color space to the HSV space, separate the brightness and color components to calculate the color temperature deviation and saturation fluctuation. Obtain the hue, saturation and brightness components through the HSV conversion formula, convert the hue value to the color temperature and compare it with the standard color temperature to obtain the deviation value, and at the same time calculate the standard deviation of the saturation values of multiple consecutive frames to characterize the fluctuation range. Solve the problem of delay in color analysis relying on cloud processing in traditional methods. Through real-time conversion and statistics on the edge side, color mutations caused by color cast due to light source changes or filter failures can be captured.

[0034] Then, to achieve the structured expression of color features, normalize the color temperature deviation and saturation fluctuation and then generate a feature vector. Eliminate the dimensional differences of different features through the Z-score standardization method, combine the standardized parameters into a two-dimensional feature vector as a quantitative representation of color anomalies. This operation solves the problem of difficult analysis caused by the unstructured traditional features. It is convenient for subsequent machine learning model training and matching with historical anomaly samples in vector form. The dimension of the feature vector is compressed to two dimensions, which reduces the data volume while retaining key color information, supports fast matching of anomaly patterns through the Euclidean distance, provides a standardized data basis for cross-frame color anomaly correlation analysis, and improves the edge-side feature storage and transmission efficiency.

[0035] Finally, brightness anomaly detection is achieved by calculating the average brightness and maximum brightness of the video frame and combining it with sliding window analysis. The average brightness and maximum brightness values are counted from the pixel data, and the brightness change rate is calculated using a 20-frame sliding window to capture brightness anomalies such as lens occlusion and sudden changes in light sources. This method solves the problem that traditional single-point brightness detection cannot identify timing changes, and improves the detection capability of gradual or sudden brightness anomalies through timing analysis. The brightness detection range covers the full dynamic range and can identify slight brightness changes. The response time to scenes such as sudden strong light interference meets the requirements of real-time monitoring, and the false alarm rate is controlled within the acceptable range of the industry.

[0036] In one embodiment, the step of generating a corresponding video frame quality feature matrix according to the clarity index, color distribution feature and brightness fluctuation feature, and constructing a dynamic quality feature sequence based on timestamp information includes: S201, assigning a unique timestamp to each frame of video, and sorting the quality feature matrix according to the order of the timestamps; S202, obtaining the time interval between adjacent frames, and performing sliding window processing on the quality feature matrix based on the time interval to generate a local quality feature subsequence; S203, calculating the mean, variance and change rate of each local quality feature subsequence to obtain a dynamic quality feature vector; S204, combining the dynamic quality feature vector with the original quality feature matrix to generate a dynamic quality feature sequence; Perform discrete Fourier transform on the dynamic quality feature sequence to extract periodic quality fluctuation characteristics; S205. Merge the periodic quality fluctuation characteristics with the dynamic quality characteristic sequence to form a complete dynamic quality characteristic sequence.

[0037] In summary, the present invention establishes a temporal correlation basis for video frames by assigning a unique timestamp to each frame of video and arranging the quality feature matrix in order. The timestamp is extracted from the video stream, a time identifier is generated for each frame, and then the single-frame quality feature matrix (including clarity, color, and brightness features) is arranged in ascending order by the timestamp to form an ordered temporal matrix. The problem of lack of time dimension in traditional single-frame analysis is solved, so that subsequent analysis can capture the temporal changes in video quality. The timestamp synchronization error is controlled within the industry standard range to ensure the temporal consistency of multi-frame data and provide an ordered data structure for sliding window processing and cross-frame difference analysis.

[0038] Then, a sliding window process is performed on the quality feature matrix based on the adjacent frame time interval. The window parameters are determined by calculating the frame interval, and local quality feature subsequences are generated using a fixed window size and an overlapping step length. For example, the feature matrix is intercepted with an appropriate window size and step length to form a set of continuous overlapping subsequences. This method simulates the short-term memory characteristics of human vision and can effectively identify short-term anomalies such as sudden brightness changes caused by sudden occlusions.

[0039] After that, by calculating statistics for each local quality feature subsequence, a dynamic feature vector representing short-term quality changes is generated. Specifically, the mean, variance, and change rate are calculated for each subsequence, which respectively reflect the short-term quality level, feature stability, and quality fluctuation speed. The three are combined into a dynamic feature vector. This vector extends the single-frame feature to a short-term dynamic feature, effectively quantifying the quality degradation process. For example, the slow decline trend of clarity can be identified through the change rate, and the severity of brightness fluctuations can be judged using variance analysis. The calculation of statistics combines feature aggregation in the time dimension, solving the problem that traditional single-frame analysis cannot reflect the quality change trend, enabling edge nodes to complete dynamic modeling of short-term features locally.

[0040] Next, the dynamic feature vector is combined with the original quality feature matrix to generate a spatio-temporal feature sequence, and then a discrete Fourier transform is performed. Through the Fourier transform, the time-domain features are converted into a frequency-domain representation to extract periodic fluctuation features, such as identifying periodic vibration blur caused by the operation of industrial equipment. The spatio-temporal feature sequence retains both single-frame details and short-term fluctuations, while frequency-domain analysis further analyzes long-term patterns. The combination of the two solves the problem of insufficient recognition ability of traditional methods for regular interference sources.

[0041] Finally, by screening effective frequency-domain features and fusing them with the dynamic feature sequence, a complete dynamic quality feature sequence is formed. First, a frequency threshold is set to extract meaningful periodic features, and then the frequency-domain features are converted into a time-domain representation and fused with the original sequence, so that the final sequence contains both short-term fluctuations and long-term patterns. This fusion operation solves the limitations of single-time-domain or frequency-domain analysis. For example, it can simultaneously reflect different abnormal patterns such as "current frame blur" and "blur once per hour". The incorporation of periodic features significantly reduces the false negative rate of the system for regular anomalies, and the complete dynamic feature sequence provides multi-dimensional feature support for subsequent construction of a real-time quality correlation graph.

[0042] In one embodiment, the step of performing real-time difference analysis on the quality feature sequences of adjacent video frames by an edge node to obtain a quality correlation coefficient, and obtaining an influence coefficient of adjacent video frames based on the quality correlation coefficient includes: S301. Perform pre-trained difference analysis on adjacent video frames in the edge node to obtain initialized model parameters; S302. Perform frame-by-frame difference operation on the quality feature sequences of adjacent video frames to generate a feature difference matrix; S303. Obtain the relative change rate of each feature item based on the feature difference matrix, and obtain the probability distribution of each feature item; S304. Obtain the feature anomaly entropy value according to the probability distribution; S305. Generate a quality correlation coefficient matrix according to the initialized model parameters and the feature anomaly entropy value; S306. Perform normalization processing on the quality correlation coefficient matrix to obtain the influence coefficient of adjacent video frames.

[0043] In summary, the present invention reduces cloud dependence and improves real-time performance by locally deploying a lightweight model at edge nodes. In specific implementation, the edge node loads a pre-trained convolutional neural network (CNN), which includes three 3×3 convolutional layers, and the number of channels changes from 64→128→256, and the output feature dimension is 256. The model parameters are initialized by the Xavier initialization method to avoid the problem of gradient disappearance or explosion, and are pre-distributed to the edge node through the 5G network and stored in the NVRAM to ensure fast loading. This method reduces the computational overhead through a lightweight CNN, ensuring that the edge node completes feature extraction within a value less than the preset value.

[0044] Then, perform frame-by-frame difference operation on the quality feature sequences of adjacent frames to generate a feature difference matrix. After specifically obtaining the quality feature sequences of two adjacent frames, calculate the Euclidean distance of the feature vectors at each time point to construct a matrix reflecting the degree of feature difference between frames. The dimension of this matrix corresponds to the time series length and the feature dimension, and its element value represents the feature difference intensity between adjacent frames at the corresponding time point. In this way, the abstract quality feature change is transformed into structured matrix data, which is convenient for subsequent analysis of the temporal pattern and abnormal fluctuation of feature changes.

[0045] After that, calculate the relative change rate of each feature item and statistically analyze its probability distribution to quantify the degree and law of feature changes. For each feature dimension of the difference matrix, calculate the relative change rate of adjacent time points to reflect the temporal fluctuation of feature differences. By statistically analyzing the occurrence frequency of each change rate, a probability distribution is generated to characterize the uncertainty and regular pattern of feature changes. This process transforms the numerical difference matrix into probability statistical features, effectively describing the randomness and trend of feature changes, and solving the problem that traditional methods only focus on absolute difference values and ignore change laws.

[0046] Calculate the feature anomaly entropy value based on the probability distribution of the feature item change rate to quantify the uncertainty of feature changes. The entropy value calculation is based on the occurrence probability of each change rate in the probability distribution and is obtained through the information entropy formula. The larger the value, the higher the uncertainty of feature changes, that is, the greater the possibility of anomaly. This entropy value, as a quantitative index of the feature anomaly degree, effectively integrates the amplitude of feature changes and the characteristics of probability distribution, overcoming the limitations of traditional single-threshold judgment. Through entropy value analysis, edge nodes can evaluate the anomaly degree of feature changes from the perspective of information theory.

[0047] Then, combining the pre-trained model parameters and the feature anomaly entropy value, this step generates a quality correlation coefficient matrix to characterize the quality correlation strength between adjacent video frames. By weighted fusion of the feature difference representation extracted by the pre-trained model and the anomaly uncertainty reflected by the entropy value, a multi-dimensional correlation coefficient matrix is constructed. The matrix element values comprehensively consider the absolute strength, change law, and anomaly possibility of feature differences, forming a comprehensive quantification of the inter-frame quality correlation. This matrix not only reflects the direct feature differences between adjacent frames but also incorporates the temporal correlation patterns learned by the model, solving the problem of single-dimensional correlation analysis in traditional methods. The quality correlation coefficient matrix provides a basis for the calculation of subsequent influence coefficients, enabling edge nodes to evaluate the inter-frame quality influence from multiple dimensions and providing more accurate parameter support for anomaly propagation analysis.

[0048] Finally, normalize the quality correlation coefficient matrix to generate standardized adjacent video frame influence coefficients, which are convenient for cross-scene comparison and application. By mapping the element values of the correlation coefficient matrix to a reasonable range, the dimension difference is eliminated, forming a unified influence coefficient quantification index. The normalization process makes the influence coefficients comparable and can be directly used to construct the edge weights of the real-time quality correlation graph, supporting the analysis of anomaly propagation paths based on graph theory. This step solves the analysis difficulties caused by different dimensions of traditional correlation parameters, enabling the influence coefficients generated by edge nodes to be widely applicable to different monitoring scenarios. The normalized influence coefficients accurately reflect the quality influence degree between adjacent frames, providing a standardized quantification basis for subsequent anomaly source location and propagation path identification.

[0049] In one embodiment, the step of constructing a real-time quality correlation graph based on the influence coefficient includes: S401. Represent each video frame as a node and represent the influence coefficient between adjacent video frames as an edge weight; S402. Construct an initial correlation graph based on the edge weights and mark the positions of the abnormal video frame sequences of each edge; S403. Obtain the correlation path of each abnormal video frame by according to the position of the abnormal video frame sequence; S404. Obtain the total delay of each correlation path, obtain the path with the minimum propagation delay, and use it as the critical path; S405. Generate a real-time quality correlation graph through topological structure analysis based on the critical path.

[0050] In summary, the present invention constructs a basic graph structure for video quality correlation by abstracting video frames as graph nodes and influence coefficients as edge weights. Specifically, each video frame is mapped to a graph node, and the node attributes include timestamps and quality features. The directed edge weight between adjacent frames is set as the influence coefficient output in step S306, forming a weighted directed graph model. The graph structure is stored using an adjacency matrix to ensure efficient node query and edge weight update. This operation transforms the temporal correlation of video frames into the relationship between nodes and edges in graph theory, solving the problem that traditional temporal analysis lacks visualization and quantitative representation, and enabling the degree of quality influence between frames to be intuitively reflected through edge weights. After that, an initial correlation graph is constructed based on edge weights and historical abnormal positions are marked. Nodes and edges are added sequentially according to the video frame time sequence to form a chain-like initial graph. Then, the historical abnormal records are traversed, the timestamp positions of the edges causing the anomalies are marked, and the anomaly occurrence probability of the edges is calculated as an attribute. The initial graph reflects the inter-frame correlation under normal conditions, and the historical anomaly markings provide prior knowledge for subsequent path search, enabling the graph structure to contain both current quality correlation and historical anomaly patterns.

[0051] Then, possible abnormal correlation paths are obtained through graph traversal. First, locate the node position of the abnormal frame in the graph, and use breadth-first search to traverse backward from this node, restricting the search depth and filtering low-weight edges, only retaining paths with influence coefficients greater than the threshold. The backward search strategy conforms to the logic of abnormal traceability, inferring possible sources from the current anomaly. The depth and weight filtering balance the search efficiency and effectiveness, avoiding interference from invalid paths. It solves the problem of low efficiency in traditional manual analysis of abnormal propagation paths, automatically generates a set of possible propagation paths through algorithms, provides candidates for subsequent critical path determination, and enables the acquisition of abnormal correlation paths to be automated and real-time.

[0052] Next, the critical path is determined by calculating the propagation delay. The transmission delays of each edge are accumulated for each valid path, and the Dijkstra algorithm is used to select the path with the minimum delay as the critical path. If there are paths with equal delays, they are sorted according to the edge anomaly probability. The propagation delay reflects the actual propagation speed of the anomaly, and the path with the shortest delay is most likely to be the real propagation chain. The anomaly probability is used as an auxiliary indicator to improve reliability. It solves the problem of fuzzy path priority judgment in traditional methods, clearly defines the critical path through quantitative indicators, makes the positioning of the abnormal source more accurate, provides a reliable path basis for subsequent root cause analysis, and improves the efficiency and accuracy of anomaly handling.

[0053] Finally, in this step, a real-time quality correlation graph is generated through topology analysis. Topological features of the critical path are extracted, and the graph structure is maintained in combination with a dynamic update strategy. Newly added frames are added in sequence. When there is an abnormal update, the critical path is recalculated and visually marked. The generated graph structure includes the quality status of nodes and critical path annotations, and visually displays quality anomalies and the degree of influence through node colors and edge widths. Topology analysis refines the structural features of the critical path, and dynamic updates ensure that the graph reflects the quality correlation of the video stream in real time. The visual output facilitates the operation and maintenance personnel to quickly locate anomalies. This step solves the problem that traditional correlation analysis lacks intuitive display. The generated real-time quality correlation graph provides visual support for video quality root cause location, upgrades anomaly propagation analysis from data-driven to graph visualization-driven, and greatly improves the efficiency of anomaly handling.

[0054] In one embodiment, the step of obtaining the video frame data to be evaluated according to the real-time quality correlation graph includes: S406. Obtain the video frames of the critical path according to the real-time quality correlation graph, and record them as candidate frames to be evaluated; S407. Analyze the original annotation data of the candidate frames to be evaluated to obtain a frame marker sequence; S408. Obtain the shooting scene, device type, and content features of the video frame, and generate video frame type information according to the shooting scene, device type, and content features of the video frame; S409. Perform local sliding window processing on the dynamic quality feature sequence of the candidate frames to be evaluated to extract time correlation features; S410. Combine the confidence values of the frame marker sequence to generate a marker confidence feature vector; S411. Fuse the time correlation features and the marker confidence feature vector to construct a dynamic feature correlation matrix; S412. Obtain the video frame data to be evaluated according to the video frame type information and the dynamic feature correlation matrix.

[0055] In summary, the present invention extracts the video frames on the critical propagation path from the real-time quality correlation graph as candidate frames to be evaluated, so as to focus on high-risk anomaly propagation links. Specifically, by locating the critical path (such as the shortest delay path) in the correlation graph, all video frame nodes on the path are obtained, and the candidate frames are sorted in descending order according to the influence coefficient weights of the nodes, and the video frames with high influence coefficients are preferentially included. This operation is based on the temporal correlation of anomaly propagation, reduces the number of frames to be evaluated to a small proportion of the original video stream, solves the problem of low efficiency of traditional full-scale evaluation, enables high-risk frames to be processed first, and ensures the efficiency of anomaly traceability while greatly reducing the computational amount.

[0056] Then, by parsing the original annotation data of the candidate frame to be evaluated, read the manual annotation or historical automatic marking results from the edge node cache, and extract the marking sequence (such as statuses like normal / blurry, etc.) and the corresponding confidence scores (if any). By parsing the annotation data, clarify the historical quality status of the candidate frame, and at the same time use the confidence to quantify the reliability of the annotation, avoiding misjudgment caused by incorrect annotation. This process provides a reference basis for subsequent evaluation, ensures the effectiveness of the marking information, and improves the accuracy of the evaluation results.

[0057] Next, generate video frame type information by integrating multi-dimensional metadata to adapt to the evaluation requirements of different scenarios. Extract the shooting scenario (such as industrial production line, urban road) and device type from the video stream, use lightweight CNN to extract content features (such as moving objects, texture complexity), and convert the scenario, device, and content features into encoded vectors. The generation of type information enables the evaluation criteria to be dynamically adjusted according to the scenario (for example, higher clarity requirements in industrial scenarios), solves the problem that traditional evaluations ignore scenario differences, and improves the adaptability of evaluations in different environments.

[0058] After that, perform a sliding window process on the dynamic quality feature sequence of the candidate frame to be evaluated to extract temporal correlation features. Use a sliding window of appropriate size to traverse the feature sequence, calculate the Pearson correlation coefficient of the features within the window, generate a correlation matrix, and then obtain the feature vector through principal component analysis for dimensionality reduction. This operation effectively identifies gradual anomalies such as a gradual decrease in clarity by analyzing the feature correlation degree between adjacent frames. Compared with traditional single-frame analysis, it significantly improves the ability to capture temporal anomalies.

[0059] Then, combine the marking information with confidence quantification to generate a feature vector representing the reliability of the marking. Standardize the marking confidence values, encode the marking status (normal / suspicious / abnormal) as a vector, and fuse the two according to weights to form a comprehensive feature vector. Through standardization and encoding fusion, quantify the credibility of the annotation, avoid the misleading of low-confidence markings, solve the problem of marking uncertainty in traditional evaluations, make the representation of the marking information more comprehensive, and provide reliable marking dimension data for subsequent feature fusion.

[0060] Next, fuse the temporal correlation features and the marking confidence vector to construct a dynamic feature association matrix. After aligning the two types of feature dimensions, arrange them in chronological order to form a matrix, and set weights for different feature columns to highlight the key dimensions. The matrix structure can represent the temporal association and marking reliability of the candidate frame, support the efficient identification of cross-frame abnormal patterns, and significantly improve the identification efficiency of abnormal patterns compared with single-feature analysis, providing structured feature association data for subsequent comprehensive evaluation.

[0061] Finally, fuse the video frame type information with the dynamic feature correlation matrix to generate standardized data to be evaluated. Concatenate the type encoding vector and the feature matrix, perform standardization processing, and encapsulate them into a unified format. The fusion of multi-dimensional features enables the evaluation data to include information such as scene, device, time series, and markings, solving the problem of single-dimensional traditional evaluation features. The standardized encapsulation ensures that the data can be processed efficiently, ultimately improving the accuracy of anomaly evaluation.

[0062] In one embodiment, the step of obtaining the video frame evaluation coefficient according to the frame marking sequence and the video frame type information includes: S501. Obtain marking information from the frame marking sequence, where the marking information includes normal marking, suspicious marking, and abnormal marking; S502. Classify and encode the video frame type information to generate a type feature vector; S503. Construct an evaluation input matrix based on the marking information and the type feature vector; S504. Generate a preliminary evaluation coefficient according to the evaluation input matrix; S505. Obtain the final video frame evaluation coefficient according to the preliminary evaluation coefficient.

[0063] In summary, the present invention extracts marking information from the original annotation data of the video frame to be evaluated, including three marking states of normal, suspicious, and abnormal and the corresponding confidence scores. By analyzing the annotation data obtained in step S407, the marking state is converted into a 3D vector (for example, the normal marking corresponds to [1, 0, 0]), and at the same time, the confidence value is extracted to quantify the annotation reliability. This operation converts the historical quality evaluation results into structured features, providing a time series reference for the current frame evaluation and avoiding misjudgment caused by low-confidence annotations. Through marking state encoding and confidence filtering, historical marking information is effectively integrated, improving the reliability of the evaluation basis.

[0064] After that, classify and encode the shooting scene, device type, and content features of the video frame to generate a type feature vector. Obtain the scene (such as an industrial production line), device model (such as a certain model of Hikvision), and content features (such as the proportion of moving objects) from step S408, and convert them into feature vectors of a unified dimension through one-hot encoding and standardization processing. The type feature encoding defines the quality evaluation criteria in different scenarios. For example, industrial scenarios have higher requirements for clarity, and the device type affects the setting of the brightness threshold. This process enables the evaluation model to adapt to the requirements of different scenarios and reduces misjudgment caused by scene differences.

[0065] Then, fuse the marked information with the type feature vector to construct a multi-dimensional evaluation input matrix. Through dimension alignment processing, splice the 3D marked state vector and the N-dimensional type feature vector into a matrix of unified dimension, and set initial weights for each feature column according to historical data (for example, the abnormal marking weight is higher than the ordinary type feature). The evaluation matrix integrates the marked time series and scene attributes to form a multi-dimensional evaluation basis, and the weight distribution reflects the contribution differences of different features to the quality evaluation. This matrix structure breaks the limitations of traditional single-feature evaluation, covers more evaluation dimensions, provides comprehensive feature inputs for subsequent intelligent evaluation, and improves the accuracy and comprehensiveness of the evaluation results.

[0066] Next, use a lightweight neural network to process the evaluation input matrix to generate a preliminary evaluation coefficient. Adopt a two-layer fully connected neural network (input layer = N + 3, hidden layer = 8, output layer = 1), and through the ReLU activation function and Sigmoid mapping, convert the matrix features into preliminary coefficients in the range of [0, 1]. The neural network automatically learns the non-linear association between the marked and type features. For example, the combination of industrial scenarios and continuous abnormal markings will increase the evaluation coefficient. The model is fine-tuned and optimized through historical data, with cross-entropy as the loss function. Compared with traditional weighted summation, the recognition accuracy of abnormal frames is significantly improved, and the inference delay meets the real-time requirements.

[0067] Finally, combine the scene weight and the environmental adjustment factor to optimize the preliminary evaluation coefficient into the final evaluation coefficient. Query the scene weight table according to the video frame type information (such as the industrial production line weight is 1.2), and combine the real-time network status and the edge node load to adjust the environmental factor, and obtain the final coefficient through weighted operation. Compare the final coefficient with the dynamic threshold (such as the mean + 2 times the standard deviation) to generate the evaluation result. The scene weight realizes the adaptive adjustment of the evaluation criteria in different scenarios, and the environmental factor compensates for the influence of external factors, so that the evaluation system maintains a stable high accuracy in different scenarios.

[0068] Embodiment 2 This application also provides a real-time quality detection system based on 5G edge computing, as Figure 2 shown, including: A video processing module 1, configured to obtain dynamic video stream data of real-time video monitoring, and perform low-latency feature extraction on the dynamic video stream data to obtain video frame feature extraction data, where the video frame feature extraction data includes a clarity index, a color distribution feature, and a brightness fluctuation feature; A sequence construction module 2, configured to generate a corresponding video frame quality feature matrix according to the clarity index, the color distribution feature, and the brightness fluctuation feature, and construct a dynamic quality feature sequence based on timestamp information; A difference analysis module 3 is configured to perform real-time difference analysis on the quality feature sequences of adjacent video frames through edge nodes to obtain quality correlation coefficients, and obtain influence coefficients of adjacent video frames based on the quality correlation coefficients; An association graph construction module 4 is configured to construct a real-time quality association graph based on the influence coefficients, and obtain video frame data to be evaluated according to the real-time quality association graph, wherein the video frame data to be evaluated includes a frame marker sequence and video frame type information; An evaluation decision module 5 is configured to obtain a video frame evaluation coefficient according to the frame marker sequence and the video frame type information, and determine whether the abnormal video frame evaluation coefficient is less than a preset threshold; If it is less than or equal to, it is determined that the video frame data to be evaluated is normal; If it is greater than, it is determined that the video frame data to be evaluated is abnormal.

[0069] In one embodiment, the video processing module 1 includes: A clarity extraction unit is configured to obtain resolution, frame rate, and contrast data of consecutive frames according to a real-time video stream, and obtain a calculated clarity index based on the resolution, frame rate, and contrast data; A color analysis unit is configured to perform RGB to HSV color space conversion according to the color distribution characteristics of the real-time video stream, and obtain a color temperature deviation value and a saturation fluctuation range; A feature generation unit is configured to generate a color distribution feature vector based on the color temperature deviation value and the saturation fluctuation range; A brightness analysis unit is configured to obtain an average brightness value and a maximum brightness value of a video frame according to the real-time video stream, and obtain a brightness fluctuation feature according to the average brightness value and the maximum brightness value.

[0070] Embodiment 3 As Figure 3 shown, Embodiment 3 of the present invention provides a computer-readable storage medium, on which a program is stored, and when the program is executed by a processor, the real-time quality detection method based on 5G edge computing provided in Embodiment 1 is implemented.

[0071] The above embodiments only represent several implementation manners of the present application, and their descriptions are relatively specific and detailed, but should not be construed as limiting the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A real-time quality detection method based on 5G edge computing, characterized in that, The method includes: Obtain the dynamic video stream data of real-time video monitoring, and perform video frame low-latency feature extraction on the dynamic video stream data to obtain video frame feature extraction data, where the video frame feature extraction data includes clarity index, color distribution feature, and brightness fluctuation feature; Generate a corresponding video frame quality feature matrix according to the clarity index, color distribution feature, and brightness fluctuation feature, and construct a dynamic quality feature sequence based on the timestamp information; Perform real-time difference analysis on the quality feature sequences of adjacent video frames through an edge node to obtain a quality correlation coefficient, and obtain an influence coefficient of adjacent video frames based on the quality correlation coefficient; Construct a real-time quality association graph based on the influence coefficient, and obtain video frame data to be evaluated according to the real-time quality association graph, where the video frame data to be evaluated includes a frame marker sequence and video frame type information; Obtain a video frame evaluation coefficient according to the frame marker sequence and video frame type information, and determine whether the abnormal video frame evaluation coefficient is less than a preset threshold; If it is less than or equal to, determine that the video frame data to be evaluated is normal; If it is greater than, determine that the video frame data to be evaluated is abnormal.

2. The real-time quality detection method based on 5G edge computing according to claim 1, wherein The step of performing video frame low-latency feature extraction on the dynamic video stream data to obtain video frame feature extraction data includes: Obtain the resolution, frame rate, and contrast data of consecutive frames according to the real-time video stream, and obtain a calculated clarity index based on the resolution, frame rate, and contrast data; Perform RGB to HSV color space conversion according to the color distribution feature of the real-time video stream, and obtain the color temperature deviation value and saturation fluctuation range; Generate a color distribution feature vector based on the color temperature deviation value and saturation fluctuation range; Obtain the average brightness value and maximum brightness value of the video frame according to the real-time video stream, and obtain the brightness fluctuation feature according to the average brightness value and maximum brightness value.

3. The real-time quality detection method based on 5G edge computing according to claim 1, wherein The step of generating a corresponding video frame quality feature matrix according to the clarity index, color distribution feature, and brightness fluctuation feature, and constructing a dynamic quality feature sequence based on the timestamp information includes: Assign a unique timestamp to each video frame, and sort the quality feature matrix in timestamp order; Obtain the time interval between adjacent frames, and perform a sliding window process on the quality feature matrix based on the time interval to generate a local quality feature subsequence; Calculate the mean, variance, and change rate for each local quality feature subsequence to obtain a dynamic quality feature vector; Combine the dynamic quality feature vector with the original quality feature matrix to generate a dynamic quality feature sequence; Perform discrete Fourier transform on the dynamic quality feature sequence to extract periodic quality fluctuation features; Fuse the periodic quality fluctuation features with the dynamic quality feature sequence to form a complete dynamic quality feature sequence.

4. The real-time quality detection method based on 5G edge computing according to claim 1, characterized in that, The step of performing real-time difference analysis on the quality feature sequences of adjacent video frames through an edge node to obtain a quality correlation coefficient, and obtaining an influence coefficient of adjacent video frames based on the quality correlation coefficient includes: Perform loaded pre-trained difference analysis on adjacent video frames in the edge node to obtain initialized model parameters; Perform frame-by-frame difference operation on the quality feature sequences of adjacent video frames to generate a feature difference matrix; Obtain the relative change rate of each feature item based on the feature difference matrix, and obtain the probability distribution of each feature item; Obtain the feature anomaly entropy value according to the probability distribution; Generate a quality correlation coefficient matrix according to the initialized model parameters and the feature anomaly entropy value; Perform normalization processing on the quality correlation coefficient matrix to obtain the influence coefficient of adjacent video frames.

5. The real-time quality detection method based on 5G edge computing according to claim 1, characterized in that, The step of constructing a real-time quality correlation graph based on the influence coefficient includes: Represent each video frame as a node, and represent the influence coefficient of adjacent video frames as edge weights; Construct an initial correlation graph based on the edge weights, and mark the positions of the abnormal video frame sequences of each edge; Obtain the correlation path of each abnormal video frame by according to the position of the abnormal video frame sequence; Obtain the total delay of each correlation path, obtain the path with the minimum propagation delay, and use it as the critical path; Generate a real-time quality correlation graph through topological structure analysis according to the critical path.

6. The real-time quality detection method based on 5G edge computing according to claim 1, characterized in that The step of obtaining the video frame data to be evaluated according to the real-time quality correlation graph includes: Obtain the video frames of the critical path according to the real-time quality correlation graph, and record them as candidate frames to be evaluated; Parse the original annotation data of the candidate frames to be evaluated to obtain a frame tag sequence; Obtain the shooting scene, device type and content features of the video frame, and generate video frame type information according to the shooting scene, device type and content features of the video frame; Perform local sliding window processing on the dynamic quality feature sequence of the candidate frames to be evaluated to extract time correlation features; Combine the confidence values of the frame tag sequence to generate a tag confidence feature vector; Fuse the time correlation features and the tag confidence feature vector to construct a dynamic feature correlation matrix; Obtain the video frame data to be evaluated according to the video frame type information and the dynamic feature correlation matrix.

7. The real-time quality detection method based on 5G edge computing according to claim 1, characterized in that The step of obtaining the video frame evaluation coefficient according to the frame tag sequence and the video frame type information includes: Obtain tag information from the frame tag sequence, where the tag information includes normal tags, suspicious tags and abnormal tags; Perform classification coding on the video frame type information to generate a type feature vector; Construct an evaluation input matrix based on the tag information and the type feature vector; Generate a preliminary evaluation coefficient according to the evaluation input matrix; Obtain the final video frame evaluation coefficient according to the preliminary evaluation coefficient.

8. A real-time quality detection system based on 5G edge computing, characterized in that, The system includes: A video processing module for obtaining dynamic video stream data of real-time video monitoring and performing video frame low-latency feature extraction on the dynamic video stream data to obtain video frame feature extraction data, where the video frame feature extraction data includes a clarity index, a color distribution feature and a brightness fluctuation feature; A sequence construction module for generating a corresponding video frame quality feature matrix according to the clarity index, the color distribution feature and the brightness fluctuation feature, and constructing a dynamic quality feature sequence based on the timestamp information; A difference analysis module for performing real-time difference analysis on the quality feature sequences of adjacent video frames through edge nodes to obtain quality correlation coefficients, and obtaining the influence coefficients of adjacent video frames based on the quality correlation coefficients; The correlation graph construction module is used to construct a real-time quality correlation graph based on the influence coefficient, and obtain video frame data to be evaluated according to the real-time quality correlation graph, wherein the video frame data to be evaluated includes a frame marking sequence and video frame type information; The evaluation decision module is used to obtain a video frame evaluation coefficient according to the frame marking sequence and video frame type information, and determine whether the abnormal video frame evaluation coefficient is less than a preset threshold; If it is less than or equal to, it is determined that the video frame data to be evaluated is normal; If it is greater than, it is determined that the video frame data to be evaluated is abnormal.

9. A real-time quality detection system based on 5G edge computing according to claim 8, characterized in that, The video processing module includes: The clarity extraction unit is used to obtain the resolution, frame rate and contrast data of consecutive frames according to the real-time video stream, and obtain a calculated clarity index based on the resolution, frame rate and contrast data; The color analysis unit is used to perform RGB to HSV color space conversion according to the color distribution characteristics of the real-time video stream, and obtain the color temperature deviation value and the saturation fluctuation range; The feature generation unit is used to generate a color distribution feature vector based on the color temperature deviation value and the saturation fluctuation range; The brightness analysis unit is used to obtain the average brightness value and the maximum brightness value of the video frame according to the real-time video stream, and obtain the brightness fluctuation feature according to the average brightness value and the maximum brightness value.

10. A computer-readable storage medium, characterized in that, A program is stored thereon, and when the program is executed by a processor, the real-time quality detection method based on 5G edge computing according to any one of claims 1-7 is implemented.

Images