A real-time video analysis method based on edge computing

Abstract

The present invention relates to the field of video processing technology, and in particular to a real-time video analysis method based on edge computing. The content includes: collecting raw video data, and preprocessing the raw video data to obtain an image frame sequence; segmenting the image frame sequence through an adaptive time slice adjustment mechanism to generate an adaptive time slice; based on the adaptive time slice, compressing and reconstructing through a dynamic feature-guided compression algorithm, extracting features from the obtained reconstruction matrix of the adaptive time slice, and obtaining a reconstructed feature matrix of the adaptive time slice; based on the reconstructed feature matrix of the adaptive time slice, performing video analysis. It solves the problem that the traditional time slice method usually adopts a fixed length and is difficult to cope with the dynamic changes of the video content; the existing compression algorithm is difficult to balance the reduction of the video data volume and the retention of key features during the compression process, resulting in a decrease in the quality of the compressed video and affecting the subsequent analysis accuracy.

Description

Technical Field

[0001] The present invention relates to the field of video processing technology, and in particular to a real-time video analysis method based on edge computing. Background Art

[0002] In the field of video processing technology, traditional video analysis methods usually rely on cloud computing platforms, which transmit large amounts of video data to the cloud for processing and analysis. However, with the popularity of high-definition cameras and the increase in real-time requirements, the bandwidth requirements for cloud processing and the delay in data transmission have become increasingly prominent. Therefore, edge computing technology has gradually become an important direction for video analysis. Edge computing reduces data transmission delays and bandwidth usage by delegating some computing tasks to devices closer to the data source, thereby improving processing efficiency. In recent years, adaptive time slicing technology and feature-oriented compression algorithms based on edge computing have received widespread attention. By adjusting the time slice length and compression ratio in real time, the processing of video data is optimized.

[0003] Although edge computing technology has made significant progress in video analysis, the above technology still has some challenges and shortcomings: first, the traditional time slicing method usually adopts a fixed length, which is difficult to cope with the dynamic changes of video content, resulting in errors and inefficiencies when processing complex scenes; second, the traditional compression algorithm is difficult to balance the reduction of video data volume and the retention of key features during the compression process, resulting in a decrease in the quality of the compressed video, affecting the subsequent analysis accuracy; in addition, the traditional video analysis method still relies on a large amount of historical data when dealing with emergencies and identifying complex scenes, and cannot achieve a good balance between real-time and accuracy. Therefore, further optimizing the adaptability of time slicing and compression algorithms, and improving the accuracy and response speed of real-time video analysis are important directions for current technological development, and they urgently need to be improved through a new algorithm and processing technology. Summary of the invention

[0004] The present invention provides a real-time video analysis method based on edge computing to solve the following technical problems: the traditional time slicing method usually adopts a fixed length, which is difficult to cope with the dynamic changes of video content, resulting in errors and low efficiency when processing complex scenes; the traditional compression algorithm is difficult to balance the reduction of video data volume and the retention of key features during the compression process, resulting in a decrease in the quality of the compressed video, affecting the subsequent analysis accuracy; the traditional video analysis method still relies on a large amount of historical data when processing emergencies and identifying complex scenes, and cannot achieve a good balance between real-time and accuracy.

[0005] A real-time video analysis method based on edge computing of the present invention specifically includes the following technical solutions:

[0006] A real-time video analysis method based on edge computing includes the following steps:

[0007] S1: Collecting and preprocessing the original video data to obtain an image frame sequence; segmenting the image frame sequence through an adaptive time slice adjustment mechanism to generate adaptive time slices;

[0008] S2: Compress and reconstruct the adaptive time slices through a dynamic feature-guided compression algorithm to obtain a reconstruction matrix of the adaptive time slices; perform feature extraction on the reconstruction matrix of the adaptive time slices to obtain a reconstructed feature matrix of the adaptive time slices; perform video analysis based on the reconstructed feature matrix of the adaptive time slices.

[0009] Preferably, the S1 specifically includes:

[0010] The steps for implementing the adaptive time slice adjustment mechanism are as follows:

[0011] The first step is to set the initial time slice length, and segment the image frame sequence according to the initial time slice length to obtain time slices;

[0012] The second step is to calculate the dynamic change rate of the time slice based on the time slice;

[0013] The third step is to dynamically adjust the length of the time slice based on the dynamic change rate of the time slice to generate an adaptive time slice.

[0014] Preferably, the S1 specifically includes:

[0015] Based on time slicing, the logarithmic function and normalization operation are introduced to calculate the dynamic change rate of the time slice. The calculation formula of the dynamic change rate is as follows:

[0016]

[0017] Among them, R k is the dynamic change rate of the kth time slice; ΔT k is the length of the kth time slice; is the feature matrix of the kth time slice The characteristic element of the p-th row and q-th column in ; is the feature matrix of the k-1th time slice The characteristic element of the p-th row and q-th column in ; ∈ is a positive number that avoids the denominator being zero.

[0018] Preferably, the S1 specifically includes:

[0019] Based on the dynamic change rate of the time slice, the length of the time slice is adjusted to obtain the adjusted time slice length; based on the adjusted time slice length, the image frame sequence is segmented to generate adaptive time slices; features are extracted from the image frames in the adaptive time slice to obtain a feature matrix of the adaptive time slice.

[0020] Preferably, the S2 specifically includes:

[0021] In the process of implementing the dynamic feature-oriented compression algorithm, the feature weight matrix of the adaptive time slice is calculated based on the feature matrix of the adaptive time slice.

[0022] Preferably, the S2 specifically includes:

[0023] Based on the feature weight matrix of the adaptive time slice, the feature matrix of the adaptive time slice is differentially compressed to obtain the compression matrix of the adaptive time slice. The specific formula is as follows:

[0024]

[0025] in, is the compression matrix of the kth adaptive time slice; α is the feature amplification coefficient; β is the non-important feature adjustment coefficient; τ is the weight threshold; γ is the normalization coefficient; W k represents the feature weight matrix of the kth adaptive time slice; D k，mn The feature matrix D represents the kth adaptive time slice k The characteristic element of the mth row and nth column in W k，mn is the feature weight matrix W of the kth adaptive time slice k The feature weight of the mth row and nth column in; min elem (W k ,τ) is the element-by-element minimum operation; max elem (W k ,τ) is an element-wise maximum operation.

[0026] Preferably, the S2 specifically includes:

[0027] Nonlinear operation is introduced to reconstruct the compression matrix of the adaptive time slice to obtain the reconstructed matrix of the adaptive time slice; through image processing technology, feature extraction is performed on the reconstructed matrix of the adaptive time slice to obtain the reconstructed feature matrix of the adaptive time slice.

[0028] Preferably, the S2 specifically includes:

[0029] Based on the reconstructed feature matrix of adaptive time slices, logarithmic and square root operations are introduced to capture feature changes and calculate the dynamic change index.

[0030] Preferably, the S2 specifically includes:

[0031] Based on the historical video data in the existing database, an event feature template matrix is ​​generated; a dynamic change threshold and an event recognition threshold are set. When the dynamic change index exceeds the preset dynamic change threshold, an event recognition algorithm is introduced to calculate the similarity value between the reconstructed feature matrix of the adaptive time slice and the event feature template matrix; when the similarity value between the reconstructed feature matrix of the adaptive time slice and the event feature template matrix exceeds the event recognition threshold, a response mechanism is triggered to obtain an event recognition result; the event recognition result is recorded, and the efficiency and accuracy of video analysis are optimized.

[0032] The beneficial effects of the technical solution of the present invention are:

[0033] 1. An adaptive time slice adjustment mechanism is introduced to make the video analysis process more efficient and accurate. Traditional video analysis methods are usually unable to adapt to the dynamic changes of video content in real time. The embodiments of the present invention can not only retain the time series information of the video data by adjusting the time slice length in real time, but also improve the flexibility and accuracy of video processing by optimizing the time slice length.

[0034] 2. The dynamic feature-guided compression algorithm is used to compress and reconstruct the adaptive time slices, which can selectively retain key features. The dynamic feature-guided compression algorithm calculates the feature weight matrix of the adaptive time slices and performs differential compression according to the feature weights of the adaptive time slices, thereby reducing the amount of video data while ensuring the effective transmission and storage of key image information. Compared with traditional compression algorithms, the DFOC algorithm can better retain important information in the video data and ensure the accuracy and effectiveness of the subsequent video analysis process, especially in the edge computing environment with limited resources, which greatly improves the restoration quality of the compressed image.

[0035] 3. By calculating the dynamic change index of the reconstructed feature matrix of the adaptive time slice and the similarity value between it and the predefined event feature template matrix, it is possible to more accurately capture important events and abnormal situations in the video. By introducing nonlinear operations such as logarithms and square roots, the dynamic change index is more sensitive to changes in key features, which can effectively amplify important changes and suppress noise, reducing the risk of misjudgment. At the same time, the similarity value calculation method measures the degree of matching between the adaptive time slice and the preset event feature template, thereby achieving accurate identification of specific events, improving the event recognition capability in complex video scenes, and making the embodiments of the present invention widely applicable in edge computing environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1This is a flow chart of a real-time video analysis method based on edge computing described in the present invention. DETAILED DESCRIPTION

[0037] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0038] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0039] The following is a detailed description of a real-time video analysis method based on edge computing provided by the present invention in conjunction with the accompanying drawings.

[0040] Refer to the attached Figure 1 , which shows a flow chart of a real-time video analysis method based on edge computing provided by an embodiment of the present invention, the method comprising the following steps:

[0041] S1: Collecting and preprocessing the original video data to obtain an image frame sequence; segmenting the image frame sequence through an adaptive time slice adjustment mechanism to generate adaptive time slices;

[0042] First, the edge computing device receives the original video data from the camera device or video source in real time, and preprocesses the original video data to obtain an image frame sequence. Specifically, the preprocessing step includes using Gaussian filtering to denoise the original video data to obtain denoised data; using the Sobel operator to perform edge detection on the denoised data to obtain edge detected data; performing adaptive histogram equalization on the edge detected data to obtain video data; and finally using video decoding technology to decompose the video data to obtain an image frame sequence. The technologies used in the preprocessing step are all existing technologies and will not be elaborated here;

[0043] Then, the adaptive time slice adjustment mechanism is used to perform time slice operation on the image frame sequence, and the time slice length is adjusted in real time according to the dynamic changes of the video data, so that the generated adaptive time slice not only retains the time series information of the video data, but also dynamically adapts to the changes of the video content, significantly improving the processing efficiency of the video data and the accuracy of the analysis;

[0044] The process of the adaptive time slice adjustment mechanism is as follows:

[0045] First, the initial time slice length is set, and the image frame sequence is segmented according to the initial time slice length to obtain time slices. Each time slice contains image frames within the time period defined by the time slice length. The time slice expression is as follows:

[0046]

[0047] t k =t0+k·ΔT0

[0048] in, is the kth time slice; F(t) is the image frame sequence generated at time t; t k is the starting time of the kth time slice; t0 represents the starting time of the 0th time slice; t k+1 is the start time of the k+1th time slice and the end time of the kth time slice; k is the index of the time slice, k=0, 1, 2, ..., M-1; is the total number of time slices; ΔT0 is the initial time slice length, which is set by expert experience according to the specific implementation scenario; T is the total time length of the video data;

[0049] Then, in order to make the time slices adaptive to the dynamic changes of the video data, the dynamic change rate of the time slices is calculated; the dynamic change rate formula introduces logarithmic functions and normalization operations to improve the sensitivity to changes in video data. Logarithmic processing can better capture subtle but important changes, and normalization avoids the influence of eigenvalues ​​that are too large or too small, effectively improving the accuracy of video data analysis in complex scenarios. The dynamic change rate formula is as follows:

[0050]

[0051] Among them, R k is the dynamic change rate of the kth time slice; ΔT k is the length of the kth time slice; is the feature matrix of the k-th time slice, which is generated by extracting features from the image frame in the k-th time slice through image processing techniques such as edge detection and brightness gradient calculation. The image processing technology is a prior art and will not be described in detail here; is the feature matrix of the kth time slice The characteristic element of the p-th row and q-th column in ; is the feature matrix of the k-1th time slice The characteristic element of the p-th row and q-th column in ; ∈ is a positive number used to avoid the denominator being zero and is set according to the specific scenario;

[0052] According to the dynamic change rate of the time slice, the length of the subsequent time slice is adjusted to obtain the adjusted time slice length. The adjustment formula is as follows:

[0053]

[0054] Where, ΔT k+1 is the length of the k+1th time slice; θ is the preset rate of change threshold;

[0055] Calculate the adjusted time slice length ΔT k+1 After that, according to ΔT k+1 Segment the image frame sequence to generate adaptive time slices S k+1 , the above image processing technology is used to extract features from the image frames in the k+1th adaptive time slice, and the feature matrix D of the k+1th adaptive time slice is obtained. k+1 , using the dynamic change rate formula to evaluate the degree of change of the current adaptive time slice, and according to the calculation result of the dynamic change rate formula, the length of the next time slice is calculated by adjusting the formula, and through the iterative process, the length of the time slice is dynamically adjusted to generate an adaptive time slice.

[0056] S2: Compress and reconstruct the adaptive time slices through a dynamic feature-guided compression algorithm to obtain a reconstruction matrix of the adaptive time slices; perform feature extraction on the reconstruction matrix of the adaptive time slices to obtain a reconstructed feature matrix of the adaptive time slices; perform video analysis based on the reconstructed feature matrix of the adaptive time slices.

[0057] After generating the adaptive time slice, the edge computing device performs real-time image compression processing on the adaptive time slice to improve processing efficiency in an edge computing environment with limited resources. The embodiment of the present invention proposes a dynamic feature-oriented compression algorithm (DFOC), which enhances the ability to retain key features during the compression process. The key features represent feature elements in the feature matrix whose feature weights are greater than a preset weight threshold. Specifically, the DFOC algorithm prioritizes retaining key features to ensure that key features are effectively retained and transmitted during the compression process, which not only reduces the amount of video data, but also improves the restoration quality of the compressed image, providing more accurate and reliable data support for subsequent video analysis processes.

[0058] The specific process of the DFOC algorithm is as follows:

[0059] The feature matrix D based on the kth adaptive time slice k, calculate the feature weight matrix of the kth adaptive time slice, the feature weight matrix of the adaptive time slice represents the feature weights of the feature elements in the feature matrix of the adaptive time slice. The feature weight calculation process is as follows:

[0060]

[0061] Among them, W k，mn is the feature weight matrix W of the kth adaptive time slice k The feature weight of the mth row and nth column in ; and Represents the feature matrix D of the kth adaptive time slice respectively k The gradients of the characteristic elements in the mth row and nth column in the horizontal and vertical directions;

[0062] Furthermore, after obtaining the feature weight matrix of the adaptive time slice, the feature matrix of the adaptive time slice is compressed to obtain the compressed matrix of the adaptive time slice. Specifically, for the key features in the feature matrix of the adaptive time slice, the feature amplification coefficient is used to achieve a low compression ratio to retain more detail information; for the unimportant features in the feature matrix of the adaptive time slice, the unimportant feature adjustment coefficient is used to achieve a high compression ratio to reduce the storage and transmission volume of unimportant features; unimportant features represent feature elements in the feature matrix whose feature weights are less than or equal to a preset weight threshold. The specific values ​​of the weight threshold, feature amplification coefficient, and unimportant feature adjustment coefficient are determined by expert experience;

[0063] The compression matrix calculation formula is as follows:

[0064]

[0065] in, is the compression matrix of the kth adaptive time slice; α is the feature amplification coefficient; β is the non-important feature adjustment coefficient; τ is the weight threshold; γ is the normalization coefficient, which is set according to the expert experience method; min elem (W k ,τ) is an element-by-element minimum operation, which means that the feature weight matrix W of the kth adaptive time slice k Each element in is compared with the weight threshold τ to get the minimum value, and the result is a feature weight matrix W corresponding to the kth adaptive time slice. k Matrices of the same dimension; max elem (W k ,τ) is an element-by-element maximum operation, which means that the feature weight matrix W of the kth adaptive time slice k Each element in is compared with the weight threshold τ to get the maximum value, and the result is a feature weight matrix W corresponding to the kth adaptive time slice. kMatrices of the same dimensions;

[0066] Finally, the compression matrix of the adaptive time slice is used to reconstruct the matrix and obtain the reconstructed matrix of the adaptive time slice. The compression matrix of the adaptive time slice is weightedly amplified using the reconstruction formula, and the elements with larger weights in the feature weight matrix of the adaptive time slice are restored preferentially. The key features that retain more detailed information are restored preferentially, providing accurate and rich feature information for subsequent video analysis, which is suitable for low-latency, high-precision real-time video analysis. The reconstruction formula introduces nonlinear operations such as logarithms and square roots to enhance the ability to adjust the feature weight matrix of the adaptive time slice and the compression matrix of the adaptive time slice, thereby improving the reducibility and robustness of the reconstruction matrix of the adaptive time slice to the original video data;

[0067] The reconstruction formula is as follows:

[0068]

[0069] Among them, D′ k is the reconstruction matrix of the kth adaptive time slice; ζ is the normalization coefficient; δ is the adjustment parameter used to adjust the feature matrix D of the kth adaptive time slice k The feature elements with lower feature weights are appropriately suppressed to reduce the influence of feature elements with lower feature weights in the reconstruction process; ζ and δ are obtained through experimental tuning.

[0070] Then, image processing techniques such as edge detection and brightness gradient calculation are used to extract features from the reconstruction matrix of the kth adaptive time slice to obtain the reconstructed feature matrix of the kth adaptive time slice. The dynamic change index is calculated by comparing the reconstructed feature matrices of two consecutive adaptive time slices. The dynamic change index is a numerical value that quantifies the degree of change between the current adaptive time slice and the previous adaptive time slice. It can identify changes or abnormalities in the video, which helps to capture important events or actions, such as sudden acceleration of moving objects, switching of picture scenes, etc., and provides a basis for event recognition.

[0071] The dynamic change index formula introduces logarithmic and square root operations so that it can more sensitively capture changes in key features during calculation. Compared with traditional differential formulas, it can amplify important changes while suppressing noise, smoothing numerical fluctuations, reducing the risk of misjudgment, and improving the accuracy and reliability of the dynamic change index.

[0072] The dynamic change index formula is as follows:

[0073]

[0074] Among them, V kis the dynamic change index of the reconstructed feature matrix of the kth adaptive time slice; F′ k is the reconstructed feature matrix of the kth adaptive time slice; F′ k-1 is the reconstructed feature matrix of the k-1th adaptive time slice; when calculating the dynamic change index of the reconstructed feature matrix of the first adaptive time slice, the reconstructed feature matrix F′ of the k-1th adaptive time slice is k-1 Set to zero matrix or the reconstructed feature matrix F′ of the kth adaptive time slice k The same matrix is ​​selected according to the specific implementation scenario;

[0075] Next, based on the dynamic change index of the reconstructed feature matrix of the adaptive time slice, event detection and recognition are performed. The dynamic change threshold and event recognition threshold are set according to the expert experience method. When the dynamic change index V of the reconstructed feature matrix of the kth adaptive time slice is k When the preset dynamic change threshold is exceeded, the event recognition algorithm is triggered. The event recognition algorithm determines whether the image frame corresponding to the adaptive time slice matches a known event or scene by calculating the similarity value between the reconstructed feature matrix of the adaptive time slice and the predefined event feature template matrix. The event feature template matrix extracts features of known events or scenes in historical video data in an existing database, and uses the statistical average value or cluster center as a template to generate a feature matrix representing the event;

[0076] The similarity value calculation formula is as follows:

[0077]

[0078] in, is the event feature template matrix; ||F′ k || and are the reconstructed feature matrix F′ of the kth adaptive time slice. k and event feature template matrix, the norm; Sim k is the reconstructed feature matrix F′ of the kth adaptive time slice k With event feature template matrix The similarity value ranges from [0 to 1]. The closer the similarity value is to 1, the more similar the feature elements in the reconstructed feature matrix of the current adaptive time slice are to the feature elements in the predefined event feature template matrix, indicating that the probability of the event occurring is greater.

[0079] Finally, the event recognition threshold is set. When the similarity value between the reconstructed feature matrix of the adaptive time slice and the event feature template matrix exceeds the event recognition threshold, the response mechanism is triggered, such as alarm, recording video data, and obtaining the event recognition result. At the same time, the event recognition result is recorded to facilitate the subsequent adjustment of parameters (such as dynamic change threshold, event recognition threshold, weight threshold, feature amplification coefficient, non-important feature adjustment coefficient) to optimize the efficiency and accuracy of video analysis.

[0080] In summary, a real-time video analysis method based on edge computing is completed.

[0081] The order of the embodiments of the invention is for description only and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0082] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0083] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. A real-time video analysis method based on edge computing, characterized in that: The following steps are involved: S1: Collecting original video data and preprocessing the original video data to obtain an image frame sequence; segmenting the image frame sequence through an adaptive time slice adjustment mechanism to generate adaptive time slices; S2: compress and reconstruct the adaptive time slices through a dynamic feature-guided compression algorithm to obtain a reconstruction matrix of the adaptive time slices; extract features from the reconstruction matrix of the adaptive time slices to obtain a reconstructed feature matrix of the adaptive time slices; Reconstructed feature matrix based on adaptive time slicing for video analysis.

2. A real-time video analysis method based on edge computing according to claim 1, characterized in that: The S1 specifically includes: The steps for implementing the adaptive time slice adjustment mechanism are as follows: The first step is to set the initial time slice length, and segment the image frame sequence according to the initial time slice length to obtain time slices; The second step is to calculate the dynamic change rate of the time slice based on the time slice; The third step is to dynamically adjust the length of the time slice based on the dynamic change rate of the time slice to generate an adaptive time slice.

3. A real-time video analysis method based on edge computing according to claim 2, characterized in that: The S1 specifically includes: Based on time slicing, the logarithmic function and normalization operation are introduced to calculate the dynamic change rate of the time slice. The calculation formula of the dynamic change rate is as follows: Among them, R k is the dynamic change rate of the kth time slice; ΔT k is the length of the kth time slice; is the feature matrix of the kth time slice The characteristic element of the p-th row and g-th column in ; is the feature matrix of the k-1th time slice The characteristic element of the p-th row and q-th column in ; ∈ is a positive number that avoids the denominator being zero.

4. The real-time video analysis method based on edge computing according to claim 3 is characterized in that: The S1 specifically includes: Based on the dynamic change rate of the time slice, the length of the time slice is adjusted to obtain the adjusted time slice length; based on the adjusted time slice length, the image frame sequence is segmented to generate adaptive time slices; features are extracted from the image frames in the adaptive time slice to obtain a feature matrix of the adaptive time slice.

5. A real-time video analysis method based on edge computing according to claim 4, characterized in that: The S2 specifically includes: In the process of implementing the dynamic feature-oriented compression algorithm, the feature weight matrix of the adaptive time slice is calculated based on the feature matrix of the adaptive time slice.

6. A real-time video analysis method based on edge computing according to claim 5, characterized in that: The S2 specifically includes: Based on the feature weight matrix of the adaptive time slice, the feature matrix of the adaptive time slice is differentially compressed to obtain the compression matrix of the adaptive time slice. The specific formula is as follows: in, is the compression matrix of the kth adaptive time slice; α is the feature amplification coefficient; β is the non-important feature adjustment coefficient; τ is the weight threshold; γ is the normalization coefficient; W k represents the feature weight matrix of the kth adaptive time slice; D k，mn The feature matrix D represents the kth adaptive time slice k The characteristic element of the mth row and nth column in W k，mn is the feature weight matrix W of the kth adaptive time slice k The feature weight of the mth row and nth column in; min elem (W k ,τ) is the element-by-element minimum operation; max elem (W k ,τ) is an element-wise maximum operation.

7. A real-time video analysis method based on edge computing according to claim 6, characterized in that: The S2 specifically includes: Nonlinear operation is introduced to reconstruct the compression matrix of the adaptive time slice to obtain the reconstructed matrix of the adaptive time slice; through image processing technology, feature extraction is performed on the reconstructed matrix of the adaptive time slice to obtain the reconstructed feature matrix of the adaptive time slice.

8. The real-time video analysis method based on edge computing according to claim 7 is characterized in that: The S2 specifically includes: Based on the reconstructed feature matrix of adaptive time slices, logarithmic and square root operations are introduced to capture feature changes and calculate the dynamic change index.

9. The real-time video analysis method based on edge computing according to claim 8, characterized in that: The S2 specifically includes: Generate an event feature template matrix based on historical video data in an existing database; set a dynamic change threshold and an event recognition threshold. When the dynamic change index exceeds the preset dynamic change threshold, introduce an event recognition algorithm to calculate the similarity between the reconstructed feature matrix of the adaptive time slice and the event feature template matrix; When the similarity value between the reconstructed feature matrix of the adaptive time slice and the event feature template matrix exceeds the event recognition threshold, the response mechanism is triggered to obtain the event recognition result; the event recognition result is recorded, and the efficiency and accuracy of video analysis are optimized.

Images