An archive management system based on distributed storage

Abstract

This invention relates to the field of archival management technology, specifically to an archival management system based on distributed storage. The invention matches the archival type of pre-stored video archives in a smart campus with corresponding storage nodes, obtains archival features of the pre-stored video archives based on image feature recognition algorithms, and uses the archive continuity similarity and video archive feature data length as the criteria for determining distributed storage segmentation points to facilitate subsequent distributed storage. By traversing the similarity of the segmented pre-stored video archives, the invention obtains the first keyframe video archive features and the second keyframe video archive features as the basis for data backup and data restoration, and performs distributed storage based on archive type. This method improves the security of smart campus video archives, and, based on image recognition algorithm feature selection and distributed storage, ensures the integrity of video backup and restoration.

Description

Technical Field

[0001] This invention relates to the field of archives management technology, specifically to an archives management system based on distributed storage. Background Technology

[0002] Distributed storage is a data storage method that divides a single piece of data into multiple data blocks and sends them to multiple nodes for storage. Distributed storage is mainly used to solve the problem of storing massive amounts of data that are growing explosively today. For large amounts of data such as TB, PB, and ZB, the limited transfer speed of hard drives means that splitting these large amounts of data into smaller amounts can effectively improve storage efficiency. At the same time, distributed storage also solves the problem of data loss of single pieces of data. Backups are performed during the splitting process, so even if data is lost on a single node, it can be restored through backups, significantly improving the reliability and fault tolerance of data storage.

[0003] Smart campuses utilize advanced computer technologies such as the Internet of Things and big data to achieve intelligent and digital transformation of teaching, services, and management within the campus. The digital security management of campus archives is a crucial aspect of this process. Besides traditional archives like student records, faculty and staff personnel files, and employee files, new types of archival data, such as teaching videos and lecture videos, campus security surveillance footage, and event records from campus meetings, also need to be preserved and managed. Compared to traditional archives, video archives are characterized by larger individual data sets and greater continuity, making the security and integrity checks of video archives more challenging during storage.

[0004] Existing technologies have certain limitations and low applicability in addressing the security and integrity issues of video archive storage within smart campuses. Therefore, a distributed storage-based archive management system is proposed. Summary of the Invention

[0005] The purpose of this invention is to provide an archive management system based on distributed storage. This system matches the archive types of pre-stored video archives in a smart campus to corresponding storage nodes. It obtains the archive features of the pre-stored video archives based on image feature recognition algorithms, and uses archive continuity similarity and video archive feature data length as criteria for determining distributed storage segmentation points to facilitate subsequent distributed storage. By traversing the similarity of the segmented pre-stored video archives, it obtains the features of the first and second keyframe video archives, which serve as the basis for data backup and restoration, and performs distributed storage based on archive type. This method improves the security of smart campus video archives, and the feature selection based on image recognition algorithms and distributed storage ensures the integrity of video backup and restoration.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A distributed storage-based archive management system includes a smart campus video archive acquisition module, a smart campus video archive segmentation and judgment module, a smart campus video archive first segmentation module, a smart campus video archive second segmentation module, a smart campus video archive storage module, a smart campus primary feature backup module, and a smart campus secondary feature backup module, specifically:

[0008] The smart campus video archive acquisition module obtains the file type of pre-stored video archives and matches multiple corresponding storage nodes, filters the storage nodes, and generates a target storage node;

[0009] Each storage node corresponds to a file type, and there are multiple groups of storage nodes of the same file type, with each group containing multiple storage nodes; the file types include personal video files, laboratory video files, open course video files, surveillance video files, event video files, meeting video files, temporary video files, and unclassified video files;

[0010] Furthermore, the number of multiple groups of storage nodes of the same file type is at least three groups, and each group of storage nodes is divided according to functional type. The multiple groups of storage nodes include a video original file storage group, a video feature file backup group, and a video feature file restoration group.

[0011] The smart campus video archive segmentation and judgment module extracts multiple image archive data from pre-stored video archives to obtain video archive feature data, and calculates the archive continuity similarity of the video archive feature data;

[0012] Furthermore, the video archive feature data and the archive continuity similarity are determined through a smart campus video archive feature model; the smart campus video archive feature model includes an archive video data input unit, an archive video data feature extraction unit, an archive video data similarity calculation unit, an archive video data similarity marking unit, an archive video feature output unit, an archive video similarity output unit, and an archive video feature restoration unit;

[0013] The archive video data input unit classifies and marks the archive types of the pre-stored video archives, and preprocesses the pre-stored video archives to generate archive video input data and archive video type tags;

[0014] Specifically, the archive video similarity data is used to mark the pre-stored video archives as follows: the archive video similarity data is used to mark the pre-stored video archives for each timestamp.

[0015] If the similarity data of the archive video at the t-th timestamp is less than the continuity similarity of the archive, a first mark is made;

[0016] Calculate the difference θ between the archival video similarity data at the t-th timestamp and the archival video similarity data at the (t+1)-th timestamp. t1 Calculate the difference θ between the archive video similarity data at the t-th timestamp and the archive video similarity data at the (t-1)-th timestamp. t2 ;

[0017] If θ t1 or θ t2 If the difference in similarity between the files is greater than the file similarity difference, then a second labeling is performed;

[0018] The archival video data feature extraction unit calls the corresponding video feature extraction module according to the archival video type label, and extracts the features of the archival video input data through the video feature extraction module to generate archival video feature data;

[0019] The video feature extraction module consists of an encoding layer and a decoding layer. The encoding layer is composed of an improved YOLOv5 acquisition layer and a Transformer encoder. The video feature extraction module only uses the YOLOv5 backbone layer and neck layer. The video feature extraction module includes position feature extraction and color feature extraction. The position feature extraction uses the YOLOv5 backbone layer and neck layer. The color feature extraction uses the improved backbone layer and neck layer based on dilated convolution. The structure of the color feature extraction is as follows:

[0020] output TCN=neck(backbone) TCN (input));

[0021]

[0022] Where, output TCN For output features, neck refers to the neck layer, backbone refers to the backbone. TCN The backbone layer is an improvement based on the dilated convolution, where input is the input data; Conv6*6 is a convolutional layer with a kernel size of 6*6, TCNV1, TCNV2, TCNV3, TCNV4, TCNV... m This is a feature extraction layer based on the dilated convolution, where SPPF is the SPPF layer used for pooling in YOLOv5, and TCNS is... m For the m-th dilated convolution group, TCN11 m The m-th CSP structure based on dilated convolution;

[0023] The TCNS m and the TCN11 m The structure is as follows:

[0024]

[0025] Among them, ReLU is the activation layer, BN is the BN layer, TCN is the dilated convolutional layer, and CNN1*1 is the convolutional layer with a 1*1 kernel.

[0026] The formula for the kernel size of the dilated convolutional layer is as follows:

[0027] KT λ =λ×(KT0-1)+1;

[0028] λ j =max(λ) j+1 -2λ j ,λ j+1 -2(λ j+1 -λ j ),λ j );

[0029] Among them, KT λ KT0 is the size of the dilated convolution kernel, λ is the dilation factor, KT0 is the set kernel size, and max is the maximum value. j Let λ be the expansion factor of the j-th layer. j+1 The expansion factor of the (j+1)th layer;

[0030] After feature extraction, the data is input into the Transformer encoder to generate positional and color features;

[0031] The decoding layer is restored using a Transformer decoder and an inverse dilated convolution.

[0032] The kernel size and image size of the inverse dilation convolution are calculated as follows:

[0033]

[0034] λ k =max(λ) k+1 -2λ k ,λ k+1 -2(λ k+1 -λ k ),λ k );

[0035]

[0036] Among them, KT A λ is the size of the inverse dilated convolution kernel. A KT is the inverse expansion factor. C λ is the size of the inverse convolution kernel. k λ is the expansion factor of the k-th layer. k+1 is the dilation factor of the (k+1)th layer, outa is the output image size, ina is the input image size, padding is the padding size, and d is the step size.

[0037] The archive video data similarity calculation unit sorts all the archive video feature data in chronological order and calculates the similarity of adjacent archive video feature data in turn to generate archive video similarity data.

[0038] Specifically, the similarity data of the archive videos is calculated as follows:

[0039]

[0040] Where, γ sim The archive video similarity data is given, where N is the number of features and P is the number of features. i t For the i-th color feature of the archive video feature data at the t-th timestamp, P i t-1 Let i be the i-th color feature of the archive video feature data at the (t-1)-th timestamp. Let i be the i-th location feature of the archive video feature data at timestamp t. For the i-th location feature of the archive video feature data at the (t-1)-th timestamp, α c and α l The weights for calculating the similarity data of the archive videos;

[0041] The archive video data similarity marking unit marks the pre-stored video archives based on the archive video similarity data, and generates marked pre-stored video archive data;

[0042] The archive video feature output unit outputs and stores the archive video feature data sorted according to the time order.

[0043] The archive video similarity output unit outputs and stores the marked pre-stored video archive data;

[0044] The archive video feature restoration unit restores the archive video feature data;

[0045] Segmentation points are generated based on the continuity similarity of the archives and the length of the video archive feature data.

[0046] Furthermore, the timestamp of the first frame of the video is used as the start timestamp; the segmentation point is determined by the continuity similarity of the archives and the length of the video archive feature data;

[0047] If the file continuity similarity is less than the continuity threshold, then the timestamp corresponding to the file continuity similarity being less than the continuity threshold is used as the segmentation point; if the length of the video file feature data exceeds the segmentation continuity threshold, then the timestamp corresponding to the segmentation continuity threshold length is used as the segmentation point.

[0048] The first segmentation module of the smart campus video archive segments the pre-stored video archive through the segmentation point, obtains the video archive feature data corresponding to the segmented video archives, and calculates multiple first keyframe video archive features by combining the archive continuity similarity.

[0049] Furthermore, the first keyframe video file feature is obtained by acquiring the number of frames of the pre-stored video file and the feature data of the segmented video file;

[0050] Based on the similarity of the video archive feature data and the number of frames, the positions of multiple first keyframe video archive features are obtained from the video archive feature data:

[0051]

[0052]

[0053] Where, η F φ represents the location of the first keyframe video file feature. F and φ V For the pre-selected position, φ videoThe number of frames; For calculation and The position of the closest value, k n The number of similarities calculated for the pre-selected locations. The similarity of the feature data of the i-th video file is... For k n The similarity of the pre-selected locations, where min is the minimum value and max is the maximum value;

[0054] The smart campus video archive second segmentation module acquires the video archive feature data corresponding to the segmentation point, and generates multiple second keyframe video archive features by combining the average file continuity similarity of the feature data of adjacent segmented video archives. The acquisition of the second keyframe video archive features involves acquiring and storing the video archive feature data of segmentation point A; and acquiring and storing the average file continuity similarity of the segmentation point within the similarity ranges (M-1, M) and (M, M+1). Here, M is the segmentation point, M-1 is the previous segmentation point with the smallest distance from M frames, and M+1 is the next segmentation point with the smallest distance from M frames.

[0055] The smart campus video archive storage module stores the segmented pre-stored video archive into the target storage node;

[0056] The smart campus feature backup module stores the segmented video file feature data into the target storage node;

[0057] The smart campus secondary feature backup module stores the first keyframe video file feature and the second keyframe video file feature into the target storage node;

[0058] The smart campus archive storage module, the smart campus primary feature backup module, and the smart campus secondary feature backup module also include:

[0059] The smart campus archive storage module stores similarity data of video archive feature data corresponding to pre-stored video archives; the smart campus primary feature backup module and the smart campus secondary feature backup module store similarity data of corresponding video archive feature data;

[0060] The smart campus primary feature backup module and the smart campus secondary feature backup module restore feature data through the archive video feature restoration unit of the smart campus video archive feature model.

[0061] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0062] 1. This invention obtains feature data of pre-stored video archives through image feature algorithms, and segments the data based on the similarity of archive continuity between feature data and the length of video archive feature data, so as to preserve the integrity of video archive data to the greatest extent, reduce data size, reduce the required storage capacity space, and facilitate subsequent distributed storage; at the same time, it retains data feature data, providing a basis for subsequent data backup and restoration work, and ensuring the implementation of subsequent video archive data security processing.

[0063] 2. To ensure the accuracy of feature similarity analysis, this invention calculates the similarity between adjacent feature data by extracting all feature data from pre-stored video files. By judging color features and the positional features of related content, the accuracy of similarity analysis of video files is improved, ensuring the integrity of video file content during segmentation. Even with distributed storage, the video content will not be damaged, ensuring the coherence and local integrity of the segmented video content. Even in the worst-case scenario where a small portion of data is missing, the coherence and integrity of other content can be guaranteed as much as possible. At the same time, the characteristics of distributed storage can also be combined to ensure the security of video file data.

[0064] 3. This invention, based on segmented feature data and similarity data of neighboring feature data, obtains the feature position of the first keyframe video file of the segmented pre-stored video through similarity data and frame count data, and generates the first keyframe video file as the basis for backup and restoration. This ensures that data can be restored in case of loss, and also ensures that the content of the restored video file is coherent, thus guaranteeing the integrity of the video file. At the same time, the feature data also ensures that the content of the video file cannot be understood through the feature data when the backup data is leaked, thereby improving the security of the backup data. Attached Figure Description

[0065] Figure 1 This is a schematic diagram of the system flow of the present invention;

[0066] Figure 2 This is a schematic diagram of the smart campus video archive feature model process of the present invention;

[0067] Figure 3 This is a flowchart illustrating the video feature extraction module of the smart campus video archive feature model of the present invention.

[0068] Figure 4 This is a schematic diagram of the color feature extraction part in the video feature extraction module of the present invention;

[0069] Figure 5 TCNS in the video feature extraction module of the present invention m and TCN11 mA structural diagram. Detailed Implementation

[0070] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0071] With the gradual maturation of new computer technologies such as big data, smart campuses, implemented through these technologies, can achieve highly efficient digital management of traditional teaching and service operations within the campus. Security management has always been a key focus of digital management. Besides the security of traditional archival data such as personnel or faculty records, the security of video archives, as a new type of archival data, cannot be ignored. Compared to traditional archival data, video archives have a larger capacity and more continuous information. Using traditional distributed algorithms for storage can easily compromise the integrity of video archives. Furthermore, video archives are more complex than traditional archival data, making it impossible to ensure the same level of security. Therefore, this invention provides an archive management system based on distributed storage, with the following technical solution:

[0072] Reference Figure 1 As shown, the smart campus video archive acquisition module obtains the file type of the pre-stored video archive and matches multiple corresponding storage nodes, filters the storage nodes, and generates a target storage node;

[0073] The smart campus video archive segmentation and judgment module extracts multiple image archive data from pre-stored video archives to obtain video archive feature data, calculates the archive continuity similarity of the video archive feature data, and generates segmentation points based on the archive continuity similarity and the length of the video archive feature data.

[0074] The first segmentation module of the smart campus video archive segments the pre-stored video archive through the segmentation point, obtains the video archive feature data corresponding to the segmented video archives, and calculates multiple first keyframe video archive features by combining the archive continuity similarity.

[0075] The second segmentation module of the smart campus video archive obtains the video archive feature data corresponding to the segmentation point, and generates multiple second keyframe video archive features by combining the average file continuity similarity of the feature data of adjacent segmented video archives.

[0076] The smart campus video archive storage module stores the segmented pre-stored video archive into the target storage node;

[0077] The smart campus feature backup module stores the segmented video file feature data into the target storage node;

[0078] The smart campus secondary feature backup module stores the first keyframe video file features and the second keyframe video file features into the target storage node.

[0079] To ensure the integrity and security of all video archive data in the smart campus;

[0080] First, to facilitate storage and management, the file types of storage nodes are matched with the file types of pre-stored video files, and storage nodes are filtered to generate target storage nodes for pre-stored files, so that pre-stored videos can be placed in the corresponding storage nodes, facilitating distributed storage and management.

[0081] Secondly, in order to ensure the continuity of distributed storage of video archive data, the image data of each timestamp in the video archive is extracted, and the image data is obtained through image feature recognition algorithm for similarity analysis. Based on the continuity similarity of video archives and the length of video archive feature data as the judgment criteria, the video archive data is distributed and partitioned. This can ensure the continuity of video archives to the greatest extent on the basis of the original data, prevent the destruction of video archive information, and improve the local integrity after distributed storage while ensuring the security of distributed storage.

[0082] Finally, for the corresponding video archive feature data of the segmented video archive data, multiple first keyframe video archive features are obtained through continuous similarity calculation. The continuous similarity between the video archive feature data of the segmentation point and the adjacent segmented archive data is used as the second keyframe video archive features. By storing and backing up the video feature data, on the one hand, it can improve the ability to restore data when data is missing, ensuring data integrity. On the other hand, these feature data can also ensure that video archive information cannot be obtained after data leakage, thus improving data security.

[0083] Example 1

[0084] To illustrate this in detail, the present invention will be described in the context of a single file type. For specific illustration, this embodiment will use a campus meeting video file as an example:

[0085] A distributed storage-based archive management system includes a smart campus video archive acquisition module, a smart campus video archive segmentation and judgment module, a smart campus video archive first segmentation module, a smart campus video archive second segmentation module, a smart campus video archive storage module, a smart campus primary feature backup module, and a smart campus secondary feature backup module, specifically:

[0086] The smart campus video archive acquisition module obtains the file type of pre-stored video archives, matches multiple corresponding storage nodes, filters storage nodes, and generates target storage nodes.

[0087] Each storage node corresponds to a file type. There are multiple groups of storage nodes of the same file type, and each group contains multiple storage nodes. File types include personal video files, laboratory video files, open course video files, surveillance video files, event video files, meeting video files, temporary video files, and unclassified video files.

[0088] In the specific implementation process, since the stored video files are campus meeting videos, all storage nodes of the type of campus meeting videos will be matched during the traversal and filtering process; the implementation method can take three forms.

[0089] In the first implementation, the storage node is identified as the corresponding storage node if it stores campus meeting videos.

[0090] In the second implementation method, storage nodes that store campus meeting videos and account for 50% or more of the total storage capacity are identified as the corresponding storage nodes.

[0091] The third implementation method involves dividing the space into dedicated storage nodes for storing campus meeting videos;

[0092] To reduce the storage pressure on each storage server while ensuring data security, distributed storage divides the data into multiple parts, with each part stored on a different storage server. To facilitate data retrieval and management, different file types are assigned to each storage server. By matching the stored campus meeting videos with the corresponding storage server file type as candidate storage nodes, the situation of scattered and difficult-to-manage data can be prevented.

[0093] Furthermore, the number of storage nodes of the same file type is at least three groups. Each group of storage nodes is divided according to its function. The multiple groups of storage nodes include the video original file storage group, the video feature file backup group, and the video feature file restoration group. In the specific implementation process, all storage nodes corresponding to the campus meeting video after matching will be divided into at least three groups according to their functions, namely the video original file storage group, the video feature file backup group, and the video feature file restoration group.

[0094] In the actual implementation process, it may be further divided. For example, according to the type of meeting, the video original archive storage group may be divided into the video original archive storage group for important school meetings, the video original archive storage group for routine school meetings, etc.; or according to the date, it may be further divided into the video original archive storage group for the first quarter meetings, the video original archive storage group for the second quarter meetings, etc.

[0095] In order to accurately divide the data, in addition to dividing it according to data type, we also divided it into video original archive storage group, video feature archive backup group, and video feature archive restoration group. By dividing the functional data, we can ensure that the corresponding data can be accurately located when performing original data storage, data backup and data restoration, improve data processing speed, and avoid data mixing, which may lead to data restoration failure or data loss during restoration.

[0096] The smart campus video archive segmentation and judgment module extracts multiple image archive data from pre-stored video archives to obtain video archive feature data, and calculates the archive continuity similarity of the video archive feature data;

[0097] Furthermore, the video archive feature data and archive continuity similarity are determined using a smart campus video archive feature model. This model includes an archive video data input unit, an archive video data feature extraction unit, an archive video data similarity calculation unit, an archive video data similarity marking unit, an archive video feature output unit, an archive video similarity output unit, and an archive video feature restoration unit. Figure 2 As shown;

[0098] The archival video data input unit classifies and labels the archive types of pre-stored video archives, and preprocesses the pre-stored video archives to generate archival video input data and archival video type labels;

[0099] Specifically, the archive video similarity data is used to tag pre-stored video archives as archive video similarity data for each timestamp of the pre-stored video archive;

[0100] If the similarity data of the archive video at the t-th timestamp is less than the archive continuity similarity, then the first mark is made;

[0101] Calculate the difference θ between the archival video similarity data at timestamp t and the archival video similarity data at timestamp t+1. t1 Calculate the difference θ between the archival video similarity data at timestamp t and the archival video similarity data at timestamp t-1. t2 ;

[0102] If θ t1 or θ t2If the difference in similarity is greater than the file similarity value, a second mark is made; the default similarity difference is 5%, which can be modified according to the specific implementation situation, the file type of the actual video data, or expert experience;

[0103] After determining the continuity similarity of video archives, in order to improve the data processing speed and facilitate subsequent processing of related video archive data, marking the similarity can accurately obtain the corresponding data. At the same time, additional marking of data with large similarity differences can facilitate the confirmation of subsequent segmentation points, improve the data processing speed, and record relevant data can also assist in data query and reconstruction, improve the integrity of data during query and reconstruction. In addition, similarity data can also serve as one of the indicators for data security verification.

[0104] The archival video data feature extraction unit calls the corresponding video feature extraction module based on the archival video type label, and extracts the features of the archival video input data through the video feature extraction module to generate archival video feature data; in this embodiment, the video feature extraction module trained based on campus video conferencing archival data is called.

[0105] The video feature extraction module consists of an encoding layer and a decoding layer, as shown in the reference. Figure 3 As shown, Figure 3 In the diagram, output1 is the feature output, and output2 is the feature reconstruction output. The encoding layer consists of an improved YOLOv5 acquisition layer and a Transformer encoder. The video feature extraction module only uses the YOLOv5 backbone and neck layers. The video feature extraction module includes position feature extraction and color feature extraction. Position feature extraction uses the YOLOv5 backbone and neck layers; color feature extraction uses a bankbone and neck layer improved based on dilated convolution. (Refer to...) Figure 4 As shown; the composition of color feature extraction is as follows:

[0106] output TCN =neck(backbone) TCN (input));

[0107]

[0108] Where, output TCN For output features, neck is the neck layer, backbone TCN This is a backbone layer based on dilated convolution, where input is the input data; Conv6*6 is a convolutional layer with a kernel size of 6*6, TCNV1, TCNV2, TCNV3, TCNV4, TCNV... mThis is a feature extraction layer based on dilated convolution, with SPPF being the SPPF layer used for pooling in YOLOv5, and TCNS. m For the m-th dilated convolution group, TCN11 m The m-th CSP structure based on dilated convolution;

[0109] TCNS m and TCN11 m The structure, refer to Figure 5 As shown below:

[0110]

[0111] Among them, ReLU is the activation layer, BN is the BN layer, TCN is the dilated convolutional layer, and CNN1*1 is the convolutional layer with a 1*1 kernel.

[0112] The formula for the kernel size of a dilated convolutional layer is as follows:

[0113] KT λ =λ×(KT0-1)+1;

[0114] λ j =max(λ) j+1 -2λ j ,λ j+1 -2(λ j+1 -λ j ),λ j );

[0115] Among them, KT λ KT0 is the size of the dilated convolution kernel, λ is the dilation factor, KT0 is the set kernel size, and max is the maximum value. j Let λ be the expansion factor of the j-th layer. j+1 is the dilation factor for the (j+1)th layer; the initial kernel size of the dilated convolution is adjusted based on the training data of the campus conference video.

[0116] After feature extraction, the data is input into the Transformer encoder to generate positional and color features;

[0117] The decoding layer restores the original structure using the Transformer decoder and inverse dilated convolution.

[0118] The kernel size and image size for inverse dilation convolution are calculated as follows:

[0119]

[0120] λ k =max(λ) k+1 -2λ k ,λ k+1 -2(λk+1 -λ k ),λ k );

[0121]

[0122] Among them, KT A λ is the size of the inverse dilated convolution kernel. A KT is the inverse expansion factor. C λ is the size of the inverse convolution kernel. k Let λ be the expansion factor of the k-th layer. k+1 is the dilation factor of the (k+1)th layer, outa is the output image size, ina is the input image size, padding is the padding size, and d is the stride.

[0123] The archival video data similarity calculation unit sorts all archival video feature data in chronological order and calculates the similarity of adjacent archival video feature data in turn to generate archival video similarity data.

[0124] The smart campus video archive feature model enables similarity calculation, labeling, and related data storage. This model allows for data segmentation according to a defined process when inputting data of varying durations and sizes, improving data processing speed. Combined with image feature recognition algorithms, it can quickly and accurately extract corresponding features from the video archive data, calculate the corresponding similarity, accurately grasp the continuous relationships between video archive data, ensure the coherence of video archive data during segmentation, and guarantee the local and global integrity of the video.

[0125] Specifically, the similarity data of the archived videos is calculated as follows:

[0126]

[0127] Where, γ sim For archival video similarity data, N is the number of features, and P is the number of features. i t Let P be the i-th color feature of the archival video feature data at the t-th timestamp. i t-1 Let i be the i-th color feature of the archive video feature data at timestamp t-1. Let i be the i-th location feature of the archival video feature data with the t-th timestamp. Let α be the i-th location feature of the archival video feature data at timestamp t-1. c and α l Weights are used to calculate the similarity data of archival videos;

[0128] α c and αl The default values ​​are 0.4 and 0.6. These values ​​were obtained by analyzing simulated campus meeting data. Each weight was calculated 30 times, and the average was taken after removing the highest and lowest values. The average was then compared with the actual data. The same data was used for all weights, as shown in Table 1.

[0129] Table 1α c and α l Comparison of similarity scores calculated with different weights and actual similarity scores

[0130] <![CDATA[α c ]]> <![CDATA[α l ]]> Similarity after calculation Actual similarity Difference 0.7 0.3 92.3％ 94.6％ 2.3％ 0.6 0.4 94.1％ 94.6％ 0.5％ 0.5 0.5 93.7％ 94.6％ 0.9％ 0.4 0.6 93.2％ 94.6％ 1.4％ 0.3 0.7 92.5％ 94.6％ 2.1％

[0131] In the specific implementation process, α c and α l Modifications can be made based on the data type and size of historical video archives, combined with expert experience;

[0132] Considering the content of the video itself, in order to ensure that the similarity calculation can achieve a high degree of accuracy, both color features and position features are calculated. Color features are obtained by simulating a wide-area view to obtain the features of the entire image, and position features are determined by simulating a local view to determine the position features of different key objects in the current frame. Through the above calculation, the similarity between adjacent frames can be accurately calculated, ensuring the continuity and integrity of similar data during segmentation to the greatest extent.

[0133] The archival video data similarity tagging unit tags pre-stored video files based on archival video similarity data, generating tagged pre-stored video file data;

[0134] The archive video feature output unit outputs and stores archive video feature data sorted in chronological order.

[0135] The archive video similarity output unit outputs tags pre-stored video archive data and stores it;

[0136] Segmentation points are generated based on the continuity similarity of the archives and the length of the video archive feature data.

[0137] Furthermore, the timestamp of the first frame of the video is used as the start timestamp; the segmentation point is determined by the similarity of the file continuity and the length of the video file feature data.

[0138] If the file continuity similarity is less than the continuity threshold, the timestamp corresponding to the file continuity similarity being less than the continuity threshold is used as the segmentation point; if the length of the video file feature data exceeds the segmentation continuity threshold, the timestamp corresponding to the segmentation continuity threshold length is used as the segmentation point.

[0139] The continuity threshold was determined through extensive data analysis and expert experience, and is set to 90% by default. This threshold can be modified based on the specific circumstances on campus. The continuity similarity ratio of adjacent file feature data is compared to the continuity threshold. The continuity similarity ratio is calculated as follows:

[0140]

[0141] Where, ω γ This is the ratio of continuous similarity. Let be the continuity similarity of the t-th timestamp. Let be the continuity similarity of the (t+1)th timestamp;

[0142] The segmentation continuity threshold is based on the current video frame rate. For example, a 120-frame campus meeting video has a segmentation continuity threshold of 120, which means the 120 images in the campus meeting video are split into frames; a 60-frame campus meeting video has a segmentation continuity threshold of 60, which means the 60 images in the campus meeting video are split into frames.

[0143] In distributed storage, there may be instances where content appears similar for extended periods. For example, during a faculty meeting on campus, when a teacher or leader is speaking, there might be instances where the content appears highly similar for a prolonged period. If segmentation is based solely on similarity, a segmentation could become excessively large, defeating the purpose of distributed storage. Therefore, a segmentation continuity threshold is added. If the current data length exceeds this threshold, segmentation is also performed. This ensures the continuity of the segmented data, guaranteeing local integrity, while also preventing excessively long data from becoming too large. This ensures the implementation of distributed storage and guarantees data security.

[0144] The first segmentation module of the smart campus video archive segmentation module segments the pre-stored video archives by segmentation points, obtains the video archive feature data corresponding to the segmented video archives, and obtains the video archive features of multiple first keyframes by combining the archive continuity similarity calculation.

[0145] Furthermore, the first keyframe video file feature is to obtain the number of frames of the pre-stored video file and the feature data of the segmented video file;

[0146] Based on the similarity and frame number of the video archive feature data, the positions of multiple first keyframe video archive features are obtained from the video archive feature data:

[0147]

[0148] Where, η FThe location of the first keyframe video file feature, φ F and φ V For the pre-selected position, φ video For the number of frames; For calculation and The position of the closest value, k n The number of similarities calculated for pre-selected locations. Let be the similarity of the feature data of the i-th video file. For k n The similarity of the pre-selected locations, where min is the minimum value and max is the maximum value;

[0149] 20 is the minimum number of frames per second that the human eye can observe in continuous video; if it is less than this, stuttering will occur. In actual video capture, 30, 60, 120 frames per second and above are more common. Since existing video data generally uses 60 frames per second and above, this invention uses 60 frames per second as the reference, hence using 3 as the dividing point. 30 frames per second is retained because the difference from 20 frames per second is small. Taking a campus meeting video as an example, if the captured campus meeting video has a frame rate φ... video ≤60; For example, a campus meeting video has 30 frames, which is 1.5 frames even according to the formula. However, this number cannot be obtained in actual data acquisition. Since the difference between 30 frames and 20 frames is small, each frame is rounded down as the first keyframe video file feature. No position calculation is performed. All 30 frames are used as the first keyframe video file feature.

[0150] To ensure the security of data backup and the integrity of data restoration, the location of the first keyframe video archive feature is obtained. This allows for the acquisition of the most critical data among all feature data. To ensure continuity, calculations are performed with 20 frames as the minimum integrity level. By selecting the first keyframe video archive feature, data security can be guaranteed in the event of leakage. Even in the event of significant data loss, the restored data will still exhibit minimum integrity and smoothness. Furthermore, the limited amount of feature data saves space on the distributed storage server, improving space utilization efficiency, both in terms of the data itself and the amount of data.

[0151] The second segmentation module of the smart campus video archive obtains the video archive feature data corresponding to the segmentation point, and generates multiple second keyframe video archive features by combining the average file continuity similarity of the feature data of adjacent segmented video archives. The acquisition of second keyframe video archive features involves obtaining and storing the video archive feature data of segmentation point A; and obtaining and storing the average file continuity similarity of the segmentation point within the similarity range (M-1, M) and (M, M+1). Here, M is the segmentation point, M-1 is the previous segmentation point with the smallest number of frames away from M, and M+1 is the next segmentation point with the smallest number of frames away from M. During the restoration process, in order to facilitate the query of segmented data, similarity data can be quickly associated with the original data. At the same time, by backing up the segmentation points, the integrity and smoothness of the data during connection can be guaranteed. In addition, storing based on feature data can reduce the capacity and improve data security, preventing the acquisition of key information in the event of data leakage.

[0152] The smart campus video archive storage module stores the segmented, pre-stored video archives into the target storage node;

[0153] The smart campus feature backup module stores the segmented video archive feature data into the target storage node.

[0154] The smart campus secondary feature backup module stores the features of the first keyframe video file and the features of the second keyframe video file into the target storage node.

[0155] The smart campus archive storage module, the smart campus primary feature backup module, and the smart campus secondary feature backup module also include:

[0156] The smart campus archive storage module stores similarity data of video archive feature data corresponding to pre-stored video archives; the smart campus primary feature backup module and the smart campus secondary feature backup module store similarity data of corresponding video archive feature data.

[0157] The primary feature backup module and the secondary feature backup module of the smart campus restore feature data through the archive video feature restoration unit of the smart campus video archive feature model.

[0158] By storing this similarity data, key information for data search can be improved and matching speed can be accelerated when data is retrieved; at the same time, the corresponding backup data can be accurately searched and restored when data is restored, which can also reduce the number of data traversals and improve efficiency.

[0159] Example 2

[0160] To provide a detailed explanation, this embodiment is illustrated using the scenario of having only multiple file types. For specific illustration, this embodiment uses open courses, campus meeting videos, and surveillance videos as examples, and the selected campus meeting videos are the same as those in Embodiment 1:

[0161] A distributed storage-based archive management system includes a smart campus video archive acquisition module, a smart campus video archive segmentation and judgment module, a smart campus video archive first segmentation module, a smart campus video archive second segmentation module, a smart campus video archive storage module, a smart campus primary feature backup module, and a smart campus secondary feature backup module, specifically:

[0162] The smart campus video archive acquisition module obtains the file type of pre-stored video archives, matches multiple corresponding storage nodes, filters storage nodes, and generates target storage nodes.

[0163] Each storage node corresponds to a file type. There are multiple groups of storage nodes of the same file type, and each group contains multiple storage nodes. File types include personal video files, laboratory video files, open course video files, surveillance video files, event video files, conference video files, temporary video files, and unclassified video files. In this embodiment, since it includes open course videos, campus conference videos, and surveillance videos, the storage nodes corresponding to open course video files, surveillance video files, and conference video files will be searched and stored.

[0164] Furthermore, the number of storage nodes of the same file type is at least three groups. Each group of storage nodes is divided according to its functional type. The multiple groups of storage nodes include a video original file storage group, a video feature file backup group, and a video feature file restoration group. Each file of open course video files, surveillance video files, and conference video files corresponds to three groups of functions: original file storage group, video feature file backup group, and video feature file restoration group.

[0165] The smart campus video archive segmentation and judgment module extracts multiple image archive data from pre-stored video archives to obtain video archive feature data, and calculates the archive continuity similarity of the video archive feature data;

[0166] Furthermore, the video archive feature data and archive continuity similarity are judged through the smart campus video archive feature model; the smart campus video archive feature model includes an archive video data input unit, an archive video data feature extraction unit, an archive video data similarity calculation unit, an archive video data similarity marking unit, an archive video feature output unit, an archive video similarity output unit, and an archive video feature restoration unit;

[0167] The archival video data input unit classifies and labels the archive types of pre-stored video archives, and preprocesses the pre-stored video archives to generate archival video input data and archival video type labels;

[0168] Specifically, the archive video similarity data is used to tag pre-stored video archives as archive video similarity data for each timestamp of the pre-stored video archive;

[0169] If the similarity data of the archive video at the t-th timestamp is less than the archive continuity similarity, then the first mark is made;

[0170] Calculate the difference θ between the archival video similarity data at timestamp t and the archival video similarity data at timestamp t+1. t1 Calculate the difference θ between the archival video similarity data at timestamp t and the archival video similarity data at timestamp t-1. t2 ;

[0171] If θ t1 or θ t2 If the difference in similarity between the files is greater than the file similarity difference, then a second labeling is performed;

[0172] The archival video data feature extraction unit calls the corresponding video feature extraction module based on the archival video type label, and extracts the features of the archival video input data through the video feature extraction module to generate archival video feature data; in this embodiment, it calls the video feature extraction module trained based on campus video conference archival data, the video feature extraction module trained based on surveillance video archival data, and the video feature extraction module trained based on open course video archival data;

[0173] The video feature extraction module consists of an encoding layer and a decoding layer. The encoding layer comprises an improved YOLOv5 acquisition layer and a Transformer encoder. The video feature extraction module only uses the YOLOv5 backbone layer and neck layer. The video feature extraction module includes position feature extraction and color feature extraction. Position feature extraction uses the YOLOv5 backbone layer and neck layer; color feature extraction uses an improved bankbone layer and neck layer based on dilated convolution. The structure of color feature extraction is as follows:

[0174] output TCN =neck(backbone) TCN (input));

[0175]

[0176]

[0177] Where, output TCN For output features, neck is the neck layer, backbone TCN This is a backbone layer based on dilated convolution, where input is the input data; Conv6*6 is a convolutional layer with a kernel size of 6*6, TCNV1, TCNV2, TCNV3, TCNV4, TCNV... m This is a feature extraction layer based on dilated convolution, with SPPF being the SPPF layer used for pooling in YOLOv5, and TCNS. m For the m-th dilated convolution group, TCN11 m The m-th CSP structure based on dilated convolution;

[0178] TCNS m and TCN11 m The structure is as follows:

[0179]

[0180] Among them, ReLU is the activation layer, BN is the BN layer, TCN is the dilated convolution layer, and CNN1*1 is the convolution layer with a 1*1 kernel. The initial kernel size of the dilated convolution is adjusted based on the training data of the campus meeting videos, open courses, and surveillance videos.

[0181] The formula for the kernel size of a dilated convolutional layer is as follows:

[0182] KT λ =λ×(KT0-1)+1;

[0183] λ j =max(λ) j+1 -2λ j ,λ j+1 -2(λ j+1 -λ j ),λ j );

[0184] Among them, KT λ KT0 is the size of the dilated convolution kernel, λ is the dilation factor, KT0 is the set kernel size, and max is the maximum value. j Let λ be the expansion factor of the j-th layer. j+1 The expansion factor of the (j+1)th layer;

[0185] After feature extraction, the data is input into the Transformer encoder to generate positional and color features;

[0186] The decoding layer restores the original structure using the Transformer decoder and inverse dilated convolution.

[0187] The kernel size and image size for inverse dilation convolution are calculated as follows:

[0188]

[0189] λ k =max(λ) k+1 -2λ k ,λ k+1 -2(λ k+1 -λ k ),λ k );

[0190]

[0191] Among them, KT A λ is the size of the inverse dilated convolution kernel. A KT is the inverse expansion factor. C λ is the size of the inverse convolution kernel. k Let λ be the expansion factor of the k-th layer. k+1 is the dilation factor of the (k+1)th layer, outa is the output image size, ina is the input image size, padding is the padding size, and d is the stride.

[0192] The archival video data similarity calculation unit sorts all archival video feature data in chronological order and calculates the similarity of adjacent archival video feature data in turn to generate archival video similarity data.

[0193] Specifically, the similarity data of the archived videos is calculated as follows:

[0194]

[0195] Where, γ sim For archival video similarity data, N is the number of features, and P is the number of features. i t Let P be the i-th color feature of the archival video feature data at the t-th timestamp. i t-1 Let i be the i-th color feature of the archive video feature data at timestamp t-1. Let i be the i-th location feature of the archival video feature data with the t-th timestamp. Let α be the i-th location feature of the archival video feature data at timestamp t-1. c and α l Weights are used to calculate the similarity data of archival videos;

[0196] alpha of open courses and surveillance videos c and α lDifferent values ​​were used: 0.3 and 0.7 for public courses, and 0.5 and 0.5 for surveillance videos. Each simulated dataset was calculated 30 times, and the highest and lowest values ​​were removed before averaging. This average was then compared with the actual data. The same data was used for all actual data with different weights; see Tables 2 and 3 for reference.

[0197] Table 2α c and α l Comparison of similarity scores calculated with different weights and actual similarity scores for open courses

[0198] <![CDATA[α c ]]> <![CDATA[α l ]]> Similarity after calculation Actual similarity Difference 0.7 0.3 90.2％ 93.2％ 3.0％ 0.6 0.4 91.8％ 93.2％ 1.5％ 0.5 0.5 92.2％ 93.2％ 1.0％ 0.4 0.6 92.6％ 93.2％ 0.6％ 0.3 0.7 92.8％ 93.2％ 0.4％ 0.2 0.85 92.3％ 93.2％ 0.9％

[0199] Table 3α c and α l Comparison of similarity scores calculated with different weights and actual similarity scores in surveillance videos.

[0200] <![CDATA[α c ]]> <![CDATA[α l ]]> Similarity after calculation Actual similarity Difference 0.7 0.3 94.7％ 96.9％ 2.2％ 0.6 0.4 95.0％ 96.9％ 1.9％ 0.5 0.5 96.1％ 96.9％ 0.8％ 0.4 0.6 95.9％ 96.9％ 1.0％ 0.3 0.7 94.3％ 96.9％ 2.6％

[0201] The archival video data similarity tagging unit tags pre-stored video files based on archival video similarity data, generating tagged pre-stored video file data;

[0202] The archive video feature output unit outputs and stores archive video feature data sorted in chronological order.

[0203] The archive video similarity output unit outputs tags pre-stored video archive data and stores it;

[0204] Segmentation points are generated based on the continuity similarity of the archives and the length of the video archive feature data.

[0205] Furthermore, the timestamp of the first frame of the video is used as the start timestamp; the segmentation point is determined by the similarity of the file continuity and the length of the video file feature data.

[0206] If the file continuity similarity is less than the continuity threshold, the timestamp corresponding to the file continuity similarity being less than the continuity threshold is used as the segmentation point; if the length of the video file feature data exceeds the segmentation continuity threshold, the timestamp corresponding to the segmentation continuity threshold length is used as the segmentation point.

[0207] Because it includes various types of data, the continuity thresholds are set at 90% by default for campus meeting videos, 85% for open courses, and 95% for surveillance videos, after combining a large amount of data with expert experience. Specific settings can be modified according to the actual needs and circumstances of each campus.

[0208] The continuity threshold is set according to the frame rate standard of each video file. For example, if a public course has 30 frames, then the continuity threshold for the public course is 30; if a surveillance video has 60 frames, then the continuity threshold for the surveillance video is 60.

[0209] The first segmentation module of the smart campus video archive segmentation module segments the pre-stored video archives by segmentation points, obtains the video archive feature data corresponding to the segmented video archives, and obtains the video archive features of multiple first keyframes by combining the archive continuity similarity calculation.

[0210] Furthermore, the first keyframe video file feature is to obtain the number of frames of the pre-stored video file and the feature data of the segmented video file;

[0211] Based on the similarity and frame number of the video archive feature data, the positions of multiple first keyframe video archive features are obtained from the video archive feature data:

[0212]

[0213] Where, η F The location of the first keyframe video file feature, φ F and φ V For the pre-selected position, φ video For the number of frames; For calculation and The position of the closest value, k n The number of similarities calculated for pre-selected locations. Let be the similarity of the feature data of the i-th video file. For k n The similarity of the pre-selected locations, where min is the minimum value and max is the maximum value;

[0214] In the specific implementation process, regardless of the video type, 20 is the lowest number of consecutive frames that the human eye can observe in a video. Therefore, if the calculated value of a public course, surveillance video, or campus meeting video is less than 3, then each frame is the first keyframe video file.

[0215] The second segmentation module of the smart campus video archive acquires the video archive feature data corresponding to the segmentation point, and generates multiple second keyframe video archive features by combining the average archive continuity similarity of the feature data of adjacent segmented video archives. The acquisition of second keyframe video archive features involves storing the video archive feature data of segmentation point A; and storing the average archive continuity similarity of the segmentation point within the similarity ranges (M-1, M) and (M, M+1). Here, M is the segmentation point, M-1 is the previous segmentation point with the smallest distance from M frames, and M+1 is the next segmentation point with the smallest distance from M frames.

[0216] The smart campus video archive storage module stores the segmented, pre-stored video archives into the target storage node;

[0217] The smart campus feature backup module stores the segmented video archive feature data into the target storage node.

[0218] The smart campus secondary feature backup module stores the features of the first keyframe video file and the features of the second keyframe video file into the target storage node.

[0219] The smart campus archive storage module, the smart campus primary feature backup module, and the smart campus secondary feature backup module also include:

[0220] The smart campus archive storage module stores similarity data of video archive feature data corresponding to pre-stored video archives; the smart campus primary feature backup module and the smart campus secondary feature backup module store similarity data of corresponding video archive feature data.

[0221] The primary feature backup module and the secondary feature backup module of the smart campus restore feature data through the archive video feature restoration unit of the smart campus video archive feature model.

[0222] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A file management system based on distributed storage, characterized in that, include: The smart campus video archive acquisition module obtains the file type of pre-stored video archives and matches multiple corresponding storage nodes, filters the storage nodes, and generates a target storage node; The smart campus video archive segmentation and judgment module extracts multiple image archive data from pre-stored video archives to obtain video archive feature data, calculates the archive continuity similarity of the video archive feature data, and generates segmentation points based on the archive continuity similarity and the length of the video archive feature data. The first segmentation module of the smart campus video archive segments the pre-stored video archive through the segmentation point, obtains the video archive feature data corresponding to the segmented video archives, and calculates multiple first keyframe video archive features by combining the archive continuity similarity. The second segmentation module of the smart campus video archive obtains the video archive feature data corresponding to the segmentation point, and generates multiple second keyframe video archive features by combining the average file continuity similarity of the feature data of adjacent segmented video archives. The smart campus video archive storage module stores the segmented pre-stored video archive into the target storage node; The smart campus feature backup module stores the segmented video file feature data into the target storage node; The smart campus secondary feature backup module stores the first keyframe video file features and the second keyframe video file features into the target storage node.

2. The archive management system based on distributed storage according to claim 1, characterized in that, The storage node corresponds to a file type, and there are multiple groups of storage nodes of the same file type, with each group containing multiple storage nodes; the file types include personal video files, laboratory video files, open course video files, surveillance video files, event video files, meeting video files, temporary video files, and unclassified video files.

3. The archive management system based on distributed storage according to claim 2, characterized in that, The number of storage nodes of the same file type is at least three groups. Each group of storage nodes is divided according to its functional type. The multiple groups of storage nodes include a video original file storage group, a video feature file backup group, and a video feature file restoration group.

4. The archive management system based on distributed storage according to claim 1, characterized in that, The segmentation criteria based on the continuity similarity of the archives and the length of the video archive feature data include: The start timestamp is the timestamp of the first frame of the video; the segmentation point is determined by the continuity similarity of the archives and the length of the video archive feature data. If the file continuity similarity is less than the continuity threshold, then the timestamp corresponding to the file continuity similarity being less than the continuity threshold is used as the segmentation point; if the length of the video file feature data exceeds the segmentation continuity threshold, then the timestamp corresponding to the segmentation continuity threshold length is used as the segmentation point. The next start timestamp is determined by the split point, and the process continues until the entire video has been traversed.

5. The archive management system based on distributed storage according to claim 1, characterized in that, The video archive feature data and the archive continuity similarity are determined using a smart campus video archive feature model, including: The smart campus video archive feature model includes an archive video data input unit, an archive video data feature extraction unit, an archive video data similarity calculation unit, an archive video data similarity marking unit, an archive video feature output unit, an archive video similarity output unit, and an archive video feature restoration unit; The archive video data input unit classifies and marks the archive types of the pre-stored video archives, and preprocesses the pre-stored video archives to generate archive video input data and archive video type tags; The archival video data feature extraction unit calls the corresponding video feature extraction module according to the archival video type label, and extracts the features of the archival video input data through the video feature extraction module to generate archival video feature data; The archive video data similarity calculation unit sorts all the archive video feature data in chronological order and calculates the similarity of adjacent archive video feature data in turn to generate archive video similarity data. The archive video data similarity marking unit marks the pre-stored video archives based on the archive video similarity data, and generates marked pre-stored video archive data; The archive video feature output unit outputs and stores the archive video feature data sorted according to the time order. The archive video similarity output unit outputs and stores the marked pre-stored video archive data; The archive video feature restoration unit restores the archive video feature data.

6. The archive management system based on distributed storage according to claim 5, characterized in that, Tagging the pre-stored video archives based on the video similarity data includes: For each timestamp, the pre-stored video archive is tagged with archive video similarity data; If the similarity data of the archive video at the t-th timestamp is less than the continuity similarity of the archive, a first mark is made; Calculate the difference θ between the archival video similarity data at the t-th timestamp and the archival video similarity data at the (t+1)-th timestamp. t1 Calculate the difference θ between the archive video similarity data at the t-th timestamp and the archive video similarity data at the (t-1)-th timestamp. t2 ; If θ t1 and / or θ t2 If the difference in similarity between the files is greater than the file similarity difference, then a second labeling is performed.

7. A file management system based on distributed storage according to claim 5, characterized in that, The calculation of the archive video similarity data includes: Where, γ sim The archive video similarity data is given, where N is the number of features and P is the number of features. i t For the i-th color feature of the archive video feature data at the t-th timestamp, P i t-1 Let i be the i-th color feature of the archive video feature data at the (t-1)-th timestamp. Let i be the i-th location feature of the archive video feature data at timestamp t. For the i-th location feature of the archive video feature data at the (t-1)-th timestamp, α c and α l The weights for calculating the similarity data of the archive videos are denoted as .

8. The archive management system based on distributed storage according to claim 1, characterized in that, The first keyframe video file feature acquisition includes: Obtain the frame count of the pre-stored video file and the feature data of the segmented video file; Based on the similarity of the video archive feature data and the number of frames, the positions of multiple first keyframe video archive features are obtained from the video archive feature data: Where, η F φ represents the location of the first keyframe video file feature. F and φ V For the pre-selected position, φ video The number of frames; For calculation and The position of the closest value, k n The number of similarities calculated for the pre-selected locations. The similarity of the feature data of the i-th video file is... For k n The similarity of the pre-selected locations, where min is the minimum value and max is the maximum value.

9. A file management system based on distributed storage according to claim 1, characterized in that, The second keyframe video file features are obtained as follows: The video archive feature data of the segmentation points are obtained and stored. The file continuity similarity mean values ​​of the similarity range (M-1, M) and (M, M+1) of the segmentation point are obtained and stored; where M is the segmentation point, M-1 is the previous segmentation point with the smallest distance from M frames, and M+1 is the next segmentation point with the smallest distance from M frames.

10. A file management system based on distributed storage according to claim 5, characterized in that, The smart campus archive storage module, the smart campus primary feature backup module, and the smart campus secondary feature backup module also include: The smart campus archive storage module stores similarity data of video archive feature data corresponding to pre-stored video archives; the smart campus primary feature backup module and the smart campus secondary feature backup module store similarity data of corresponding video archive feature data; The smart campus primary feature backup module and the smart campus secondary feature backup module restore feature data through the archive video feature restoration unit of the smart campus video archive feature model.

Images