A multi-source video streaming method based on cache-assisted super-resolution technology

Abstract

The present invention discloses a multi-source video streaming transmission method with cache-assisted super-resolution technology, which improves video quality while minimizing the overall system delay. The method runs in a proxy server, which is located between a client and an edge server in a network. After detecting a designated video playback request from a client, the proxy server makes a set-oriented video segment caching decision, determines a first video segment segment that matches the target video, and caches it. During video playback, for segments that are not detected in the proxy server, a second video segment segment in a designated edge server is obtained, wherein the resolution of the first video segment segment is greater than that of the second video segment segment; the resolution of the second video segment segment is reconstructed to the resolution of the first video segment segment, and a target video stream is generated based on the reconstructed second video segment segment and the first video segment segment; the target video stream is sent to a user terminal to play the target video.

Description

Technical Field

[0001] The present invention belongs to the field of video transmission technology, and in particular relates to a method for designing a collaborative storage and computing resource enhancement method for video transmission performance under a multi-level distributed edge service network architecture. Background Art

[0002] With the rapid development of multimedia technology, video streaming has become an indispensable service in people's daily lives. Simultaneously, the demand for video resolution has gradually evolved from standard definition (SD) and high definition (HD) to ultra-high definition (UHD) and other standards. To achieve higher image quality and provide users with a better viewing experience, transmitting UHD video typically requires high-speed and stable downlink bandwidth. For example, the transmission bitrate of 4K video (resolution of 3840×2160) is typically between 20-50Mbps, while the transmission bitrate of 8K video (resolution of 7680×4320) exceeds 100Mbps. With the continuous increase in the number of users and viewing time in the internet video industry, the demand for this high-speed and stable bandwidth has exploded, posing a significant challenge to the existing network environment. Consequently, users often experience video freezes and interruptions when watching UHD videos due to insufficient bandwidth, resulting in a poor user experience.

[0003] New network architectures, such as cloud-native 5G networks, aim to provide high bandwidth, low latency, and multiple connections, bringing changes and upgrades to all aspects of the ultra-high-definition video industry chain. Deploying video streaming transmission services in such a multi-level distributed network architecture can collaboratively utilize computing and storage resources and transmit in a multi-source mode, ultimately enhancing the user experience quality of video playback services. Traditional multi-source video streaming transmission solutions usually store all quality versions of the entire video in multiple servers. Compared with traditional single-source transmission, more storage resources will be consumed in the network to support multi-source transmission. In addition, the bandwidth within the network fluctuates violently, and it is difficult for the client to accurately predict the bandwidth of multiple end-to-end devices. Direct transmission of high-resolution videos can easily cause video freezes. Therefore, providing a video transmission solution that efficiently utilizes various resources within the network is a technical problem that needs to be solved urgently by those skilled in the art. Summary of the Invention

[0004] The present invention provides a multi-source video streaming media transmission method using a cache-assisted super-resolution technology, which improves video quality while minimizing overall system delay.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A multi-source video streaming method using cache-assisted super-resolution technology comprises the following steps:

[0007] (1) Video generation and deployment:

[0008] A video is divided into a series of N segments of equal length in time. Each video segment i is spatially divided into a series of J non-overlapping tiles of equal length and width. All tiles of a video segment are divided into M sets, where the sum of the tile data sizes within each set is kept as close as possible. To reduce the amount of control signaling and narrow the decision space, tile sets are used as the minimum unit for storage and transmission. There are M edge servers in the network, and a low-resolution version of each set is stored on one edge server.

[0009] (2) Cache decision of video segment segment set:

[0010] The proxy server is located between the client and the edge server in a multi-level distributed network architecture, close to the client. Acting as a proxy for the client, the proxy server receives video streaming requests from the client. Upon detecting a video streaming request for a target video, the proxy server pre-caches a set of high-resolution video clip segments. These segments are the first video clip segments that match the target video.

[0011] (3) Video playback process:

[0012] Upon receiving a request for video segment i from a client, the proxy server performs the following steps:

[0013] ① Searching for a high-resolution tile set of segment i cached in the proxy server and determining a tile identifier of the first video segment tile;

[0014] ② Sending a video segment segment set request to each edge server, wherein the video segment segment set request is used to instruct each edge server to send a second video segment segment corresponding to a segment identifier other than the segment identifier of the first video segment segment to the proxy server, wherein the resolution of the second video segment segment is lower than the resolution of the first video segment segment;

[0015] ③ Using two super-resolution technologies, reconstruct the second video segment transmitted from the edge server into a high-resolution segment with the resolution value of the first video segment;

[0016] ④ Obtain all the segments, and splice the reconstructed second video segment segments and the first video segment segments into a whole video segment, generate a target video stream, and transmit it to the client.

[0017] Furthermore, the proxy server constructs a cache decision problem based on a set of video segment segments through mathematical modeling. For each video segment i, the transmission delay l between each edge server and the proxy server is known. im. After the low-resolution video segment slices (second video segment slices) are transmitted to the proxy server, the proxy server uses two super-resolution technologies to reconstruct different slices: (1) the traditional interpolation method, which has poor reconstruction performance but short running time and can be executed only on the CPU; (2) the method based on deep learning, which has good reconstruction performance but long running time and needs to be executed on the GPU. Some slices with simple content (such as slices that form the background of night, blue sky, etc.) can achieve good performance using only the interpolation method without deep learning. The amount of data for these slices is usually small, so this method determines which super-resolution technology to use based on the data size of the slice. It is known that the time d to reconstruct a slice using two different super-resolution technologies is ij and d' ij , we can get the total time g for reconstructing all the segments of video segment i on the proxy server i The overall delay f of a video segment i is obtained after making different caching decisions. i =l im +g i The user's attention to the video clip is used as the weight of the video clip v i ,The problem objective is to minimize the estimated weighted overall delay and meet the caching capacity upper limit of the proxy ,server.

[0018] Furthermore, when solving the set-oriented video segmentation caching decision problem, the proxy server has the following steps:

[0019] ① The minimization of the estimated weighted overall delay in the original problem is equivalently transformed into maximizing the difference between the maximum system delay and the actual delay;

[0020] ② Based on the proxy server's cache capacity limit, construct all possible cache policy allocation schemes and generate an optional set;

[0021] ③ Find the optimal cache strategy allocation scheme within the set of options. If there is a conflict in allocation, use the auction concept to differentiate the expected benefits of the cache strategy for different video clips. Repeat the solution until all conflicts are resolved. Cache the set of video clip segments according to the optimal scheme.

[0022] The beneficial effects of the present invention are:

[0023] The present invention takes into account the reduction of the number of signaling transmissions and the narrowing of the decision space of the problem, and uses the collection of video clip blocks as the minimum unit of transmission and decision-making; during the transmission process, the computing and storage resources in the network are simultaneously utilized to enhance the transmission performance and improve the quality of experience of video users. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0025] Figure 1 Flow chart of the method of the present invention.

[0026] Figure 2 This is a network architecture diagram of the present invention.

[0027] Figure 3 The present invention is a flowchart for solving the problem. DETAILED DESCRIPTION

[0028] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0029] Combined with attachment Figure 1-2 A multi-source video streaming method using cache-assisted super-resolution technology comprises the following steps:

[0030] (1) Video generation and deployment:

[0031] A video is divided into a series of N segments of equal length along the temporal dimension. Each video segment i is spatially divided into a series of J non-overlapping tiles of equal length and width. All tiles of a video segment are divided into M sets, where the sum of the tile data sizes within each set is kept as close as possible. To reduce the amount of control signaling and narrow the decision space, tile sets are used as the minimum unit for storage and transmission. There are M edge servers in the network, and the low-resolution video segment tiles (second video segment tiles) of each set are stored on one edge server.

[0032] (2) Cache decision of video segment segment set:

[0033] The proxy server, located between the client and the edge server in a multi-level distributed network architecture, acts as a proxy for the client and receives video streaming requests. Before streaming, the proxy server pre-cache a set of high-resolution video segments (the first video segment segments).

[0034] (3) Video playback process:

[0035] Upon receiving a request for video segment i from a client, the proxy server performs the following steps:

[0036] ① Confirm the first video segment of segment i cached in the proxy server and cut it into pieces;

[0037] ② Requesting at least one corresponding edge server for a set of second video segment segments that are not cached in the proxy server;

[0038] ③ Using two super-resolution technologies, reconstruct the second video segment transmitted from the edge server into a high-resolution segment with the resolution value of the first video segment;

[0039] ④ Obtain all the segments, and splice the reconstructed second video segment segments and the first video segment segments into a whole video segment, generate a target video stream, and transmit it to the client.

[0040] In this embodiment, the total duration of the video is known in step (1), and the video is divided into a series of N video segments of equal and fixed duration in the time dimension, and the index number of the video segment is i. Each video segment is divided into a series of blocks of equal length and width in the spatial dimension, and the index number of the block is j. All the blocks of a video segment are divided into M groups E im , when forming a set, ensure that the sum of the sizes of the block data between sets is as close as possible. Let α ijm =1 means that the slice j of video segment i belongs to the set E im There are M edge servers in the network, and a low-resolution version of each set (the second video segment) is stored on one edge server. Based on this, when the client sequentially requests video segments from the proxy server, the proxy server rewrites each request and sends a request to the M edge servers simultaneously, requesting each edge server for the set of segments it stores.

[0041] In this embodiment, the proxy server has certain computing and storage capabilities. Due to resource competition between different services in the network, the network bandwidth between the proxy server and the edge server fluctuates violently, making it difficult to accurately predict each end-to-end bandwidth. Therefore, this method deploys super-resolution technology at the proxy server. The video server on the edge server only needs to transmit a set of low-resolution tiles to the proxy server. The proxy server then uses super-resolution technology to reconstruct the low-resolution tiles into high-resolution tiles and transmit the high-resolution video to the client. At the same time, this method uses the storage resources of the proxy server to cache a portion of the high-resolution tiles in advance to reduce the overall time required for transmission and reconstruction and ensure the quality of user experience.

[0042] Before a video session begins, this method determines the caching strategy of the video by estimating the overall time of transmission and reconstruction, that is, step (2). First, consider two super-resolution technologies to reconstruct different blocks in parallel: one is the traditional interpolation method, which has poor reconstruction performance but short running time and can be executed only on the CPU; the other is a method based on deep learning, which has good reconstruction performance but long running time and needs to be executed on the GPU. Some blocks with simple content (such as blocks that form the background of night, blue sky, etc.) can achieve good performance using only interpolation methods without deep learning. The amount of data for these blocks is usually small, so this method determines which super-resolution technology to use based on the data size of the block. Let e ij =1 means that the slice j needs to be reconstructed by the method based on the deep learning model. It is known that the time to reconstruct a slice using two different super-resolution techniques is d. ij and d' ij , we can get the total time g for reconstructing all the segments of video segment i on the proxy server i for

[0043] g i (0)=∑ j d ij e ij +∑ j d′ ij (1-e ij ).

[0044] In the proxy server, each tile is reconstructed sequentially. If the total reconstruction time is too long, it may cause video lag. This method uses the proxy server's own storage resources to cache some high-resolution tiles in the proxy server before the video session begins, reducing the reconstruction processing time. To narrow the decision space, the set of video tiles stored on an edge server is considered as a whole, and the decision on whether to pre-cache a set of video tiles is made. Let the control variable x im =1 means caching the high-resolution version of the segments in the set m of video segment i in the proxy server, x im =0 means no caching. After introducing caching, the total time for reconstructing video segment i is

[0045] g i (Σ m x im )=Σ j Σ m (1-x im )(α ijm de ij +α ijm d′(1-e ij ))

[0046] The proxy server still needs to request the corresponding edge server at low resolution for the slice sets that are not cached in advance, and then reconstruct them. For each video segment i, the transmission delay l between each edge server and the proxy server is known. im . Reorder the M proxy servers according to the transmission delay so that l i1 ≥l i2 ≥…≥l iM , the overall delay of a video segment i can be obtained:

[0047]

[0048] Obviously, the more slice sets a video can cache in the proxy server in advance, the lower its overall latency, but the proxy server's storage capacity is limited. Let C represent the upper limit of the number of slice sets that can be cached in the proxy server, then the following constraints apply:

[0049] ∑ i ∑ m x im ≤C.

[0050] In addition, since the content presented during video playback is constantly changing, users usually pay different attention to different video clips. For example, users tend to pay more attention to scenes where the storyline reaches a critical moment (such as the moment of a goal in a football game) and scenes where users need to obtain information (such as the moment when the camera shows the scoreboard in a basketball game). In addition, users pay less attention to transition clips that present the background. Let v i It represents the user's attention to video segment i, which serves as the weight of the video segment and affects the caching decision.

[0051] In summary, the cache decision problem in step (2) is modeled as follows:

[0052]

[0053] st∑ i ∑ m x im ≤C,

[0054]

[0055] By solving problem P, this method derives a cache decision allocation that minimizes the weighted overall delay, while not exceeding the specified cache limit. After solving the problem, the proxy server requests a high-resolution video segment slice (the first video segment slice) from the remote server for caching according to the caching strategy, and then proceeds with the video playback process, i.e., step (3).

[0056] Please see the attached Figure 3 The present invention provides a solution to the set-oriented tiling cache decision problem, including the following steps:

[0057] (1) Problem transformation. The original problem P, which belongs to nonlinear integer programming, is equivalently transformed to construct an M+1-dimensional weighted delay difference array for each video segment i:

[0058] {τ i，0 , τ i，1 ,...,τ i，M}={0,[l i1 +g i (0)-l i2 -g i (1)]υ i ,...,[l i1 +g i (0)-l im -g i (m-1)]v i ,...,[l i1 +g i (0)-l iM -g i (M-1)]υ i ,(l i1 +g i (0))υ i},

[0059] where τ i,m =[l i1 +g i (0)-l im+1 -g i (m)]v i It represents the total transmission and reconstruction delay of video segment i that is expected to be reduced by multiplying the weight value when m-1 sets are cached in the proxy server, that is, the difference with the maximum weighted delay (i.e., the case of no caching). Let k i =∑ m x im Minimizing the original problem P is equivalent to maximizing the reduced weighted overall delay, which is as follows:

[0060]

[0061] stk i -Σ m x im ,

[0062] ∑ i ∑ m x im =C,

[0063]

[0064] (2) Construct a set of cache policy allocation schemes. Based on the known upper limit of the proxy server's cache capacity, construct an optional set of cache policy allocation schemes. Introduce the auxiliary variable y k , represents the number of video clips cached for k sets of tiles, Define {y1, y2, ..., y M} is a possible cache allocation scheme, and the following equation can be obtained:

[0065]

[0066] The first equation indicates that the sum of the number of slice sets cached for all video segments is equal to the cache upper limit, and the second equation indicates that the sum of the number of video segments assigned with cache decisions does not exceed the total number of video segments.

[0067] All possible cache strategy allocation schemes can be obtained through iterative search. First, define y k The upper limit of allocable value is

[0068]

[0069] The steps to construct the solution set algorithm are as follows:

[0070] ① Initialize {y1, y2, ..., y M} = {C, 0, ..., 0};

[0071] ② If under this cache strategy allocation scheme If {y1, y2, ..., y M} is a feasible solution; for y2, jump to step ③;

[0072] ③If Then y k ←y k +1, otherwise y k ←0;

[0073] ④ For {y3, ..., y M} Judge in sequence: If y k-1 is set to 0, then y k Jump to step ③;

[0074] ⑤ Assign value to y1, Skip to step ②.

[0075] (3) Obtain the best decision from the set of available options. First, based on the weighted delay difference array, a greedy strategy is used to determine the allocation scheme that maximizes the weighted delay difference. In this scheme, there may be a conflict between two caching strategies matching the same video clip, and each video clip can only have one caching strategy k. i In order to eliminate the conflict, first extract the set of conflicting matches {N [c] ,M [c]}, where N [c] Represents a set of video clips, M [c] Represents a cache strategy set. At the same time, an auction price p for competition is introduced for each video clip. i And initialized to 0; for each cache strategy, introduce an auction revenue τ' for competition i,k =τ i,k -p i , represents the revenue obtained by cache strategy k after being allocated to video segment i through auction competition. Initialize the auction revenue array τ' to the weighted delay difference array τ, while ensuring τ' i,k ≥ 0. Calculate the sum of auction revenue V k =sum_top(τ'(:,k),y k ), Represents the front y k The sum of the maximum auction revenue and the removal of {τ' i,k y before} k The sum of the maximum auction revenues. Auction price p i The variations are as follows:

[0076]

[0077] Update the value of τ'(i,:) to max{0,τ(i,:)-p i By continuously increasing the auction price, a cache allocation scheme with greater bidding power in the conflict set is found. The above process is repeated until the conflict set is empty, that is, all video clips have at most one cache strategy assigned. At this point, an optimal conflict-free cache strategy for an allocation scheme Y is obtained. Calculate the objective function of problem Q. If the objective function value is greater than the objective function values ​​of all cache decisions obtained previously, then is the final decision result. When all cycles are completed, the optimal decision result is obtained.

[0078] The following is an explanation of the symbols in the above formula:

[0079] Number of video segments N

[0080] Video clip index i

[0081] Number of video segments J

[0082] Video segmentation index j

[0083] Number of edge servers M

[0084] The set E of all video segments of video segment i stored in server m im

[0085] Edge server index m

[0086] Transmission delay between edge server and proxy server im

[0087] The time d of the two super-resolution techniques for the reconstruction of a certain block ij and d' ij

[0088] The total time g for reconstructing all the segments of video segment i on the proxy server i

[0089] The overall delay f of video segment i i

[0090] Define the weight v of video segment i according to the user's attention to video segment i i

[0091] Indicates whether the segment j of video segment i is in the video segment set E im Middle a ijm

[0092] Indicates whether the segment j of video segment i is reconstructed using the deep learning model e ij

[0093] Indicates whether the set m of video segment i caches the high-resolution version x im

[0094] The overall delay f of video segment i i

[0095] The upper limit C of the number of segments that a video can cache in the proxy server

[0096] When m sets are cached in the proxy server, the total transmission and reconstruction delay of video segment i is expected to be reduced by the weight value τ compared to the maximum weighted delay (i.e., no caching). i,m

[0097] The number k of video clips i cached by the proxy server i

[0098] The number of video clips y cached for k sets of slices k

[0099] y k The allocatable upper limit

[0100] Video clip collection N [c]

[0101] Cache strategy set M [c]

[0102] The competitive auction price p for video clip i i

[0103] The payoff τ' assigned to video clip i by video caching strategy k in competitive auction i,k

[0104] before y k The sum of the maximum auction revenues V k

[0105] Remove {τ' i，k y before} k The sum of the maximum auction revenues

[0106] The optimal conflict-free buffering strategy for a certain allocation scheme Y

[0107] The present invention adopts the computer simulation mode to carry out the experiment.

[0108] In practice, the present invention is deployed in a proxy server close to the client, connecting the client and the edge server. Software simulation is performed on the proxy server to initialize the proxy server capacity and the transmission delay between servers. Assume that there are 6 edge servers participating in the transmission in the network, which are distributed in different geographical locations. The transmission delay between each edge server and the proxy server is quite different and is randomly generated between 1-100 milliseconds. Bicubic interpolation is used as a simple reconstruction method using the CPU and runs on an Intel i9 processor; the NAS model is used as a deep learning model reconstruction method using the GPU and runs on a Nivdia GTX3090 graphics card. Experiments are conducted on two videos, "Surf" (205 seconds long, sports category) and "Rhinos" (292 seconds long, documentary category). FFmpeg is used to split the video into video segments with a duration of 1 second in time and into blocks of 4 rows x 6 columns in space. The video is reconstructed from a resolution of 540p to 4K (one block is reconstructed from a resolution of 160x135 to a resolution of 640x540), and the processing time of each block is recorded. All blocks of each video clip are evenly divided into 6 sets according to the data size, and their low-resolution versions are stored in 6 edge servers respectively. According to a certain ratio of the data size of all blocks, each block is determined to be reconstructed using the bicubic interpolation method or the NAS model method. Taking visual saliency as the degree of user attention to the video, the visual saliency of each video frame is calculated, and the average of all frames belonging to a video clip is calculated, and then normalized as the degree of attention of each video clip. Other personalized user interest metrics are also applicable to the present invention.

[0109] First, a set-oriented video segmentation cache decision is made, and the cache information is simulated based on the decision result. Then, the entire playback process is simulated. The client requests video segments from the proxy server in sequence in the form of streaming media. After receiving the request, the proxy server rewrites the request and requests a set of uncached low-resolution video segments (second video segment segments) from the designated edge server. The edge server sends the set, and after the proxy server receives the segments, it uses two super-resolution methods to reconstruct the low-resolution segments into high-resolution segments. After obtaining all the high-resolution segments, the proxy server splices them into the entire video segment and then transmits it to the client. The experimental evaluation criteria are the quality of the video viewed by the user and the total freeze time of the video playback. The experimental results show that this method can reduce the total freeze time during playback while ensuring video quality.

[0110] The various embodiments of the present invention are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0111] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-source video streaming method using cache-assisted super-resolution technology, characterized in that: The method is performed by a proxy server, and includes: If it is determined that a video playback request of a target video is detected, determining a first video segment segment matching the target video; If all first video segment segments are not detected locally on the proxy server, obtaining second video segment segments matching the target video from at least one designated edge server, wherein the second video segment segments are video segment segments that are not detected locally, and a resolution of the first video segment segments is greater than a resolution of the second video segment segments; Reconstructing the resolution of the second video segment slice to the resolution of the first video segment slice using two super-resolution technologies, and generating a target video stream based on the reconstructed second video segment slice and the first video segment slice; Send the target video stream to the user terminal to play the target video; When solving the cache decision problem for a tiled set, the proxy server has the following steps, including: (1) The original problem of minimizing the estimated weighted overall delay is equivalently transformed into maximizing the difference between the maximum system delay and the actual delay; (2) Based on the upper limit of the proxy server's cache capacity, construct all possible cache policy allocation schemes and generate an optional set; (3) Find the optimal cache strategy allocation scheme in the optional set. If there is an allocation conflict, use the auction competition idea to distinguish the expected benefits brought by the cache strategy for different video clips. Repeat the solution until all conflicts are eliminated, and cache the video clip block set according to the optimal solution.

2. The multi-source video streaming method using cache-assisted super-resolution technology according to claim 1, characterized in that: When determining a first video segment segment matching the target video, the method includes: Performing spatial segmentation on each video segment corresponding to the target video to obtain first and second video segment segments; Cutting each second video segment into pieces and sending them to each edge server for storage; At least one set of target first video segment segments is determined, and the first video segment segments are stored in the proxy server.

3. The multi-source video streaming method using cache-assisted super-resolution technology according to claim 2, characterized in that: Slicing each video segment corresponding to the target video to obtain each second video segment segment includes: Slicing a target video segment corresponding to the target video, and grouping each second video segment segment obtained by slicing to obtain at least two segment sets corresponding to the target video segment; Cutting each second video segment into pieces and sending them to each edge server for storage includes: Send each set of slices to each edge server for storage; wherein a difference between the sum of the sizes of the slice data of the second video segment corresponding to the slice sets stored in each edge server is less than or equal to a preset difference threshold; Determining at least one set of target first video segment segments includes: Determining a network delay of an edge server storing each set of slices, and determining a total reconstruction time required for super-resolution reconstruction of each slice of the second video segment in each set of slices; Based on the network delay and the sum of reconstruction time corresponding to each tile set, with the goal of minimizing the total time of the entire video, a cache decision problem for the tile set is solved, and the target tile set is determined in each tile set. Each first video segment block in the target segment block set is used as a target first video segment block.

4. The multi-source video streaming method using cache-assisted super-resolution technology according to claim 1, characterized in that: Obtaining a second video segment segment matching the target video from at least one designated edge server includes: Determining a segmentation identifier for segmenting the first video segment; A video segment segment acquisition instruction is sent to each edge server, wherein the video segment segment acquisition instruction is used to instruct each edge server to send second video segment segments corresponding to segment identifiers other than the segment identifier of the first video segment segment to the proxy server.

5. The multi-source video streaming method using cache-assisted super-resolution technology according to claim 1, characterized in that: The two super-resolution technologies include: (1) traditional interpolation method, which has poor reconstruction performance but short running time and can be executed only on CPU; (2) deep learning-based method, which has good reconstruction performance but long running time and needs to be executed on GPU. The choice of the two super-resolution technologies is determined by the size of the data slice.

6. The multi-source video streaming method using cache-assisted super-resolution technology according to claim 1 or 2, characterized in that: The proxy server is located between the client and the edge server, close to the client.

Images