CN112953922B

CN112953922B - Self-adaptive streaming media control method, system, computer equipment and application

Info

Publication number: CN112953922B
Application number: CN202110146565.0A
Authority: CN
Inventors: 栾浩; 穆康乐; 郑金凯; 承楠
Original assignee: Xidian University
Current assignee: Xidian University
Priority date: 2021-02-03
Filing date: 2021-02-03
Publication date: 2022-09-16
Anticipated expiration: 2041-02-03
Also published as: CN112953922A

Abstract

The invention belongs to the technical field of self-adaptive streaming media, and discloses a self-adaptive streaming media control method, a self-adaptive streaming media control system, computer equipment and application, wherein the self-adaptive streaming media control system comprises the following components: the system comprises an edge server, user equipment, a video analysis module and a decision-making module based on deep reinforcement learning. The invention considers the network throughput and the size of the buffer area by introducing the edge server; by adopting the deep reinforcement learning method, the video playing can be automatically optimized according to the network environment, and the video playing can be deployed in various network environments after the neural network training is finished, so that the method has better practicability. The invention creatively increases the new dimension of playing speed adjustment, and reduces the requirement of video transmission on network bandwidth by reducing the playing speed; through the video content analysis module, most users can not perceive the change of the playing speed, and the high-bit-rate video playing can be continuously maintained, so that the user experience is improved to a great extent.

Description

Self-adaptive streaming media control method, system, computer equipment and application

Technical Field

The invention belongs to the technical field of self-adaptive streaming media, and particularly relates to a self-adaptive streaming media control method, a self-adaptive streaming media control system, computer equipment and application.

Background

Currently, streaming video transmission technology is developed to provide high-quality multimedia services to users. In the case of Streaming video services, the most widely used protocol at present is Dynamic Adaptive Streaming over HTTP (DASH). Streaming video services using the DASH protocol first segment the video into several video blocks (chunks), each video block being 2 or tens of seconds long. And each video block is coded according to different code rates, so that one video block has a plurality of versions with different definitions. In the using process of a user, the user side firstly sends a request for requesting video resources to a remote server for storing videos, and then after receiving the request, the server sends video blocks to the user side one by one according to a time sequence. According to the network environment, the video block with high code rate or low code rate can be selected for transmission. When the user receives the video blocks from the server, the video blocks are placed into a play buffer area and then played from the buffer area one by one. If the network environment is not good, a video block with low code rate needs to be sent, so that the dynamic adjustment of the code rate is realized. If the play-out buffer is exhausted, a stuck occurs.

When a user watches a video by using a mobile device, the throughput of a network generally fluctuates greatly due to the instability of a wireless channel, so that the video is jammed in the playing process, or the user can only select to watch the video with a low bit rate, thereby affecting the user experience.

The existing solution mainly adopts an Adaptive Bitrate adjustment Algorithm (Adaptive Bitrate Algorithm or ABR) at a client, and adjusts the Bitrate of a video in real time according to the network throughput condition, so that a video with a high Bitrate is provided as much as possible under the condition of limited bandwidth, and video blockage is reduced as much as possible.

The ABR algorithms deployed at the client are mainly classified into three categories: 1. an ABR algorithm based on network throughput; 2. an ABR algorithm based on the play buffer; 3. ABR algorithm combining both.

And adjusting the code rate of the video according to the throughput of the network by using the ABR algorithm based on the network throughput. Generally, when the network throughput is low, the algorithm switches to a low video rate, and vice versa switches to a high rate. The method has the disadvantages that the network throughput fluctuation is large, the network throughput is difficult to predict under normal conditions, and the uncertainty is large, so that the scheme has frequent adjustment on the video playing code rate, and frequent code rate switching brings bad experience to users. Users prefer a stable video bitrate.

And adjusting the video code rate according to the size of the video playing buffer area based on the ABR algorithm of the playing buffer area. When a user watches videos, the user side firstly downloads the videos from the remote server to the local, then the videos are placed into the playing buffer area, and the contents of the buffer area are sequentially played to the user in a first-in first-out mode. Generally, when the content of a buffer area is insufficient, the ABR algorithm based on a play buffer area can reduce the code rate to prevent the buffer area from being exhausted, thereby causing video playing pause; when the buffer content is sufficient, a higher code rate is selected. The method and the ABR algorithm based on the network throughput are relatively conservative in selecting the code rate, for example, when the video starts to play, a low code rate can be selected because the buffer area is empty, and even though the network throughput is high, a low code rate can still be selected, so that a part of network resources are wasted.

There are also methods to combine the two, taking into account both network throughput and buffer size. Such algorithms are more complex and often have many prior assumptions about the network environment, and thus in the face of different network environments, once these assumptions are no longer true, they need to be further adjusted in parameters before they can be used.

The existing solutions mainly have the following disadvantages: 1. the video rate is the only degree of freedom of adjustment. 2. The ABR algorithm deployed at the user side can obtain limited network environment information, only throughput information of the network can be obtained from an application layer, and detailed understanding of the network state is lacked, so that the existing algorithm is difficult to respond to network changes in time. 3. The user terminal mobile device (such as a mobile phone) has limited calculation power, and is difficult to deploy a higher-level artificial intelligence algorithm, so that the ABR algorithm of the mobile terminal has limited complexity and is difficult to deal with different network environments (mobile cellular networks or WiFi networks). Therefore, a new adaptive streaming media system and a control method thereof are needed.

Through the above analysis, the problems and defects of the prior art are as follows:

(1) the ABR algorithm based on the network throughput has the problems that the network throughput is large in fluctuation, the network throughput is difficult to predict under normal conditions, and the uncertainty is large, so that the adjustment of the scheme on the video playing code rate is frequent, and the frequent code rate switching brings bad experience to users.

(2) The ABR algorithm based on the playing buffer area can reduce the code rate to prevent the buffer area from being exhausted when the content of the buffer area is insufficient, so that video playing is blocked and a part of network resources are wasted.

(3) Algorithms that take into account both network throughput and buffer size are more complex and often have many prior assumptions about the network environment, so in the face of different network environments, once these assumptions do not hold, these algorithms need to adjust the parameters further before they can be used.

(4) In the existing solution, the video bitrate is the only adjustment freedom.

(5) Network environment information which can be acquired by an ABR algorithm deployed at a user side is limited, throughput information of a network can be acquired only from an application layer, and a network state is lack of more detailed understanding, so that the existing algorithm is difficult to respond to network changes in time.

(6) The calculation power of the user terminal mobile equipment is limited, and a higher-level artificial intelligence algorithm is difficult to deploy, so the complexity of the ABR algorithm of the mobile terminal is limited, and the ABR algorithm is difficult to deal with different network environments.

The difficulty in solving the above problems and defects is:

the first problem is that: the modeling difficulty of the real network environment is high by using the traditional method, and the artificial intelligence algorithm makes the problem solved possible. However, even with artificial intelligence algorithms, predicting future throughput using only relatively coarse historical throughput as an input is difficult to do because throughput does not accurately reflect the quality of the channel. Therefore, the CSI is introduced in the invention, and the future throughput is accurately predicted by using an artificial intelligence method.

The second problem is that: to solve the problem, the ABR algorithm must consider the network throughput, otherwise, it is difficult to make a decision only by the buffer size.

The third problem is that: how to adjust parameters quickly in a short time is very challenging, because only after enough information is collected can the network environment be more accurately known, but at a greater cost in time.

The fourth problem is that: it is difficult to find other tuning dimensions.

The fifth problem is that: the same problem is solved.

Problem six: solving this problem requires a substantial increase in mobile device performance. At present, mobile equipment has large and uneven performance difference, so that the challenge of deploying a complex artificial intelligence algorithm on the mobile equipment is large.

The significance of solving the problems and the defects is as follows: users will be able to get a better video viewing experience, without requiring the user to use high performance mobile devices, reducing the user's demand for network bandwidth.

Disclosure of Invention

Aiming at the problems in the prior art, the invention provides a self-adaptive streaming media control method, a self-adaptive streaming media control system, computer equipment and application.

The invention is realized in such a way that a self-adaptive streaming media control method comprises the following steps:

the edge server pre-caches videos that the user may watch locally at the edge server from a server that remotely stores the videos over the internet. Therefore, the video service can be directly provided for the user, and the network congestion possibly occurring on the backbone network is avoided in the video transmission. The pre-caching of the video can reserve time for an edge server to analyze video content, namely, a play rate adjustment reference of each video block is calculated;

when a user needs to watch the video, the edge server analyzes the video content, determines an adjustment reference for adjusting the video playing rate and provides video service for the user. The step ensures that the user does not feel uncomfortable due to the change of the video rate, so that the user cannot easily perceive the speed change of the video;

when video blocks are sent to users one by one, the edge server determines the code rate and the video playing rate of the next video block to be sent according to the network environment. The step enables the user to obtain the proper video code rate and playing speed, namely, the video with high code rate is provided as far as possible, but the video is prevented from being blocked as far as possible;

the edge server takes out the video block with the corresponding code rate which is cached in advance according to the code rate decision, and then adjusts the playing rate according to the playing rate decision;

after the rate adjustment is completed, the edge server sends the video block to the user equipment;

the user equipment plays the received video blocks one by one, and feeds the network state and the playing state back to the edge server in real time, so that the edge server can make the next code rate decision and the next playing speed decision. This step provides the edge server with the necessary state information for the decision.

Further, the edge server analyzes the video content, including:

(1) determining the adjustment tolerance of the user to the playing rates of different video clips by utilizing the fact that the user has higher slow-playing tolerance to scenes with high content change speed and has lower tolerance to scenes with low content change speed;

(2) using the frame size to infer similarity between adjacent frames of the video, and further determining how fast the video content changes; the lower the similarity between adjacent frames is, the faster the change of the video picture is, and the greater the adjustment tolerance of the user on the playing rate is;

(3) and according to the adjustment tolerance, carrying out pre-analysis setting and dynamic adjustment on the playing rate of each video block at the edge server side.

Further, in the step (2), the similarity of the picture between the adjacent frames is measured by using an index SSIM for measuring the similarity of the two pictures; wherein SSIM ═ 1 indicates that the two pictures are completely the same, and SSIM ═ 0 indicates that the two pictures are completely different;

the SSIM estimation method comprises the following steps:

1) selecting partial segments in a video, calculating SSIM between every two adjacent frames of the partial segments, and extracting the frame size of each frame of the video;

2) respectively processing SSIM and frame size data by using a low-pass filter, and extracting a low-frequency signal by using an exponentially weighted moving average filter EWMA;

3) performing first-order linear regression on the SSIM and the frame size data after passing through the low-pass filter to obtain a linear relation between the frame size and the SSIM;

4) the SSIM of the entire video is estimated from the linear relationship between the frame size and the SSIM.

Further, in the step (2), the video playing rate is adjusted by taking the video block as an adjustment object, that is, within one video block, the playing rate is the same, so that the average variation speed of the picture within one video block needs to be determined; the average value of the SSIM of all adjacent frames within a video block is used as a measure of the average degree of change of the picture of the video block, and since a linear relationship between the SSIM and the frame size is already obtained, the frame size is used to directly estimate the average value of the SSIM using the relationship.

Further, the edge server analyzes the video content, and further includes: quantitatively determining the maximum tolerance of the user to the adjustment of the playing rate of each video block, comprising:

(1) linearly mapping the average SSIM to the interval [ a, b ], where a < b, and a and b belong to the interval [0,1], i.e., max (SSIM) → a, min (SSIM) → b;

(2) a is the maximum playback rate adjustment strength for the video block with the slowest content change. For example, a ═ 0.1 indicates that the playback rate cannot be lower than 0.9 times the speed for the video block whose content changes the slowest;

(3) b is the maximum playing rate adjustment strength of the video block with the fastest content change, and b is 0.25, which means that the playing rate of the video block with the fastest content change cannot be lower than 0.75 times;

(4) the values of a and b can be determined according to personal preference of a user, according to the subjective test result, the reference value of a is 0.1, the reference value of b is 0.25, the values are only used for reference, and the values are adjusted in actual use.

Further, the edge server is configured to obtain detailed channel information, i.e., CSI; the throughput prediction of a future period of time is realized by utilizing CSI information through a convolutional neural network;

determining the code rate and the video playing rate of the next video block to be sent by using a deep reinforcement learning method;

the deep reinforcement learning comprises three elements, namely state, action and reward;

a state representing a network state and a client play state before a next video block is transmitted; wherein the network state comprises throughput of the network, future network throughput predicted using the CSI; the playing state of the client is the size of the current buffer area; and other information related to the video itself;

behavior for representing a rate decision and a play rate decision for a next video block; the code rate decision corresponds to the available code rate of the video one by one, and the play rate decision has three options, namely no adjustment, half-speed adjustment and full-speed adjustment; wherein, the non-adjustment is played according to the original speed; the full-speed regulation refers to regulation according to the play rate regulation tolerance given by the video analysis module; the half-speed regulation refers to regulating according to half of the tolerance of the playing rate regulation;

the reward is determined by four factors, namely the reward is higher when the selected code rate is higher, the reward is higher when the difference between the selected code rate and the code rate selected last time is smaller, the reward is higher when the playing pause time caused by the current selection is shorter, and the reward is higher when the selected playing speed is closer to the original speed; in the training process of deep reinforcement learning, the algorithm automatically evolves towards the direction of obtaining more rewards, and finally evolves to a video playing strategy with high code rate, low code rate fluctuation and less pause;

the playing state is the size of the playing buffer.

It is a further object of the invention to provide a computer device comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the steps of:

the edge server caches videos which are possibly watched by a user in the edge server in advance from a server for storing the videos remotely through the Internet;

when a user needs to watch a video, the edge server analyzes the video content, determines an adjustment reference for adjusting the video playing rate and provides a video service for the user;

when video blocks are sent to users one by one, the edge server determines the code rate and the video playing rate of the next video block to be sent according to the network environment;

the user equipment plays the received equipment one by one, and feeds the network state and the playing state back to the edge server in real time, so that the edge server can make the next code rate decision and the next playing speed decision.

Another object of the present invention is to provide an information data processing terminal, which is used for implementing the adaptive streaming media control method.

Another object of the present invention is to provide an adaptive streaming control system implementing the adaptive streaming control method, the adaptive streaming control system comprising:

the edge server is used for caching videos which are possibly watched by a user in the edge server from the servers for storing the videos remotely through the Internet, and when the user needs to watch the videos, the edge server can directly provide video services for the user;

the user equipment is used for sending video watching requirements, playing the received equipment one by one, and feeding back the network state and the playing state to the edge server in real time so that the edge server can make a next code rate decision and a next playing rate decision;

the video analysis module is used for processing video resources which are cached in the edge server in advance;

the decision module based on deep reinforcement learning is used for making a code rate decision and a play rate decision on a video block to be sent according to the current network state and the play condition of a user side video;

the edge server further comprises: when video blocks are sent one by one for users, determining the code rate and the video playing rate of the next video block to be sent according to the network environment; taking out the video block with the corresponding code rate cached in advance according to the code rate decision, and then adjusting the playing rate according to the playing rate decision; and after the rate adjustment is finished, sending the video block to the user equipment through the edge server.

Another objective of the present invention is to provide an adaptive streaming media terminal, where the adaptive streaming media terminal is configured to implement the adaptive streaming media control method.

By combining all the technical schemes, the invention has the advantages and positive effects that: the self-adaptive streaming media control system provided by the invention can provide intelligent video streaming service for users, so that the occurrence of blocking is greatly reduced while the users keep higher video quality when watching videos, and better user experience can be obtained.

The present invention creatively adds this new dimension to the playout rate adjustment. The existing method can only adjust the code rate, so that when the network conditions are poor, the existing method can only maintain the fluency of video playing by reducing the video code rate, and the invention can reduce the requirement of video transmission on the network bandwidth by reducing the playing rate.

The invention uses a deep reinforcement learning method which has strong robustness. The existing method needs a lot of assumptions about the network environment, which may no longer be true once the network environment changes (such as switching from a cellular network to a WiFi network), and thus needs a lot of parameter adjustment, which is not practical. The invention can be deployed in various network environments after the neural network training is finished, and has stronger practicability.

The invention also has the following advantages:

(1) not only is the code rate adjusted, but also the adjustment of the video playing speed rate is creatively introduced. The slow playing of the video can reduce the requirement of the video on network bandwidth, thereby reducing the occurrence of blocking. The edge server can cache the video resources which the user wants to watch from the internet in advance and store the video resources in the local part of the edge server to directly serve the user, thereby avoiding the influence of the fluctuation of a backbone network on the video playing. The edge server can analyze the content of the pre-cached video and calculate the tolerance of the user to the slow playing of the video of different video clips, so that different playing rates can be applied to different video clips in a targeted manner, and the user cannot perceive the video.

(2) The edge server can acquire detailed Channel State Information (CSI), so that a wireless Channel between a user and the edge server can be more meticulously and accurately known, the meticulous degree of the CSI is far higher than the throughput measured at an application layer, and the CSI can reflect the real-time change condition of a network. The present invention uses CSI information so that the present invention can respond to network fluctuations faster.

(3) Compared with the mobile device, the edge server has abundant computing resources, so that the video code rate and the playing rate can be adjusted by using a more advanced artificial intelligence algorithm. Note that the computing task in the present invention is performed by the edge server, thereby greatly reducing the computing power requirements on the user side.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments of the present invention will be briefly described below, and it is obvious that the drawings described below are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without creative efforts.

Fig. 1 is a flowchart of an adaptive streaming media control method according to an embodiment of the present invention.

Fig. 2 is a schematic diagram of an adaptive streaming media control system according to an embodiment of the present invention.

Fig. 3 is a schematic diagram of a linear relationship between a frame size and an SSIM according to an embodiment of the present invention.

Fig. 4 is a schematic diagram of a method for estimating SSIM by using frame size according to an embodiment of the present invention.

Fig. 5 is a schematic diagram of a decision module based on deep reinforcement learning according to an embodiment of the present invention.

Fig. 6 is a diagram illustrating simulation comparison results with other mainstream adaptive code rate algorithms according to an embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

In view of the problems in the prior art, the present invention provides a method, a system, a computer device and an application for adaptive streaming media control, and the present invention is described in detail below with reference to the accompanying drawings.

As shown in fig. 1, the adaptive streaming media control method provided in the embodiment of the present invention includes the following steps:

s101, caching videos which are possibly watched by a user in an edge server from a server for storing the videos remotely through the Internet in the local edge server in advance by the edge server;

s102, when a user needs to watch a video, the edge server analyzes the video content, determines an adjustment reference for adjusting the video playing rate, and further provides a video service for the user;

s103, when sending video blocks for users one by one, the edge server determines the code rate and the video playing rate of the next video block to be sent according to the network environment;

s104, the edge server takes out the video block with the corresponding code rate cached in advance according to the code rate decision, and then adjusts the playing rate according to the playing rate decision;

s105, after the rate adjustment is completed, the edge server sends the video block to the user equipment;

and S106, the user equipment plays the received video blocks one by one, and feeds the network state and the playing state back to the edge server in real time, so that the edge server can make the next code rate decision and the next playing speed decision.

Those skilled in the art can also implement the adaptive streaming media control method provided by the present invention by using other steps, and the adaptive streaming media control method provided by the present invention in fig. 1 is only one specific embodiment.

As shown in fig. 2, an adaptive streaming media system for edge calculation according to an embodiment of the present invention includes: the system comprises an edge server, user equipment, a video analysis module and a decision-making module based on deep reinforcement learning.

The edge server is used for caching videos which are possibly watched by a user in the edge server from the servers for storing the videos remotely through the Internet, and when the user needs to watch the videos, the edge server can directly provide video service for the user;

and the decision module based on deep reinforcement learning is used for making a code rate decision and a play rate decision for the video block to be sent according to the current network state and the play condition of the video of the user side.

The technical solution of the present invention will be further described with reference to the following explanation of terms.

(1) And (3) edge calculation: the method is characterized in that a nearest-end service is provided nearby by adopting an open platform integrating network, computing, storage and application core capabilities on one side close to an object or a data source. The invention uses the edge device as the server of the streaming media to provide the multimedia service for the nearby users.

(2) Streaming media: streaming media (streaming media) refers to a technology and process of compressing a series of media data, sending the data in segments on the internet, and transmitting video and audio on the internet in real time for viewing. Such as Youtube, you ku, bilibili.

(3) Self-adaptation: in the invention, self-adaptation means that the code rate (i.e. definition) and the video playing rate of a video can be automatically adjusted according to the network environment, so that better video experience (i.e. less video jamming and higher definition) is provided for a user.

The technical solution of the present invention is further described with reference to the following examples.

1. Overview of the System

The whole system consists of an edge server and user equipment. First, the edge server pre-caches videos that the user may watch locally at the edge server from a server that remotely stores the videos over the internet. Then, when the user needs to watch the video, the edge server can directly provide the video service for the user. Because the video resources are cached on the edge server, the edge server analyzes the video content first and determines an adjustment reference for video playback rate adjustment. When video blocks are sent one by one for users, the code rate and the video playing rate of the next video block to be sent are determined according to the network environment. The process determines the code rate and the video playing rate of the video by using a deep reinforcement learning method. And then, the edge server takes out the video block with the corresponding code rate cached in advance according to the code rate decision, and then adjusts the playing rate according to the playing rate decision. After the rate adjustment is completed, the edge server sends the video block to the user equipment. The user equipment plays the received equipment one by one, and then feeds back the network state and the playing state (namely the size of a playing buffer) to the edge server in real time, so that the edge server can make the next code rate decision and the next playing speed decision.

2. Video analysis module

The module is responsible for processing video resources which are cached in the edge server in advance. The goal is to determine the size of the slowest playback rate that the user can tolerate for each segment of each video. In order to ensure the experience of watching the video, the invention ensures that the video does not feel uncomfortable to the user due to too slow playing through the module.

The invention utilizes the fact that the user has higher slow release tolerance to the scene with fast content change speed (such as fierce fighting scene); and a lower tolerance for scenes with slower content change rates (e.g., conversational scenes) to determine the user's accommodation tolerance for different video segments.

In the present invention, to determine how fast the video content changes, the present invention uses the frame size to infer the similarity between adjacent frames of the video. The lower the similarity between adjacent frames, the faster the video frames change, and the greater the tolerance of the user to adjustments in the playback rate.

The invention uses the widely accepted index SSIM (structure Similarity index) for measuring the Similarity of two photos to measure the Similarity of pictures between adjacent frames (SSIM 1 means that two pictures are completely the same, SSIM 0 means that two photos are completely different). However, the direct calculation of SSIM between all adjacent frames is huge in calculation amount and lacks of practicability, so that the invention provides a method for estimating SSIM by using frame size.

At present, a mainstream video coding mode is Variable bit Encoding (VBR), that is, under the condition that the overall video coding rate is kept constant, a higher coding rate is allocated to some segments to keep more details, and a lower coding rate is allocated to some simple pictures to save storage space. Under the VBR coding mode, a segment with fast picture change will obtain a higher code rate, i.e. the frame size after coding is larger. Therefore, the speed of the picture change has a strong correlation with the frame size. However, the frame size is also limited by other factors, such as the switching of scene, the switching of frame format (I frame, P frame, B frame), etc. directly affect the frame size. Therefore, the present invention needs to remove these influencing factors and strip the relationship between the frame size and the SSIM. It should be noted that the change of SSIM in a video is relatively slow, because the rhythm of a segment is generally stable and belongs to a low-frequency signal, and the fluctuation of the frame size caused by the switching of the frame format is relatively frequent and belongs to a high-frequency signal. Therefore, in the present invention, the present invention uses this feature to eliminate the interference signal in the high frequency part using the low pass filter, so that the relationship between the SSIM and the frame size can be restored. Similarly, the SSIM is also affected by factors such as scene change, so that it is difficult to truly reflect the degree of change of the picture, and such interference is mainly concentrated in the high frequency part and can be removed by a low pass filter. The specific process is as follows:

(1) and calculating the SSIM between every two adjacent frames of the video. The frame size of each frame of the video is extracted.

(2) The SSIM and frame size data are processed separately using low pass filters. An Exponentially Weighted Moving Average (EWMA) filter may be used to extract the low frequency signal.

(3) A first order linear regression is performed on the SSIM and frame size data after passing through the low pass filter to obtain a linear relationship between the frame size and the SSIM (see fig. 3).

Next, because the present invention has already obtained a linear relationship between SSIM and frame size, the present invention can use low pass filter processed frame size data to estimate SSIM.

As shown in fig. 4, in the present invention, the adjustment of the video playback rate takes the video block as the adjustment target, that is, within one video block, the playback rate is the same. Therefore, the present invention needs to determine how fast the average change of pictures within a video block is. The present invention uses the average value of the SSIM of all adjacent frames within a video block as a measure of how fast the average change of the picture of the video block is. Because the present invention has already obtained a linear relationship between the SSIM and the frame size, the present invention uses this relationship to directly estimate the average value of the SSIM with the frame size. This greatly reduces the computational overhead of computing SSIM, since the frame size is readily available and does not involve video coding.

Now, the present invention obtains the average SSIM for each video block, and then the present invention quantitatively determines the maximum user tolerance for the adjustment of the play rate for each video block. In the present invention, this (1) linearly maps the average SSIM to the interval [ a, b ], where a < b, and a and b belong to the interval [0,1], i.e., max (SSIM) → a, min (SSIM) → b; a is the maximum playback rate adjustment strength for the video block with the slowest content change. For example, a ═ 0.1 indicates that the playback rate cannot be lower than 0.9 times the speed for the video block whose content changes the slowest; b is the maximum playing rate adjustment strength of the video block with the fastest content change, and b is 0.25, which means that the playing rate of the video block with the fastest content change cannot be lower than 0.75 times; the values of a and b can be determined according to personal preference of a user, according to the subjective test result, the reference value of a is 0.1, the reference value of b is 0.25, the values are only used for reference, and the values can be flexibly adjusted in actual use.

3. Decision module based on deep reinforcement learning

As shown in fig. 5, the module is used to make a bit rate decision and a play rate decision for a video block to be sent according to a current network state and a play situation of a user-side video. The invention uses Deep Reinforcement Learning (Deep Learning) technology, and the algorithm interacts with the environment by continuously making decision-making attempt, and finally learns a set of effective code rate and play rate regulation strategy by self.

Firstly, the invention introduces an edge server, and has the great characteristic of acquiring detailed channel information, namely CSI. The CSI may more accurately reflect the state of the network. In the invention, the throughput prediction for a period of time in the future is realized through the CSI information. The present invention uses convolutional neural networks to predict future network throughput. The predicted network throughput will be used for the next video bitrate decision and play rate decision.

In order to use the deep reinforcement learning technology, the invention needs to define three elements of deep reinforcement learning, namely state (state), action (action) and reward (reward). In the present invention, the status is defined as a network status and a client play status, etc. before the next video block is transmitted. Wherein the network state comprises throughput of the network, future network throughput predicted using the CSI; the playing state of the client is the size of the current buffer area; and other information related to the video itself, such as the size of the next video block, the number of video blocks to be transmitted, etc. Behavior is defined as the rate decision and the play rate decision of the next video block. The code rate decision corresponds to the available code rate of the video one by one, and the play rate decision has three options, namely no adjustment, half-speed adjustment and full-speed adjustment. Wherein, the non-adjustment is played according to the original speed; the full-speed regulation refers to regulating the tolerance according to the playing rate given by the video analysis module, and if the tolerance is 0.2, playing is carried out according to 0.8 times of speed; the half-speed adjustment refers to adjusting according to half of the tolerance of the playing speed adjustment, and if the tolerance is 0.2, playing is performed according to 0.9 times of speed. The reward is determined by four factors, namely the reward is higher when the selected code rate is higher, the reward is higher when the difference between the selected code rate and the code rate selected last time is smaller, the reward is higher when the playing time caused by the current selection is shorter, and the reward is higher when the selected playing speed is closer to the original speed. In the training process of deep reinforcement learning, the algorithm automatically evolves towards the direction of obtaining more rewards, and finally the video playing strategy with high code rate, low code rate fluctuation and less pause is evolved.

It is noted that the algorithm may work if future throughput is predicted by the convolutional neural network without using CSI, providing only historical throughput information to the algorithm. The speed of response to network fluctuations alone will not be timely when the CSI is not used, but can still reach a level of precedence.

Fig. 6 shows the simulation comparison result of the present invention with other mainstream adaptive code rate algorithms. The algorithms compared include throughput-based adaptive code rate algorithms, buffer-based adaptive code rate algorithms, and artificial intelligence-based adaptive code rate algorithms. It can be seen that the present invention achieves the highest video bitrate, while achieving the least pause time and the least pause times. The user experience is thus optimal.

In the above embodiments, the implementation may be wholly or partially realized by software, hardware, firmware, or any combination thereof. When used in whole or in part, can be implemented in a computer program product that includes one or more computer instructions. When loaded or executed on a computer, cause the flow or functions according to embodiments of the invention to occur, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another, for example, the computer instructions may be transmitted from one website site, computer, server, or data center to another website site, computer, server, or data center via wire (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL), or wireless (e.g., infrared, wireless, microwave, etc.)). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device, such as a server, a data center, etc., that includes one or more of the available media. The usable medium may be a magnetic medium (e.g., floppy Disk, hard Disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., Solid State Disk (SSD)), among others.

The above description is only for the purpose of illustrating the present invention and the appended claims are not to be construed as limiting the scope of the invention, which is intended to cover all modifications, equivalents and improvements that are within the spirit and scope of the invention as defined by the appended claims.

Claims

1. An adaptive streaming control method, the adaptive streaming control method comprising:

when a user needs to watch a video, the edge server analyzes the video content, determines an adjustment reference for adjusting the playing rate of a video user end and provides a video service for the user;

the user equipment plays the received video blocks one by one, and feeds the network state and the playing state back to the edge server in real time, so that the edge server can make the next code rate decision and the next playing speed decision.

2. The adaptive streaming method of claim 1, wherein the edge server analyzing the video content comprises:

(2) using the frame size to infer the similarity between adjacent frames of the video, and further determining the speed of the change of the video content; the lower the similarity between adjacent frames is, the faster the change of the video picture is, and the greater the adjustment tolerance of the user on the playing rate is;

3. The adaptive streaming media control method according to claim 2, wherein in the step (2), the similarity of pictures between adjacent frames is measured by using an index SSIM for measuring the similarity of two photos; wherein SSIM ═ 1 indicates that the two pictures are completely the same, and SSIM ═ 0 indicates that the two pictures are completely different;

the SSIM estimation method comprises the following steps:

2) respectively processing SSIM and frame size data by using a low-pass filter, and extracting a low-frequency signal; recommending an exponentially weighted moving average filter EWMA to be used;

4. The adaptive streaming media control method according to claim 2, wherein in step (2), the video playing rate is adjusted by taking the video blocks as the adjustment objects, that is, within one video block, the playing rate is the same, so that the average variation speed of the picture within one video block needs to be determined; the average value of the SSIM of all adjacent frames in a video block is used as a measure of the average degree of change of the picture of the video block, and since a linear relationship between the SSIM and the frame size is already obtained, the frame size is used to directly estimate the average value of the SSIM using the relationship.

5. The adaptive streaming media control method of claim 1, wherein the edge server analyzes video content, further comprising: quantitatively determining the maximum tolerance degree of a user to the adjustment of the playing rate of each video block, comprising the following steps:

(2) a is the maximum playing speed adjusting force for the video block with the slowest content change;

(4) the values of a and b can be determined according to personal preference of a user, and are adjusted during actual use according to subjective test results.

6. The adaptive streaming method of claim 1, wherein the edge server is configured to obtain detailed channel information (CSI); the throughput prediction of a future period of time is realized by utilizing CSI information through a convolutional neural network;

the playing state is the size of the playing buffer.

7. A computer device, characterized in that the computer device comprises a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to carry out the steps of:

the method comprises the following steps that an edge server caches videos which are possibly watched by a user in a server for storing the videos remotely from the edge server in advance through the Internet;

8. An information data processing terminal, characterized in that the information data processing terminal is used for implementing the adaptive streaming media control method according to any one of claims 1 to 6.

9. An adaptive streaming media control system for implementing the adaptive streaming media control method according to any one of claims 1 to 6, the adaptive streaming media control system comprising:

the decision module based on deep reinforcement learning is used for making a code rate decision and a play rate decision for a video block to be sent according to the current network state and the play condition of a user side video;

the edge server is further configured to: when video blocks are sent one by one for users, determining the code rate and the video playing rate of the next video block to be sent according to the network environment; taking out the video block with the corresponding code rate cached in advance according to the code rate decision, and then adjusting the playing rate according to the playing rate decision; and after the rate adjustment is finished, sending the video block to the user equipment through the edge server.

10. An adaptive streaming terminal, wherein the adaptive streaming terminal is configured to implement the adaptive streaming control method according to any one of claims 1 to 6.