Multi-path online live broadcast intelligent regulation and control management system

Abstract

The invention discloses a multipath online live broadcast intelligent regulation and control management system, and relates to the technical field of live broadcast intelligent regulation and control management. Through double analysis of a video grammar layer and a bullet screen semantic layer, a content value evaluation system is established on a protocol irrelevant layer; a dynamic path distribution mechanism deeply fuses network state perception and frame level importance decision, so that video quality degradation caused by uplink bandwidth competition is effectively relieved; according to the adaptive retransmission strategy, differential recovery is implemented based on a frame dependency relationship, the redundant transmission overhead is reduced while the visual continuity is guaranteed, and the network learning module continuously adapts to a complex network environment through online optimization, so that the self-evolution capability of the transmission strategy is formed.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent control and management of live broadcasting, and in particular to a multi-path online live broadcasting intelligent control and management system. Background Art

[0002] With the rapid development of the interactive live streaming industry, multi-path online live streaming systems have generally adopted technologies such as intelligent routing selection and dynamic bit rate adaptation to ensure transmission stability. In scenarios such as e-commerce promotions and live events, the instantaneous outburst of massive special effects barrages on the audience side forms an upstream peak, causing the control strategies originally optimized for downstream video streams to face adverse pressure.

[0003] Current mainstream systems use QoS priority marking to differentiate data streams, but the homogeneous characteristics of barrage data packets and video frames at the protocol layer make it difficult for traditional differentiated service mechanisms to accurately identify differences in business logic. Common solutions include edge computing-based traffic prediction mechanisms, which use LSTM networks to predict barrage peak periods and reserve uplink bandwidth resources in advance; or adopt dynamic weight allocation algorithms to adjust the video stream redundancy level in real time according to barrage density. Although such solutions can alleviate bandwidth competition conflicts, they do not deeply analyze the inter-frame dependencies of the video coding layer. Simply discarding B / P frames may cause GOP structure breaks. At the same time, the rendering resource consumption of special effect barrage is an order of magnitude different from that of text barrage, and the existing system lacks the ability to perceive the structured semantics of barrage content. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] The present invention provides a multi-path online live broadcast intelligent control and management system to solve the three core pain points of existing live broadcast systems in the barrage peak scenario: intensified upstream bandwidth competition, lack of video frame protection mechanism, and rough barrage type identification, which lead to key frame loss and uncontrollable degradation of picture quality.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] The embodiment of the present invention provides a multi-path online live broadcast intelligent control and management system, which includes:

[0008] Path quality monitoring module, which obtains network status parameters of each transmission path in real time;

[0009] A content analysis module, connected to the path quality monitoring module, performs syntax analysis on the video stream and extracts frame dependencies;

[0010] a dynamic scheduling decision module, receiving the output of the content analysis module and generating a path allocation instruction based on the frame type priority;

[0011] The barrage semantic parsing unit, independent of the video stream processing channel, identifies the rendering resource requirement level of the barrage data;

[0012] an adaptive retransmission controller that dynamically adjusts a data packet retransmission strategy based on the frame dependency;

[0013] The network status learning module updates the path quality assessment model based on historical scheduling records.

[0014] As a preferred solution of the multi-path online live broadcast intelligent control and management system described in the present invention, the network status parameters of the path quality monitoring module include:

[0015] Path delay fluctuation, calculated as the ratio of the delay standard deviation to the mean over the last N transmission cycles;

[0016] Packet loss correlation factor, which reflects the temporal correlation of consecutive packet loss events along a path;

[0017] Bandwidth contention coefficient, which quantifies the ratio of bullet comment data to video stream data in the uplink channel;

[0018] The packet loss correlation factor is obtained by calculating the autocorrelation coefficient of the interval time between consecutive packet losses, and a path stability warning is triggered when the autocorrelation coefficient exceeds a preset threshold.

[0019] As a preferred solution of the multi-path online live broadcast intelligent control and management system described in the present invention, the step of quantifying the relationship between the data volume ratio of barrage data and video stream in the uplink channel in the bandwidth contention coefficient includes:

[0020] Define the bandwidth contention coefficient:

[0021]

[0022] Where c represents the bandwidth contention coefficient, a represents the total bytes of bullet chat data within the statistical window, b represents the total bytes of video data within the statistical window, γ represents the burstiness weight constant, and c1 represents the coefficient of variation of the bullet chat throughput.

[0023] The coefficient of variation of the barrage throughput is:

[0024]

[0025] Among them, c1 represents the coefficient of variation of the barrage throughput, σ a Indicates the sample standard deviation of the number of bytes of bullet comment data, μ a Indicates the sample mean of the number of bytes of bullet comment data.

[0026] As a preferred solution of the multi-path online live broadcast intelligent control and management system described in the present invention, the content analysis module and the barrage semantic parsing unit share a physical storage area through a memory-mapped register, and a PCIe data channel with a priority arbitration mechanism is set between the two;

[0027] The syntax layer parsing includes decoding the sequence parameter set SPS and picture parameter set PPS in the video stream, extracting the slice_type identifier and reference frame list of each frame;

[0028] The syntax analysis of the content analysis module specifically includes:

[0029] Identify the distribution of I-frames, P-frames, and B-frames in the video stream;

[0030] Constructing a forward and backward reference relationship map for each frame within the image group;

[0031] Calculate the diffusion depth value of the impact of any frame loss on subsequent image groups.

[0032] As a preferred solution of the multi-path online live broadcast intelligent control and management system described in the present invention, the step of calculating the diffusion depth value of the impact of any frame loss on subsequent image groups during the syntax layer parsing process includes:

[0033] Identify the frame type:

[0034] s k =h k mod5,

[0035] Among them, s k Indicates the type identifier of the kth frame, 0 corresponds to I frame, 1 corresponds to P frame, 2 corresponds to B frame, h k Indicates the integer value of the slice_type field of the kth frame, where k represents the frame index;

[0036] Modeling reference relationships:

[0037] If frame i is referenced by frame j, then m i,j =1, otherwise m i,j =0,

[0038] Among them, m i,j Represents the reference relationship matrix element between frame i and frame j, where i and j represent the indexes of the referenced frame and the referencing frame respectively;

[0039] Calculate the impact diffusion depth using the formula:

[0040]

[0041] Among them, d jrepresents the maximum impact depth of the subsequent image group after the loss of frame j, p represents a directed reference path starting from frame j, represents the set of all reachable directed reference paths, and |p| represents the length of path p.

[0042] As a preferred solution of the multi-path online live broadcast intelligent control and management system described in the present invention, the path allocation instruction generation method of the dynamic scheduling decision module includes:

[0043] Assign the lowest latency path to I frames and prohibit switching transmission paths midway;

[0044] Adopt multi-path redundant transmission mode for P frames, using two or more paths in parallel;

[0045] Implement an on-demand transmission strategy for B frames, sending complete data only when the path idle bandwidth exceeds the threshold;

[0046] The redundant backup channel continuously receives the I frame check code and reconstructs the original data according to the check code when the main channel transmission fails;

[0047] The path allocation instruction generation method of the dynamic scheduling decision module also includes:

[0048] When an I-frame transmission is detected, the dynamic scheduling decision module broadcasts the frame transmission request to all available paths, selects the first path that returns a confirmation response as the main transmission channel, and the remaining paths are automatically converted into redundant backup channels.

[0049] As a preferred solution of the multi-path online live broadcast intelligent control and management system described in the present invention, the dynamic scheduling decision module generates an allocation matrix based on frame-level importance and path real-time status, including:

[0050] Calculate the composite cost for the u-th available path using the formula:

[0051] g u =θ1τ u +θ2ρ u -θ3λ u ,

[0052] Among them, g u is the composite cost of path u, θ1 represents the delay weight coefficient, τ u represents the current round-trip delay of path u, θ2 represents the packet loss weight coefficient, ρ u represents the instantaneous packet loss rate of path u, θ3 represents the bandwidth weight coefficient, λ u represents the remaining available bandwidth of path u;

[0053] The cost is converted into allocation probability using exponential mapping, and for the j-th frame we get:

[0054]

[0055] Among them, π j,u represents the normalized probability of selecting path u in the jth frame, Indicates the same frame type as s j The corresponding priority amplification factor, s j =0, 1, 2 represent I, P, and B frames respectively, U represents the total number of available paths, g v represents the composite cost of the vth path;

[0056] The number of parallel paths is determined based on the frame type and then converted into a binary allocation matrix:

[0057]

[0058] like Then A j,u =1, other cases A j,u =0,

[0059] Among them, t j Indicates the number of target paths in the jth frame, 1 {·} is an indicator function, which takes the value 1 when the condition {·} is met, otherwise it takes the value 0. j,u Assign matrix elements to frame paths, Indicates pressing π j,u Select the first t from largest to smallest j The set of paths, π j,* is the probability vector of the j-th frame on all U paths.

[0060] As a preferred solution of the multi-path online live broadcast intelligent control and management system described in the present invention, the barrage semantic parsing unit performs:

[0061] Parse the special effect triggering instruction code in the barrage text;

[0062] Distinguish between pure text bullet comments and enhanced bullet comments with particle effects and 3D models;

[0063] Generate a bandwidth requirement tag that is positively correlated with rendering complexity and inject it into the packet header;

[0064] The texture map reference identifier is a Unity engine ShaderLab format or an Unreal engine Material expression hash value;

[0065] The parsing of the special effect trigger instruction code in the barrage semantic parsing unit includes detecting whether there is a three-dimensional coordinate transformation matrix, particle emitter parameters or texture map reference identifier in the barrage text.

[0066] As a preferred solution of the multi-path online live broadcast intelligent control and management system described in the present invention, the strategy of the adaptive retransmission controller includes:

[0067] Implement unconditional and immediate retransmission of lost I-frame data;

[0068] When a P frame loss is detected, only the intra-frame predicted macroblocks of the frame are retransmitted;

[0069] If a B frame is lost and the remaining path bandwidth is below a threshold, the frame is discarded and forward frame copy compensation is enabled.

[0070] As a preferred solution of the multi-path online live broadcast intelligent control and management system described in the present invention, the network status learning module includes:

[0071] The path oscillation prediction submodule models the quality change trend of each path based on the LSTM network;

[0072] Conflict avoidance decision tree, which stores the optimal scheduling combinations under common bandwidth competition scenarios;

[0073] Model online update interface, receiving feedback on actual transmission effects and correcting path evaluation weights;

[0074] The path oscillation prediction submodule includes an FPGA-accelerated LSTM computation unit, whose weight matrix is stored in the onboard BRAM.

[0075] The decision basis of the conflict avoidance decision tree in the network status learning module includes at least two factors of the historical bandwidth utilization curve of the current period, the anchor terminal movement speed detection value, and the number of parallel microphone sessions.

[0076] The beneficial effects of the present invention are as follows: the present invention establishes a content value evaluation system at a protocol-independent level through dual analysis of the video syntax layer and the barrage semantic layer; the dynamic path allocation mechanism deeply integrates network status perception with frame-level importance decision-making, effectively alleviating the video quality degradation caused by uplink bandwidth competition; the adaptive retransmission strategy implements differentiated recovery based on frame dependencies, reducing redundant transmission overhead while ensuring visual coherence; the network learning module continuously adapts to complex network environments through online optimization, forming the self-evolution capability of the transmission strategy. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0078] Figure 1 This is a schematic diagram of the framework of the multi-path online live broadcast intelligent control and management system in Example 1. DETAILED DESCRIPTION

[0079] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0080] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0081] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0082] Example 1, with reference to Figure 1 This embodiment provides a multi-path online live broadcast intelligent control and management system, including:

[0083] Path quality monitoring module, which obtains network status parameters of each transmission path in real time;

[0084] The network status parameters of the path quality monitoring module include:

[0085] Path delay fluctuation, calculated as the ratio of the delay standard deviation to the mean over the last N transmission cycles;

[0086] Packet loss correlation factor, which reflects the temporal correlation of consecutive packet loss events along a path;

[0087] Bandwidth contention coefficient, which quantifies the ratio of bullet comment data to video stream data in the uplink channel;

[0088] The packet loss correlation factor is obtained by calculating the autocorrelation coefficient of the interval between consecutive packet losses. When the autocorrelation coefficient exceeds the preset threshold, a path stability warning is triggered;

[0089] In the bandwidth contention coefficient, the steps for quantifying the ratio of bullet chat data to video stream data in the uplink channel include:

[0090] Define the bandwidth contention coefficient:

[0091]

[0092] Where c represents the bandwidth contention coefficient, a represents the total bytes of bullet chat data within the statistical window, b represents the total bytes of video data within the statistical window, γ represents the burstiness weight constant, and c1 represents the coefficient of variation of the bullet chat throughput.

[0093] The coefficient of variation of the barrage throughput is:

[0094]

[0095] Among them, c1 represents the coefficient of variation of the barrage throughput, σ a Indicates the sample standard deviation of the number of bytes of bullet comment data, μ a Indicates the sample mean of the number of bytes of barrage data;

[0096] Specifically, this coefficient superimposes the average share with traffic fluctuations, describing both the long-term occupancy of bullet comments and the risk of sudden peaks. When the coefficient approaches 1, the scheduling layer determines that bullet comments are significantly crowding out the uplink channel, triggering rate limiting or path migration. The algorithm relies solely on byte counts and does not require parsing application-layer content. It can work in encrypted transmission scenarios. The real-time calculation sequence can be combined with packet loss indicators to identify bottlenecks caused by bullet comment bursts in advance, reducing the probability of lag and meeting the live broadcast business's pursuit of stable latency.

[0097] The content analysis module, connected to the path quality monitoring module, performs syntax analysis on the video stream and extracts frame dependencies;

[0098] The content analysis module and the barrage semantic parsing unit share physical storage areas through memory-mapped registers, and a PCIe data channel with a priority arbitration mechanism is set up between the two.

[0099] Syntax layer parsing includes decoding the sequence parameter set SPS and picture parameter set PPS in the video stream, extracting the slice_type identifier and reference frame list of each frame;

[0100] The syntax analysis of the content analysis module specifically includes:

[0101] Identify the distribution of I-frames, P-frames, and B-frames in the video stream;

[0102] Constructing a forward and backward reference relationship map for each frame within the image group;

[0103] Calculate the diffusion depth value of the impact of any frame loss on subsequent image groups;

[0104] During the syntax layer parsing process, the steps for calculating the diffusion depth value of the impact of any frame loss on subsequent picture groups include:

[0105] Identify the frame type:

[0106] s k =hk mod5,

[0107] Among them, s k Indicates the type identifier of the kth frame, 0 corresponds to I frame, 1 corresponds to P frame, 2 corresponds to B frame, h k Indicates the integer value of the slice_type field of the kth frame, where k represents the frame index;

[0108] Modeling reference relationships:

[0109] If frame i is referenced by frame j, then m i,j =1, otherwise m i,j =0,

[0110] Among them, m i,j Represents the reference relationship matrix element between frame i and frame j, where i and j represent the indexes of the referenced frame and the referencing frame respectively;

[0111] Calculate the impact diffusion depth using the formula:

[0112]

[0113] Among them, d j represents the maximum impact depth of the subsequent image group after the loss of frame j, p represents a directed reference path starting from frame j, represents the set of all reachable directed reference paths, |p| represents the length of path p;

[0114] Specifically, the frame type can be quickly calibrated by taking the modulus of slice_type and 5, avoiding conditional branches and adapting to hardware streaming implementation. The adjacency matrix is compressed into a Boolean bitmap, and the reference relationship is obtained by relying on list parsing, providing O(1) query for real-time scheduling. The depth index converts the impact of frame loss into a single value. The scheduling layer can allocate low-latency or redundant transmission strategies according to its size to achieve resource and importance matching. Pixel data is not parsed in the process. The amount of calculation is linearly related to the number of frames, which is suitable for FPGA pipeline parallel processing and still maintains low latency in high-bitrate live broadcast scenarios.

[0115] The dynamic scheduling decision module receives the output of the content analysis module and generates path allocation instructions based on the frame type priority;

[0116] The path allocation instruction generation method of the dynamic scheduling decision module includes:

[0117] Assign the lowest latency path to I frames and prohibit switching transmission paths midway;

[0118] Adopt multi-path redundant transmission mode for P frames, using two or more paths in parallel;

[0119] Implement an on-demand transmission strategy for B frames, sending complete data only when the path idle bandwidth exceeds the threshold;

[0120] The redundant backup channel continuously receives the I-frame check code and reconstructs the original data based on the check code when the main channel transmission fails;

[0121] The path allocation instruction generation method of the dynamic scheduling decision module also includes:

[0122] When an I-frame transmission is detected, the dynamic scheduling decision module broadcasts the frame transmission request to all available paths, selects the first path that returns an acknowledgment response as the primary transmission channel, and automatically converts the remaining paths into redundant backup channels;

[0123] In the dynamic scheduling decision module, an allocation matrix is generated based on the frame-level importance and the real-time status of the path, including:

[0124] Calculate the composite cost for the u-th available path using the formula:

[0125] g u =θ1τ u +θ2ρ u -θ3λ u ,

[0126] Among them, g u is the composite cost of path u, θ1 represents the delay weight coefficient, τ u represents the current round-trip delay of path u, θ2 represents the packet loss weight coefficient, ρ u represents the instantaneous packet loss rate of path u, θ3 represents the bandwidth weight coefficient, λ u represents the remaining available bandwidth of path u;

[0127] The cost is converted into allocation probability using exponential mapping, and for the j-th frame we get:

[0128]

[0129] Among them, π j,u represents the normalized probability of selecting path u in the jth frame, Indicates the same frame type as s j The corresponding priority amplification factor, s j =0, 1, 2 represent I, P, and B frames respectively, U represents the total number of available paths, g v represents the composite cost of the vth path;

[0130] The number of parallel paths is determined based on the frame type and then converted into a binary allocation matrix:

[0131]

[0132] like Then Aj,u =1, other cases A j,u =0,

[0133] Among them, t j Indicates the number of target paths in the jth frame, 1 {·} is an indicator function, which takes the value 1 when the condition {·} is met, otherwise it takes the value 0. j,u Assign matrix elements to frame paths, Indicates pressing π j,u Select the first t from largest to smallest j The set of paths, π j,* is the probability vector of the j-th frame on all U paths;

[0134] Specifically, the cost function captures three network characteristics through linear combination. The weights can be adjusted online to fit the business objectives. The exponential mapping amplifies the difference in high-quality paths. Combined with the frame-level priority coefficient, I frames are more biased towards low-cost channels, while B frames are biased towards low-cost channels. Smaller frames give way in competition. The path number formula directly enables P frames to obtain dual-channel redundancy, I frames to monopolize the fastest channel and maintain stability, and B frames to be selected only when there is sufficient margin. This automatically sacrifices the data with the least visual impact when bandwidth is tight. Binary matrices can be written to DMA descriptors. Hardware drives multiple NICs to send data in parallel based on the A matrix, reducing software intervention. The sparse rows and columns of the matrix facilitate bitmap compression and pipeline parallelism, improving scheduling throughput and reducing additional latency.

[0135] The barrage semantic parsing unit, independent of the video stream processing channel, identifies the rendering resource requirement level of the barrage data;

[0136] The bullet comment semantic parsing unit performs:

[0137] Parse the special effect triggering instruction code in the barrage text;

[0138] Distinguish between pure text bullet comments and enhanced bullet comments with particle effects and 3D models;

[0139] Generate a bandwidth requirement tag that is positively correlated with rendering complexity and inject it into the packet header;

[0140] Texture map reference identification is the Unity engine ShaderLab format or the Unreal engine Material expression hash value;

[0141] The parsing of the special effect trigger instruction code in the barrage semantic parsing unit includes detecting whether there is a three-dimensional coordinate transformation matrix, particle emitter parameters or texture map reference identifier in the barrage text;

[0142] Adaptive retransmission controller that dynamically adjusts packet retransmission strategies based on frame dependencies;

[0143] The strategies of the adaptive retransmission controller include:

[0144] Implement unconditional and immediate retransmission of lost I-frame data;

[0145] When a P frame loss is detected, only the intra-frame predicted macroblocks of the frame are retransmitted;

[0146] If a B frame is lost and the remaining path bandwidth is lower than the threshold, the frame is discarded and forward frame copy compensation is enabled;

[0147] Network status learning module, which updates the path quality assessment model based on historical scheduling records;

[0148] The network status learning module includes:

[0149] The path oscillation prediction submodule models the quality change trend of each path based on the LSTM network;

[0150] Conflict avoidance decision tree, which stores the optimal scheduling combinations under common bandwidth competition scenarios;

[0151] Model online update interface, receiving feedback on actual transmission effects and correcting path evaluation weights;

[0152] The path oscillation prediction submodule includes an FPGA-accelerated LSTM computation unit, whose weight matrix is stored in the onboard BRAM.

[0153] The decision basis of the conflict avoidance decision tree in the network status learning module includes at least two factors among the historical bandwidth utilization curve of the current period, the anchor terminal movement speed detection value, and the number of parallel microphone sessions.

[0154] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A multi-path online live broadcast intelligent control and management system, characterized in that: include, Path quality monitoring module, which obtains network status parameters of each transmission path in real time; A content analysis module, connected to the path quality monitoring module, performs syntax analysis on the video stream and extracts frame dependencies; a dynamic scheduling decision module, receiving the output of the content analysis module and generating a path allocation instruction based on the frame type priority; The barrage semantic parsing unit, independent of the video stream processing channel, identifies the rendering resource requirement level of the barrage data; an adaptive retransmission controller that dynamically adjusts a data packet retransmission strategy based on the frame dependency; The network status learning module updates the path quality assessment model based on historical scheduling records.

2. A multi-path online live broadcast intelligent control and management system according to claim 1, characterized in that: The network status parameters of the path quality monitoring module include: Path delay fluctuation, calculated as the ratio of the delay standard deviation to the mean over the last N transmission cycles; Packet loss correlation factor, which reflects the temporal correlation of consecutive packet loss events along a path; Bandwidth contention coefficient, which quantifies the ratio of bullet comment data to video stream data in the uplink channel; The packet loss correlation factor is obtained by calculating the autocorrelation coefficient of the interval time between consecutive packet losses, and a path stability warning is triggered when the autocorrelation coefficient exceeds a preset threshold.

3. A multi-path online live broadcast intelligent control and management system according to claim 2, characterized in that: In the bandwidth contention coefficient, the step of quantifying the data volume ratio of the bullet screen data and the video stream in the uplink channel includes: Define the bandwidth contention coefficient: Where c represents the bandwidth contention coefficient, a represents the total bytes of bullet chat data within the statistical window, b represents the total bytes of video data within the statistical window, γ represents the burstiness weight constant, and c1 represents the coefficient of variation of the bullet chat throughput. The coefficient of variation of the barrage throughput is taken as: Among them, c1 represents the coefficient of variation of the barrage throughput, σ a Indicates the sample standard deviation of the number of bytes of bullet comment data, μ a Indicates the sample mean of the number of bytes of bullet comment data.

4. A multi-path online live broadcast intelligent control and management system according to claim 1, characterized in that: The content analysis module and the barrage semantic parsing unit share a physical storage area through a memory-mapped register, and a PCIe data channel with a priority arbitration mechanism is set between the two; The syntax layer parsing includes decoding the sequence parameter set SPS and picture parameter set PPS in the video stream, extracting the slice_type identifier and reference frame list of each frame; The syntax analysis of the content analysis module specifically includes: Identify the distribution of I-frames, P-frames, and B-frames in the video stream; Constructing a forward and backward reference relationship map for each frame within the image group; Calculate the diffusion depth value of the impact of any frame loss on subsequent image groups.

5. A multi-path online live broadcast intelligent control and management system according to claim 4, characterized in that: In the syntax layer parsing process, the step of calculating the diffusion depth value of the impact of any frame loss on subsequent image groups includes: Identify the frame type: s k= h kmod5, Among them, s k Indicates the type identifier of the kth frame, 0 corresponds to I frame, 1 corresponds to P frame, 2 corresponds to B frame, h k Indicates the integer value of the slice_type field of the kth frame, where k represents the frame index; Modeling reference relationships: If frame i is referenced by frame j, then m i,j =1, otherwise m i,j =0, Among them, m i,j Represents the reference relationship matrix element between frame i and frame j, where i and j represent the indexes of the referenced frame and the referencing frame respectively; Calculate the impact diffusion depth using the formula: Among them, d j represents the maximum impact depth of the subsequent image group after the loss of frame j, p represents a directed reference path starting from frame j, represents the set of all reachable directed reference paths, and |p| represents the length of path p.

6. A multi-path online live broadcast intelligent control and management system according to claim 1, characterized in that: The path allocation instruction generation method of the dynamic scheduling decision module includes: Assign the lowest latency path to I frames and prohibit switching transmission paths midway; Adopt multi-path redundant transmission mode for P frames, using two or more paths in parallel; Implement an on-demand transmission strategy for B frames, sending complete data only when the path idle bandwidth exceeds the threshold; The redundant backup channel continuously receives the I-frame check code and reconstructs the original data based on the check code when the main channel transmission fails; The path allocation instruction generation method of the dynamic scheduling decision module also includes: When an I-frame transmission is detected, the dynamic scheduling decision module broadcasts the frame transmission request to all available paths, selects the first path that returns a confirmation response as the main transmission channel, and the remaining paths are automatically converted into redundant backup channels.

7. A multi-path online live broadcast intelligent control and management system according to claim 6, characterized in that: In the dynamic scheduling decision module, an allocation matrix is generated based on the frame-level importance and the real-time status of the path, including: Calculate the composite cost for the u-th available path using the formula: g u =θ1τ u +θ2ρ u -θ3λ u , Among them, g u is the composite cost of path u, θ1 represents the delay weight coefficient, τ u represents the current round-trip delay of path u, θ2 represents the packet loss weight coefficient, ρ u represents the instantaneous packet loss rate of path u, θ3 represents the bandwidth weight coefficient, λ u represents the remaining available bandwidth of path u; The cost is converted into allocation probability using exponential mapping, and for the j-th frame we get: Among them, π j,u represents the normalized probability of selecting path u in the jth frame, Indicates the same frame type as s j The corresponding priority amplification factor, s j =0, 1, 2 represent I, P, and B frames respectively, U represents the total number of available paths, g v represents the composite cost of the vth path; The number of parallel paths is determined based on the frame type and then converted into a binary allocation matrix: like Then A j,u =1, other cases A j,u =0, Among them, t j Indicates the number of target paths in the jth frame, 1 {·} is an indicator function, which takes the value 1 when the condition {·} is met, otherwise it takes the value 0. j,u Assign matrix elements to frame paths, Indicates pressing π j,u Select the first t from largest to smallest j The set of paths, π j,* is the probability vector of the j-th frame on all U paths.

8. The multi-path online live broadcast intelligent control and management system according to claim 1, characterized in that: The barrage semantic parsing unit performs: Parse the special effect triggering instruction code in the barrage text; Distinguish between pure text bullet comments and enhanced bullet comments with particle effects and 3D models; Generate a bandwidth requirement tag that is positively correlated with rendering complexity and inject it into the packet header; The texture map reference identifier is a Unity engine ShaderLab format or an Unreal engine Material expression hash value; The parsing of the special effect trigger instruction code in the barrage semantic parsing unit includes detecting whether there is a three-dimensional coordinate transformation matrix, particle emitter parameters or texture map reference identifier in the barrage text.

9. The multi-path online live broadcast intelligent control and management system according to claim 1, characterized in that: The strategy of the adaptive retransmission controller includes: Implement unconditional and immediate retransmission of lost I-frame data; When a P frame loss is detected, only the intra-frame predicted macroblocks of the frame are retransmitted; If a B frame is lost and the remaining path bandwidth is below a threshold, the frame is discarded and forward frame copy compensation is enabled.

10. The multi-path online live broadcast intelligent control and management system according to claim 1, characterized in that: The network status learning module includes: The path oscillation prediction submodule models the quality change trend of each path based on the LSTM network; Conflict avoidance decision tree, which stores the optimal scheduling combinations under common bandwidth competition scenarios; Model online update interface, receiving feedback on actual transmission effects and correcting path evaluation weights; The path oscillation prediction submodule includes an FPGA-accelerated LSTM computation unit, whose weight matrix is stored in the onboard BRAM. The decision basis of the conflict avoidance decision tree in the network status learning module includes at least two factors of the historical bandwidth utilization curve of the current period, the anchor terminal movement speed detection value, and the number of parallel microphone sessions.

Images