A video dynamic code rate intelligent adjusting method and system

Abstract

The present application relates to the technical field of network communication, and more particularly to a video dynamic code rate intelligent adjustment method and system, which comprises the following steps: collecting encoding parameters, transmission state data and playing state data comprehensively to provide basis for intelligent adjustment and ensure that the decision is in line with the actual scene; deconstructing the encoding parameters, extracting features and fusing to generate a content-aware feature vector, so that the code rate adjustment is in line with the video characteristics and the visual quality is effectively improved; predicting the network quality change trend and evaluating the playing risk index to provide a forward-looking guide for code rate adjustment, timely adjust the code rate and other parameters, and reduce the possibility of playing interruption; generating an optimal decision vector by comprehensively considering multiple targets, balancing the visual quality, the probability of freezing and the switching smoothness, and realizing the overall optimization of video transmission and playing performance; and adjusting the parameters of each link of the video through cross-layer self-adaptation to fully exert the advantages of different levels and ensure smooth and high-quality video playing.

Description

Technical Field

[0001] This invention relates to the field of network communication technology, and in particular to a method and system for intelligent adjustment of dynamic video bitrate. Background Technology

[0002] With the widespread adoption of mobile networks and video streaming services, video transmission systems need to respond in real time to dynamically changing network conditions and diverse video content. Adaptive bitrate technology, as the core of ensuring a smooth playback experience, aims to dynamically adjust the video encoding bitrate under limited and fluctuating network bandwidth to balance visual quality and playback continuity.

[0003] Existing technologies face fundamental limitations in complex real-world scenarios: First, the adjustment decisions rely on a single basis. Methods based on network state may lead to quality overload or waste during sudden content changes, while buffer-based methods are slow to respond to sudden network congestion. Second, there is a lack of cross-layer collaboration. Bitrate switching at the encoding layer, congestion control and protection strategies at the transport layer, and buffer prefetching at the playback layer operate independently, and policy conflicts may lead to overall efficiency losses. Existing methods struggle to achieve a dynamic optimal balance between visual quality, playback smoothness, and bitrate stability under multiple objectives such as time-varying network constraints, differentiated content requirements, and limited terminal resources. As a result, in complex environments such as wireless mobile or shared networks, video services still frequently face problems such as sudden changes in image quality, playback stuttering, or buffering delays, severely impacting user experience. To address these issues, this invention proposes a method and system for intelligent dynamic bitrate adjustment in video. Summary of the Invention

[0004] The purpose of this invention is to solve the problems in the background art, and to propose a video dynamic bitrate intelligent adjustment method and system.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: The first aspect of this invention provides a method for intelligent adjustment of dynamic bitrate in video, comprising: S1. Real-time acquisition of encoding parameters of the target video stream, transmission status data of the network transmission link, and playback status data of the terminal player; wherein, encoding parameters include quantization parameters and frame type sequence; transmission status data includes real-time bandwidth, round-trip time, packet loss rate, and jitter intensity; playback status data includes current buffer length, historical stuttering records, decoding frame rate, and rendering frame rate; S2. Deconstruct the encoding parameters of the target video stream, extract scene semantic features and temporal motion features respectively, and then fuse them to generate a content-aware feature vector; S3. Based on the transmission status data of the network transmission link, predict the network quality change trend in the future time window; at the same time, combine the playback status data of the terminal player with the predicted network quality change trend to assess the playback risk index in the future time window. S4. The content-aware feature vector, network quality change trend and playback risk index are combined as inputs and calculated by a multi-objective optimization decision-maker to generate a bitrate adjustment decision vector. S5. Perform cross-layer collaborative adjustment based on the bitrate adjustment decision vector, that is, adaptively adjust the video encoding layer, data transmission layer and terminal playback layer respectively.

[0006] Further, S2 includes: Extract keyframes from the target video stream based on the frame type sequence in the encoding parameters; Semantic segmentation is performed on the extracted keyframes to identify different semantic category regions, including at least static background regions, text regions, face regions, and highly dynamic moving object regions, and semantic segmentation results are obtained. Calculate pixel motion vectors within the same semantic category region between consecutive frames, and generate temporal motion features based on the statistical characteristics of the motion vectors; Based on the semantic segmentation results, the texture complexity and spatial frequency features of each semantic category region are calculated to generate scene semantic features; The temporal motion features and scene semantic features are weighted and fused to generate a content-aware feature vector.

[0007] Further, S3 includes: Input the transmission status data of the network transmission link into the pre-trained time series prediction submodule; The timing prediction submodule is used to analyze the short-term jitter characteristics, long-term bandwidth trends and sudden packet loss events of transmission status data, and to construct a network status evolution feature vector. The network state evolution feature vector is mapped to the bandwidth prediction, latency prediction, and packet loss rate prediction within a future preset time window, thus forming the network quality change trend. Obtain the current buffer length from the playback status data, and the bandwidth and latency prediction values ​​from the network quality change trend, and calculate the expected buffer consumption rate; Based on historical stuttering records in playback status data and packet loss rate predictions in network quality change trends, the effective arrival rate of data packets is calculated. The playback risk index is obtained by normalizing and weighting the expected buffer consumption rate and the effective arrival rate of data packets.

[0008] Further, S4 includes: A composite objective function is established. The first term of the composite objective function is the expected visual quality function, whose input includes candidate bitrate values ​​and content-aware feature vectors. The second term is the expected stuttering penalty function, whose input includes candidate bitrate values, playback risk index, and network quality change trend. The third term is the bitrate switching smoothness penalty function, whose input is the difference between the candidate bitrate value and the current actual bitrate. Finally, a set of candidate parameters is obtained, namely candidate bitrate, candidate forward error correction strategy parameters, and candidate playback buffering strategy parameters. Based on the composite objective function, a search is performed in the solution space composed of candidate parameters to find the optimal combination of solutions that makes the composite objective function value optimal, namely the optimal bitrate, the optimal forward error correction strategy parameters, and the optimal playback buffer strategy parameters. The optimal solution obtained from the search is combined and encoded into a rate adjustment decision vector.

[0009] Furthermore, in S4, the multi-objective optimization decision-maker aims to maximize the expected visual quality, minimize the probability of playback stuttering, and maintain the smoothness of bitrate switching, and outputs a bitrate adjustment decision vector containing the target bitrate value, the optimal forward error correction strategy parameters, and the optimal playback buffering strategy parameters; wherein, the target bitrate value is the optimal bitrate.

[0010] Furthermore, in S5, at the video coding layer, the coding parameters of the video coding layer are adjusted according to the target bitrate value in the bitrate adjustment decision vector; at the data transmission layer, the congestion control parameters are adjusted according to the optimal forward error correction strategy parameters of the bitrate adjustment decision vector; and at the terminal playback layer, the buffer management strategy of the terminal player is dynamically adjusted according to the optimal playback buffer strategy parameters in the bitrate adjustment decision vector.

[0011] Furthermore, in the video coding layer, adjusting the coding parameters of the video coding layer according to the target bitrate value in the bitrate adjustment decision vector includes: The target quantization parameters are determined based on the target bitrate value and the scene semantic features in the content-aware feature vector; Based on the target quantization parameters and combined with the temporal motion features in the content-aware feature vector, the decision threshold for macroblock partitioning is calibrated, and macroblock mode selection for inter-frame prediction and intra-frame prediction is performed according to the calibrated decision threshold. Based on the macroblock mode selection, the insertion interval of the next keyframe is dynamically adjusted according to the playback risk index.

[0012] Furthermore, at the data transmission layer, adjusting the congestion control parameters according to the optimal forward error correction strategy parameters of the rate adjustment decision vector includes: The proportion and timing of adding redundant data packets are determined based on the optimal forward error correction strategy parameters of the bitrate adjustment decision vector. While determining the proportion and timing of adding redundant data packets, the congestion window is dynamically adjusted based on the predicted latency and packet loss rate in the network quality change trend. Based on the playback risk index, a differentiated retransmission strategy is implemented for video data packets.

[0013] Furthermore, in the terminal playback layer, dynamically adjusting the buffer management strategy of the terminal player based on the optimal playback buffer strategy parameters in the bitrate adjustment decision vector includes: Analyze the optimal playback buffering strategy parameters to obtain the target buffer length; The current buffer length is compared and analyzed with the target buffer length. Based on the comparison results and the bandwidth prediction value in the network quality change trend, a prefetch operation is performed, that is, a request is made to the data transmission layer to prefetch subsequent video data segments. During the prefetch operation, the rate at which data is read from the buffer for decoding is fine-tuned based on the real-time decoding and rendering frame rates in the playback status data.

[0014] A second aspect of the present invention provides a video dynamic bitrate intelligent adjustment system, comprising: Multi-source data acquisition module: Real-time acquisition of encoding parameters of the target video stream, transmission status data of the network transmission link, and playback status data of the terminal player; Deep feature deconstruction and fusion module: Deconstructs the encoding parameters of the target video stream, extracts scene semantic features and temporal motion features respectively, and then fuses them to generate a content-aware feature vector; Quality Prediction and Risk Assessment Module: Based on the transmission status data of the network transmission link, it predicts the trend of network quality changes in the future time window; at the same time, it combines the playback status data of the terminal player with the predicted trend of network quality changes to assess the playback risk index in the future time window. Intelligent decision generation and optimization module: It integrates content-aware feature vectors, network quality change trends and playback risk index as joint inputs, and calculates them through a multi-objective optimization decision-maker to generate bitrate adjustment decision vectors; Multi-layer cross-domain collaborative adjustment module: Performs cross-layer collaborative adjustment based on the bitrate adjustment decision vector, that is, adaptively adjusts the video encoding layer, data transmission layer and terminal playback layer respectively.

[0015] Compared with existing technologies, the beneficial effects of the present invention in providing a video dynamic bitrate intelligent adjustment method and system are as follows: 1) By collecting target video stream encoding parameters, network transmission link status data, and terminal player playback status data in real time, comprehensive and accurate basic information is provided for subsequent intelligent adjustment. Among them, encoding parameters reflect the characteristics of the video itself, network transmission data reflects the real-time network status, and playback status data reflects the terminal playback status. The combination of multi-source data can accurately grasp the status of the entire video transmission and playback process, making adjustment decisions more in line with actual needs and avoiding decision deviations due to missing information. 2) By deconstructing the encoding parameters, extracting scene semantic features and temporal motion features and fusing them into a content-aware feature vector, we can gain a deeper understanding of the video content. Different semantic regions and motion features have different bitrate requirements. By accurately identifying features, we can provide a content-level basis for subsequent bitrate adjustment, making the bitrate adjustment more in line with the characteristics of the video content, improving visual quality, and avoiding unreasonable bitrate allocation due to insufficient understanding of the content. 3) By predicting network quality change trends and assessing playback risk index, we can grasp network dynamics and potential playback problems in advance. By analyzing historical network data to predict future trends and assessing risks in combination with playback status, we can provide forward-looking guidance for bitrate adjustment, deal with possible network fluctuations and playback stuttering in advance, reduce the risk of playback interruption, and improve user experience. 4) By integrating multi-dimensional information, a multi-objective optimization decision-maker generates a bitrate adjustment decision vector. This vector comprehensively considers multiple objectives, including expected visual quality, probability of playback stuttering, and smoothness of bitrate switching, to achieve a globally optimal decision. This balances bitrate adjustment among several important indicators, improving overall video transmission and playback performance. Furthermore, by adaptively adjusting the video encoding layer, data transmission layer, and terminal playback layer based on the bitrate adjustment decision vector, cross-layer collaborative optimization is achieved. Each layer works in concert to optimize the entire process from video generation and transmission to playback, fully leveraging the advantages of each layer to ensure smooth, high-quality video playback under different network and device environments. Attached Figure Description

[0016] Figure 1 This is a flowchart of a video dynamic bitrate intelligent adjustment method proposed in this invention.

[0017] Figure 2 This is a block diagram of a video dynamic bitrate intelligent adjustment system proposed in this invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example 1:

[0019] Please see Figure 1 This invention provides a method for intelligent adjustment of dynamic bitrate in video, comprising: S1. Real-time acquisition of encoding parameters of the target video stream, transmission status data of the network transmission link, and playback status data of the terminal player; wherein, encoding parameters include quantization parameters and frame type sequence; transmission status data includes real-time bandwidth, round-trip time, packet loss rate, and jitter intensity; playback status data includes current buffer length, historical stuttering records, decoding frame rate, and rendering frame rate; Understandably, quantization parameters and frame type sequences can be directly obtained from the video encoder; real-time bandwidth is obtained by measuring the amount of data successfully transmitted per unit time; round-trip latency is calculated by sending test data packets and recording their return times; the packet loss rate is obtained by statistically analyzing the ratio of the number of lost data packets to the total number of packets sent within a certain period; jitter intensity is determined by analyzing the changes in the arrival time interval of data packets; the current buffer length is statistically analyzed by the terminal player; historical stuttering records are recorded by the terminal player for each stuttering event; the decoding frame rate and rendering frame rate are calculated by the video decoding module and rendering module respectively, based on the number of frames processed per unit time. S2. Deconstruct the encoding parameters of the target video stream, extract scene semantic features and temporal motion features respectively, and then fuse them to generate a content-aware feature vector; S3. Based on the transmission status data of the network transmission link, predict the network quality change trend in the future time window; at the same time, combine the playback status data of the terminal player with the predicted network quality change trend to assess the playback risk index in the future time window. S4. The content-aware feature vector, network quality change trend, and playback risk index are used as joint inputs and calculated by a multi-objective optimization decision-maker to generate a bitrate adjustment decision vector. It can be understood that the multi-objective optimization decision-maker aims to maximize the expected visual quality, minimize the probability of playback stuttering, and maintain the smoothness of bitrate switching. The output bitrate adjustment decision vector includes the target bitrate value, the optimal forward error correction strategy parameters, and the optimal playback buffering strategy parameters. The target bitrate value is the optimal bitrate. S5. Perform cross-layer collaborative adjustment based on the bitrate adjustment decision vector, that is, perform adaptive adjustment on the video coding layer, data transmission layer and terminal playback layer respectively; specifically, in the video coding layer, adjust the coding parameters of the video coding layer according to the target bitrate value in the bitrate adjustment decision vector; in the data transmission layer, adjust the congestion control parameters according to the optimal forward error correction strategy parameters in the bitrate adjustment decision vector; in the terminal playback layer, dynamically adjust the buffer management strategy of the terminal player according to the optimal playback buffer strategy parameters in the bitrate adjustment decision vector.

[0020] In this embodiment of the invention, the detailed implementation steps of content deconstructing the encoding parameters of the target video stream, extracting scene semantic features and temporal motion features respectively, and then fusing them to generate a content-aware feature vector include: S21. Based on the frame type sequence in the encoding parameters, extract key frames from the target video stream. Specifically, this includes parsing the frame type information transmitted by the video coding layer, such as I-frame, P-frame, and B-frame sequences, and selecting I-frames as key frames for in-depth content analysis according to a preset sampling strategy, such as every fixed number of frames or based on scene change detection triggers. The extracted key frames carry the most complete spatial information of the video scene.

[0021] S22. Perform semantic segmentation on the extracted keyframes to identify different semantic category regions, including at least static background regions, text regions, face regions, and highly dynamic moving object regions, and obtain semantic segmentation results. Specifically, an unsupervised region segmentation and rule matching method based on multi-feature hierarchical clustering is adopted: keyframes are segmented into superpixels to obtain multiple visually coherent small regions; for each superpixel region, its color histogram features, directional gradient histogram features, and local binary pattern texture features are extracted in parallel; using all the extracted features, the superpixel regions are hierarchically clustered, and adjacent regions with high feature similarity are merged to form larger semantic candidate regions; finally, a set of predefined semantic rule bases are applied to classify the semantic candidate regions to obtain different semantic category regions. For example, if a region has high edge density, a regular rectangular outline, and a specific aspect ratio, it is classified as a text region; if its color distribution conforms to the skin color model and contains an elliptical structure, it is classified as a face region; if the motion vector amplitude of its corresponding region in the preceding and following frames is higher than the global average, it is classified as a high-dynamic moving object region; the rest are classified as static background regions.

[0022] S23. The process of calculating pixel motion vectors within the same semantic category region between consecutive frames and generating temporal motion features based on the statistical characteristics of the motion vectors is as follows: Using a pyramid optical flow method, the motion vector of each pixel in the current frame relative to the previous frame is calculated to obtain a vector field containing horizontal and vertical displacement components. Based on the identified semantic category regions, the entire vector field is partitioned according to the semantic category of the pixels. For all pixels belonging to the same semantic category, statistical calculations are performed on their motion vectors to obtain statistical features, including: calculating the average amplitude of all motion vectors in the region, i.e., the average motion velocity, calculating the distribution histogram of motion vector directions, and calculating the variance of motion vector amplitudes. Feature sub-vectors containing statistical features are generated for each semantic category region. The feature sub-vectors of all categories are concatenated in a fixed order to form a temporal motion feature vector describing the motion intensity and consistency of each category region.

[0023] S24. Based on the semantic segmentation results, calculate the texture complexity and spatial frequency features of each semantic category region to generate scene semantic features. Specifically, perform spatial feature analysis on each identified semantic category region. For static background and text regions, calculate their gray-level co-occurrence matrix and extract texture feature measures such as contrast, correlation, homogeneity, and energy from the matrix to quantify the visual complexity and regularity of the region. For face regions, perform feature point detection to analyze their spatial distribution, calculate the relative distance and angle between feature points, and form a set of features representing the spatial layout of the face. For all semantic category regions, perform discrete cosine transform on the image patch corresponding to the region, and obtain its spatial frequency features by statistically analyzing the energy distribution of its transform coefficients in the frequency domain. The texture complexity features and spatial frequency features extracted for different category regions are uniformly encoded into a comprehensive scene semantic feature vector.

[0024] S25. Weighted fusion of temporal motion features and scene semantic features generates a content-aware feature vector. Understandably, different weight coefficients are assigned to the motion feature vectors of each extracted semantic category region and the corresponding extracted scene semantic feature vectors. The weight coefficients can be set based on experience or learned through offline training. The goal of assigning weight coefficients is to make the fused features most effectively represent the impact of content on encoding difficulty and visual quality. The weighted feature vectors are concatenated into a high-dimensional comprehensive vector, namely the content-aware feature vector, which comprehensively represents the dynamic characteristics and visual complexity of video content in the temporal and spatial domains.

[0025] In this embodiment of the invention, the detailed implementation steps for predicting the network quality change trend in the future time window based on the transmission status data of the network transmission link, and simultaneously evaluating the playback risk index in the future time window by combining the playback status data of the terminal player with the predicted network quality change trend, include: S31. Input the transmission status data of the network transmission link into the pre-trained time-series prediction submodule; specifically, organize the real-time collected network transmission status data, namely real-time bandwidth, round-trip delay, packet loss rate, and jitter intensity, into a sequence according to time order and input it into the pre-trained time-series prediction submodule. The time-series prediction submodule is built based on a recurrent neural network or a temporal convolutional network and can learn the pattern of network status parameters changing over time; wherein, the training data of the time-series prediction submodule is the network status time-series data continuously collected and anonymized in historical video sessions under the same network environment.

[0026] S32. Use the time-series prediction submodule to analyze the short-term jitter characteristics, long-term bandwidth trends, and sudden packet loss events of transmission state data, and construct a network state evolution feature vector, including: Multi-scale analysis is performed on the input real-time bandwidth sequence, specifically by using moving average filters with different time window sizes: a short window filter is used to extract short-term jitter components, and a long window filter is used to extract long-term trend components. Simultaneously, the time series prediction submodule detects sudden spikes in the packet loss rate sequence and identifies sudden packet loss events. Based on the analysis results, the time series prediction submodule extracts a set of features that can comprehensively characterize the current state of the network and its recent evolution, such as short-term fluctuation variance, long-term trend slope, duration and intensity of sudden events, etc., and organizes all features into a low-dimensional dense network state evolution feature vector.

[0027] S33. The network state evolution feature vector is mapped to the predicted values ​​of bandwidth, latency, and packet loss rate within a future preset time window, forming a network quality change trend. Among them, the time series prediction submodule includes a fully connected regression output layer. This layer takes the network state evolution feature vector as input, directly maps and outputs the predicted value sequence of bandwidth, the predicted value sequence of latency, and the predicted value sequence of packet loss rate within a future preset time window. The three prediction sequences together constitute the network quality change trend to provide a quantitative prediction of future network conditions.

[0028] S34. Obtain the current buffer length from the playback status data, and the bandwidth and latency prediction values ​​from the network quality change trend, and calculate the expected buffer consumption rate; wherein, the future data reception rate is estimated based on the bandwidth prediction value; considering that data packet transmission takes time, the latency prediction value is used to calibrate the reception timeline; based on the current video playback bitrate, calculate the net data consumption rate in the buffer under the predicted bandwidth and latency conditions, i.e., the expected buffer consumption rate.

[0029] S35. Based on the historical stuttering records in the playback status data and the predicted packet loss rate in the network quality change trend, calculate the effective arrival rate of data packets; among them, analyze the historical stuttering records in the playback status data to count the frequency of recent decoding failures caused by data packet loss or excessive delay; at the same time, combine the predicted packet loss rate to estimate the proportion of valid data packets that successfully and timely arrive and can be decoded in the next time window, and this proportion is the effective arrival rate of data packets.

[0030] S36. Normalize and weight the buffer consumption rate and the effective arrival rate of data packets to obtain the playback risk index.

[0031] In this embodiment of the invention, the detailed implementation steps for generating a bitrate adjustment decision vector by fusing content-aware feature vectors, network quality change trends, and playback risk index as joint inputs and calculating them through a multi-objective optimization decision maker include: S41. Establish a composite objective function. The first term of the composite objective function is the expected visual quality function, whose input includes candidate bitrate values ​​and content-aware feature vectors; the second term is the expected stuttering penalty function, whose input includes candidate bitrate values, playback risk index, and network quality change trend; the third term is the bitrate switching smoothness penalty function, whose input is the difference between the candidate bitrate value and the current actual bitrate; finally, a set of candidate parameters is obtained, namely candidate bitrate, candidate forward error correction strategy parameters, and candidate playback buffering strategy parameters. Specifically, the expected visual quality function is dynamically adjusted based on the content-aware feature vector. The expected visual quality function Qv has the form: Qv(CR, V)=α×log(1+CR / Bb(V)); where CR is the candidate bitrate value; V represents the content-aware feature vector; α represents the normalization coefficient, which is used to normalize the output to a predetermined range; Bb(V) is the content base bitrate parameter, whose value is obtained by mapping the content-aware feature vector V through a feedforward neural network. This network simulates the bitrate baseline required for content of different complexities to achieve the same subjective quality. The expected stuttering penalty function Qs combines the playback risk index and the network quality change trend. Its specific form is: Qs(CR, P, N) = β × [P × exp(γ × max(0, CR – Bs(N)))]; where P represents the playback risk index; β and γ are sensitivity coefficients for adjusting the penalty intensity; N represents the network quality change trend vector; Bs(N) is a conservative bandwidth estimate extracted from the network quality change trend vector N, for example, the fifth percentile of the future predicted bandwidth; the expected stuttering penalty function indicates that when the selected candidate bitrate value exceeds the conservative bandwidth estimate that the network can safely carry, the penalty will be amplified exponentially due to the existence of the playback risk index; The bitrate switching smoothness penalty function Qt is expressed as: Qt(CR, CR1)=δ×tanh(η×|CR–CR1| / CR1); where CR1 is the current actual bitrate, and the hyperbolic tangent function is used so that the penalty value increases approximately linearly when the switching amplitude is small, and tends to saturate when the amplitude is large; δ and η represent the adjustment parameters; Finally, the composite objective function F is: F(CR,λ,θ)=w1×Qv+w2×Qs+w3×Qt, where λ is the candidate forward error correction strategy parameter; θ is the candidate playback buffer strategy parameter; the candidate forward error correction strategy parameter and the candidate playback buffer strategy parameter are latent variables that affect the network carrying capacity assessment in the expected stuttering penalty; w1, w2, and w3 are the weight coefficients of visual quality, stuttering penalty, and handover smoothness penalty, respectively.

[0032] S42. Based on the composite objective function, a search is performed in the solution space composed of candidate parameters to find the optimal combination of solutions that optimizes the composite objective function value. Specifically, the optimal bitrate, optimal forward error correction strategy parameters, and optimal playback buffer strategy parameters are implemented as follows: Candidate parameters are obtained, and the candidate bitrate, candidate forward error correction strategy parameters, and candidate playback buffer strategy parameters are specified as finite discrete options to form a search space. Specifically, candidate bitrate CR is selected from a preset discrete bitrate level set CR0; candidate forward error correction strategy parameters are selected from a set of predefined redundancy intensity levels λ0; and candidate playback buffer strategy parameters are selected from a set of target buffer duration levels θ0. The Cartesian product of the three sets CR0×λ0×θ0, i.e., all possible three-dimensional parameter combinations, constitutes the optimization solution space. The search process employs an improved simulated annealing algorithm. During algorithm initialization, the current solution is the currently used parameter combination (CR1, λ1, θ1), where CR1 is the current bitrate, λ1 is the current forward error correction strategy parameter, and θ1 is the current playback buffering strategy parameter. In each iteration, the algorithm generates new solutions based on the playback risk index: if the playback risk index is high, the new solution tends to perturb towards lower values ​​in the bitrate level set CR0, while simultaneously perturbing towards higher redundancy levels in the redundancy strength level set λ0; if the playback risk index is low, the search leans more towards exploring higher values ​​in the bitrate dimension to improve performance. Quality; understandably, after each new solution is generated, its corresponding objective function value is calculated; based on the probability acceptance criterion of the simulated annealing algorithm, it is decided whether to replace the current solution with the new solution; this probability acceptance criterion depends on the gradually decaying control parameter, which allows for the acceptance of worse solutions in the early stages of iteration to escape local optima, and tends to accept only better solutions in the later stages of iteration; after multiple rounds of iteration and the decay of the control parameter, the algorithm finally converges to the parameter combination that makes the composite objective function value optimal or nearly optimal, denoted as the optimal solution combination, namely the optimal bitrate CR3, the optimal forward error correction strategy parameter λ3, and the optimal playback buffer strategy parameter θ3.

[0033] S43. Encode the optimal solution obtained from the search into a rate adjustment decision vector, including: The optimal solution obtained from the search is encoded into executable unified instructions, that is, the abstract optimization result is transformed into a bitrate adjustment decision vector that can be directly parsed by each layer. The bitrate adjustment decision vector is a structured data object containing three fields: 1) Target bitrate field, which is directly assigned the optimal bitrate; 2) Forward error correction strategy field, which includes two parts: one is the redundancy ratio determined by the optimal forward error correction strategy parameters; the other is the redundancy packet transmission timing strategy dynamically generated based on the latency prediction value in the network quality change trend; 3) Playback buffer strategy field, which is the target buffer length determined by the optimal playback buffer strategy parameters. The finally generated bitrate adjustment decision vector, as a complete control instruction packet, is synchronously distributed to the video coding layer, data transmission layer, and terminal playback layer, thereby initiating cross-layer collaborative adjustment.

[0034] In this embodiment of the invention, the detailed implementation steps for performing cross-layer collaborative adjustment based on the bitrate adjustment decision vector, i.e., adaptive adjustment of the video coding layer, data transmission layer, and terminal playback layer respectively, include: A1. Adjust the coding parameters at the video coding layer according to the target bitrate value in the bitrate adjustment decision vector, including: S51. Determine the target quantization parameters based on the target bitrate value and the scene semantic features in the content-aware feature vector; Specifically, the encoding controller receives the target bitrate value, i.e., the optimal bitrate CR3, from the bitrate adjustment decision vector, and retrieves the content-aware feature vector V of the current frame (or the frame group to be encoded). The encoding controller has a built-in bitrate-quantization parameter model, which is continuously fine-tuned by an online learner. The basic form of the bitrate-quantization parameter model is CNe = Kc / QP + Co(V), where CNe is the output of the bitrate-quantization parameter model, representing the estimated bitrate, QP is the quantization parameter in the target video stream encoding parameters, Kc is a scaling factor related to image complexity, and Co(V) is the bitrate offset determined by the content-aware feature vector V. Co(V) is calculated by a linear... The combined calculation yields: Co(V) = Σ(ωi × ui), where ui is the i-th component of the content-aware feature vector V, i is the component index of the content-aware feature vector, and ωi is the weight coefficient of the corresponding component, which is obtained from training typical content sequences during the initialization of the coding controller. In order to solve the target quantization parameter, the coding controller substitutes the target bitrate value CR3, the current scaling factor Kc obtained from the statistics of the encoded frames, and the calculated bitrate offset Co(V) into the bitrate-quantization parameter model to solve for QP0 = Kc / (CR3 - Co(V)), where QP0 represents the target quantization parameter, which is the core quantization step size that guides the current encoded segment.

[0035] S52. Based on the target quantization parameters and combined with the temporal motion features in the content-aware feature vector, calibrate the decision threshold for macroblock partitioning, and perform macroblock mode selection for inter-frame prediction and intra-frame prediction according to the calibrated decision threshold; wherein, the calibration of the decision threshold for macroblock partitioning is specifically manifested as: adaptive adjustment of the key parameter used to balance distortion and bit rate in the rate-distortion optimization process, namely the Lagrange multiplier, based on the temporal motion features. Specifically, in the rate-distortion optimization process of video coding, the formula Cm=Dis+Gr×Rate is used to calculate the cost value for each candidate macroblock partitioning mode; where Cm represents the cost value; Dis represents the predicted distortion; Rate represents the estimated bit rate required for the macroblock partitioning mode; and Gr is a Lagrange multiplier, which serves as the core decision threshold to control the weight ratio of distortion and bit rate in the cost function. The Lagrange multiplier Gr is adaptively adjusted based on the temporal motion features in the content-aware feature vector. For each candidate macroblock partitioning mode, such as inter-frame prediction modes like 16x16, 16x8, 8x16, and 8x8, and corresponding intra-frame prediction modes, the local Lagrange multiplier Gr1 is calculated based on the temporal motion features corresponding to the candidate macroblock position, such as the average motion vector magnitude mv0 and directional consistency con: Gr1 = Gr0 × f(mv0, con); where Gr0 is the basic Lagrange multiplier; the function f is set as follows: when the average motion vector magnitude is large and the directional consistency is high, it indicates intense and consistent motion, so the function f value is less than 1, thereby reducing the local Lagrange multiplier and weakening the penalty on the bitrate term when calculating the cost, thus making it more inclined to choose a more refined partitioning mode; conversely, for static or chaotic regions, the function f value is greater than or equal to 1, increasing the local Lagrange multiplier and making the decision more biased towards choosing a larger block partition with a lower bitrate. The coding controller uses the local Lagrange multipliers of each candidate macroblock partitioning mode to calculate the cost value of each candidate inter-frame prediction mode and one or more candidate intra-frame prediction modes. The coding controller compares the cost values ​​of all candidate macroblock partitioning modes and selects the mode with the lowest cost value (including inter-frame and intra-frame) as the final coding scheme for the macroblock. In this way, the local Lagrange multipliers are dynamically adjusted based on temporal motion characteristics, guiding the coding controller to adaptively select the macroblock partitioning mode that best matches the motion characteristics of the current content.

[0036] S53. Based on the macroblock mode selection, dynamically adjust the insertion interval of the next key frame according to the playback risk index; whereby, the key frame, i.e. the intra-frame encoded frame, is an independent frame that does not depend on other frames for encoding and can refresh the decoding status. Specifically, the encoding controller sets the initial keyframe interval setting, i.e., the GOP length. In this step, the encoding controller monitors and evaluates the playback risk index; presets high-risk thresholds and low-risk thresholds; during the encoding process, if the playback risk index is detected to be no less than the high-risk threshold, it indicates that the playback environment has deteriorated and the risk of stuttering is high. At this time, the encoding controller forces the next frame to be encoded to be encoded as a keyframe. By forcibly inserting a keyframe, a clean synchronization point can be quickly provided to the decoding end, reducing the error propagation range caused by packet loss and enhancing robustness. Conversely, if the playback risk index is no greater than the low-risk threshold, it indicates that the playback environment is extremely stable. The encoding controller allows the actual keyframe interval to be slightly greater than the initial keyframe interval setting, saving bitrate by reducing the number of keyframes, thereby allocating more bit resources to non-keyframes to improve their quality. It can be understood that the flexible GOP control mechanism based on real-time risk assessment achieves a dynamic optimal balance between encoding efficiency and transmission robustness.

[0037] It should be noted that the video coding layer adaptively adjusts based on the bitrate adjustment decision vector, integrating the target bitrate, content features, and transmission risks into a coherent control logic, and implementing adjustments at three levels: quantization parameters, prediction mode, and frame type. The goal of step S51 is to determine the core coding parameters required to achieve the target bitrate, namely the target quantization parameters. After determining the frame-level target quantization parameters QP0 through step S51, the coding controller needs to further optimize coding efficiency at a finer granularity to adapt to the spatial texture and temporal motion changes of the video content, which leads to step S52. After completing the intra-frame coding detail optimization based on content complexity and motion features, the coding controller needs to control the robustness of the coding structure from a global perspective of inter-frame dependencies to adapt to dynamically changing transmission risks, which is achieved by step S53.

[0038] A2. At the data transmission layer, adjust congestion control parameters based on the optimal forward error correction strategy parameters and the predicted network quality change trend, including: S54. Determine the proportion and timing of adding redundant data packets based on the optimal forward error correction strategy parameters of the bit rate adjustment decision vector. Understandably, the data transmission layer reads the optimal forward error correction strategy parameters from the bitrate adjustment decision vector. These parameters specify the proportion of redundant data packets needed to combat network packet loss; for example, specifying that for every k original video data packets sent, r redundant packets need to be generated and sent. Based on the redundancy proportion, the data transmission layer selects the corresponding forward error correction coding scheme, i.e., selects the specific forward error correction coding scheme from a predefined scheme table, such as using RaptorQ coding, and determines the coding block parameters. It then groups and encodes the video data packets to be sent, generating forward error correction blocks containing both original and redundant data packets. Simultaneously, it determines... The timing for adding redundant data packets is determined by the scheduler, which refers to the predicted packet loss rate in the network quality change trend. If a sudden packet loss risk is detected, an interleaving transmission strategy is adopted for data packets in the same forward error correction block. Specifically, when the predicted packet loss rate shows a sudden characteristic, that is, when a sharp increase in the packet loss rate is predicted in a short period of time, the scheduler will interleave the original data packets and redundant data packets that were originally sent continuously and belong to the same forward error correction block on the timeline. This is to prevent sudden packet loss from destroying a complete forward error correction block in one go, thereby improving the error correction capability. Adjusting the transmission timing based on prediction is a further step in the execution of the optimal forward error correction strategy parameters.

[0039] S55. While determining the proportion and timing of adding redundant data packets, the congestion window is dynamically adjusted in conjunction with the predicted delay and packet loss rates in the network quality change trend; where congestion control is a mechanism in the transmission protocol to avoid network overload by adjusting the data transmission rate. In step S55, the congestion control algorithm, such as Cubic or BBR, continuously monitors the predicted network quality trends, latency predictions, and bandwidth predictions. When the latency prediction shows a significant upward trend while the bandwidth prediction remains stable, it is inferred that queuing accumulation is occurring in the network buffer. At this point, the algorithm performs predictive deceleration: proactively and in advance reducing the growth rate of the congestion window, for example, by reducing the growth factor in the Cubic algorithm or lowering the gain of the BBR algorithm during the bandwidth probing phase. It is understood that predictive deceleration aims to smooth data injection and avoid subsequent tail dropping when the network queue is full. At the same time, when calculating the net available bandwidth that can actually be used to carry effective video data, this step incorporates the additional bandwidth overhead generated by adding redundant data packets into the evaluation system. That is, the overhead corresponding to the current redundancy ratio is dynamically deducted from the bandwidth prediction value, and the upper limit of the congestion window is calibrated based on the net available bandwidth. This ensures that data transmission obtains forward error correction protection without self-initiating congestion due to excessive transmission of redundant data, thereby achieving a balance between reliability and network efficiency.

[0040] S56. Based on the playback risk index, implement a differentiated retransmission strategy for video data packets; wherein, the differentiated retransmission strategy includes: directly assigning the highest retransmission priority to key frame data packets; and for non-key frame data packets, if their corresponding playback risk index and playback urgency together meet the preset high priority triggering condition, then their retransmission request is marked as high priority; it can be understood that playback urgency refers to the remaining time difference between the expected playback deadline of the data packet to be retransmitted and the current system time; for retransmission data packets assigned high priority, a retransmission timeout is set and priority scheduling is given to sending; Specifically, this step handles packet loss that forward error correction failed to fully correct, i.e., fine-tuning the scheduling of retransmission resources based on the importance of the data and real-time risk; the data transmission layer maintains a priority queue of data packets to be retransmitted; priority is determined by two factors: the intrinsic type of the data packet and the external playback risk index; in this method, the classification is mainly based on the importance of the video content it carries. Keyframe data packets (i.e., data packets carrying I-frames) are automatically assigned the highest priority H because they are the independent entry point for decoding and error recovery point; for retransmission requests of non-keyframe data packets, dynamic priority evaluation is performed: the estimated playback deadline of the requested data packet is calculated, and the current buffer length and real-time playback risk are obtained. Risk index; compare the real-time playback risk index with a preset first risk threshold, and calculate the remaining time difference between the playback deadline and the current system time. Compare this remaining time difference with an urgent time threshold dynamically calculated based on the current buffer length. The first risk threshold is a quantitative critical value used to determine whether the playback environment has deteriorated to the point where an emergency retransmission mechanism needs to be triggered, and the urgent time threshold is a time margin benchmark used to determine whether a data packet urgently needs to be retransmitted. If the real-time playback risk index is higher than the first risk threshold and the remaining time difference is less than the urgent time threshold, then the data packet is determined to meet the high-priority triggering condition, and the priority of its corresponding retransmission data packet is raised to the highest priority H. For all retransmitted data packets marked with the highest priority H, the data transmission layer adopts differentiated processing: First, a shortened retransmission timeout is set for them, which is proportionally reduced based on the predicted latency value in the current network quality change trend, for example, set to 1.5 times the predicted latency, so as to achieve faster detection of packet loss and triggering of retransmission; Second, during data transmission scheduling, retransmitted data packets with the highest priority H have the right to preempt transmission, can be interrupted, and are transmitted with priority over ordinary priority data packets; Understandably, based on the dual priority scheduling mechanism of real-time risk and playback urgency, it ensures that when network resources are strained or the condition deteriorates, the data most critical to the continuity of playback can receive the most timely retransmission guarantee.

[0041] It should be noted that the forward error correction strategy configuration in step S54 sets a basic protection baseline and transmission mode for data transmission. This step determines the proportion of redundant data and the transmission timing strategy by parsing the bit rate adjustment decision vector, thereby establishing the first line of defense against network packet loss. The congestion control and transmission rate adjustment in step S55 further optimizes data transmission efficiency and network friendliness on the protection baseline set in step S54. This step dynamically adjusts the transmission rate according to network prediction and needs to deduct the redundant overhead introduced in step S54 to calculate the actual available bandwidth, ensuring that data transmission is reliable while taking into account the overall network efficiency. The differentiated retransmission recovery in step S56 performs optimal recovery for unavoidable packet loss in the transmission environment established in step S55.

[0042] A3. Dynamically adjust the buffer management strategy at the terminal playback layer based on the optimal playback buffer strategy parameters, including: S57. Parse the optimal playback buffering strategy parameters and obtain the target buffer length; wherein, the terminal player receives the bitrate adjustment decision vector and extracts the optimal playback buffering strategy parameters from it; the core component of the optimal playback buffering strategy parameters is the target buffer length, which represents the ideal data buffering duration to be maintained in the current state, denoted as Bt; the terminal player internally uses the target buffer length Bt as the direct target value for buffer length control to guide subsequent buffer management operations.

[0043] S58. Compare and analyze the current buffer length with the target buffer length, and based on the comparison results and the bandwidth prediction value in the network quality change trend, perform a prefetch operation, that is, request the data transmission layer to prefetch subsequent video data segments. Specifically, the terminal player continuously monitors the real-time status of the decoding buffer and obtains the current buffer length, denoted as Bc. It compares the current buffer length Bc with the target buffer length Bt, and simultaneously sets a buffer threshold Bl, where Bl = h × Bt, and h is a coefficient less than 1, such as 0.7. If the current buffer length is detected to be less than the buffer threshold (i.e., below the safety line), the terminal player triggers an active prefetching mechanism. This prefetching mechanism employs a bandwidth-adaptive request strategy, where the terminal player references the bandwidth prediction value in the network quality change trend to request data. The range of data requested for prefetching is denoted as Lr, and its formula is... The formula is Lr = max(Lm, αp × Br × ΔT); where Lm is the minimum request unit, Br is the bandwidth prediction value, αp is the aggressiveness coefficient, and ΔT is the predetermined prediction time window. When the bandwidth prediction value is high, the prefetch request data range formula calculates a larger request volume, thereby making full use of high bandwidth conditions and quickly filling the buffer. When the bandwidth prediction value is low, the request volume tends to be conservative, which can effectively avoid requesting too much data under narrowband conditions, thereby preventing network queue accumulation or request timeouts. It can be understood that the elastic prefetch strategy based on bandwidth prediction achieves a balance between buffer filling efficiency and network friendliness.

[0044] S59. During the prefetch operation, the rate at which data is read from the buffer is decoded is finely adjusted based on the real-time decoding frame rate and rendering frame rate in the playback status data to match the playback progress. The goal of the terminal player is to render video at a stable frame rate. The terminal player monitors two key performance indicators in real time: the decoding frame rate and the rendering frame rate. The decoding frame rate represents the number of frames output per second by the decoder, and the rendering frame rate represents the number of actual screen refreshes per second. Ideally, the decoding frame rate should be slightly greater than or equal to the rendering frame rate. The terminal player calculates the difference between the decoding and rendering frame rates to obtain the real-time frame rate difference value. It then applies a smoothing filter to this difference to obtain a smoothed frame rate difference. Based on this smoothed frame rate difference, the rate at which data is read from the buffer and sent to the decoder is fine-tuned. During the control process, if the smoothed frame rate difference remains negative and its absolute value exceeds a predetermined value, the player will be subject to further adjustments. The set difference threshold indicates that the decoder's frame output rate is lower than the renderer's frame refresh rate, which will eventually lead to buffer exhaustion. In this case, a slight deceleration signal is sent to the decoder to briefly delay the delivery of the next frame's encoded data, thereby adjusting the decoder's actual workload and allowing the decoding frame rate to slowly recover and approach the rendering frame rate. The difference threshold is used to determine whether the smooth frame rate difference reaches a level that requires adjustment of the decoder's data delivery rate. Conversely, if the smooth frame rate difference is consistently positive and large, it indicates that the decoding capacity is excessive, and the buffer may continue to grow, increasing playback latency. In this case, the terminal player appropriately speeds up the data delivery rate, prompting the decoder to consume more cached data. Example 2:

[0045] Please see Figure 2 This invention provides a video dynamic bitrate intelligent adjustment system, comprising: Multi-source data acquisition module: Real-time acquisition of encoding parameters of the target video stream, transmission status data of the network transmission link, and playback status data of the terminal player; Deep feature deconstruction and fusion module: Deconstructs the encoding parameters of the target video stream, extracts scene semantic features and temporal motion features respectively, and then fuses them to generate a content-aware feature vector; Quality Prediction and Risk Assessment Module: Based on the transmission status data of the network transmission link, it predicts the trend of network quality changes in the future time window; at the same time, it combines the playback status data of the terminal player with the predicted trend of network quality changes to assess the playback risk index in the future time window. Intelligent decision generation and optimization module: It integrates content-aware feature vectors, network quality change trends and playback risk index as joint inputs, and calculates them through a multi-objective optimization decision-maker to generate bitrate adjustment decision vectors; Multi-layer cross-domain collaborative adjustment module: Performs cross-layer collaborative adjustment based on the bitrate adjustment decision vector, that is, adaptively adjusts the video encoding layer, data transmission layer and terminal playback layer respectively.

[0046] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0047] It should be noted that all formulas in this manual are calculated by removing dimensions and taking their numerical values. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0048] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for intelligent adjustment of dynamic bitrate in video, characterized in that, include: S1. Real-time acquisition of encoding parameters of the target video stream, transmission status data of the network transmission link, and playback status data of the terminal player; wherein, encoding parameters include quantization parameters and frame type sequence; transmission status data includes real-time bandwidth, round-trip time, packet loss rate, and jitter intensity; playback status data includes current buffer length, historical stuttering records, decoding frame rate, and rendering frame rate; S2. Deconstruct the encoding parameters of the target video stream, extract scene semantic features and temporal motion features respectively, and then fuse them to generate a content-aware feature vector; S3. Based on the transmission status data of the network transmission link, predict the network quality change trend in the future time window; at the same time, combine the playback status data of the terminal player with the predicted network quality change trend to evaluate the playback risk index in the future time window; the playback risk index is obtained by joint analysis of the expected buffer consumption rate and the effective arrival rate of data packets. S4. The content-aware feature vector, network quality change trend and playback risk index are combined as inputs and calculated by a multi-objective optimization decision-maker to generate a bitrate adjustment decision vector. S5. Perform cross-layer collaborative adjustment based on the bitrate adjustment decision vector, that is, adaptively adjust the video coding layer, data transmission layer and terminal playback layer respectively. Specifically, in the video coding layer, the coding parameters of the video coding layer are adjusted according to the target bitrate value in the bitrate adjustment decision vector; in the data transmission layer, the congestion control parameters are adjusted according to the optimal forward error correction strategy parameters of the bitrate adjustment decision vector; and in the terminal playback layer, the buffer management strategy of the terminal player is dynamically adjusted according to the optimal playback buffer strategy parameters in the bitrate adjustment decision vector. The process of adjusting the encoding parameters of the video encoding layer based on the target bitrate value in the bitrate adjustment decision vector at the video encoding layer: Based on the target bitrate value and the scene semantic features in the content-aware feature vector, the target quantization parameters are determined; based on the target quantization parameters and combined with the temporal motion features in the content-aware feature vector, the decision threshold for macroblock partitioning is calibrated, and macroblock mode selection for inter-frame prediction and intra-frame prediction is performed according to the calibrated decision threshold; based on the completion of macroblock mode selection, the insertion interval of the next key frame is dynamically adjusted according to the playback risk index. The process of adjusting congestion control parameters at the data transmission layer based on the optimal forward error correction strategy parameters of the decision vector according to the bit rate adjustment: The proportion and timing of adding redundant data packets are determined based on the optimal forward error correction strategy parameters of the bitrate adjustment decision vector. While determining the proportion and timing of adding redundant data packets, the congestion window is dynamically adjusted by combining the predicted latency and packet loss rate in the network quality change trend. Based on the playback risk index, a differentiated retransmission strategy is implemented for video data packets.

2. The video dynamic bitrate intelligent adjustment method according to claim 1, characterized in that, S2 includes: Extract keyframes from the target video stream based on the frame type sequence in the encoding parameters; Semantic segmentation is performed on the extracted keyframes to identify different semantic category regions, including at least static background regions, text regions, face regions, and highly dynamic moving object regions, and semantic segmentation results are obtained. Calculate pixel motion vectors within the same semantic category region between consecutive frames, and generate temporal motion features based on the statistical characteristics of the motion vectors; Based on the semantic segmentation results, the texture complexity and spatial frequency features of each semantic category region are calculated to generate scene semantic features; The temporal motion features and scene semantic features are weighted and fused to generate a content-aware feature vector.

3. The video dynamic bitrate intelligent adjustment method according to claim 2, characterized in that, S3 includes: Input the transmission status data of the network transmission link into the pre-trained time series prediction submodule; The timing prediction submodule is used to analyze the short-term jitter characteristics, long-term bandwidth trends and sudden packet loss events of transmission status data, and to construct a network status evolution feature vector. The network state evolution feature vector is mapped to the bandwidth prediction, latency prediction, and packet loss rate prediction within a future preset time window, thus forming the network quality change trend. Obtain the current buffer length from the playback status data, and the bandwidth and latency prediction values ​​from the network quality change trend, and calculate the expected buffer consumption rate; Based on historical stuttering records in playback status data and packet loss rate predictions in network quality change trends, the effective arrival rate of data packets is calculated. The playback risk index is obtained by normalizing and weighting the expected buffer consumption rate and the effective arrival rate of data packets.

4. The video dynamic bitrate intelligent adjustment method according to claim 3, characterized in that, S4 includes: A composite objective function is established. The first term of the composite objective function is the expected visual quality function, whose input includes candidate bitrate values ​​and content-aware feature vectors. The second term is the expected stuttering penalty function, whose input includes candidate bitrate values, playback risk index, and network quality change trend. The third term is the bitrate switching smoothness penalty function, whose input is the difference between the candidate bitrate value and the current actual bitrate. Finally, a set of candidate parameters is obtained, namely candidate bitrate, candidate forward error correction strategy parameters, and candidate playback buffering strategy parameters. Based on the composite objective function, a search is performed in the solution space composed of candidate parameters to find the optimal combination of solutions that makes the composite objective function value optimal, namely the optimal bitrate, the optimal forward error correction strategy parameters, and the optimal playback buffer strategy parameters. The optimal solution obtained from the search is combined and encoded into a rate adjustment decision vector.

5. The video dynamic bitrate intelligent adjustment method according to claim 1, characterized in that, In S4, the multi-objective optimization decision-maker aims to maximize the expected visual quality, minimize the probability of playback stuttering, and maintain the smoothness of bitrate switching. It outputs a bitrate adjustment decision vector containing the target bitrate value, the optimal forward error correction strategy parameters, and the optimal playback buffering strategy parameters. The target bitrate value is the optimal bitrate.

6. The video dynamic bitrate intelligent adjustment method according to claim 1, characterized in that, The step of dynamically adjusting the buffer management strategy of the terminal player at the terminal playback layer based on the optimal playback buffer strategy parameters in the bitrate adjustment decision vector includes: Analyze the optimal playback buffering strategy parameters to obtain the target buffer length; The current buffer length is compared and analyzed with the target buffer length. Based on the comparison results and the bandwidth prediction value in the network quality change trend, a prefetch operation is performed, that is, a request is made to the data transmission layer to prefetch subsequent video data segments. During the prefetch operation, the rate at which data is read from the buffer for decoding is fine-tuned based on the real-time decoding and rendering frame rates in the playback status data.

7. A video dynamic bitrate intelligent adjustment system, characterized in that, The system, applied to the video dynamic bitrate intelligent adjustment method as described in any one of claims 1-6, comprises: Multi-source data acquisition module: Real-time acquisition of encoding parameters of the target video stream, transmission status data of the network transmission link, and playback status data of the terminal player; Deep feature deconstruction and fusion module: Deconstructs the encoding parameters of the target video stream, extracts scene semantic features and temporal motion features respectively, and then fuses them to generate a content-aware feature vector; Quality Prediction and Risk Assessment Module: Based on the transmission status data of the network transmission link, it predicts the trend of network quality changes in the future time window; at the same time, it combines the playback status data of the terminal player with the predicted trend of network quality changes to assess the playback risk index in the future time window. Intelligent decision generation and optimization module: It integrates content-aware feature vectors, network quality change trends and playback risk index as joint inputs, and calculates them through a multi-objective optimization decision-maker to generate bitrate adjustment decision vectors; Multi-layer cross-domain collaborative adjustment module: Performs cross-layer collaborative adjustment based on the bitrate adjustment decision vector, that is, adaptively adjusts the video encoding layer, data transmission layer and terminal playback layer respectively.

Images