Live streaming process dynamic collaborative management method and system for urban media convergence scene

Abstract

The application discloses a live streaming process dynamic collaborative management and control method and system for a city media convergence scene, and relates to the technical fields of edge computing and media transmission. The method acquires a front-end live streaming, extracts motion feature macroblocks and barrage burst density, and constructs a service load perception matrix containing overhead and throughput prediction values; polls idle computing power of edge nodes and residual bandwidth of slices, performs binary constraint solving in combination with the matrix, generates component offloading topology and slice routing forwarding table; accordingly, video rendering and transcoding microservices are decoupled and migrated between the edge and the cloud, and a dynamic collaborative service pipeline is constructed by binding a transmission channel; when continuously monitoring end-to-end queuing delay of the pipeline and triggering congestion boundaries, motion feature macroblocks are used to maintain a reference frame rate rendering for covered areas and asynchronous frame decimation processing is performed for uncovered areas. The system is used to implement the above method.

Description

Technical Field

[0001] This invention relates to the fields of edge computing and media transmission technology, specifically to a method and system for dynamic collaborative management and control of live streaming processes in urban converged media scenarios. Background Technology

[0002] Urban converged media has become an important form of public information dissemination and urban cultural display. Within this converged media ecosystem, multi-source live streaming, due to its strong real-time nature and immersive experience, has become a core dissemination method. Its scenarios typically include on-site acquisition nodes, cloud-based control rooms, and multi-platform distribution networks. To ensure stable transmission and image quality of multiple concurrent live streams, existing technologies mostly employ centralized resource scheduling mechanisms and fixed video processing pipelines. This involves centrally allocating computing resources and network bandwidth in the cloud to maintain the processing and distribution of multimedia data.

[0003] However, in large-scale city-level live broadcasts or breaking news scenarios, the intensity of the video content and the frequency of audience interaction are highly unpredictable and sudden. Existing management methods typically rely on pre-allocated static resources or make lagging adjustments based on simple network congestion, making it difficult to capture dynamic changes at the content level. When localized high loads occur, the rigid centralized computing power allocation and network routing mechanisms respond slowly, and heavy-load tasks such as video rendering and transcoding cannot be flexibly scheduled between front-end nodes and the cloud, resulting in a coexistence of localized resource overload and idleness. This static and fragmented management process easily leads to problems such as live broadcast stuttering, loss of important frames, and increased interactive latency. Therefore, it is still necessary to provide a method that can adaptively schedule according to the characteristics of the live broadcast content and the state of network resources to ensure the smoothness of live broadcast services and the efficiency of resource utilization in urban converged media scenarios. Summary of the Invention

[0004] In order to overcome the above-mentioned deficiencies of the prior art, embodiments of the present invention provide a method and system for dynamic collaborative management and control of live streaming processes for urban converged media scenarios to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a dynamic collaborative management and control method for live streaming processes in urban converged media scenarios, comprising the following steps: S1, acquiring multi-source heterogeneous live streaming streams from front-end acquisition nodes, extracting motion feature macroblocks and bullet screen burst densities from the multi-source heterogeneous live streaming streams; performing feature quantization and splicing of the motion feature macroblocks and bullet screen burst densities to construct a service load perception matrix, the service load perception matrix including node processing overhead prediction values ​​and link throughput demand values; S2, reading the service load perception matrix, polling the idle computing power pool and remaining bandwidth of network slices in the access network edge node group; applying binary constraints to the node processing overhead prediction values, link throughput demand values, idle computing power pools, and remaining bandwidth of network slices. S3. Based on the business component unloading topology map, perform instance-level decoupling and dynamic migration of the video rendering microservice and transcoding microservice between the front-end acquisition node and the cloud broadcasting node; allocate transmission channels between nodes according to the slice routing table, and bind the dynamically migrated microservices and transmission channels in series to form a dynamic collaborative business pipeline; S4. Monitor the end-to-end queuing latency of the dynamic collaborative business pipeline. When the end-to-end queuing latency reaches the congestion limit, call the video rendering microservice in the dynamic collaborative business pipeline, combine motion feature macroblocks, maintain the baseline frame rate rendering for the screen area covered by the motion feature macroblocks, and perform asynchronous frame extraction and downsampling processing for the uncovered areas.

[0006] In a preferred embodiment, the specific process of acquiring multi-source heterogeneous live streams from front-end acquisition nodes and extracting motion feature macroblocks and bullet screen burst density from the multi-source heterogeneous live streams is as follows: parse the frame sequences contained within the multi-source heterogeneous live streams and separate the inter-frame predictive coding data contained in the frame sequences; extract the set of motion vectors carried by the inter-frame predictive coding data, and select the directional gradient anomaly macroblock group from the motion vector set to construct motion feature macroblocks; synchronously monitor the interactive bullet screen data channel bound to the multi-source heterogeneous live streams and construct a sliding time window based on time decay attributes; statistically analyze the slope of the text bullet screen arrival rate change within the sliding time window and convert the slope of the arrival rate change into bullet screen burst density.

[0007] In a preferred embodiment, the specific process of constructing a service load perception matrix by quantizing and concatenating motion feature macroblocks and bullet burst density is as follows: The number of blocks and pixel update frequency of the motion feature macroblocks are obtained; the address is located in the computing power overhead mapping table based on the number of blocks and pixel update frequency, and the node processing overhead prediction value is output; the concurrent connection request features associated with the bullet burst density are analyzed; the packet encapsulation frequency of the underlying network protocol stack is estimated based on the concurrent connection request features, and the link throughput requirement value is output; the node processing overhead prediction value and the link throughput requirement value are converted into feature vectors according to a preset mapping dimension and tensor product operations are performed to generate the service load perception matrix.

[0008] In a preferred embodiment, the specific process of reading the service load perception matrix and polling the idle computing power pool and remaining bandwidth of the network slice of the edge node group within the access network is as follows: sending resource telemetry probes to the edge node group within the access network to collect the CPU load level and accelerator card memory fragmentation rate during the operation phase of the edge node group, and aggregating them to form an idle computing power pool; connecting to the northbound interface of the software-defined network controller to obtain the real-time flow table throughput status of the physical link between the edge node group and the cloud broadcast node; extracting the isolation bandwidth quota fed back by the network slice management unit, parsing the mapping relationship between the real-time flow table throughput status and the isolation bandwidth quota, and outputting the remaining bandwidth of the network slice.

[0009] In a preferred embodiment, the specific process of performing binary constraint solving on the predicted node processing overhead, link throughput requirements, idle computing power pool, and remaining bandwidth of network slices to generate a service component offloading topology graph and a slice routing table is as follows: A resource scheduling objective function is constructed, setting the idle computing power pool as the upper limit constraint for computing power allocation and the remaining bandwidth of network slices as the upper limit constraint for transmission bandwidth; the predicted node processing overhead and link throughput requirements are input into the resource scheduling objective function for heuristic graph search optimization; a binary constraint solution space is constructed based on the upper limit constraints for computing power allocation and transmission bandwidth, outputting a sequence of candidate service nodes and a set of available transmission paths containing positive overflow attributes of computing power surplus; the candidate service node sequence is mapped into a directed acyclic graph structure according to multimedia service dependencies to generate a service component offloading topology graph; the next-hop network card media access control address and slice service level protocol identifier are extracted from the set of available transmission paths and arranged to generate a slice routing table.

[0010] In a preferred embodiment, the specific process of performing instance-level decoupling and dynamic migration of the video rendering microservice and transcoding microservice between the front-end acquisition node and the cloud broadcasting node, based on the business component unloading topology diagram, is as follows: Extract the identifier of the computing power unloading target node marked in the business component unloading topology diagram; truncate the underlying input data stream of the video rendering microservice and transcoding microservice within the front-end acquisition node, and capture the current runtime context state of the video rendering microservice and transcoding microservice at the truncation breakpoint; encapsulate the current runtime context state into a cross-node migration image, and trigger container orchestration and deployment instructions within the cloud broadcasting node corresponding to the computing power unloading target node identifier; load the cross-node migration image to reconstruct the microservice runtime environment, and open the input ports of the video rendering microservice and transcoding microservice to receive subsequent input data streams.

[0011] In a preferred embodiment, the process of allocating inter-node transmission channels based on the slice routing table and binding the dynamically migrated microservices with the transmission channels to form a dynamic collaborative service pipeline is as follows: Parse the virtual network function node sequence and slice service level configuration parameters contained in the slice routing table; call the software-defined network control plane interface to issue the underlying flow table resource reservation instruction for the end-to-end transmission channel according to the virtual network function node sequence and slice service level configuration parameters; extract the application programming interface communication port addresses of the dynamically migrated and reconstructed video rendering microservice and transcoding microservice; execute the socket binding protocol with the ingress gateway and egress gateway of the end-to-end transmission channel to assemble and generate a dynamic collaborative service pipeline for cross-end cloud nodes.

[0012] In a preferred embodiment, step S4 includes the following steps: injecting timestamp probes into the first and last network element endpoints of the dynamic collaborative service pipeline, parsing the clock synchronization difference characteristics of the timestamp probes of the first and last network element endpoints to generate the end-to-end total transmission delay, extracting the queuing waiting time within the end-to-end total transmission delay to generate the end-to-end queuing delay; comparing the end-to-end queuing delay with the reserved congestion boundary, triggering the adaptive degradation control logic inside the video rendering microservice; extracting the pixel coordinate boundaries associated with motion feature macroblocks, defining a motion focus mask layer within the current rendering frame of the multi-source heterogeneous live stream; maintaining the base frame rate and original resolution codec resource allocation for the pixel clusters within the motion focus mask layer, performing asynchronous frame skipping and discarding in the time dimension and downsampling interpolation in the spatial dimension on the background pixel clusters outside the motion focus mask layer; fusing the pixel clusters within the motion focus mask layer with the downsampling interpolated background pixel clusters, and outputting the dimensionality-reconstructed multi-source heterogeneous live stream.

[0013] A dynamic collaborative management and control system for live streaming processes in urban converged media scenarios is used to execute the aforementioned dynamic collaborative management and control method for live streaming processes in urban converged media scenarios. It includes: a load perception module, used to acquire multi-source heterogeneous live streaming streams from front-end acquisition nodes, extract motion feature macroblocks and bullet screen burst densities from the multi-source heterogeneous live streaming streams; perform feature quantization and concatenation of the motion feature macroblocks and bullet screen burst densities to construct a service load perception matrix, which includes predicted node processing overhead values ​​and link throughput requirements; and a collaborative scheduling module, used to read the service load perception matrix, poll the idle computing power pool and remaining bandwidth of the network slice within the access network's edge node group; and perform a two-stage scheduling process on the predicted node processing overhead values, link throughput requirements, idle computing power pool, and remaining bandwidth of the network slice. The system employs several mechanisms: a meta-constraint solution to generate a business component unloading topology map and a slice routing table; a pipeline orchestration module to perform instance-level decoupling and dynamic migration of video rendering and transcoding microservices between front-end acquisition nodes and cloud broadcasting nodes based on the business component unloading topology map; allocation of inter-node transmission channels based on the slice routing table to bind the dynamically migrated microservices to the transmission channels in a dynamic collaborative business pipeline; and a degradation compensation module to monitor the end-to-end queuing latency of the dynamic collaborative business pipeline. When the end-to-end queuing latency reaches the congestion limit, the system calls the video rendering microservice within the dynamic collaborative business pipeline, combines motion feature macroblocks to maintain the baseline frame rate for rendering the screen area covered by the motion feature macroblocks, and performs asynchronous frame extraction and downsampling processing on the uncovered areas.

[0014] The present invention has the following beneficial effects:

[0015] (1) A dynamic collaborative management and control method for live streaming processes in urban converged media scenarios is proposed. Motion feature macroblocks and bullet screen burst density are extracted and feature quantization is performed to construct a service load perception matrix containing node processing overhead prediction values ​​and link throughput demand values. Subsequently, the idle computing power pool of the edge node group and the remaining bandwidth of the network slice are polled. The demand prediction values ​​and resource reserves are solved by binary constraints to generate a service component unloading topology map and a slice routing forwarding table. As a result, the problem of insufficient adaptability of static resource scheduling to sudden changes in live streaming content in the existing technology can be improved. The service load demand is accurately matched with the remaining computing and network resources of the whole network in advance, thereby improving the foresight and rationality of resource scheduling.

[0016] (2) A dynamic collaborative management and control system for live streaming processes in urban converged media scenarios performs instance-level decoupling and dynamic migration of video rendering microservices and transcoding microservices between the front end and the cloud based on the business component unloading topology diagram, and allocates transmission channels according to the slice routing forwarding table, and binds them in series as a dynamic collaborative business pipeline; when the end-to-end queuing latency reaches the congestion limit, it combines motion feature macroblocks to maintain the baseline frame rate rendering for the covered area, and performs asynchronous frame extraction and downsampling processing for the uncovered area. As a result, it can break through the computing power bottleneck caused by the fixed video processing pipeline in the existing technology, and effectively smooth the data throughput when the network is congested through the cross-node dynamic migration of microservices and adaptive degradation rendering based on the region of interest, ensuring the continuous smoothness of multi-source heterogeneous live streams and the high-quality transmission of key images.

[0017] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0018] Figure 1 This is a flowchart of the dynamic collaborative management and control method for live streaming processes in urban converged media scenarios according to the present invention.

[0019] Figure 2 This diagram illustrates the spatial distribution and candidate node optimization mapping for solving binary constraints.

[0020] Figure 3 This is a schematic diagram of the end-to-end queuing delay evolution and degradation compensation response curve for dynamic collaborative business pipelines.

[0021] Figure 4 This is a flowchart of the live streaming process dynamic collaborative management and control system for urban converged media scenarios, as described in this invention. Detailed Implementation

[0022] This application provides a method and system for dynamic collaborative management and control of live streaming processes in urban converged media scenarios. This method and system solves the problems of resource scheduling lag, system component rigidity, and inflexible degradation strategies in traditional live streaming management mechanisms when dealing with sudden content loads, which lead to live streaming lag and damage to key image quality.

[0023] The overall approach of the scheme in this application embodiment is as follows: First, a service load perception matrix is ​​constructed by extracting the motion and interaction features of the live stream to predict the processing overhead and throughput requirements of each link; second, binary constraint solution is performed by combining edge computing power and the remaining state of network slices to output the component unloading topology and routing forwarding table; then, based on this topology and forwarding table, the video processing microservice is dynamically migrated at the instance level between the edge and the cloud and the service pipeline is reconstructed; finally, when the transmission latency exceeds the limit, the motion feature macroblocks from the previous stage are linked to implement differentiated downsampling rendering control on the video image, thereby forming a closed-loop dynamic collaborative management and control mechanism from load prediction, pipeline reconstruction to congestion degradation.

[0024] Example 1; please refer to Figure 1 This invention provides a technical solution: a dynamic collaborative management and control method for live streaming processes in urban converged media scenarios, comprising the following steps: S1, acquiring multi-source heterogeneous live streams from front-end acquisition nodes, extracting motion feature macroblocks and bullet screen burst densities from the multi-source heterogeneous live streams; performing feature quantization and splicing of the motion feature macroblocks and bullet screen burst densities to construct a service load perception matrix, which includes predicted node processing overhead and required link throughput; S2, reading the service load perception matrix, polling the idle computing power pool and remaining bandwidth of the network slice of the edge node group within the access network; performing binary constraint solving on the predicted node processing overhead, required link throughput, and the idle computing power pool and remaining bandwidth of the network slice. S3. Based on the business component unloading topology map, perform instance-level decoupling and dynamic migration of the video rendering microservice and transcoding microservice between the front-end acquisition node and the cloud broadcasting node; allocate transmission channels between nodes according to the slice routing table, and bind the dynamically migrated microservices and transmission channels in series to form a dynamic collaborative business pipeline; S4. Monitor the end-to-end queuing latency of the dynamic collaborative business pipeline. When the end-to-end queuing latency reaches the congestion limit, call the video rendering microservice in the dynamic collaborative business pipeline, combine motion feature macroblocks, maintain the baseline frame rate rendering for the screen area covered by the motion feature macroblocks, and perform asynchronous frame extraction and downsampling processing for the uncovered areas.

[0025] In this implementation plan, step S1 is used to pre-quantify the future demand for calculation and transmission resources at the live streaming source. In converged media scenarios, various devices such as drones, cameras, and portable backpacks are used for streaming. These devices generate video encoding protocols and resolutions that differ, collectively referred to as multi-source heterogeneous live streams. The system extracts regions in the image where drastic pixel changes occur, i.e., motion feature macroblocks, and simultaneously calculates the density of interactive text sent by viewers within a short period, i.e., bullet screen burst density. By combining the video image complexity and user interaction frequency, a two-dimensional data model, i.e., the business load perception matrix, is constructed. This model transforms the intensity of the live streaming content into predicted values ​​of the computational processing overhead that the underlying server will consume and the throughput demand that the uplink will bear.

[0026] Step S2 coordinates computing resources and transmission paths within the edge computing network and the bearer network based on predicted load demand. The system polls the edge node group comprised of base stations or data center servers near the field site to obtain unused CPU and GPU resources. Simultaneously, it checks the remaining available bandwidth of dedicated virtual network channels (network slices) logically isolated within the core network. Subsequently, the system uses the overhead and bandwidth requirements obtained in step S1 as constraints to perform an optimal match with the actual remaining computing power and slice bandwidth—a binary constraint solution. Based on the solution results, it outputs a service component offloading topology diagram indicating which software computing modules need to be moved from the front end to the back end, and a slice routing table indicating the underlying forwarding path of data packets.

[0027] Step S3 involves decomposing the software architecture and physically reorganizing the converged media business system according to a predetermined resource allocation scheme. Following the unloading topology diagram, the system independently separates the video rendering and transcoding modules, which were originally tightly bound to the front-end acquisition devices, into separate underlying code execution environments—an instance-level decoupling. These modules are then packaged and migrated in real-time to cloud-based broadcast nodes with ample computing power—a dynamic migration. After deployment, the system configures the underlying network channels between nodes based on the routing table, connecting the various microservice nodes deployed across domains to the network channels according to the business sequence. This ultimately maps and constructs a customized end-to-end data processing pipeline—a dynamic collaborative business pipeline.

[0028] Step S4 addresses unforeseen network congestion during the operation of the business pipeline, ensuring the continuity of the live stream through differentiated spatial processing. The system continuously monitors the total waiting time of data packets in the buffer queues of each node throughout the entire business pipeline, i.e., the end-to-end queuing latency. When this latency reaches a preset congestion threshold, the system immediately activates the motion feature macroblocks extracted in the first step to lock onto the core activity area in the image. For this core area, the original refresh rate and pixel clarity are maintained. For static background areas where no movement occurs, non-critical video frames in the time series are discarded, and the spatial pixel density is reduced—an asynchronous frame-skipping and downsampling process. This significantly compresses the data packet size while preserving the core visual content, alleviating pipeline queuing pressure.

[0029] Specifically, the process of acquiring multi-source heterogeneous live streams from front-end acquisition nodes and extracting motion feature macroblocks and bullet screen burst density from these streams is as follows: Parse the frame sequences contained within the multi-source heterogeneous live streams and separate the inter-frame predictive coding data contained in the frame sequences; extract the set of motion vectors carried by the inter-frame predictive coding data, and select the directional gradient anomaly macroblock group from the motion vector set to construct motion feature macroblocks; synchronously monitor the interactive bullet screen data channel bound to the multi-source heterogeneous live streams and construct a sliding time window based on time decay attributes; statistically analyze the slope of the text bullet screen arrival rate change within the sliding time window and convert the slope of the arrival rate change into bullet screen burst density.

[0030] In this implementation scheme, after separating the inter-frame predictive coding data, the coordinates within the f-th video frame are... Extract the motion vectors of macroblocks and calculate the macroblock's spatial neighborhood. The variance of the motion vector direction changes of adjacent macroblocks within the same macroblock is used to obtain the directional gradient outliers. The calculation formula is: Where f represents the sequence number of the video frame; u and v represent the horizontal and vertical coordinates of the target macroblock, respectively; Represents the motion vector of the target macroblock; The set of coordinate parameters representing the eight neighborhoods of space; This represents the total number of macroblocks in the neighborhood; a and b represent the horizontal and vertical coordinates of the neighboring macroblocks, respectively. This represents the motion vector of the corresponding neighboring macroblock. Greater than the dynamic judgment threshold Macroblocks are assembled into motion feature macroblocks. The dynamic determination threshold is precisely determined by taking the global arithmetic mean of the outlier values ​​of the directional gradients of all macroblocks within the first P consecutive video frames, where d is the threshold index and P represents the fixed statistical number of historical backtracking frames. When synchronously monitoring the interactive bullet screen data channel, the time length is set to... The sliding time window extracts the absolute arrival timestamps of each comment within the window. Calculate the current bullet screen arrival rate with time decay attribute. Its mathematical expression is ;in, This indicates the fixed span of the sliding time window, where w is a window attribute index; This indicates the trigger time of the k-th interactive comment within the window; This represents the weighted bullet comment reach rate of the current window, and c is the current status marker. q represents the total number of bullet comments collected within the current sliding time window, where q is the quantity subscript; k represents the integer index of the discrete bullet comments; exp represents the exponential function with the natural constant as the base. This represents the information attenuation coefficient, which is the reciprocal of the average total number of bullet comments per second for the current live streaming channel throughout the historical day. This represents the current sampling reference time of the system, where n is the time feature index. Subsequently, the historical bullet screen arrival rate of the previous sliding time window is obtained. Calculate the difference between the two and divide by the sliding time step to obtain the slope of the arrival rate change. Through the logical stith function relation ; the slope of the arrival rate change Convert to dimensionless bullet burst density Where p is the historical state marker index; This represents a continuous variable of slope, with m being the slope index; This represents the final extracted output of the bullet burst density characterization quantity.

[0031] Specifically, the process of constructing a service load perception matrix by quantizing and concatenating motion feature macroblocks and bullet screen burst density is as follows: Obtain the number of blocks and pixel update frequency of the motion feature macroblocks; address the nodes in the computing power overhead mapping table based on the number of blocks and pixel update frequency; output the predicted node processing overhead value; parse the concurrent connection request features associated with the bullet screen burst density; estimate the packet encapsulation frequency of the underlying network protocol stack based on the concurrent connection request features; output the link throughput requirement value; convert the predicted node processing overhead value and the link throughput requirement value into feature vectors according to a preset mapping dimension and perform tensor product operations to generate the service load perception matrix.

[0032] In this implementation scheme, the total number of motion feature macroblocks and the average pixel update frequency of the macroblocks are obtained. These parameters are then substituted into a pre-defined hardware complexity nonlinear regression model to output a predicted value for node processing overhead. The computational mapping logic is expressed as follows: ;in, This represents the aggregated value of the number of motion feature macroblocks, where z is the subscript for the quantity statistics. This indicates the extracted pixel update frequency, where h is the frequency attribute feature index; This represents the computational complexity ratio weighting coefficient, which is dynamically calculated and determined by dividing the current edge device's graphics processor's floating-point computing power rating by the pre-allocated thread pool capacity. This represents the basic computing power loss bias constant for system kernel scheduling; This represents the predicted node processing cost output, where x is the cost attribute index. Further, the concurrent connection request feature variable associated with the bullet burst density output from the above steps is analyzed. This allows for the estimation of packet encapsulation frequency in the underlying network protocol stack. The required link throughput value is calculated by combining the payload byte length. The specific theoretical calculation model is as follows: ;in, This represents the number of concurrent connection requests parsed from the sudden peak in bullet comments, where j is the connection state type index; This represents the packet encapsulation frequency prediction parameter of the network interface controller, where s is the packet action index; Indicates the overhead amplification factor of the underlying transport protocol; This represents a minimum smoothing constant established to prevent the logarithmic base from being zero or negative; This represents the average payload bytes of a single Transmission Control Protocol (TCP) message, where i is the payload specification index. This represents the retransmission redundancy compensation rate dynamically added considering the signal-to-noise ratio of the underlying physical link. This represents the final theoretical value of the link throughput requirement, with 'o' being the throughput feature index. Finally, the predicted node processing overhead and the link throughput requirement are mapped to a two-dimensional column vector according to a preset resource allocation sensitivity, and a computational resource feature column vector is independently constructed. Network transmission feature column vector Perform a tensor outer product operation on two independent column vectors to generate a service load awareness matrix that comprehensively represents the real-time coupling status of cloud network resources. The matrix structure calculation formula is: ;in, Represents the feature vector of the computational resource dimension; A feature vector representing the dimension of network resources; This represents the tensor product operator; A represents the two-dimensional load perception matrix obtained by the outer product operation, and A is the unique identifier of the matrix.

[0033] Specifically, the process of reading the service load perception matrix and polling the idle computing power pool and remaining bandwidth of the network slice of the edge node group within the access network is as follows: Send resource telemetry probes to the edge node group within the access network to collect the CPU load level and accelerator card memory fragmentation rate during the operation phase of the edge node group, and aggregate them to form an idle computing power pool; Connect to the northbound interface of the software-defined network controller to obtain the real-time flow table throughput status of the physical link between the edge node group and the cloud broadcast node; Extract the isolation bandwidth quota fed back by the network slice management unit, parse the mapping relationship between the real-time flow table throughput status and the isolation bandwidth quota, and output the remaining bandwidth of the network slice.

[0034] In this implementation scheme, the system distributes resource telemetry probes to the edge node cluster within the access network through a lightweight daemon deployed at the infrastructure layer. All underlying hardware status data generated during the collection process is hash-encrypted and anonymized at the operating system kernel layer, stripping away the hardware firmware serial numbers of the actual devices to prevent topology privacy leaks. For specific edge nodes, the probe penetrates the host operating system to obtain the CPU time sharding ratio during container runtime and calls the graphics card driver interface to extract the memory fragmentation ratio of the accelerator card's video memory. Combined with the factory-calibrated physical baseline floating-point computing power, the equivalent of the node's idle computing power pool is calculated using the following formula: Where g represents the unique index number of the edge node; This indicates the equivalent of the idle computing power pool output. This indicates the inherent physical reference floating-point computing power of the node; This indicates the proportion of available time in the central processing unit that has been collected; This indicates the fragmentation ratio of the accelerator card's video memory; This represents the computing power reduction penalty factor introduced due to temperature overload detected by the node motherboard sensors. Its determination method is to set it to zero when the real-time temperature is below a set safe operating threshold, and increase it linearly at a fixed slope when it exceeds the threshold. Simultaneously, the system connects to the software-defined network control plane to the northbound direction to obtain the baseline bandwidth quota for the specified end-to-end physical link logical isolation slice; within the defined sampling time window, it continuously accumulates the real-time throughput bytes fed back by the physical switch flow table entries to calculate the remaining bandwidth of the network slice, using the formula: ;in, This indicates the physical link identifier between the edge node and the cloud broadcasting node; Indicates the remaining bandwidth of the output network slice; This represents the logical isolation baseline bandwidth quota issued by the control plane; This represents the total distance between the sliding time window and the step length for the flow table status acquisition. Represents a single discrete-time sampled sequence within the window; Indicates at the sampling point The real-time throughput rate is obtained by parsing the flow table.

[0035] Specifically, the process of performing binary constraint solving on the predicted node processing overhead, link throughput requirements, idle computing power pool, and remaining bandwidth of network slices to generate a service component offloading topology graph and a slice routing table is as follows: A resource scheduling objective function is constructed, setting the idle computing power pool as the upper limit constraint for computing power allocation and the remaining bandwidth of network slices as the upper limit constraint for transmission bandwidth; the predicted node processing overhead and link throughput requirements are input into the resource scheduling objective function for heuristic graph search optimization; a binary constraint solution space is constructed based on the upper limit constraints for computing power allocation and transmission bandwidth, outputting a sequence of candidate service nodes containing positive overflow attributes of computing power surplus and a set of available transmission paths; the candidate service node sequence is mapped into a directed acyclic graph structure according to multimedia service dependencies, generating a service component offloading topology graph; the next-hop network card media access control address and slice service level protocol identifier are extracted from the set of available transmission paths, and a slice routing table is generated.

[0036] In this implementation plan, the system constructs a binary constraint space centered on latency and resource utilization, namely, the upper limit constraint of the mandatory computing power allocation is the equivalent of the idle computing power pool. The node processing overhead is greater than the predicted value of the business component, and the mandatory transmission bandwidth upper limit constraint is the remaining bandwidth of the network slice. The link throughput requirement exceeds the business flow's demand. Within this security constraint space, the system employs a heuristic graph search to optimize the mapping association between eligible physical nodes and links across the entire network, calculating the comprehensive fitness score for each candidate combination. The calculation formula is as follows: ;in, This represents the overall fitness evaluation score of the candidate node and its corresponding transmission link combination; This represents the weight for guiding the optimization of computing resources; The optimization guidance weights for network bandwidth are determined dynamically based on the current stage of the multimedia service pipeline. When the live stream is in the high-density encoding and compression stage, the weights are set accordingly. The value is greater than Set during the streaming and distribution phase The value is greater than And the sum of the two is always one; This represents the predicted processing overhead value of a specific business component node, as parsed from the business load perception matrix. This represents the parsed link throughput requirement; `max` represents the maximum value function to avoid division by zero. The system traverses the binary constraint space and filters out... The nodes with the highest scores and satisfying the positive overflow characteristic of computing power margin are formed into a candidate service node sequence, and the link with the highest score is extracted to generate a set of available transmission paths. If the optimization space is empty, an anti-paralysis fallback mechanism is automatically triggered, bypassing heuristic search and forcibly downgrading to a pre-configured cloud backbone disaster recovery cluster. Subsequently, based on the temporal dependencies of video acquisition, rendering, transcoding, and distribution, the candidate service node sequence is mapped into a directed acyclic graph structure. In this graph, vertices correspond to computing physical nodes, and directed edges correspond to business data flows, forming a business component offloading topology graph. Finally, the next-hop network interface media access control address and slice service level protocol identifier of the underlying software-defined switch are extracted hop-by-hop from the set of available transmission paths, and these two are encapsulated, mapped, and injected into the forwarding plane to generate a slice routing forwarding table. Figure 2 As shown in the figure, the three-dimensional coordinate system intuitively displays the spatial distribution of the edge node group under the constraints of the service load perception matrix and physical resources. The X-axis and Y-axis represent the predicted node processing overhead and the required link throughput, respectively, while the Z-axis represents the comprehensive fitness score obtained through heuristic graph search. The scattered points distributed in the high fitness interval (high segment of the Z-axis) represent the preferred service nodes that simultaneously satisfy the positive overflow of computing power margin and sufficient bandwidth. The system extracts the candidate node sequence based on the aggregation and optimization distribution pattern of the scattered points in this three-dimensional space, and then generates the service component offloading topology map.

[0037] Specifically, based on the business component unloading topology diagram, the specific process of performing instance-level decoupling and dynamic migration of the video rendering microservice and transcoding microservice between the front-end acquisition node and the cloud broadcasting node is as follows: Extract the target node identifier marked in the business component unloading topology diagram; truncate the underlying input data stream of the video rendering microservice and transcoding microservice within the front-end acquisition node, and capture the current runtime context state of the video rendering microservice and transcoding microservice at the truncation breakpoint; encapsulate the current runtime context state into a cross-node migration image, and trigger container orchestration and deployment instructions within the cloud broadcasting node corresponding to the target node identifier; load the cross-node migration image to reconstruct the microservice runtime environment, and open the input ports of the video rendering microservice and transcoding microservice to receive subsequent input data streams.

[0038] In this implementation, the system parses the target node identifier for computing power offloading issued in the business component offloading topology diagram and issues a suspension command to the container engine of the front-end acquisition node. The underlying daemon process truncates the input data stream buffers of the video rendering microservice and the transcoding microservice, and uses a checkpoint recovery mechanism to scan the container's memory space, capturing dirty pages, CPU register context, and active socket states at the time of the truncation breakpoint. When packaging the cross-node migration image, the system calculates the estimated microservice migration latency. Its calculation model is ;in, Indicates the expected microservice migration delay time; This represents the memory page compression ratio, which is determined by performing information entropy sampling on the video frame buffer data of the microservice that is currently in an active state, and obtaining the corresponding lossless compression ratio value by forward lookup table according to the information entropy dispersion. This represents the total number of dirty memory pages detected within the microservice container; The index number representing the traversal of the dirty page; Indicates the first The measured size in bytes of a dirty memory page; Indicates the register state serialization coefficients; The base byte size representing the runtime context register state; This indicates the available physical bandwidth of the current cross-platform cloud migration dedicated channel; This indicates the fixed startup time for the cloud-based production node corresponding to the target node of the computing power offloading to start and orchestrate the new container. The system pushes the packaged cross-node migration image to the cloud-based production node through a dedicated channel. The cloud node loads the image and allocates an independent network namespace to reconstruct the microservice runtime environment. Subsequently, it releases the program from its suspended state and opens the remapped input ports to ensure that the video stream data after the breakpoint is seamlessly connected to the migrated microservice for continued processing.

[0039] Specifically, the process of allocating inter-node transmission channels based on the slice routing table and binding the dynamically migrated microservices with the transmission channels to form a dynamic collaborative service pipeline is as follows: Parse the virtual network function node sequence and slice service level configuration parameters contained in the slice routing table; call the software-defined network control plane interface to issue the underlying flow table resource reservation instructions for the end-to-end transmission channel according to the virtual network function node sequence and slice service level configuration parameters; extract the application programming interface communication port addresses of the dynamically migrated and reconstructed video rendering microservice and transcoding microservice; execute the socket binding protocol with the ingress gateway and egress gateway of the end-to-end transmission channel to assemble and generate a dynamic collaborative service pipeline for cross-end cloud nodes.

[0040] In this implementation, the system reads the slice routing table, sequentially parses the virtual network function node sequence and slice service level configuration parameters, including the target bit rate and maximum tolerable latency, within the table. The system calculates the flow table matching priority of the underlying data plane. The calculation formula is: ;in, This indicates the priority value for the underlying switch to perform flow table matching and queuing forwarding. This represents the weighting coefficient for the bandwidth of the service level. These represent the penalty weight coefficients for route hop count decay. The method for determining these two weight coefficients is to linearly normalize the allocation based on the ratio of image quality requirements to latency sensitivity in the current live streaming scenario, ensuring that the sum of the two is constant at one. This represents the target bitrate extracted from the slice service level configuration parameters; This indicates the underlying physical base bit rate of the default routing channel; It represents exponential operations with the natural constant as the base; This indicates the total number of forwarding hops contained in the sequence of virtual network function nodes; Indicates the maximum topology addressing depth allowed by the physical architecture of a software-defined network; This represents the floor function. After calculation, the system calls the software-defined network control plane interface to match the flow table with priority. Along with the underlying flow table resource reservation instructions, the system sends them to each underlying physical switch. Subsequently, the system extracts the newly allocated application programming interface communication port addresses of the cloud video rendering microservice and transcoding microservice after dynamic migration and reconstruction. It then encapsulates the communication port addresses into IPv6-based segment routing extension headers using network address translation rules. Finally, it performs transmission control protocol socket binding actions at the ingress and egress gateways of the end-to-end transmission channel and uses the underlying tunneling mechanism to assemble and generate a dynamic collaborative business pipeline that runs through each microservice node.

[0041] Specifically, step S4 includes the following steps: injecting timestamp probes into the first and last network element endpoints of the dynamic collaborative service pipeline, parsing the clock synchronization difference characteristics of the timestamp probes of the first and last network element endpoints to generate the end-to-end total transmission delay, extracting the queuing waiting time within the end-to-end total transmission delay to generate the end-to-end queuing delay; comparing the end-to-end queuing delay with the reserved congestion boundary, triggering the adaptive degradation control logic within the video rendering microservice; extracting the pixel coordinate boundaries associated with motion feature macroblocks, defining a motion focus mask layer within the current rendering frame of the multi-source heterogeneous live stream; maintaining the base frame rate and original resolution codec resource allocation for the pixel clusters within the motion focus mask layer, performing asynchronous frame skipping and discarding in the time dimension and downsampling interpolation in the spatial dimension for the background pixel clusters outside the motion focus mask layer; fusing the pixel clusters within the motion focus mask layer with the downsampling interpolated background pixel clusters, and outputting the dimensionality-reduced and reconstructed multi-source heterogeneous live stream.

[0042] In this implementation scheme, at the physical network interface layer of the dynamic collaborative service pipeline, a precise time protocol is used to inject probe packets with hardware timestamps into the underlying data stream. The end-to-end queuing delay is generated by parsing the clock synchronization difference characteristics of the timestamp probes at the beginning and end network element endpoints; its mathematical expression is as follows: ;in, This represents the end-to-end queuing delay, where q is the queuing feature index. This indicates the absolute arrival timestamp at which the probe was captured at the media access control layer of the tail element; This indicates the absolute transmission timestamp injected by the probe into the first-end network element; This represents the master-slave clock synchronization offset measured through precise time protocol message exchange. This represents the theoretical reference propagation delay, derived by dividing the physical fiber optic link length of the pipeline by the propagation speed of the optical signal medium. This indicates the total number of microservice nodes included in the dynamic collaborative business pipeline; This represents the discrete traversal index of a microservice node; Indicates the probe at the The system will handle the fixed processing latency for each microservice node to reside within and complete protocol stack parsing. With reserved congestion limits Perform real-time comparison, when Greater than At this time, the adaptive degradation control logic within the video rendering microservice is triggered; this reserved congestion limit... The method for determining the threshold is as follows: extract hardware count samples of normal queuing latency in the edge network under no network congestion conditions over the past natural month, fit the cumulative probability distribution curve using a kernel density estimation function, and select the absolute latency value corresponding to a 95% confidence level as the determined boundary constant. After triggering the degradation control logic, the video rendering microservice immediately extracts the pixel coordinate boundaries associated with motion feature macroblocks. Within the luminance and chroma channels of the current rendering frame of the multi-source heterogeneous live stream, the coordinate boundaries are mapped to a binarized motion focus mask layer. For pixel clusters with a value of one within the motion focus mask layer, the system allocates an independent hardware encoder queue to maintain complete resource allocation between the baseline frame rate and the original resolution; for background pixel clusters with a value of zero outside the motion focus mask layer, joint downsampling suppression in both time and space dimensions is performed. The downsampling interpolation coefficients in the spatial dimension are defined as follows: Its calculation model is ;in, This indicates the spatial resolution preservation ratio when performing downsampling interpolation on the background pixel cluster; This refers to a hard minimum resolution retention constant set to avoid complete failure of background image resolution, serving as a fallback mechanism to prevent visual decoding failure. Represents the natural exponential function; This represents the spatial degradation penalty coefficient, which is determined by taking the natural logarithm of the ratio of the physical bandwidth of the bottleneck graphics processor to the current actual memory bandwidth utilization rate through real-time telemetry of the physical bandwidth of the edge node graphics processor. and These represent the real-time queuing latency and reserved congestion limit output from the aforementioned steps, respectively; `max` represents the maximum value operation function. Meanwhile, in the time dimension, the video rendering microservice... As an asynchronous frame skipping drop rate, non-reference frames are forcibly removed from consecutive background frames on the timeline, blocking non-critical frames from entering the subsequent compression pipeline. Finally, the microservice calls the on-chip system's video post-processing unit to perform memory copying and alignment merging of pixel clusters within the motion-focused mask layer (maintaining the original rendered image quality) and background pixel clusters (after asynchronous frame skipping and downsampling interpolation) under the same hardware timestamp reference, outputting a dimensionality-reduced, reconstructed multi-source heterogeneous live stream. For example... Figure 3 As shown, the solid line reflects the dynamic evolution of end-to-end queuing latency in the dynamic collaborative service pipeline during runtime, while the horizontal dashed line defines the congestion limit reserved by the system. In stages T1 to T5, with the increase in sudden traffic, queuing latency rises sharply and exceeds the congestion limit (as marked at the highest point). The system precisely triggers degradation control logic at the moment of exceeding the limit, performing asynchronous frame-by-frame downsampling processing on non-motion focus areas by calling the video rendering microservice, thus rapidly reducing the data packet size. As shown in stages T6 to T10, the system throughput pressure is effectively mitigated, and queuing latency quickly returns to a safe level, forming a complete self-closing control loop.

[0043] Example 2; please refer to Figure 4A dynamic collaborative management and control system for live streaming processes in urban converged media scenarios is provided to execute the dynamic collaborative management and control method for live streaming processes in urban converged media scenarios described in the embodiments. The system includes: a load perception module, used to acquire multi-source heterogeneous live streaming streams from front-end acquisition nodes, extract motion feature macroblocks and bullet screen burst densities from the multi-source heterogeneous live streaming streams; perform feature quantization and concatenation of the motion feature macroblocks and bullet screen burst densities to construct a service load perception matrix, which includes predicted node processing overhead values ​​and link throughput requirements; and a collaborative scheduling module, used to read the service load perception matrix, poll the idle computing power pool and remaining bandwidth of the network slices in the access network edge node group; and combine the predicted node processing overhead values, link throughput requirements, idle computing power pools, and remaining bandwidth of the network slices. The system performs binary constraint solving to generate a business component unloading topology graph and a slice routing table. The pipeline orchestration module, based on the business component unloading topology graph, performs instance-level decoupling and dynamic migration of video rendering and transcoding microservices between front-end acquisition nodes and cloud-based broadcasting nodes. It allocates inter-node transmission channels according to the slice routing table and binds the dynamically migrated microservices to these channels as a dynamic collaborative business pipeline. The degradation compensation module monitors the end-to-end queuing latency of the dynamic collaborative business pipeline. When the end-to-end queuing latency reaches the congestion limit, it calls the video rendering microservice within the dynamic collaborative business pipeline. Combining motion feature macroblocks, it maintains the baseline frame rate for rendering the image area covered by the motion feature macroblocks and performs asynchronous frame-skipping and downsampling processing on the uncovered areas.

[0044] In this implementation scheme, the load perception module serves as the system's input and analysis front-end unit, responsible for performing preliminary quantitative calculations of service pressure at the live streaming front-end. This module directly accesses the data link of multi-source heterogeneous live streams, parallelly extracting inter-frame data and interactive channel text. By calculating the directional anomaly gradient of video frame macroblocks and statistically analyzing the slope of bullet screen arrival changes within a time window, it obtains motion feature macroblocks and bullet screen burst density. Subsequently, the module performs quantization and concatenation of the extracted spatial complexity parameters and network concurrent connection parameters through tensor product operations, ultimately outputting a service load perception matrix containing predicted node processing overhead and link throughput requirements. This provides benchmark data input for subsequent system resource scheduling, encompassing both computational overhead and network bandwidth perspectives.

[0045] The collaborative scheduling module, acting as the central hub for global resource management and optimization, receives baseline data from the load awareness module and performs supply-demand matching calculations. This module first sends telemetry probes to the underlying hardware and connects to the northbound network interface, summarizing in real-time the equivalent idle computing power of the edge node clusters within the access network and the remaining bandwidth of the network slices of physical links. After obtaining the global resource view, the module uses the idle computing power and remaining bandwidth as mandatory upper limit constraints, substitutes the demand values ​​from the service load awareness matrix into the resource scheduling objective function, and performs a heuristic graph search. Through solving binary constraints, it filters out node sequences and link paths whose comprehensive fitness satisfies positive overflow, thereby generating a service component offloading topology map and slice routing forwarding table to guide the underlying system reconfiguration.

[0046] The pipeline orchestration module is responsible for performing physical-level cross-domain reorganization of the converged media business system architecture based on the strategy file output by the scheduling center. This module reads the business component unloading topology diagram and, by truncating the front-end data flow and encapsulating the runtime context image, decouples the video rendering microservice and transcoding microservice from the underlying front-end acquisition node, and hot-deploys and migrates them to the cloud-based broadcasting node with ample computing power. Simultaneously, this module parses the underlying flow table and service level parameters within the slice routing forwarding table, calls the control plane interface to issue physical resource reservation instructions, and binds the dynamically migrated and reconstructed microservice communication ports to the gateway execution socket of the end-to-end transmission channel. This connects and maps a complete dynamic collaborative business pipeline between physically dispersed end-cloud network elements.

[0047] As a fault-tolerant closed-loop unit ensuring the continuity of the live stream, the degradation compensation module continuously performs status monitoring and low-level rendering intervention during the operation of the dynamically coordinated business pipeline. This module injects timestamp probe messages into the first and last network element endpoints of the business pipeline to calculate the end-to-end queuing latency and strictly compares and monitors it against the normal congestion threshold. When the queuing latency touches the threshold and triggers a congestion alarm, this module directly takes over the internal control logic of the video rendering microservice, extracts the motion feature macroblock coordinates generated by the load perception module in the early stage, and defines the focus mask layer in the current frame; then, it maintains the original resource allocation for the core area within the mask layer, and forces asynchronous frame skipping and spatial downsampling interpolation to be performed on the external background area, and finally merges and outputs the dimensionality-reconstructed live stream to instantly smooth the throughput pressure of the pipeline.

[0048] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0049] This invention is described with reference to flowchart illustrations and / or block diagrams of systems, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0050] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0051] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0052] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0053] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for dynamic coordination and control of a live broadcast process for a city media integration scene, characterized in that, Includes the following steps: S1. Obtain multi-source heterogeneous live streams from the front-end acquisition nodes, and extract motion feature macroblocks and bullet screen burst densities from the multi-source heterogeneous live streams; perform feature quantization and splicing on the motion feature macroblocks and bullet screen burst densities to construct a business load perception matrix, which includes node processing overhead prediction values ​​and link throughput requirements. S2. Read the service load awareness matrix and poll the idle computing power pool and remaining bandwidth of the network slice of the edge node group within the access network. Perform binary constraint solving on the predicted node processing overhead, link throughput requirements, idle computing power pool, and remaining bandwidth of network slices to generate a service component offloading topology map and slice routing forwarding table. S3. Based on the business component unloading topology diagram, perform instance-level decoupling and dynamic migration of the video rendering microservice and transcoding microservice between the front-end acquisition node and the cloud broadcasting node. Based on the slice routing table, allocate the transmission channel between nodes, and bind the dynamically migrated microservices and the transmission channel together to form a dynamic collaborative business pipeline; S4. Monitor the end-to-end queuing latency of the dynamic collaborative service pipeline. When the end-to-end queuing latency reaches the congestion limit, call the video rendering microservice in the dynamic collaborative service pipeline. Combine motion feature macroblocks to maintain the baseline frame rate rendering for the screen area covered by the motion feature macroblocks, and perform asynchronous frame downsampling processing for the uncovered area.

2. The live streaming process dynamic coordination and management method for the urban media convergence scene according to claim 1, characterized in that: The specific process of acquiring multi-source heterogeneous live streams from front-end acquisition nodes and extracting motion feature macroblocks and bullet screen burst density from these streams is as follows: Analyze the frame sequences contained in multi-source heterogeneous live streams and separate the inter-frame predictive coding data contained in the frame sequences; Extract the set of motion vectors carried by the inter-frame predictive coding data, and select macroblocks with directional gradient anomalies from the set of motion vectors to construct motion feature macroblocks; Simultaneously monitor the interactive bullet screen data channels bound to multi-source heterogeneous live streams and construct a sliding time window based on time decay attributes; The slope of the text bullet comment arrival rate within the sliding time window is statistically analyzed and converted into bullet comment burst density.

3. The live streaming process dynamic coordination and management method for the urban media scene according to claim 1 or 2, characterized in that: The specific process of constructing the service load perception matrix by quantizing and concatenating motion feature macroblocks and bullet screen burst densities is as follows: Obtain the number of blocks and pixel update frequency of the motion feature macroblock, address it in the computing power overhead mapping table based on the number of blocks and pixel update frequency, and output the node processing overhead prediction value. Analyze the concurrent connection request characteristics related to the burst density of bullet comments, estimate the packet encapsulation frequency of the underlying network protocol stack based on the concurrent connection request characteristics, and output the required link throughput value. The predicted node processing overhead and link throughput requirements are transformed into feature vectors according to a preset mapping dimension, and tensor product operations are performed to generate a service load awareness matrix.

4. The live streaming process dynamic coordination and management method for the urban media scene, according to claim 1, characterized in that: The specific process of reading the service load awareness matrix and polling the idle computing power pool and remaining bandwidth of the network slices of the edge node group within the access network is as follows: Send resource telemetry probes to the edge node group within the access network to collect the CPU load level and accelerator card memory fragmentation rate during the operation of the edge node group, and aggregate them to form an idle computing power pool. Interact with the northbound interface of the software-defined network controller to obtain the real-time flow table throughput status of the physical link between the edge node group and the cloud broadcast node; Extract the isolation bandwidth quota fed back by the network slice management unit, parse the mapping relationship between the real-time flow table throughput status and the isolation bandwidth quota, and output the remaining bandwidth of the network slice.

5. The live streaming process dynamic coordination and management method and system for the urban media scene, according to claim 1, characterized in that: The specific process of performing binary constraint solving on the predicted node processing overhead, link throughput requirements, idle computing power pool, and remaining bandwidth of network slices to generate the service component offloading topology map and slice routing table is as follows: Construct a resource scheduling objective function, set the idle computing power pool as the upper limit constraint for computing power allocation, and set the remaining bandwidth of the network slice as the upper limit constraint for transmission bandwidth. The predicted node processing overhead and the required link throughput are input into the resource scheduling objective function for heuristic graph search optimization. A binary constraint solution space is constructed based on the upper limit constraints of computing power allocation and the upper limit constraints of transmission bandwidth. The output includes a sequence of candidate service nodes and a set of available transmission paths containing positive overflow attributes of computing power margin. The candidate service node sequence is mapped into a directed acyclic graph structure according to the multimedia business dependency relationship, and a business component unloading topology graph is generated. Extract the next-hop network interface card's media access control address and slice service level protocol identifier from the set of available transmission paths, and arrange them to generate a slice routing forwarding table.

6. The live streaming process dynamic coordination and management method for the urban media scene according to claim 1, characterized in that: Based on the business component unloading topology diagram, the specific process of performing instance-level decoupling and dynamic migration of the video rendering microservice and transcoding microservice between the front-end acquisition node and the cloud broadcasting node is as follows: Extract the identifiers of the target nodes for unloading computing power from the unloading topology diagram of the business components; Cut off the underlying input data stream of the video rendering microservice and transcoding microservice within the front-end acquisition node, and capture the current runtime context state of the video rendering microservice and transcoding microservice at the cutoff point; The current runtime context state is encapsulated as a cross-node migration image, and container orchestration and deployment instructions are triggered in the cloud broadcast node corresponding to the target node identifier of the computing power unloading target node. Load the cross-node migration image to reconstruct the microservice runtime environment, and open the input ports of the video rendering microservice and transcoding microservice to receive subsequent input data streams.

7. The live streaming process dynamic coordination and management method for the urban media scene according to claim 1, characterized in that: The process of allocating inter-node transmission channels based on the slice routing table and binding the dynamically migrated microservices with the transmission channels to form a dynamic collaborative business pipeline is as follows: Parse the sequence of virtual network function nodes and the slice service level configuration parameters contained in the slice routing table; The software-defined network control plane interface is invoked to issue underlying flow table resource reservation instructions for the end-to-end transmission channel according to the virtual network function node sequence and slice service level configuration parameters; Extract the application programming interface communication port addresses of the video rendering microservice and the transcoding microservice after dynamic migration and reconstruction; By executing a socket binding protocol between the application programming interface communication port address and the ingress and egress gateways of the end-to-end transmission channel, a dynamic collaborative service pipeline for cross-end cloud nodes is generated. 8.The method and system for dynamic coordination and management of live streaming process of urban media convergence scene according to claim 1, characterized in that: Step S4 includes the following steps: Inject timestamp probes into the first and last network element endpoints of the dynamic collaborative service pipeline, analyze the clock synchronization difference characteristics of the timestamp probes at the first and last network element endpoints to generate the total end-to-end transmission delay, and extract the queuing waiting time within the total end-to-end transmission delay to generate the end-to-end queuing delay. By comparing the end-to-end queuing latency with the reserved congestion limit, the adaptive degradation control logic within the video rendering microservice is triggered. Extract the pixel coordinate boundaries associated with motion feature macroblocks and define a motion focus mask layer within the current rendering frame of the multi-source heterogeneous live stream; For pixel clusters within the motion focus mask layer, maintain the base frame rate and original resolution codec resource allocation; for background pixel clusters outside the motion focus mask layer, perform asynchronous frame skipping and dropping in the temporal dimension and downsampling interpolation in the spatial dimension. The pixel clusters within the motion-focused mask layer are merged with the background pixel clusters after downsampling and interpolation, and the multi-source heterogeneous live stream is output after dimensionality reduction and reconstruction.

9. The live broadcast process dynamic collaborative management system for the urban media convergence scene, used for executing the live broadcast process dynamic collaborative management method for the urban media convergence scene according to any one of claims 1-8, characterized in that, include: The load perception module is used to acquire multi-source heterogeneous live streams from front-end acquisition nodes, extract motion feature macroblocks and bullet screen burst density from the multi-source heterogeneous live streams, perform feature quantization and splicing of motion feature macroblocks and bullet screen burst density, and construct a business load perception matrix. The business load perception matrix includes the predicted value of node processing overhead and the value of link throughput requirement. The collaborative scheduling module is used to read the service load awareness matrix and poll the idle computing power pool and remaining bandwidth of the network slice of the edge node group within the access network. Perform binary constraint solving on the predicted node processing overhead, link throughput requirements, idle computing power pool, and remaining bandwidth of network slices to generate a service component offloading topology map and slice routing forwarding table. The pipeline orchestration module is used to perform instance-level decoupling and dynamic migration of video rendering microservices and transcoding microservices between the front-end acquisition nodes and the cloud broadcasting nodes, based on the business component unloading topology diagram. Based on the slice routing table, allocate the transmission channel between nodes, and bind the dynamically migrated microservices and the transmission channel together to form a dynamic collaborative business pipeline; The degradation compensation module is used to monitor the end-to-end queuing latency of the dynamic collaborative service pipeline. When the end-to-end queuing latency reaches the congestion limit, it calls the video rendering microservice in the dynamic collaborative service pipeline. Combined with motion feature macroblocks, it maintains the baseline frame rate rendering for the screen area covered by the motion feature macroblocks and performs asynchronous frame extraction and downsampling processing for the uncovered area.

Images