Intelligent network control method for low-delay video return and related equipment

Abstract

The invention relates to the field of multimedia communication and network control, in particular to an intelligent network control method for low-delay video return and related equipment. The intelligent network control method comprises the following steps: acquiring network key indexes including bandwidth, delay, jitter and packet loss rate of a network link in real time, and providing real-time network environment data support for transmission strategy adjustment. According to the method, an intelligent control mechanism combining network state perception and video content feature recognition is constructed, so that the technical problem of low-delay video return in a complex network environment is effectively solved. Specifically, key indexes of a network link are collected in real time, and a lightweight CNN model is introduced to analyze the video content activeness, so that dual perception capabilities for a network environment and content features are formed, data support is provided for dynamic adjustment of coding parameters, and accurate balance between video quality and network adaptability is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of multimedia communication and network control, in particular to an intelligent network control method for low-latency video backhaul and related equipment. BACKGROUND

[0002] With the popularization of high-definition video acquisition devices and the growth of remote control demand for unmanned vehicles, unmanned aerial vehicles, robots and the like, video transmission in complex mobile network environments faces multiple challenges. Traditional video backhaul schemes rely on fixed bandwidth allocation strategies and are difficult to dynamically adapt to network congestion, bandwidth fluctuations and resource competition problems in multi-path video concurrent transmission, resulting in excessively high end-to-end delay and insufficient video smoothness, which seriously affects the real-time performance of remote control. In the prior art, mainstream transmission protocols such as RTMP and RTSP lack deep perception of network state and are unable to dynamically adjust encoding parameters and transmission priorities in combination with video content features, resulting in a technical gap in the cooperative optimization of source encoding and channel transmission. In addition, in the face of different network failure modes such as random packet loss and continuous packet loss, traditional anti-packet loss mechanisms are difficult to achieve intelligent balance between bandwidth overhead and recovery capability, further exacerbating the stability problem of video transmission. Therefore, the technical problem to be solved by the present application is: how to construct an end-to-end optimization mechanism that integrates network state perception, video content feature recognition and intelligent resource scheduling in a complex and dynamic network environment, to realize low-latency and high-reliability transmission of multi-path video streams, while also taking into account encoding efficiency and network adaptability. SUMMARY

[0003] The present disclosure proposes an intelligent network control method for low-latency video backhaul and related equipment, aiming to overcome at least one of the deficiencies in the prior art.

[0004] To achieve the above-mentioned purpose, the technical solutions disclosed by the present application are as follows: According to one aspect of the present disclosure, an intelligent network control method for low-latency video backhaul is provided, comprising the following steps: Real-time acquisition of network link key indicators including bandwidth, delay, jitter and packet loss rate, to provide real-time network environment data support for transmission strategy adjustment; Extraction of inter-frame motion vectors of video frames through a lightweight CNN model to determine video content activity and evaluate the frequency and amplitude of content changes between adjacent frames, to provide content feature basis for encoding strategy optimization; Dynamic adjustment of video encoding parameters, including resolution, code rate and frame rate, based on the network key indicators and video content activity, to balance video quality and network adaptability; According to the priority and resource occupation of multi-path video streams, network bandwidth resources are allocated and transmission parameters of each video stream are controlled to ensure the efficiency of multi-path video transmission; According to the real-time network environment, the packet loss type is intelligently identified, the FEC redundancy parameter is dynamically adjusted, and redundant data is added in network transmission to resist packet loss problems, so as to balance the packet loss recovery capability and bandwidth overhead.

[0005] Further, the network key indicator collection step includes: The uplink available bandwidth is estimated by sending a probe packet and analyzing the response time, the bandwidth change trend is tracked in real time to identify network congestion, and the peak bandwidth in a unit time is counted; The round-trip delay and one-way transmission delay of the data packet are measured, the stability of the delay jitter is analyzed to determine the size of the decoding end buffer, and the size of the decoding end buffer is determined. The instantaneous packet loss rate in a short time window and the average packet loss rate in a long time period are counted, the mode of continuous packet loss or random packet loss is identified, and the basis for adjusting the FEC redundancy parameter is provided.

[0006] Further, the video inter-frame motion vector extraction step includes: The calculation complexity and resource consumption of the lightweight CNN model are reduced for real-time video analysis scenes; Based on the pixel difference and motion trajectory of adjacent video frames, the motion vector feature representing content change is extracted, the dynamic degree of video content is quantified, and the video content activity evaluation result is generated to drive the dynamic adjustment of the encoding parameter.

[0007] Further, the step of dynamically adjusting the video encoding parameter includes: When the network bandwidth is sufficient, high code rate and high resolution encoding are used to improve the video detail quality, and when the network bandwidth is limited, the code rate and resolution are automatically reduced to ensure transmission continuity; For high dynamic video content, the frame rate and resolution are increased to retain motion details, and for static or low dynamic content, the frame rate and code rate are reduced to reduce data volume; When the network is congested or the bandwidth drops sharply, a fast code rate adjustment mechanism is triggered, and key I frames are preferentially transmitted to ensure continuous display at the video decoding end.

[0008] Further, the step of allocating network bandwidth resources includes: A video stream priority evaluation model is established, and priority is allocated according to the service type and content importance of the video stream; Based on the real-time network bandwidth and the priority of each video stream, scheduling instructions are issued to the network transmission layer through a standard interface, the transmission bandwidth quota and transmission priority of each video stream are dynamically adjusted, and resource contention of multiple video streams is avoided.

[0009] Further, the step of dynamically adjusting the FEC redundancy parameter includes: When the network packet loss rate is detected to be high and presents a continuous packet loss mode, the FEC redundancy is increased to improve the packet loss recovery capability, and when the network packet loss rate is low and presents a random packet loss, the FEC redundancy is reduced to reduce the bandwidth overhead; According to the delay and jitter characteristics of the network link, the block size and redundancy ratio of the FEC encoding are dynamically adjusted to balance the anti-packet loss performance and the encoding processing delay, so as to ensure the real-time transmission of the video.

[0010] Furthermore, the network control method is optimized jointly by the source and the channel for low-delay transmission: the adaptive encoding control module is used on the source side to reduce the redundant data, the FEC anti-packet loss module and the intelligent scheduling decision module are used on the channel side to improve the transmission reliability and the bandwidth utilization, and the feedback strategy is combined to form the dynamic optimization of the end-to-end.

[0011] According to another aspect of the present disclosure, there is provided an intelligent network control system for low-delay video backhaul, which is used to implement the intelligent network control method for low-delay video backhaul as described above, and comprises: A network state detection module is configured to collect network key indicators in real time to provide data support for transmission strategy adjustment, wherein the network key indicators include the bandwidth, delay, jitter and packet loss rate of the network link. A video content analysis module is configured to extract the motion vector between video frames by using a lightweight CNN model, judge the activity of the video content, and output a content feature evaluation result. An adaptive encoding control module is configured to receive the network key indicators and the content feature evaluation result, dynamically adjust the video encoding parameters, and balance the video quality and network adaptability. An intelligent scheduling decision module is configured to allocate the network bandwidth resources according to the priority and resource occupation of the multiple video streams, and control the transmission parameters of each video stream. A feedback adjustment mechanism module is configured to periodically collect the module running data, evaluate the system performance, and update the control strategy. An FEC anti-packet loss module is configured to intelligently identify the packet loss type according to the real-time network condition, dynamically adjust the FEC redundancy parameters, and add redundant data in the network transmission to resist the packet loss problem.

[0012] According to still another aspect of the present disclosure, there is provided a video conference device, which comprises: An encoder device is disposed at the video acquisition end and is configured to perform image acquisition and preprocessing. A decoder device is disposed at the video receiving end and is configured to perform network data decoding and image display. The intelligent network control system for low-delay video backhaul as described above is used to connect the encoder device and the decoder device, and implement the low-delay backhaul control of the video stream.

[0013] According to another aspect of the present disclosure, a computer readable storage medium is provided, having stored thereon a computer program which, when executed by a processor, implements the steps of the intelligent network control method for low-latency video backhaul as described above.

[0014] The present application has the following advantages: The present application effectively solves the technical problem of low-latency video backhaul in complex network environment by constructing an intelligent control mechanism combining network state perception and video content feature recognition. Specifically, by real-time collection of key indicators of network link and introduction of a lightweight CNN model to analyze video content activity, a dual perception capability of network environment and content feature is formed, providing data support for dynamic adjustment of encoding parameters, and realizing accurate balance of video quality and network adaptability.

[0015] Further, the intelligent scheduling decision module allocates bandwidth resources of multiple video streams based on a priority evaluation model, avoiding performance degradation caused by resource contention and improving overall efficiency of multi-path transmission; the FEC anti-packet loss module dynamically adjusts the redundancy strategy according to real-time packet loss type, optimizing bandwidth utilization while ensuring transmission reliability. The feedback regulation mechanism forms a closed-loop control through periodic strategy update, continuously adapting to network environment changes. The adaptive encoding control on the source side and the intelligent scheduling and anti-packet loss technology on the channel side are jointly optimized, significantly reducing end-to-end delay and improving video smoothness, especially in network fluctuation scenarios.

[0016] Further, the present application breaks through the limitations of traditional fixed strategies through the intelligent control system of multiple modules in cooperation, providing a comprehensive solution for video backhaul in remote control scenarios with real-time, reliability and resource efficiency, effectively meeting the stringent demand of low-latency video transmission for remote control of unmanned devices.

[0017] The above description is only a summary of the technical solutions of the present application. In order to more clearly understand the technical means of the present application and can be implemented according to the content of the specification, the following describes the preferred embodiments of the present application with reference to the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 Flow chart of low-latency video backhaul control method in an embodiment of the present application; Figure 2 Principle diagram of low-latency video backhaul control method in an embodiment of the present application; Figure 3 Spatiotemporal evolution diagram of network link key indicators in an embodiment of the present application; Figure 4 Video inter-frame motion vector field diagram in an embodiment of the present application; Figure 5A coding parameter adaptive decision curve diagram in an embodiment of the present application; Figure 6 An FEC redundancy strategy performance analysis diagram in an embodiment of the present application; Figure 7 A multi-path video stream intelligent bandwidth scheduling diagram in an embodiment of the present application; Figure 8 An architecture diagram of a low-latency video backhaul control system in an embodiment of the present application. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application.

[0020] The present application provides the following preferred embodiments: Embodiment one: in order to solve the problem of insufficient adaptation of network environment dynamic change and video content characteristics in low-latency video backhaul, the present embodiment provides an intelligent network control method for low-latency video backhaul, which realizes the optimization of transmission efficiency and quality through the collaborative perception and dynamic regulation of network state and video content. As shown in Figure 1 and Figure 2 , the flow of the control method is as follows: S100: real-time collection of network link key indicators including bandwidth, delay, jitter and packet loss rate, to provide real-time network environment data support for transmission strategy adjustment.

[0021] S200: extraction of video frame motion vectors through a lightweight CNN model, judgment of video content activity, and evaluation of the frequency and amplitude of content change between adjacent frames, to provide content feature basis for encoding strategy optimization.

[0022] S300: dynamic adjustment of video encoding parameters including resolution, code rate and frame rate based on network key indicators and video content activity, to balance video quality and network adaptability.

[0023] S400: overall allocation of network bandwidth resources according to the priority and resource occupation of multi-path video streams, control of the transmission parameters of each video stream, to ensure the efficiency of multi-path video transmission.

[0024] S500: intelligent identification of packet loss type according to real-time network environment, dynamic adjustment of FEC redundancy parameters, and addition of redundant data in network transmission to combat packet loss problems, to balance packet loss recovery capability and bandwidth overhead.

[0025] Further, in the network state perception link, multi-dimensional indicators such as bandwidth, delay, jitter and packet loss rate of the network link are continuously collected. As shown inFigure 3 As shown, the time series sliding window records the bandwidth fluctuation characteristics, and identifies the network congestion event at a specific period such as t=28-35 seconds, which provides support for policy decision through real-time quantitative spatio-temporal evolution data. It needs to be understood that the delay heat map also reflects the stability characteristics of the network path, which plays a fundamental role in subsequent transmission parameter control.

[0026] Further, the motion vector between video frames is extracted by a lightweight CNN model. As shown, Figure 4 Based on the streamline diagram, the pixel displacement vector field is visualized, and the red motion trajectory marks the motion characteristics of the fast-moving object. The spatial distribution uniformity of the motion vector, such as the contrast between sparse and dense areas, and its instantaneous intensity are represented by color mapping, which together constitute the quantitative basis for the activity of video content, providing a basis for content dynamic characteristic analysis for encoding optimization.

[0027] Further, in the dynamic adjustment of encoding parameters, the network state and content activity collected in advance are fused for decision-making. As shown in the three-dimensional decision surface, Figure 5 The available bandwidth (X-axis) and content activity (Y-axis) are used as input variables to establish a nonlinear mapping relationship to the target code rate (Z-axis). In the implementation process, the encoding parameter combination is dynamically selected along the black optimization path, for example, when the bandwidth is limited (Bw=2Mbps), the resolution is preferentially reduced rather than the frame rate, so as to maintain the recognizability of key motion information.

[0028] Further, for multi-path video stream transmission, bandwidth resource scheduling is implemented according to the preset service priority. As shown in the hierarchical filling diagram, Figure 7 Under the condition of total bandwidth constraint, the high-priority stream is implemented to guarantee allocation, and the medium and low-priority streams are implemented to perform elastic compression. The congestion event at a specific time such as t=28 seconds triggers the dynamic resource reallocation mechanism, which temporarily limits the bandwidth of low-priority streams to ensure the transmission continuity of key video streams.

[0029] Further, in the packet loss control strategy, according to the real-time network monitoring data, random packet loss and continuous packet loss types are distinguished. As shown in the twin graph analysis, Figure 6 For random packet loss scenarios, a linear increase in FEC redundancy (5-30%) is adopted, while for burst continuous packet loss, an exponential redundancy strategy is adopted, such as 25% redundancy when the packet loss length is >8. This strategy establishes a packet loss type identification module in the control system kernel, dynamically generates redundancy parameters that match the current network damage characteristics.

[0030] This embodiment forms a multi-dimensional collaborative dynamic adjustment mechanism by establishing a control link of network state-content characteristics-transmission strategy, which improves the network resource utilization rate on the premise of guaranteeing the basic quality of video.

[0031] Embodiment two: In order to solve the problem of transmission parameter adaptation deviation caused by insufficient network state monitoring precision in video backhaul, the embodiment further optimizes the collection process of network key indicators, and constructs an accurate network environment model through multi-dimensional indicator collaborative analysis. The following will be described in detail in combination with the drawings: Specifically, in the bandwidth monitoring stage, the uplink available bandwidth is estimated by periodically sending probe packets and analyzing the response time. As shown in Figure 3 , the time series sliding window (X axis is time) records the bandwidth fluctuation, in which the bandwidth curve and the congestion event mark capture the real-time change trend. After identifying the characteristics of the sudden drop in bandwidth in a specific period (t=28-35 seconds), the peak bandwidth in the unit time window is counted to form a dynamic view of bandwidth capacity. It should be understood that this process indirectly verifies the estimation accuracy by calculating RTT x BDP (bandwidth delay product), and provides the space-time evolution basis for congestion identification.

[0032] Further, in the delay measurement link, the round-trip delay and one-way transmission delay values of the data packets are synchronously collected. As shown in Figure 3 , the color depth of each data point in the delay heat map (Y axis is delay value) maps the jitter variance j 2 , and the spatial distribution of low jitter area and high jitter area reflects the network path stability. By analyzing the statistical characteristics of the heat map, such as the change of jitter standard deviation with time, the adjustment threshold of the decoding end buffer size is determined. It can be understood that the buffer configuration needs to match the delay jitter stability curve to avoid playback jitter.

[0033] Further, for packet loss analysis, a double-time scale statistical model is set: the short-term window (second level) calculates the instantaneous packet loss rate loss , and the long-term window (minute level) analyzes the average packet loss rate. As shown in Figure 3 , the bottom subgraph, the packet loss rate histogram separates random packet loss and continuous packet loss, and identifies the pattern characteristics based on burst packet loss events such as continuous packet loss pulse at t=68-73 seconds. The pattern classification result is directly output to the subsequent FEC parameter controller to provide network damage type input for redundancy strategy.

[0034] The benefits of this embodiment are: through the multi-dimensional indicator space-time evolution collaborative collection mechanism as shown in Figure 3 , efficient modeling of network state is realized. The coupling analysis of bandwidth, delay and packet loss pattern provides a joint input space for transmission decision, avoiding strategy adaptation deviation caused by single indicator distortion.

[0035] Embodiment three: To solve the problem of excessive consumption of computing resources leading to increased delay in real-time video analysis, this embodiment further optimizes the extraction process of the motion vector between video frames, and realizes accurate quantification of the dynamic characteristics of video content through a lightweight neural network model, so as to adapt the intelligent adjustment of the coding parameters.

[0036] To solve the problem of computing complexity and resource consumption, this embodiment adopts a lightweight CNN model, reduces the parameter size through model compression techniques such as pruning and quantization, and makes it suitable for resource-constrained embedded devices. As shown in Figure 4 , the motion vector field of the video frame is obtained by feature extraction on the pixel difference of adjacent frames, and the streamline diagram is generated based on the color mapping motion intensity calculated from the u, v components of the x and y grid points. It should be understood that this model uses the pixel gradient information of the input frame to construct a motion trajectory diagram, such as the red curve in Figure 4 , by analyzing the continuity of the trajectory to capture the object motion pattern, thereby reducing the computational overhead of real-time inference.

[0037] Further, the quantification process of the motion vector feature includes deriving the video content activity evaluation result from the vector field. As shown in Figure 4 , the comparison of the fast moving object trajectory and the motion intensity of other areas generates a normalized activity index Av. This index is calculated by statistical analysis of the overall distribution characteristics of the vector field, such as the average speed size and direction change rate. It can be understood that the activity evaluation combines the inter-frame pixel change amplitude and the motion vector consistency, for example, the object trajectory at t=10 point position shows fast motion, and Av significantly increases, providing input for subsequent encoding control.

[0038] Further, the activity evaluation result directly drives the dynamic adjustment of the encoding parameters. As shown in Figure 5 , the content activity Av is one of the key input parameters (Y axis), which cooperates with the network bandwidth Bw in the adaptive decision surface to generate the optimized value of the target code rate Rc. In implementation, the spatio-temporal evolution data of Av is output in real time through a lightweight CNN, forming part of the encoder input space. It can be understood that this mechanism avoids repeated analysis of the original frame and realizes efficient content perception through motion vector features.

[0039] The benefits of this embodiment are that through the linkage mechanism of Figure 4 and Figure 5 , the motion vector extraction and encoding decision form a closed-loop control. The lightweight CNN model reduces resource consumption while ensuring the real-time reliability of the activity index, avoiding the lag of encoding parameter adjustment behind the video content changes.

[0040] Example 4: In order to solve the dual impact of network fluctuations and video content dynamics on encoding parameters, this embodiment further optimizes the dynamic encoding adjustment mechanism. Figure 5 The decision surface shown achieves dual adaptation of network status and content characteristics.

[0041] When it is detected that the network bandwidth resources are sufficient, Figure 5 In the region where the Bw value is higher than 10Mbps, this embodiment improves the quantization accuracy and sampling density to enhance the ability to restore video details. Figure 5 As shown in the right half of the decision surface, under conditions of high Bw and Av, such as Bw = 15Mbps and Av > 0.6, the encoding bitrate Rc increases nonlinearly to over 2000kbps. It is important to understand that this strategy improves image quality by increasing motion compensation accuracy and reducing macroblock granularity while avoiding overloading the encoder's computational resources.

[0042] Furthermore, the dynamic degree of video content is measured by Figure 4 The motion vector field is quantized to adjust the driving parameters. Figure 4 The strong motion area corresponding to the fast-moving object trajectory is orange-red in the color map (value range > 1.2). Figure 5 The gradient change of the longitudinal dimension Av axis of the decision surface is determined by actively increasing the frame sampling rate to 60fps and using full resolution encoding. It can be understood that when the vector field presents a low-intensity blue distribution, such as Figure 4 upper right corner <0.3 area), the frame rate reduction mechanism is triggered and temporal downsampling is enabled.

[0043] In a network congestion scenario, such as Figure 3 In the event of a sudden bandwidth drop of 28-35 seconds, this embodiment activates the fast bit rate adjustment mechanism. This mechanism includes three levels of response: First, according to Figure 3 Real-time Bw monitoring value will drop Rc to the lowest area of ​​the surface. Figure 5 Rc<500kbps in the lower left corner; then the non-key frame encoding is interrupted and only the I frame carrying complete information is transmitted; finally Figure 7 As shown in the figure, the bandwidth allocation module allocates a high priority channel for the I frame stream, the red "high priority" bandwidth area. It should be understood that this process uses Figure 3 The RTT×BDP product value is used to predict congestion risks, and the bit rate reduction plan is initiated 300ms before the bandwidth falls below the threshold (for example, <4Mbps).

[0044] The benefits of this embodiment are: Figure 5 The decision surface constructs a coordinated control mechanism between Bw and Av to avoid image quality fluctuations caused by single factor decisions. Figure 7The multi-path flow scheduling is shown in the dynamic resource allocation to maintain the Rc stability of high priority video streams such as monitoring alarm pictures, while integrating the content activity assessment into the longitudinal dimension of the curved surface, i.e. Figure 5 the Av axis, to achieve a quantitative balance between motion details and bandwidth consumption.

[0045] Embodiment five: To solve the problem of transmission quality degradation caused by resource competition in concurrent transmission of multiple video streams, this embodiment further optimizes the global scheduling mechanism of network bandwidth resources, realizes efficient collaborative allocation of network resources by establishing a service-driven priority assessment model and a dynamic quota adjustment system. The specific implementation is carried out in combination with Figure 4 motion vector analysis and Figure 7 intelligent scheduling architecture.

[0046] Firstly, a video stream priority assessment model is constructed, which realizes priority quantitative classification through double-dimensional feature extraction. The service type dimension uses deep packet detection technology to identify the application scene characteristics of video streams, and marks video conference and video backhaul control and other key business streams as red high-priority channels; the content importance dimension is based on the inter-frame motion field analysis technology shown in Figure 4 , determines the emergency level of specific regions in continuous video frames by calculating the motion vector amplitude and the duration thereof. It should be understood that when the vector amplitude of a specific region continuously exceeds the set threshold, the system automatically upgrades the video stream to the emergency transmission queue.

[0047] Further, the scheduling engine issues dynamic resource allocation instructions to the underlying transmission stack through a standardized southbound interface. As shown in Figure 7 , under the constraint of total available bandwidth, the system constructs a three-layer bandwidth allocation channel of red / orange / green: high-priority streams occupy the uppermost layer of the transmission channel (red area), ensuring the continuous transmission of key business video frames; medium-priority streams (orange area) and low-priority streams (green area) are dynamically contracted in proportion. It can be understood that this scheduling algorithm monitors the network congestion state characteristics in real time, and when the competition signal of the resource contention area is detected, such as the scheduling event at 28s, the system preferentially compresses the low-priority stream quota and maintains the stable bandwidth occupation of high-priority streams. Figure 7

[0048] The benefits of this embodiment are that Figure 4 content analysis and Figure 7 ​The organic synergy of the scheduling model establishes an accurate mapping path from business semantics to network resources. The bandwidth isolation mechanism based on flow priority is implemented in the transmission layer, which converts resource conflicts in network congestion into a controlled hierarchical allocation process, effectively avoiding the risk of transmission interruption of high-importance video streams. The dynamic quota model ensures the scientific allocation of bandwidth resources among different priority flows by real-time sensing of network state changes, providing a stable quality guarantee foundation for multi-channel video transmission.

[0049] Embodiment six: To solve the problem of video transmission quality degradation caused by unstable network channel, this embodiment further refines the control strategy of forward error correction mechanism, realizes the unification of transmission reliability guarantee and bandwidth efficiency through the fine identification of network packet loss characteristics and the coordinated regulation of FEC parameters. The specific implementation is combined Figure 3 with the network link monitoring data and Figure 6 the effectiveness analysis of the redundancy strategy to build a closed-loop optimization system.

[0050] Further, based on the network link multi-dimensional monitoring system shown in Figure 3 , including bandwidth mutation events, delay change trend and jitter heat map, real-time capture of packet loss rate magnitude and its time sequence distribution characteristics. When the packet loss pattern recognition module detects a network congestion event as marked at 28 seconds in Figure 3 , which is characterized by a sudden drop in bandwidth and a red dense area in the delay heat map, the system determines that there is a risk of continuous packet loss. At this time, the block-like redundancy enhancement strategy shown in the right side of Figure 6 is automatically triggered, which increases the FEC redundancy ratio to the interval of 15%-25%. The technical point is: using a sliding window to analyze the continuous packet loss length characteristics, such as Figure 6 indicated by the horizontal axis, when the packet loss length exceeds 3 packets, the redundancy allocation coefficient is increased linearly according to the packet loss sequence length, and at this time the FEC encoding block size is expanded to the interval of 32-64 packets, ensuring the recovery ability of long sequence packet loss.

[0051] Further, in the network channel stable stage, such as the interval of 0-20 seconds in Figure 3 , the jitter heat map is displayed as blue-green distribution and the packet loss rate is less than 2%, and the system automatically identifies it as a random packet loss mode. In this state, the fine control strategy corresponding to the performance curve of the left side of Figure 6 is activated: the FEC redundancy is compressed to the interval of 5%-10%, and the encoding block size is reduced to 8-16 packets. It can be understood that this operation is realized through the overlapping characteristics of multiple curves in Figure 6 : when the random packet loss rate is less than 5%, the redundancy of 5%-10% has met the recovery demand of more than 95%, avoiding the waste of bandwidth caused by high redundancy.

[0052] Further, the system is based on Figure 3The color intensity of the bottom delay jitter heat map changes, and the delay stability index of the transmission path is calculated in real time. When the jitter intensification event in the yellow to red area is monitored, the system automatically implements the block scale compression strategy: the upper limit of the FEC encoding block is constrained within 16 packets. It needs to be emphasized that this strategy aims at the processing delay problem caused by large block FEC encoding, reduces the data packaging time by limiting the encoding block size, and ensures that the encoding and decoding process is completed under the 100ms delay constraint. Figure 3 At 85 seconds, the automatic block scale compression strategy is implemented: the upper limit of the FEC encoding block is constrained within 16 packets. It needs to be emphasized that this strategy aims at the processing delay problem caused by large block FEC encoding, reduces the data packaging time by limiting the encoding block size, and ensures that the encoding and decoding process is completed under the 100ms delay constraint.

[0053] Through this embodiment: Figure 3 The network state awareness and Figure 6 The performance model forms a linkage regulation system, and establishes a closed-loop control logic of packet loss feature recognition→parameter decision→performance verification. In the random packet loss scenario, the redundant overhead is reduced to save bandwidth resources, the error correction capability is enhanced under the burst packet loss condition to guarantee the transmission continuity, and the contradiction between the anti-packet loss performance and the processing delay is balanced through the block size dynamic adaptation mechanism. The hierarchical control model makes the FEC redundant parameters always in the optimal matching state of the network channel characteristics, providing technical support for the end-to-end reliable transmission of video stream.

[0054] To solve the problem that video encoding redundancy and network transmission efficiency are difficult to be considered, this embodiment refines the source and channel cooperative control mechanism, and forms a double-level optimization system through Figure 5 The adaptive encoding control module, Figure 6 The FEC anti-packet loss module and Figure 7 The intelligent scheduling decision module. The specific implementation is as follows: Based on Figure 4 The dynamic analysis of the inter-frame motion vector field of the video, when a large range of directional motion is detected, the high motion compression strategy of Figure 5 The decision surface is triggered. In the high motion scene with content activity>0.6, corresponding to Figure 4 The right side of the bright yellow area, the system automatically reduces Rs along the gradient descent direction of the surface, and switches the GOP structure to IBBP mode. It can be understood that this reduces the amount of redundant data by reducing the intra refresh rate, and the motion compensation encoding retains the key motion information, so that the source output bit rate is reduced by about 30%.

[0055] Further, in the network transmission layer, Figure 3 The spatio-temporal evolution data and Figure 6 The performance model forms a joint decision basis: When Figure 3 The packet loss rate curve detects a continuous packet loss event, the block redundancy strategy of Figure 6 Is activated, and the encoding block is amplified to 48 packets in the 25%-30% redundancy interval according to the packet loss length.

[0056] Figure 7 The bandwidth scheduling module receives the multi-channel stream characteristics output by the encoder in real time. When the high-priority stream triggers the low-resolution mode, the released bandwidth resources are dynamically reallocated to the medium and low-priority streams. Figure 7 As shown, at t=28 seconds, the bandwidth demand of the high-priority flow drops by 1.2 Mbps, and the resource is immediately allocated to the medium-priority flow in the green area.

[0057] Furthermore, the QoE evaluation data at the receiving end triggers dual parameter correction via the feedback channel: When the decoding end motion blur index is greater than 0.4, it is forced to increase Figure 5 The Rs parameter of the decision surface is adjusted down when the packet retransmission rate is greater than 8%. Figure 6 The FEC block size is increased to 24 packets. This feedback mechanism enables the channel control parameters of the signal source to continuously approach the optimal working point, forming Figure 5 Surface and Figure 6 Dynamic coupling of curves.

[0058] Through this embodiment: the redundant control of the source coding link provides flexibility for channel transmission, while the FEC and bandwidth scheduling strategy on the channel side reversely guarantee the source quality. This two-way collaborative mechanism avoids the problem of disconnection between coding parameters and network status in traditional solutions, allowing the video stream to automatically trigger the graded degradation strategy when the bandwidth is limited, and obtain the optimal anti-loss protection level when the network is congested. Especially in Figure 7 In the scheduling strategy, priority diversion and resource reallocation form a dynamic balance system to ensure that critical video streams still meet end-to-end delay constraints when the link fluctuates.

[0059] Example 8: To solve the control problem of dynamic matching of multi-dimensional parameters and video content features in a complex network environment, this embodiment constructs an intelligent network control system architecture with a hierarchical coordination mechanism, such as Figure 8 As shown in the figure, precise control of low-latency video return is achieved through data interaction and strategy linkage between modules.

[0060] Furthermore, the network status detection module constructs a network parameter monitoring system in the time and space dimensions through distributed sampling nodes. The bandwidth detection unit uses a time window sliding algorithm to estimate the available bandwidth in real time, and forms a statistical feature by periodically sending a sequence of detection packets and calculating the statistical characteristics of the sequence response interval. Figure 3The bandwidth change curve shown in the upper half, where the network congestion event occurs, such as the red filled area in the 28-35 second interval, the algorithm automatically switches to the high-density sampling mode to capture the bandwidth sudden change characteristics. The delay monitoring unit generates a bidirectional probe packet pair based on the UDP protocol, and constructs a delay jitter heat map by calculating the probability density distribution of the round-trip time (RTT), which takes the time axis as the vertical axis and the jitter amplitude as the horizontal axis, clearly presenting the time sequence characteristics of the delay fluctuation. The packet loss statistics unit then identifies the packet loss type such as random packet loss or continuous packet loss by maintaining the packet number sequence in the sliding window, and outputs a feature vector containing parameters such as packet loss rate, packet loss continuity, and burst length, providing basic data support for subsequent FEC strategy adjustment. It should be understood that each monitoring unit shares the underlying transmission channel through time division multiplexing mechanism, which reduces the system overhead while ensuring data collection accuracy.

[0061] Further, for the computing power constraint of the edge computing node, the embodiment optimizes the video content analysis module, as shown in Figure 4 The lightweight CNN processing unit replaces the traditional convolutional layer with a deep separable convolution, compresses the model parameter quantity to less than 30% of the original ResNet18, and introduces a 15x15 normalized grid division at the input layer to control the single-frame processing time within 20ms. The motion vector extraction unit calculates the pixel-level motion vector based on the adjacent frame difference operation through the Farneback optical flow algorithm to generate a vector field containing direction and speed information, as shown in Figure 4 On this basis, the system performs block processing on the video frames, and when the motion vector density in a single block exceeds a predetermined threshold, it determines that the region is high activity content, and finally generates a global content feature evaluation result through weighted average of regional activity. This hierarchical processing mechanism not only retains the detailed features of video dynamic changes, but also avoids the performance bottleneck caused by full-frame intensive calculation.

[0062] Further, as shown in Figure 5As shown, the adaptive coding control module constructs a three-dimensional decision space based on network bandwidth and content activity. The parameter adjustment algorithm unit receives the available bandwidth value (X-axis) output by the network state detection module and the content activity (Y-axis) output by the video content analysis module, and calculates the optimal coding parameter (Z-axis represents the target code rate) through a pre-set nonlinear mapping model. This model uses a power function fitting method to represent the code rate adjustment strategy as a joint function of network resources and content characteristics, for example, when the available bandwidth improves and the content activity is high, the system automatically increases the coding rate to retain more detailed information; while when the bandwidth is limited or the content tends to be static, the transmission fluency is maintained by reducing the code rate and adjusting the resolution parameter. The generation process of the decision surface combines historical transmission data training and real-time parameter calibration, ensuring that the coding strategy can dynamically adapt to the dual changes of network and content, forming an optimal balance between video quality and transmission delay.

[0063] Further, for the concurrent transmission scene of multiple video streams, the intelligent scheduling decision module constructs a bandwidth allocation model based on priority, as shown in Figure 7 . The system presets the quality of service (QoS) level for each video stream, and high-priority streams such as unmanned vehicle control video backhaul enjoy resource reservation rights in bandwidth allocation, while low-priority streams such as non-real-time recording videos adopt elastic allocation strategies. The scheduling algorithm monitors the fluctuation curve of the total available bandwidth as shown in Figure 7 and the resource occupation of each stream, and generates the optimal allocation scheme through dynamic programming method: when the bandwidth is sufficient, allocate resources according to the peak demand of each stream; when a bandwidth bottleneck occurs, compress the code rate or adjust the transmission interval of low-priority streams according to the priority coefficient, to ensure the transmission stability of critical video streams. It should be noted that this module introduces a prediction mechanism, based on historical scheduling data to predict the bandwidth demand trend in the future period, to adjust the transmission parameters of each stream in advance to reduce decision delay, forming a forward-looking resource management strategy.

[0064] Further, the FEC anti-packet loss module realizes precise control by establishing a mapping relationship between packet loss types and redundancy parameters. The redundancy strategy decision unit analyzes the packet loss mode characteristics output by the network state detection module, as shown in Figure 6As shown, when the random packet loss rate is detected to rise, the redundancy is adjusted using a linear increasing strategy to ensure effective recovery at a lower redundancy overhead; and in the face of continuous packet loss events, the FEC block size is dynamically increased according to the packet loss burst length to expand the protection range of redundant data and improve the recovery success rate. In a specific implementation, the system maintains a pre-set redundancy strategy library, which contains the optimal redundancy and block size combination under different packet loss scenarios, and the decision unit quickly retrieves the corresponding strategy through a pattern matching algorithm to avoid the increase in time delay caused by online calculation. This scenario-based differentiated processing mechanism enables the FEC scheme to maintain high efficiency when the network packet loss characteristics change, reducing the additional occupation of bandwidth resources by redundant data.

[0065] Further, the feedback adjustment mechanism module periodically collects module operation data such as encoding parameter adjustment frequency, scheduling decision response time, FEC recovery success rate, etc., and analyzes the effectiveness of the current control strategy through a pre-set performance evaluation index system. When the key indicators deviate from the pre-set threshold, the system triggers a strategy update process: for example, if the code rate adjustment of the adaptive encoding module lags behind the network changes, the parameter update period of the decision model is automatically optimized; if there is a resource allocation conflict in the intelligent scheduling module, the flow priority weights are recalibrated. This closed-loop adjustment mechanism enables the system to continuously adapt to changes in network environment and business demand in the long run, and gradually optimizes the control strategy through a data-driven approach.

[0066] Through the modular design and collaborative control architecture of the embodiment, the system realizes intelligent processing of the whole process from network state perception, content feature extraction to transmission strategy generation. Each module has an independent functional implementation unit, and forms a strategy linkage through a standardized data interface, ensuring that video quality, transmission delay and resource utilization can be dynamically balanced under complex network conditions, providing a systematic solution for low-latency video backhaul scenarios.

[0067] Embodiment Nine: To solve the problem of delay control and quality guarantee in the multi-device collaborative transmission of the video conference scene, the embodiment provides a video conference device integrated with intelligent network control function, which realizes end-to-end low-latency video backhaul control through the deep coupling design of the encoder device, the decoder device and the intelligent control system. The following aspects are described in detail: device hardware architecture, module collaborative mechanism and data processing flow, etc.

[0068] The video conference device of the embodiment adopts an end-to-end layered architecture, the encoder device is deployed at the video acquisition end, such as a conference room camera terminal, the decoder device is deployed at the video receiving end, such as a conference terminal display screen, and the two achieve bidirectional data interaction through an intelligent network control system. It should be understood that the intelligent network control system can be integrated as an independent hardware module in the device or deployed on an edge computing node in a software containerized manner to form a distributed control architecture. This deployment mode takes into account the real-time requirements of local processing of the device and the synergy requirements of network global strategies, reduces the load of end-side computing power, and ensures the dynamic adaptability of the control strategy.

[0069] Further, the core functional units of the encoder device include an image acquisition module and a preprocessing module. The image acquisition module adopts a progressive scan CMOS sensor, supports dynamic switching of 1080P to 4K resolution, and cooperates with an automatic white balance and exposure control algorithm to ensure the color restoration degree and brightness uniformity of the original video stream. The preprocessing module integrates a lightweight video content analysis function, based on a lightweight CNN model such as Figure 4 The motion vector extraction mechanism shown in the figure, performs difference operation on consecutive frames to generate an inter-frame motion vector field. In specific implementation, the module divides the video frame into 16x16 pixel macroblock units, calculates the motion vector density and direction distribution of each macroblock, and outputs content activity evaluation results in real time, such as high dynamic region proportion, object moving speed average, and other feature parameters. These parameters are directly input to the adaptive encoding control module of the intelligent network control system, providing content feature basis for dynamically adjusting encoding parameters such as target code rate, GOP structure, and entropy encoding mode. It should be noted that the preprocessing module adopts a pipeline processing architecture, parallelizes image noise reduction, resolution scaling, and motion feature extraction, and controls the single-frame processing delay within 15ms, meeting the timing requirements of low-latency transmission.

[0070] Further, the decoder device undertakes the dual functions of network data decoding and image display. In the decoding link, it supports adaptive switching of multiple encoding formats such as H.264, H.265, and AV1, and automatically configures decoding engine parameters by analyzing parameter set information in the code stream. In view of the possible packet loss or error code problems in network transmission, the decoder integrates a forward error correction (FEC) check module, which forms a closed loop with the FEC anti-packet loss module of the intelligent control system: when continuous packet loss is detected to cause decoding failure, the decoder sends packet loss position and type information, such as random packet loss or burst packet loss, to the control system through the feedback channel, triggering dynamic adjustment of FEC redundancy parameters, such as Figure 6The redundancy strategy performance analysis mechanism is shown. In the image display link, the decoded video stream is first processed by the frame rate synchronization module, which dynamically adjusts the display frame rate through the buffer queue to eliminate the frame rate fluctuations caused by network transmission; then enters the resolution adaptation module, which performs image scaling according to the physical resolution of the receiving end display screen, such as 1920x1080 or 3840x2160, and uses a bicubic interpolation algorithm to improve the quality of the scaled image. In addition, the decoder device is provided with a real-time monitoring interface to continuously collect performance indicators such as decoding delay, buffer occupancy rate, and image freezing rate, and feed back to the feedback regulation mechanism module of the intelligent control system, providing real measurement data for the optimization of transmission strategies.

[0071] Further, the intelligent network control system builds a multi-level data interaction channel between the encoder and the decoder. The network state detection module collects key indicators such as bandwidth, delay, jitter, and packet loss rate of the transmission link in real time, and Figure 3 The spatio-temporal evolution monitoring mechanism is shown, which provides network layer basic data for encoding parameter adjustment and FEC strategy configuration. Specifically, when the available bandwidth decreases, the control system synchronously triggers the rate control module of the encoder, which adjusts the target code rate according to the content activity evaluation result, and adjusts the quantization parameter to maintain the relative stability of the visual quality; if a burst packet loss event is detected, the FEC anti-packet loss module immediately switches to a high redundancy strategy by increasing the size and number of redundant data blocks, such as Figure 5 The adaptive decision surface model is shown, which dynamically reduces the target code rate under the constraint of the content activity evaluation result, and adjusts the quantization parameter to maintain the relative stability of the visual quality; if a burst packet loss event is detected, the FEC anti-packet loss module immediately switches to a high redundancy strategy by increasing the size and number of redundant data blocks, such as Figure 6 The continuous packet loss scenario processing mechanism is shown to improve the anti-packet loss capability of the code stream in a poor network environment.

[0072] Further, in a multi-path video stream concurrent scenario such as a multi-party video conference, the intelligent scheduling decision module plays a key coordination role. This module assigns priority labels to each video stream, such as setting the speaker video to the highest priority and the participant video to the standard priority, based on Figure 7 The intelligent bandwidth scheduling model is shown, which calculates the bandwidth allocation scheme for each stream in real time. When the total available bandwidth is insufficient, the system prioritizes the transmission quality of high-priority streams, and releases bandwidth resources by reducing the encoding resolution or increasing the I-frame interval of low-priority streams; when the network state is stable, the load balancing strategy is used to optimize the overall resource utilization. This hierarchical scheduling mechanism forms a synergistic effect with the encoding control and anti-packet loss strategies, ensuring that video streams of different importance can meet the delay and quality requirements when sharing network resources.

[0073] Further, to improve the environmental adaptability of the device, the embodiment introduces a number of optimization designs at the hardware level: the encoder device integrates a temperature sensor and a fan speed regulation module, which ensures the stability of long-time high-load operation by dynamically adjusting the cooling strategy; the decoder device is configured with a dual-network-port redundancy design, which supports automatic switching between wired and wireless networks and seamlessly switches to the standby link in the event of a single link failure. At the software level, the intelligent control system adopts a microservice architecture, encapsulating functions such as network detection, content analysis, and encoding control as independent service units, and achieving low-latency communication between modules through a message queue, facilitating subsequent function expansion and algorithm upgrade. In addition, the device supports a remote management interface, through which operation and maintenance personnel can monitor the device's running state and configure policy parameters in real time, forming a centralized device management system.

[0074] Through the device architecture design of the embodiment, the video conference system realizes intelligent control of the entire chain from video capture, encoding and transmission to decoding and display. The encoder and decoder form a policy closed loop through the intelligent network control system, enabling the front-end capture and processing and the back-end display to adapt to changes in network status and content characteristics in real time, effectively reducing end-to-end transmission latency in complex network environments while ensuring the clarity and smoothness of video content. This deep integration solution at the device level provides an engineering implementation path for low-latency application scenarios such as remote video conferencing and real-time collaboration, ensuring stable and reliable operation performance of the system in different deployment environments through collaborative optimization between modules and deep coupling of hardware and software.

[0075] Embodiment Ten: To solve the portability and efficient execution of the low-latency video backhaul method on different computing platforms, the embodiment provides a specific implementation of a computer-readable storage medium, which converts the method steps into an executable code system through structured program design, ensuring the optimal adaptation of software logic and hardware resources.

[0076] Specifically, the computer-readable storage medium includes but is not limited to solid-state storage devices (SSDs), disk arrays, cloud storage units, and other non-volatile storage carriers, and the computer program stored thereon adopts a modular architecture corresponding to the steps of the intelligent network control method for low-latency video backhaul. The program main body is divided into six functional modules: network status monitoring, video content analysis, encoding strategy control, intelligent scheduling decision, anti-packet loss processing, and feedback regulation, and each module realizes data interaction and logic collaboration through standardized interfaces. It should be understood that the storage medium is compatible with mainstream processor architectures such as x86 and ARM, and supports Docker containerized deployment, encapsulating runtime dependencies through container images to ensure seamless migration of the program on different hardware platforms such as edge computing nodes and server clusters.

[0077] Further, the program implementation of the network state monitoring module adopts asynchronous I / O technology, and parameters such as bandwidth, delay, jitter and packet loss rate of the network link are collected through multi-thread concurrent collection, and the data collection frequency can be dynamically adjusted through a configuration file (for example, the default is 50 ms / time). The video content analysis module integrates the inference code of the lightweight CNN model, the model file is stored in the medium in the form of PROTOBUF, and is loaded into the memory at runtime and calls the GPU acceleration library (such as CUDA) to realize the rapid extraction of inter-frame motion vectors. The content activity features output by the module are transmitted in real time to the encoding strategy control module through a shared memory mechanism. The encoding strategy control module includes code implementation of a nonlinear mapping algorithm, and a three-dimensional decision logic is constructed based on network state parameters and content features. The encoding parameters such as target code rate and GOP structure are dynamically adjusted through conditional judgment statements, and the related parameters are transmitted to the encoder device through a socket interface.

[0078] Further, the intelligent scheduling decision module manages multiple video streams using a priority queue data structure, and dynamically allocates bandwidth resources through a dynamic programming algorithm. A preset quality of service level mapping table is included in the algorithm logic, and a user-defined priority strategy configuration file can be read. The anti-packet loss processing module includes index code of an FEC redundancy strategy library, and matches a preset strategy according to real-time packet loss characteristics, and dynamically generates and checks redundant data through bit operations. The feedback adjustment module periodically collects the running logs of each module, evaluates the strategy performance through statistical analysis code, and triggers the automatic update process of the configuration file when detecting abnormal performance indicators, to realize closed-loop optimization of the control strategy.

[0079] Further, an error handling mechanism is embedded in the program code, and a capture function is set for abnormal conditions such as network connection interruption and model loading failure, and the abnormal stack information is recorded through a log system. The storage medium supports incremental update function, and when the algorithm module needs to be upgraded, only the corresponding code file needs to be replaced without affecting the overall program running, improving the maintainability of the system. It should be noted that the memory paging mechanism is used to manage data caching during program execution, and the network state data and encoding parameters with high frequency access are stored in the cache area to reduce the memory access delay of the processor and ensure the real-time execution requirements of the method steps.

[0080] Through the computer readable storage medium of the embodiment, the intelligent network control method for low-latency video backhaul is converted into a software entity that can be independently deployed, and the modular design and hardware-independent architecture provide a standardized solution for system integration in different application scenarios. The efficient execution of the program in the storage medium depends on the optimization design of the data interaction interface and the completeness of the exception handling mechanism, ensuring that the logical integrity of the method steps and the dynamic adaptability of the control strategy can still be maintained in a complex computing environment, forming an intelligent video transmission solution with software and hardware collaboration.

[0081] While the application has been particularly described with reference to the preferred embodiments thereof, it is understood that the application is not limited to the embodiments described above but various modifications and changes can be made by those skilled in the art without departing from the spirit of the application, and it is intended to fall within the scope of the appended claims and their equivalents.

Claims

1. An intelligent network control method for low-latency video backhaul, characterized by, The method comprises the following steps: Real-time collection of network key indicators including bandwidth, delay, jitter and packet loss rate of network link, providing real-time network environment data support for transmission strategy adjustment; Extracting inter-frame motion vector of video frame through lightweight CNN model, judging video content activity, evaluating frequency and amplitude of content change between adjacent frames, providing content feature basis for encoding strategy optimization; Comprehensive network key indicators and video content activity, dynamically adjusting video encoding parameters including resolution, code rate and frame rate, balancing video quality and network adaptability; According to priority and resource occupation of multi-path video stream, overall allocation of network bandwidth resources, control of transmission parameters of each video stream, to ensure the efficiency of multi-path video transmission; According to real-time network environment, intelligent identification of packet loss type, dynamic adjustment of FEC redundancy parameters, addition of redundant data in network transmission to resist packet loss problem, to balance packet loss recovery ability and bandwidth overhead. 2.The intelligent network control method of low-latency video backhauling of claim 1, wherein, The collection step of the network key indicators comprises: Estimating uplink available bandwidth by sending probe packets and analyzing response time, tracking bandwidth change trend in real time to identify network congestion, and counting peak bandwidth per unit time; Measuring round-trip delay and one-way transmission delay of data packets, analyzing stability of delay jitter to determine decoding end buffer size; Statistical analysis of instantaneous packet loss rate in short time window and average packet loss rate in long time period, identification of continuous packet loss or random packet loss mode, providing basis for FEC redundancy parameter adjustment. 3.The intelligent network control method of low-latency video backhauling of claim 1, wherein, The extraction step of the inter-frame motion vector of the video frame comprises: Reducing the calculation complexity and resource consumption of lightweight CNN model for real-time video analysis scene; Based on the pixel difference and motion trajectory of adjacent video frames, extracting motion vector features representing content change, quantifying the dynamic degree of video content, and generating video content activity evaluation results to drive the dynamic adjustment of the encoding parameters. 4.The intelligent network control method of low-latency video backhauling of claim 1, wherein, The step of dynamically adjusting the video encoding parameters comprises: When sufficient network bandwidth is detected, high code rate and high resolution encoding are used to improve video detail quality, and when network bandwidth is limited, code rate and resolution are automatically reduced to ensure transmission continuity; For high dynamic video content, frame rate and resolution are increased to retain motion details, and for static or low dynamic content, frame rate and code rate are reduced to reduce data volume; When network congestion or bandwidth drops, trigger fast code rate adjustment mechanism, preferentially transmit key I frame to ensure continuous display of video decoding end. 5.The intelligent network control method of low-latency video backhauling of claim 1, wherein, The step of overall allocation of network bandwidth resources comprises: Establishing a video stream priority evaluation model, assigning priority according to the service type and content importance of the video stream; Based on real-time network bandwidth and priority of each video stream, issuing scheduling instructions to network transmission layer through standard interface, dynamically adjusting transmission bandwidth quota and transmission priority of each video stream, avoiding resource contention of multi-path video stream. 6.The intelligent network control method of low-latency video backhauling of claim 1, wherein, The step of dynamically adjusting FEC redundancy parameters comprises: When detecting that network packet loss rate increases and presents continuous packet loss mode, increase FEC redundancy to improve packet loss recovery ability, when network packet loss rate is low and presents random packet loss, reduce FEC redundancy to reduce bandwidth overhead; According to the delay and jitter characteristics of the network link, the block size and redundancy ratio of FEC encoding are dynamically adjusted to balance the anti-packet loss performance and encoding processing delay, so as to ensure real-time video transmission. 7.The intelligent network control method of low-latency video backhauling of claim 1, wherein, The network control method jointly optimizes the source and channel to transmit with low delay: at the source side, the adaptive encoding control module reduces redundant data, at the channel side, the FEC anti-packet loss module and the intelligent scheduling decision module improve transmission reliability and bandwidth utilization, and the feedback strategy is combined to form end-to-end dynamic optimization.

8. An intelligent network control system for low-latency video backhauling, configured to implement the intelligent network control method for low-latency video backhauling according to any one of claims 1 to 7, characterized in that Comprise: a network state detection module for real-time collection of network key indicators to provide data support for transmission strategy adjustment, the network key indicators including bandwidth, delay, jitter and packet loss rate of network link; a video content analysis module for extracting inter-frame motion vectors of video frames through a lightweight CNN model to judge the activity of video content and output content feature evaluation results; an adaptive encoding control module for receiving the network key indicators and content feature evaluation results to dynamically adjust video encoding parameters and balance video quality and network adaptability; an intelligent scheduling decision module for allocating network bandwidth resources according to the priority and resource occupation of multiple video streams and controlling the transmission parameters of each video stream; a feedback adjustment mechanism module for periodically collecting module running data, evaluating system performance and updating control strategies; an FEC anti-packet loss module for intelligently identifying packet loss types according to real-time network conditions, dynamically adjusting FEC redundancy parameters, and adding redundant data in network transmission to resist packet loss problems.

9. A video conferencing device, characterized by Comprise: an encoder device deployed at the video acquisition end for image acquisition and preprocessing; a decoder device deployed at the video receiving end for network data decoding and image display; The intelligent network control system for low-delay video backhaul of claim 8 is used to connect the encoder device and the decoder device to realize low-delay backhaul control of video stream.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the steps of the intelligent network control method for low-delay video backhaul of any one of claims 1-7.

Images