Network transmission optimization method in video monitoring scene

Abstract

The invention relates to a network transmission optimization method in a video monitoring scene, and belongs to the field of network transmission optimizing.The method comprises the steps that multi-dimensional feature data of the monitoring scene are collected, and the multi-dimensional feature data comprise terminal performance features, network dynamic features and monitoring picture features; performing multi-feature fusion on the terminal performance feature, the network dynamic feature and the monitoring picture feature, and dynamically adjusting the weight of each feature based on a monitoring scene; dynamically adjusting a buffering strategy of video monitoring so as to control the priority of a monitoring picture; optimizing the monitoring bandwidth of the weak network scene; and setting priorities of the monitoring tasks, and performing lightweight scheduling on the monitoring tasks according to the priorities. According to the method, the adaptation precision of bandwidth prediction to monitoring video transmission can be improved, and the picture identifiability is guaranteed.

Description

Technical Field

[0001] This invention belongs to the field of network transmission optimization, and in particular relates to a network transmission optimization method in video surveillance scenarios. Background Technology

[0002] Video surveillance transmission is characterized by high bandwidth, high real-time requirements, and highly flexible bandwidth demands. Its transmission optimization requires adapting to various aspects, including encoding, protocols, networks, and front-end intelligence, to suit the monitoring scenario. Therefore, one of the core aspects of video transmission optimization technology is monitoring scenario adaptation. However, existing video transmission optimization technologies all have certain shortcomings in terms of monitoring scenario adaptation.

[0003] The invention patent application CN120583251A discloses a CDN-based video downlink weak network adaptive image quality optimization method and system. This solution suffers from keyframe loss issues in weak network conditions, and its redundancy scheme cannot adapt to low bandwidth monitoring. Its indiscriminate task migration leads to core decoding latency. This solution is based solely on fixed threshold adjustments for bandwidth / buffering and is geared towards general video downlink, failing to consider the monitoring requirements of "no lost keyframes" and latency sensitivity. The forward error correction redundancy requires an additional 30% bandwidth, making it unsuitable for weak monitoring networks (such as remote road monitoring with bandwidth <2Mbps).

[0004] The invention patent with publication number CN118158089A discloses a method of using multiple network and video conferencing devices. This solution is only for video conferencing servers and only for video conferencing, focusing on server resource allocation, which is inconsistent with the resource-constrained characteristics of edge monitoring terminals (such as cameras).

[0005] The invention patent application CN120151531A discloses a method for video stream decoding delay prediction and dynamic quality optimization based on multi-feature fusion. This scheme can only predict decoding delay and uses a fixed model. This scheme focuses on decoding delay prediction, does not link bandwidth prediction and buffer optimization, and lacks a dedicated protection strategy for key frames, resulting in the inability to identify targets after frame loss in the monitoring screen.

[0006] In summary, existing video transmission optimization technologies can only support single scenarios, and their effectiveness is insufficient for optimizing weak networks. There are still technological gaps in addressing the specific pain points of the aforementioned monitoring scenarios. Summary of the Invention

[0007] In view of the shortcomings of the prior art, the purpose of the invention is to provide a network transmission optimization method in video surveillance scenarios, which can improve the accuracy of bandwidth prediction for the adaptation of surveillance video transmission and ensure the recognizability of the image.

[0008] This invention proposes a network transmission optimization method for video surveillance scenarios, including:

[0009] S1, collect multi-dimensional feature data of the monitoring scene, including: terminal performance characteristics, network dynamic characteristics and monitoring screen characteristics;

[0010] S2, perform multi-feature fusion on the terminal performance characteristics, network dynamic characteristics and monitoring screen characteristics, and dynamically adjust the weight of each feature based on the monitoring scenario;

[0011] S3 dynamically adjusts the video surveillance buffering strategy to control the priority of the monitored images and outputs predicted bandwidth adapted to the monitoring scenario, including:

[0012] S31, Let the initial prediction bandwidth at time t0 be BW_Pred(t0), and according to the relationship at time t0...

[0013] Keyframe density, set the buffer threshold Buf_Target to:

[0014] Buf_Target=Buf_Base×(1+K);

[0015] Where Buf_Base is the base buffer threshold and K is the correction coefficient for the keyframe density;

[0016] S32, when it is detected that the current buffer occupancy Buf_Cur is lower than the buffer threshold Buf_Target, priority is given to ensuring the filling of the key frame buffer, and the non-key frames are temporarily compressed to ensure that the monitoring screen corresponding to the key frame is output first.

[0017] S33, at time t1 in the next cycle, adjust BW_Pred(t0) according to the dynamically adjusted weight of buffer usage rate Buf_Usage and the latest buffer status, and output the predicted bandwidth BW_Pred adapted to the monitoring scenario.

[0018] S4, optimize the monitoring bandwidth for weak network scenarios, including: first, detect whether the predicted bandwidth or packet loss rate meets the weak network judgment criteria; if it does, optimize the monitoring bandwidth for weak network scenarios by extending the key frame interval, lightweighting and compressing non-key frames, and supplementing packet loss.

[0019] S5 allows you to set the priority of monitoring tasks and schedule each monitoring task in a lightweight manner based on the priority.

[0020] Furthermore, in S1, the terminal performance characteristics include: CPU decoding load rate, GPU video memory utilization rate, and buffer utilization rate;

[0021] The network dynamic characteristics include: current bandwidth, packet loss rate, and bandwidth fluctuation range;

[0022] The features of the monitoring footage include: keyframe density and target motion intensity.

[0023] Furthermore, in S2, the dynamic adjustment of the weights of each feature based on the monitoring scenario includes:

[0024] When the edge terminal type is detected as an edge camera, the weight of CPU decoding load rate is automatically reduced;

[0025] When the intensity of the detected moving target is higher than the preset intensity, it is judged as a high-motion target, and the weight of the keyframe density is automatically increased;

[0026] When the current network is detected to be weak, the weight of the current bandwidth is automatically increased;

[0027] When the density of the keyframes is detected to be higher than the preset frame density, the weight of the packet loss rate is automatically increased;

[0028] When static monitoring is detected, the weight of the target motion intensity is automatically reduced;

[0029] When a video surveillance delay is detected to be higher than a preset delay value, the weight of the buffer utilization rate is automatically increased.

[0030] Furthermore, when the number of CPU cores is ≤4, the edge terminal type is determined to be an edge camera;

[0031] If the current bandwidth BW_Cur is less than 2Mbps, the current network is considered a weak network.

[0032] Furthermore, in S31, when the keyframe density is ≤10 frames / second, K = 0.1; when the keyframe density is >10 frames / second, K = 0.2.

[0033] Furthermore, in S4, the weak network judgment criteria are: predicted bandwidth BW_Pred≤1.5Mbps or packet loss rate Loss_Rate>10%.

[0034] Furthermore, in S32, the higher the keyframe density value, the larger the K value and the higher the buffer threshold Buf_Target value, indicating that more buffer data needs to be prepared to meet the keyframe transmission requirements in order to prevent it from being lost due to insufficient buffering.

[0035] Furthermore, in S4, the extension of the key frame interval includes: in a static monitoring scenario, the key frame interval is increased from 10 frames to 15 frames; in a dynamic monitoring scenario, the key frame interval remains at 10 frames.

[0036] The lightweight compression of non-critical frames includes: using a differentiated compression method for monitoring screen areas, increasing the compression rate of the background area of ​​the screen, and maintaining a compression rate of 20% for the target area;

[0037] The packet loss compensation includes: retransmitting lost key frames using incremental retransmission, compensating for non-key frames using monitoring target interpolation, and reducing retransmission bandwidth.

[0038] Furthermore, in S5, the priority of the monitoring task is set as follows:

[0039] Set the keyframe decoding task to high priority; set the non-keyframe decoding task to medium priority; and set the rendering preprocessing task to low priority.

[0040] Furthermore, in S5, the lightweight scheduling of each monitoring task according to priority includes:

[0041] When edge terminal decoding resources are insufficient, high-priority tasks should not be discarded; only medium-priority and low-priority tasks should be discarded.

[0042] If the CPU decoding load rate of a monitoring task is >80% and it is a medium-priority task, the monitoring task will be discarded.

[0043] The beneficial effects of this invention are as follows:

[0044] The method described in this invention integrates four-dimensional features of "terminal decoding - network transmission - buffer status - monitoring screen features (keyframe density)" in the monitoring scenario to improve the accuracy of bandwidth prediction for monitoring video transmission; it designs a dynamic buffering strategy that links "buffering - bandwidth - keyframe density" to provide an optimization scheme with low bandwidth overhead and keyframe priority, ensuring screen recognizability.

[0045] 1. Monitoring scenario-based multi-feature fusion prediction:

[0046] This invention is the first to integrate monitoring screen features (keyframe density / target motion intensity) into bandwidth prediction. The linkage between monitoring screen features and bandwidth prediction requires a deep understanding of the "identification priority" requirement of monitoring, and dynamically adjusts feature weights based on edge terminal type and the motion state of the monitored target (such as dynamically increasing the weight of keyframe density in monitoring). The design focuses on the impact of keyframes on target recognition in the monitoring scenario. The prediction error of this invention is reduced by 25% compared with traditional methods, and the recognition rate of 1080P monitoring screens is improved by 30%.

[0047] This invention improves the accuracy of bandwidth prediction for monitoring video transmission by integrating four-dimensional features of "terminal decoding, network transmission, buffer status, and monitoring screen features (keyframe density)" in a monitoring scenario.

[0048] 2. Dynamic buffering with linkage between buffering, bandwidth, and keyframes:

[0049] This invention breaks through the traditional method of "adjusting a fixed threshold based solely on bandwidth / buffer". It introduces a keyframe density correction coefficient, prioritizing keyframe buffer filling while differentially compressing non-keyframe background areas when buffering is insufficient. Dynamic buffer adjustment ensures high-definition surveillance video latency ≤1 second, meeting real-time monitoring standards. It satisfies the ≤1-second latency requirement while avoiding recognition failures caused by keyframe loss. The number of stutters is reduced by 50% compared to traditional methods, and the keyframe integrity rate reaches 99.5%.

[0050] This invention achieves dual guarantees of real-time monitoring and identifiability: latency ≤ 1 second, keyframe integrity rate ≥ 99.5%, target identification rate ≥ 95%, solving the problem of keyframe loss under weak network conditions in traditional methods. This invention implements a linkage between bandwidth prediction and buffer optimization; the bandwidth-buffer linkage needs to consider both transmission and storage, making it more complex than simple latency prediction.

[0051] 3. Lightweight optimization for monitoring weak networks:

[0052] For monitoring in weak networks (bandwidth <1.5Mbps), a "no additional bandwidth overhead" strategy is designed: static monitoring extends the keyframe interval, dynamic monitoring uses differentiated compression for different areas, and keyframes are retransmitted incrementally. This improves the smoothness (24FPS) of 1080P video by 40% and reduces bandwidth usage by 25% under weak network conditions. For monitoring scenarios in weak networks, an optimization scheme of "low bandwidth overhead + keyframe priority" is provided to ensure image recognizability.

[0053] This invention optimizes bandwidth utilization in weak network conditions without additional redundant overhead. The lightweight strategy for weak network conditions requires a balance between compression rate and recognition rate, adapting to low-bandwidth monitoring. Bandwidth utilization is improved by 25% in weak network environments, saving 30% of bandwidth compared to traditional methods. This invention exhibits strong weak network adaptability, maintaining high-definition video smoothness above 24 FPS in weak network scenarios, meeting the monitoring requirements of "high-definition, identifiable, and stutter-free" performance.

[0054] The weak network optimization of this invention has no additional bandwidth overhead. It uses incremental retransmission for key frames in weak networks. Unlike traditional packet loss compensation, the incremental retransmission of this invention requires parsing the key frame structure.

[0055] 4. Priority resource coordination for edge monitoring terminals:

[0056] This invention introduces "monitoring task priority" into terminal resource scheduling for the first time, allowing only non-core tasks to migrate, thus avoiding the core decoding delay caused by the "indiscriminate task migration" of traditional methods. The priority migration in this invention needs to be combined with the characteristics of the monitoring tasks, rather than simple load balancing. Due to the significant differences in resources between edge terminals and servers, this invention adopts a dedicated lightweight design, rather than a simple migration solution.

[0057] This invention can improve the 1080P decoding efficiency of edge terminals (Snapdragon 865 level) by 30% and the resource load balancing rate by 25%. This invention is adapted to the resource-constrained characteristics of edge monitoring terminals (such as cameras) and realizes lightweight collaborative scheduling of CPU / GPU.

[0058] This invention improves edge terminal adaptability with lightweight computing (dynamic weight simplification + priority migration), reducing terminal CPU usage by 30% and GPU memory usage by 25%, and adapting to resource-constrained devices such as cameras;

[0059] 5. Strong scene scalability: It can be adapted to various monitoring scenarios such as static (park), dynamic (road), and edge (remote area), while traditional methods only support a single scenario.

[0060] 6. Improved smoothness: The simplified multi-feature fusion prediction error is less than 15%, and the dynamic buffering strategy reduces the number of stutters in high-definition videos by 40%-60%.

[0061] 7. High resource utilization: Terminal CPU / GPU collaboration balances resource load and simplifies prediction models, reducing terminal computing overhead by 40%, making it more suitable for edge monitoring devices to process high-definition video streams;

[0062] 8. Dynamic weights for monitoring scenarios: Dynamic weights adapt to changes in monitoring scenarios, making them more targeted than fixed models. Attached Figure Description

[0063] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. It is obvious that the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings.

[0064] Figure 1 This is a flowchart illustrating a network transmission optimization method in a video surveillance scenario according to an embodiment of the present invention;

[0065] Figure 2 This is a schematic diagram of a network transmission optimization method in a video surveillance scenario according to an embodiment of the present invention. Detailed Implementation

[0066] To enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. It should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0067] Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts disclosed in this invention.

[0068] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance. The terms "installed," "connected," and "linked" should be interpreted broadly; for example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0069] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of methods and systems consistent with some aspects of the invention as detailed in the appended claims.

[0070] like Figure 1 and Figure 2 As shown, this invention proposes a network transmission optimization method for video surveillance scenarios, including:

[0071] S1 collects multi-dimensional feature data of the monitoring scene, including: terminal performance characteristics, network dynamic characteristics, and monitoring screen characteristics.

[0072] (1) Terminal performance characteristics include: CPU decoding load rate, GPU memory load and buffer usage rate, etc.

[0073] Among them, CPU decoding load rate (Load_CPU) refers to the proportion of resources used by the monitoring terminal's CPU to process video encoding and decoding tasks, with a value range of 0-100%.

[0074] GPU memory utilization (Load_GPU_Mem): This refers to the proportion of GPU memory used to store video rendering data (such as frame images and rendering effects) on the monitoring terminal, with a value range of 0-100%. It reflects the GPU's rendering capacity margin.

[0075] Buffer Usage (Buf_Usage): This refers to the actual percentage of video buffer space occupied by the monitoring terminal. The calculation formula is "current buffer capacity (Buf_Cur) / maximum buffer capacity (Buf_Max)", with a value range of 0-100%.

[0076] (2) Network dynamic characteristics include: current bandwidth, packet loss rate and bandwidth fluctuation range, etc.

[0077] Current bandwidth (BW_Cur): refers to the real-time network data transmission rate between the monitoring terminal and the central platform / display terminal, in Mbps.

[0078] Bandwidth fluctuation amplitude (BW_Fluct): refers to the relative rate of change of the current bandwidth compared to the bandwidth at the previous moment. The calculation formula is "|BW_Cur-BW_Last| / BW_Last", and the result is expressed as a percentage.

[0079] Packet loss rate (Loss_Rate): refers to the proportion of data packets lost during network transmission out of the total number of packets sent, with a value ranging from 0 to 100%.

[0080] (3) Monitoring screen features include: key frame density and target motion intensity, etc., which are monitoring-specific features.

[0081] Keyframe density (KeyFrame_Density): The number of keyframes per unit time (e.g., 10 frames / second), reflecting the richness of detail in the monitored image (e.g., the keyframe density needs to be increased during peak hours).

[0082] Target Motion Intensity: The intensity value (0-100) calculated based on the motion trajectory of the target (person / vehicle) in the monitoring screen. The stronger the motion, the higher the bit rate is required to ensure clarity.

[0083] S2 performs multi-feature fusion on terminal performance characteristics, network dynamic characteristics, and monitoring screen characteristics, and dynamically adjusts the weight of each feature based on the monitoring scenario.

[0084] This invention is the first to integrate "monitoring screen features (i.e., keyframe density and target motion intensity)" into bandwidth prediction, and dynamically adjusts the feature weights based on edge terminal type and the motion state of the monitored target.

[0085] First, basic feature processing is performed, including: for monitoring scenarios, when the target motion intensity Motion_Intensity>70 (i.e., high-motion target), the weight of key frame density features is automatically increased; when the monitoring terminal is an edge camera (i.e., CPU / GPU resources <50% of mainstream devices), the computational complexity of terminal load features is reduced, and only the core indicators of CPU decoding load rate Load_CPU and GPU memory usage Load_GPU_Mem are retained.

[0086] Then, dynamic adaptation rules are designed to dynamically adjust the weights of each feature. The weight allocation is not fixed, but dynamically adapted based on the monitoring scenario, as shown in Table 1. The dynamic weight logic designed in this invention is specifically designed for monitoring scenarios, requires no complex models, and adapts to the computing capabilities of edge terminals.

[0087] Table 1 Weight Dynamic Adaptation Rules

[0088]

[0089]

[0090] Specifically, the weights of each feature are dynamically adjusted based on the monitoring scenario, including:

[0091] (1) When the edge terminal type is detected as an edge camera, the weight of the CPU decoding load rate is automatically reduced. For example, the weight of the CPU decoding load rate is reduced from the base weight of 15% to 10%.

[0092] In this invention, when the number of CPU cores is ≤4, the edge terminal type is determined to be an edge camera.

[0093] (2) When the motion intensity of a moving target is detected to be higher than the preset intensity, it is judged as a high-motion target, and the weight of the keyframe density is automatically increased. For example, when Motion_Intensity>70, it is judged as a high-motion target, and the weight of the keyframe density is increased from the base weight of 15% to 20%.

[0094] (3) When the current network is detected to be weak, the weight of the current bandwidth is automatically increased. For example, the weight of the current bandwidth is increased from the basic weight of 30% to 40%. In this invention, when the current bandwidth BW_Cur is detected to be less than 2Mbps, the current network is determined to be weak.

[0095] (4) When the key frame density is detected to be higher than the preset frame density, the weight of the packet loss rate is automatically increased. For example, when the key frame density > 12 frames / second, the weight of the key frame density is increased from the basic weight of 15% to 20% to ensure the transmission of key frames.

[0096] (5) When it is detected that the current is in static monitoring (such as a fixed camera position in a park), the weight of the target motion intensity is automatically decreased. For example, the weight of the target motion intensity is decreased from the basic weight of 10% to 5%.

[0097] (6) When the video monitoring delay is detected to be higher than the preset delay value, the weight of the buffer utilization rate is automatically increased. For example, when the delay of video monitoring approaches 2 seconds (the monitoring upper limit), the weight of the buffer utilization rate is increased from the basic weight of 15% to 20%.

[0098] S3. Dynamically adjust the buffer strategy of video monitoring to control the priority of the monitoring screen and output the predicted bandwidth adapted to the monitoring scenario, including:

[0099] S31. Set the initial predicted bandwidth BW_Pred(t〇) at time t〇, and set the buffer threshold Buf_Target according to the key frame density at time t〇 as the delay constraint of the monitoring scenario. The buffer threshold Buf_Target is:

[0100] Buf_Target = Buf_Base × (1 + K)

[0101] Where, Buf_Base is the basic buffer threshold, and K is the correction coefficient of the key frame density.

[0102] In the present invention, when the key frame density ≤ 10 frames / second, K = 0.1; when the key frame density > 10 frames / second, K = 0.2. The basic buffer threshold Buf_Base = 2 seconds, which strictly matches the monitoring delay upper limit.

[0103] The present invention can avoid the loss of key frames caused by insufficient buffer by associating the key frame density of monitoring. Among them, the higher the value of the key frame density, the larger the K value, and the higher the value of the buffer threshold Buf_Target, indicating that more buffer data needs to be prepared to meet the key frame transmission requirements to prevent it from being lost due to insufficient buffer.

[0104] S32. When it is detected that the current buffer occupancy Buf_Cur is lower than the buffer threshold Buf_Target (i.e., Buf_Cur < Buf_Target), give priority to ensuring the buffer filling of key frames and temporarily compress non-key frames (the compression rate is increased to 30%) to ensure the priority output of the monitoring screen corresponding to key frames.

[0105] S33, at the next cycle t1, adjust BW_Pred(t0) according to the dynamically adjusted weight of buffer usage rate Buf_Usage and the latest buffer status, and output the predicted bandwidth BW_Pred adapted to the monitoring scenario.

[0106] This invention introduces a keyframe density correction coefficient to prioritize keyframe buffer filling, while differentially compressing non-keyframe background areas when the buffer is insufficient. This satisfies the requirement of monitoring latency ≤1 second and avoids recognition failure caused by keyframe loss.

[0107] The process of buffer dynamic control and output prediction bandwidth is explained below.

[0108] Premise: The system continuously runs a closed loop of data acquisition, prediction, and adjustment at a certain period (e.g., per second).

[0109] 1. Initial prediction and buffer target setting:

[0110] Based on the data collected at time t0, the initial prediction bandwidth BW_Pred(t0) is set by multi-feature fusion.

[0111] Simultaneously, based on the keyframe density at time t0, using the formula...

[0112] Buf_Target = Buf_Base × (1 + K)

[0113] Calculate the target buffer threshold (Buf_Target) that should be maintained at the current time.

[0114] For example, when the keyframe density is high, the K value increases and the Buf_Target increases, indicating that more buffered data needs to be prepared to meet the possible keyframe transmission requirements and prevent them from being lost due to insufficient buffering.

[0115] 2. Buffer status monitoring and dynamic correction of predicted bandwidth:

[0116] At time t1 (the next cycle), data is collected again. At this time, the buffer utilization rate (Buf_Usage), as an important buffer state feature, is input into the multi-feature fusion prediction model.

[0117] According to the dynamic weight adjustment rule described in step S2: "When the delay approaches 2 seconds (the monitoring limit), increase to 20%." Here, "delay" is directly related to the buffer depth. The higher the Buf_Usage, the greater the risk of playback delay.

[0118] Direct feedback from buffer dynamic control: The "buffer dynamic control" rules in step S32 (e.g., prioritizing keyframes when the buffer is insufficient) directly affect network request behavior. For example, when the system prioritizes requesting keyframes, this keyframe data enters the buffer, changing the Buf_Usage and the buffer content structure. These changes are captured in the next data acquisition cycle and, as new Buf_Usage and potential network request pattern features, are input again into the prediction model, affecting the BW_Pred output in the next round.

[0119] 3. Output the predicted bandwidth ultimately used to guide transmission:

[0120] After the above process, the final output predicted bandwidth BW_Pred is a composite value that takes into account the current network conditions, terminal performance, monitoring screen requirements (keyframe density), and the latest buffer status. This BW_Pred value will be sent to the video encoder or flow control module to dynamically adjust the video encoding bitrate or data transmission rate, thereby optimizing network transmission.

[0121] In summary, buffer strategy adjustment achieves real-time dynamic correction of bandwidth prediction output by changing the value of the core feature "buffer usage (Buf_Usage)" and its weight in the prediction model, making it more consistent with the current buffer health status and the real-time requirements of the monitoring scenario. This is a closed-loop feedback control process.

[0122] S4 optimizes the monitoring bandwidth for weak network scenarios, including: first, checking whether the predicted bandwidth or packet loss rate meets the weak network judgment criteria; if it does, then optimizing the monitoring bandwidth for weak network scenarios by extending the key frame interval, lightweighting and compressing non-key frames, and supplementing packet loss.

[0123] First, weak network conditions are determined by monitoring bandwidth thresholds:

[0124] In this invention, for 1080P surveillance video, the weak network judgment criteria are: predicted bandwidth BW_Pred≤1.5Mbps (i.e., 75% of the 1080P base bitrate of 2Mbps) or packet loss rate Loss_Rate>10%.

[0125] It should be noted that the current bandwidth BW_Cur < 2Mbps mentioned above is also a standard for judging weak networks. The BW_Pred predicted bandwidth standard mentioned in this step is a relatively clear and more formal standard for judging weak networks.

[0126] Then, a lightweight optimization strategy without additional bandwidth overhead is adopted to optimize the monitoring bandwidth in weak network scenarios. Specifically, this involves extending the key frame interval, lightweight compression of non-key frames, and compensating for packet loss.

[0127] Extending the keyframe interval includes increasing the keyframe interval from 10 frames to 15 frames in static monitoring scenarios (i.e., when the target motion intensity Motion_Intensity < 30); and keeping the keyframe interval at 10 frames in dynamic monitoring scenarios (e.g., roads), thereby avoiding bandwidth waste caused by indiscriminate redundancy.

[0128] Lightweight compression of non-critical frames includes: employing a differentiated compression method for different areas of the monitored image, increasing the compression rate only for the background area (40%), while maintaining a compression rate of 20% for the target area (people / vehicles). This invention has the capability to detect monitored targets, achieves differentiated compression output for different areas of the monitored image, and identifies the differentiated compressed areas (e.g., target area coordinates).

[0129] Packet loss compensation includes: retransmitting lost key frames using incremental retransmission and compensating for non-key frames using monitoring target interpolation, thereby reducing retransmission bandwidth by 50%.

[0130] Specifically, incremental retransmission is used for key frame packet loss, only retransmitting the lost frame segments; non-key frames use monitoring target interpolation, which is based on the target position of the preceding and following frames, rather than general pixel interpolation.

[0131] This invention designs a strategy for monitoring weak networks (bandwidth <1.5Mbps) without additional bandwidth overhead: extending the key frame interval for static monitoring, differentiating compression for dynamic monitoring areas, and incremental retransmission of key frames.

[0132] S5 allows you to set the priority of monitoring tasks and schedule each monitoring task in a lightweight manner based on the priority.

[0133] It should be noted that priority scheduling, as a resource management mechanism, operates continuously throughout the entire process.

[0134] Specifically, this invention enables terminal resource collaboration and performs load monitoring (monitoring task priority):

[0135] In this invention, the priority of the monitoring task is set as follows: the key frame decoding task is set to high priority (P1); the non-key frame decoding task is set to medium priority (P2); and the rendering preprocessing task is set to low priority (P3).

[0136] Configure resource allocation rules to schedule each monitoring task in a lightweight manner based on priority, including:

[0137] When edge terminal decoding resources are insufficient, high-priority (P1) tasks should not be discarded. Only medium-priority (P2) and low-priority (P3) tasks should be discarded to avoid keyframe decoding delay caused by "discarding without priority".

[0138] When the CPU decoding load rate of a monitoring task is >80% and it is a medium-priority (P2) task, the monitoring task is dropped to avoid failing to distinguish task priorities, which could easily lead to resource contention for core tasks (such as keyframe decoding).

[0139] This invention introduces "monitoring task priority" in terminal resource scheduling for the first time, prohibiting the migration of high-priority (P1) tasks and only allowing the migration of non-core tasks (P2 / P3), thus avoiding the core decoding delay caused by the traditional indiscriminate task migration.

[0140] The network transmission optimization method for video surveillance scenarios of the present invention addresses the core requirements of "real-time performance (latency ≤ 1 second), identifiability (complete keyframes), and lightweight edge terminals" in monitoring scenarios such as residential areas, roads, and parks. By combining network bandwidth prediction and video buffer optimization methods based on terminal device performance and network dynamic characteristics, it can solve the problem of insufficient adaptability of existing technologies in monitoring scenarios and improve the smoothness and real-time performance of monitoring videos in weak network and dynamic network environments.

[0141] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the protection scope of the present invention.

Claims

1. A network transmission optimization method in a video monitoring scene, characterized in that, The method comprises the following steps: S1, collecting multi-dimensional feature data of the monitoring scene, wherein the multi-dimensional feature data comprises terminal performance features, network dynamic features and monitoring picture features; S2, performing multi-feature fusion on the terminal performance features, network dynamic features and monitoring picture features, and dynamically adjusting the weights of the features based on the monitoring scene; S3, dynamically adjusting the buffer strategy of the video monitoring to control the priority of the monitoring picture, and outputting a predicted bandwidth adapted to the monitoring scene, comprising: S31, setting an initial predicted bandwidth BW_Pred(t0) at time t0, and setting a buffer threshold Buf_Target according to the key frame density at time t0 as follows: Buf_Target = Buf_Base * (1 + K); wherein Buf_Base is a basic buffer threshold, and K is a correction coefficient of the key frame density; S32, when it is detected that the current buffer occupancy Buf_Cur is lower than the buffer threshold Buf_Target, preferentially guaranteeing key frame buffer filling and temporarily compressing non-key frames to preferentially output the monitoring picture corresponding to the key frames; S33, at the next period t1, adjusting BW_Pred(t0) according to the dynamically adjusted weight of the buffer usage Buf_Usage and the latest buffer state, and outputting a predicted bandwidth BW_Pred adapted to the monitoring scene; S4, optimizing the monitoring bandwidth of the weak network scene, comprising: first detecting whether the predicted bandwidth or the packet loss rate meets the weak network judgment standard, and if so, using the ways of prolonging the key frame interval, lightweight compressing the non-key frames and supplementing the lost packets to optimize the monitoring bandwidth of the weak network scene; S5, setting the priority of the monitoring task, and performing lightweight scheduling on each monitoring task according to the priority.

2. The network transmission optimization method in a video monitoring scene according to claim 1, characterized in that, In S1, the terminal performance features comprise CPU decoding load rate, GPU video memory occupancy rate and buffer usage rate; The network dynamic features comprise current bandwidth, packet loss rate and bandwidth fluctuation amplitude; The monitoring picture features comprise key frame density and target motion intensity.

3. The network transmission optimization method in a video monitoring scene according to claim 2, characterized in that, In S2, the weights of the features are dynamically adjusted based on the monitoring scene, comprising: When it is detected that the edge terminal type is an edge camera, the weight of the CPU decoding load rate is automatically reduced; When it is detected that the target motion intensity of the moving target is higher than a preset intensity, the moving target is judged as a high motion target, and the weight of the key frame density is automatically increased; When it is detected that the current network is a weak network, the weight of the current bandwidth is automatically increased; When it is detected that the key frame density is higher than a preset frame density, the weight of the packet loss rate is automatically increased; When it is detected that the current monitoring is static, the weight of the target motion intensity is automatically reduced; When it is detected that the video monitoring delay is higher than a preset delay value, the weight of the buffer usage rate is automatically increased.

4. The network transmission optimization method in a video monitoring scene according to claim 3, characterized in that, When it is detected that the number of CPU cores is less than or equal to 4, the edge terminal type is judged as an edge camera; When it is detected that the current bandwidth BW_Cur is less than 2 Mbps, the current network is judged as a weak network.

5. The network transmission optimization method in a video monitoring scene according to claim 1, characterized in that, In S31, when the key frame density is less than or equal to 10 frames per second, K = 0.1; when the key frame density is greater than 10 frames per second, K = 0.

2.

6. The network transmission optimization method in a video monitoring scene according to claim 1, characterized in that, In S32, the higher the value of the key frame density, the greater the K value and the value of the buffer threshold Buf_Target, indicating that more buffer data needs to be prepared to meet the key frame transmission requirements to prevent loss due to insufficient buffer.

7. The network transmission optimization method in a video monitoring scene according to claim 1, characterized in that, In S4, the weak network judgment criteria are: predicted bandwidth BW_Pred≤1.5 Mbps or packet loss rate Loss_Rate>10%.

8. The network transmission optimization method in a video monitoring scene according to claim 1, characterized in that, In S4, The extended key frame interval includes: in a static monitoring scene, the key frame interval is increased from 10 frames to 15 frames; in a dynamic monitoring scene, the key frame interval remains 10 frames. The lightweight compression of non-key frames includes: using a monitoring picture area differential compression method to improve the compression rate of the picture background area, and keeping the compression rate of the target area at 20%. The packet loss compensation includes: using an incremental retransmission method to retransmit the lost key frames, and using a monitoring target interpolation method to compensate for non-key frames, thereby reducing the retransmission bandwidth.

9. The network transmission optimization method in a video monitoring scene according to claim 1, characterized in that, In S5, the priority of the monitoring task is set as: The key frame decoding task is set as high priority, the non-key frame decoding task is set as medium priority, and the rendering preprocessing task is set as low priority.

10. The network transmission optimization method in a video monitoring scene according to claim 1, characterized in that, In S5, the lightweight scheduling of each monitoring task according to the priority includes: When the edge terminal decoding resource is insufficient, high-priority tasks are prohibited from being discarded, and only medium-priority and low-priority tasks are allowed to be discarded; When the CPU decoding load rate of the monitoring task is >80% and the task is a medium-priority task, the monitoring task is discarded.

Images