An AI-based method and system for routing, gatewaying, and camera traffic scheduling

Abstract

This invention provides an AI-based method and system for traffic scheduling of routers, gateways, and cameras in the field of communication technology. The method includes: Step S1, the gateway or router collects data traffic from each network device in the local area network in real time; Step S2, using an AI recognition model to analyze the behavioral characteristics of the data traffic to identify video streams generated by cameras and distinguish their service types; Step S3, based on the service type, current network status, and user policies, generating a dynamic bandwidth allocation scheme through an AI scheduling decision model; Step S4, according to the dynamic bandwidth allocation scheme, sending a bitrate adjustment command to the corresponding camera through a private signaling channel; Step S5, the camera responds to the bitrate adjustment command and dynamically adjusts its own video encoding parameters to achieve coordinated scheduling of network traffic. The advantages of this invention are: in complex multi-device environments, it prioritizes the service quality of critical audio and video services and improves the overall network resource utilization efficiency.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and in particular to an AI-based method and system for scheduling traffic to routers, gateways, and cameras. Background Technology

[0002] With the rapid development of IoT and smart home technologies, the number of devices connected to home and small office networks has surged, especially smart cameras, voice assistants, and mobile terminals, which continuously generate audio and video data streams with high bandwidth and low latency requirements. In such scenarios, home routers or home gateways act as network hubs, undertaking the critical tasks of data forwarding, device management, and internet access.

[0003] However, traditional network traffic management technologies are proving inadequate when faced with complex environments involving multiple concurrent devices and dynamically changing traffic loads. Current mainstream solutions have several limitations: 1. Static bandwidth allocation strategy: Presets a fixed bandwidth limit for specific devices or ports. This method lacks flexibility, cannot adapt to real-time fluctuations in network usage, and is prone to wasting bandwidth resources or failing to meet the needs of critical services during peak traffic periods.

[0004] 2. Protocol- or port-based traffic identification: This method distinguishes application types by identifying TCP / UDP port numbers (e.g., identifying RTSP ports as video streams). However, many applications today (especially those for security and performance reasons) widely use encrypted transmission protocols (such as HTTPS and QUIC), rendering port-based identification methods largely ineffective and unable to accurately determine the true content and business importance of the traffic.

[0005] 3. Centralized Cloud-Based Scheduling: This method reports network status information to a cloud platform for analysis and decision-making. However, it inherently suffers from response latency, is limited by external network connection quality, and performs poorly in real-time scheduling scenarios requiring rapid responses. Furthermore, it exposes users' local network data to the external network, posing a privacy risk.

[0006] 4. Indiscriminate congestion control: When network congestion occurs, the rate of all devices within the local area network is uniformly limited. This "one-size-fits-all" approach can severely impact the service quality of critical services (such as security surveillance video), causing video stuttering, delays, or even interruptions, resulting in a poor user experience.

[0007] In summary, existing technologies generally suffer from problems such as inaccurate traffic identification, delayed scheduling decisions, low resource utilization efficiency, and a lack of effective collaboration among multiple devices. The root cause lies in the fact that traditional methods lack the ability to deeply perceive encrypted traffic, predict network conditions in real time, and make rapid, intelligent local decisions using edge computing capabilities.

[0008] Therefore, how to provide an AI-based method and system for routing traffic to routers, gateways, and cameras to achieve accurate business perception and network status prediction of encrypted traffic locally, and to perform dynamic and differentiated bandwidth scheduling accordingly, so as to prioritize the service quality of critical audio and video services and improve the overall network resource utilization efficiency in complex multi-device environments, has become an urgent technical problem to be solved. Summary of the Invention

[0009] The technical problem to be solved by this invention is to provide an AI-based method and system for scheduling traffic for routers, gateways and cameras, which enables accurate service perception and network status prediction of encrypted traffic locally, and performs dynamic and differentiated bandwidth scheduling accordingly. In complex multi-device environments, this prioritizes the quality of service for critical audio and video services and improves the overall efficiency of network resource utilization.

[0010] In a first aspect, the present invention provides an AI-based method for scheduling traffic to routers, gateways, and cameras, comprising the following steps: Step S1: The gateway or router collects the data traffic of each network device in the local area network in real time; Step S2: Utilize the AI ​​recognition model deployed on the gateway or router to perform behavioral feature analysis on the collected data traffic in order to identify the video stream generated by the camera and distinguish its service type; Step S3: Based on the service type, current network status, and preset user policies, generate a dynamic bandwidth allocation scheme through an AI scheduling decision model deployed on the gateway or router; Step S4: The gateway or router sends a bitrate adjustment command to the corresponding camera through a private signaling channel according to the dynamic bandwidth allocation scheme. Step S5: The camera responds to the bitrate adjustment command and dynamically adjusts its own video encoding parameters to achieve coordinated scheduling of network traffic.

[0011] Furthermore, in step S2, the AI ​​recognition model is a lightweight deep learning model, which identifies the video stream generated by the camera and the service type by analyzing one or more temporal features among the five-tuple, packet length sequence and time interval of the data traffic without parsing the encrypted content of the data traffic. The service types include at least one of real-time preview, video playback, and alarm upload.

[0012] Furthermore, in step S3, the AI ​​scheduling decision model is a reinforcement learning model, and its decision-making basis also includes the prediction results of short-term network traffic demand, so as to achieve forward-looking scheduling. The user policy can be configured through a graphical interface, including setting device priority, time-based policy, maximum bandwidth threshold, minimum bandwidth threshold, and selecting one or more predefined scene modes; The scenario modes include security priority mode and game acceleration mode.

[0013] Furthermore, in step S4, the private signaling channel is constructed based on the CoAP protocol or the MQTT-SN protocol.

[0014] Furthermore, in step S5, the video encoding parameters include one or more of resolution, frame rate, and bit rate; The method further includes step S6: The AI ​​scheduling decision model adaptively optimizes its scheduling strategy based on the user's historical scheduling records and feedback.

[0015] Secondly, the present invention provides an AI-based router, gateway, and camera traffic scheduling system, comprising the following modules: The data traffic acquisition module is used by gateways or routers to collect data traffic from various network devices within the local area network in real time. The behavioral feature analysis module is used to perform behavioral feature analysis on the collected data traffic using an AI recognition model deployed on a gateway or router, in order to identify the video stream generated by the camera and distinguish its service type. The bandwidth allocation scheme generation module is used to generate a dynamic bandwidth allocation scheme based on the service type, current network status, and preset user policies, through an AI scheduling decision model deployed on the gateway or router. A bitrate adjustment instruction sending module is used by a gateway or router to send a bitrate adjustment instruction to the corresponding camera through a private signaling channel according to the dynamic bandwidth allocation scheme. The collaborative scheduling module is used by the camera to respond to the bitrate adjustment command and dynamically adjust its own video encoding parameters in order to achieve collaborative scheduling of network traffic.

[0016] Furthermore, in the behavior feature analysis module, the AI ​​recognition model is a lightweight deep learning model, which identifies the video stream generated by the camera and the service type by analyzing one or more temporal features among the five-tuple, packet length sequence and time interval of the data traffic without parsing the encrypted content of the data traffic. The service types include at least one of real-time preview, video playback, and alarm upload.

[0017] Furthermore, in the bandwidth allocation scheme generation module, the AI ​​scheduling decision model is a reinforcement learning model, and its decision-making basis also includes the prediction results of short-term network traffic demand, so as to achieve forward-looking scheduling. The user policy can be configured through a graphical interface, including setting device priority, time-based policy, maximum bandwidth threshold, minimum bandwidth threshold, and selecting one or more predefined scene modes; The scenario modes include security priority mode and game acceleration mode.

[0018] Furthermore, in the rate adjustment instruction sending module, the private signaling channel is constructed based on the CoAP protocol or the MQTT-SN protocol.

[0019] Furthermore, in the collaborative scheduling module, the video encoding parameters include one or more of resolution, frame rate, and bit rate; The method further includes a model optimization module, used for: The AI ​​scheduling decision model adaptively optimizes its scheduling strategy based on the user's historical scheduling records and feedback.

[0020] The advantages of this invention are: 1. Real-time data traffic from various network devices within the local area network is collected via gateways or routers. AI recognition models are used to analyze the behavioral characteristics of the collected data traffic to identify video streams generated by cameras and distinguish their service types. Then, based on the service type, current network status, and preset user policies, an AI scheduling decision model generates a dynamic bandwidth allocation scheme. According to the dynamic bandwidth allocation scheme, the gateway or router sends a bitrate adjustment command to the corresponding camera via a private signaling channel. The camera responds to the bitrate adjustment command and dynamically adjusts its own video encoding parameters to achieve coordinated network traffic scheduling. In other words, by deploying an AI model locally on the gateway or router, a lightweight deep learning model is first used to analyze the external behaviors of encrypted data traffic, such as the 5-tuple and packet length sequence. The system accurately identifies the video stream and service type of cameras without cracking the content. Then, a reinforcement learning model combines this identification result with real-time network status and user policies to predict short-term traffic and generate a forward-looking dynamic bandwidth allocation scheme. Subsequently, through a private signaling channel based on CoAP / MQTT-SN, a bitrate adjustment command is sent to a specific camera to dynamically adjust its encoding parameters such as resolution and frame rate. This forms a closed-loop control system within the local area network without cloud intervention, enabling accurate service perception and network status prediction of encrypted traffic locally. Based on this, dynamic and differentiated bandwidth scheduling is performed, prioritizing the quality of service for critical audio and video services in complex multi-device environments and improving the overall network resource utilization efficiency.

[0021] 2. By deploying AI recognition models and AI scheduling decision models on gateways or routers, real-time behavioral characteristic analysis and dynamic bandwidth allocation of data traffic within the local area network are achieved. This AI-based scheduling method can accurately identify the service type of camera video streams (such as real-time preview and video playback) and generate optimization schemes based on the current network status and user policies, thereby effectively reducing network congestion, improving bandwidth utilization, and significantly enhancing the automation level and efficiency of network management, meeting the needs of modern smart homes or enterprise networks for efficient resource scheduling.

[0022] 3. The AI ​​recognition model adopts a lightweight deep learning model, which identifies video streams by analyzing time-series features such as the five-tuple and packet length sequence of data traffic. It does not require parsing encrypted content. This design avoids direct access to sensitive user data, reduces the risk of privacy leakage, and complies with increasingly stringent data protection regulations (such as GDPR), improving the security and credibility of the system. It is particularly suitable for privacy-sensitive scenarios such as security cameras.

[0023] 4. User policies can be configured through a graphical interface, including setting device priorities, time-based policies, maximum / minimum bandwidth thresholds, and predefined scenario modes (such as security priority mode). This highly customizable design allows non-professional users to easily adjust scheduling policies to meet personalized needs in different scenarios, enhancing the product's ease of use and market adaptability, thereby improving user satisfaction and stickiness.

[0024] 5. Sending rate adjustment commands to the camera via a private signaling channel based on the CoAP or MQTT-SN protocol. This dedicated channel design ensures the confidentiality and integrity of communication and prevents external interference or eavesdropping. This not only improves the reliability of scheduling command transmission but also reduces the risk of network attacks, making it suitable for IoT environments with high security requirements.

[0025] 6. The AI ​​scheduling decision model adopts reinforcement learning technology, which can adaptively optimize the scheduling strategy based on the user's historical scheduling records and feedback. This self-learning ability enables the system to dynamically adapt to changes in the network environment and user behavior patterns, achieve long-term performance improvement, reduce the need for manual intervention, and demonstrate the technology's forward-looking nature and potential for sustainable improvement.

[0026] 7. The deployment of lightweight deep learning and reinforcement learning models optimizes the use of computing resources, enabling the solution to run efficiently on resource-constrained gateways or routers. This low-overhead design facilitates large-scale deployment and expansion, and is suitable for networks of different sizes, from home to enterprise, with high industrial applicability and cost-effectiveness.

[0027] 8. By collecting and analyzing LAN data traffic in real time, a lightweight AI model is used to identify the service type of camera video streams, and reinforcement learning is combined for dynamic bandwidth allocation, significantly improving the intelligence and efficiency of network scheduling. Meanwhile, its non-intrusive traffic analysis avoids the parsing of encrypted content, enhancing user privacy and compliance. The graphical interface supports personalized policy configurations, such as time-based priorities and scene modes, ensuring a flexible user experience. Furthermore, a private signaling channel ensures secure and reliable command transmission, and the adaptive learning mechanism continuously optimizes scheduling strategies based on historical feedback. Combined with a lightweight design that reduces resource overhead, the solution possesses good scalability and practicality, achieving efficient, secure, and user-friendly traffic management overall. Attached Figure Description

[0028] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0029] Figure 1 This is a flowchart of an AI-based router, gateway, and camera traffic scheduling method according to the present invention.

[0030] Figure 2 This is a schematic diagram of the structure of an AI-based router, gateway, and camera traffic scheduling system according to the present invention. Detailed Implementation

[0031] The technical solution in this application embodiment follows the general idea as follows: By deploying an AI model locally on the gateway or router, the AI ​​recognition model first analyzes the external behavioral characteristics of data traffic, such as the five-tuple and packet length sequence, to accurately identify the video stream and its service type from the camera without cracking the content. Then, the AI ​​scheduling decision model combines this recognition result with real-time network status and user policies to predict short-term traffic and generate a forward-looking dynamic bandwidth allocation scheme. Subsequently, a bitrate adjustment command is sent to a specific camera through a private signaling channel to dynamically adjust its video encoding parameters, thereby forming a closed-loop control system within the local area network that does not require cloud intervention. This enables accurate service perception and network status prediction of encrypted traffic locally, and dynamic, differentiated bandwidth scheduling is performed accordingly. In complex multi-device environments, this prioritizes the quality of service for critical audio and video services and improves the overall network resource utilization efficiency.

[0032] Please refer to Figures 1 to 2 As shown, a preferred embodiment of the AI-based router, gateway, and camera traffic scheduling method of the present invention includes the following steps: Step S1: The gateway or router collects the data traffic of each network device in the local area network in real time; Step S2: Utilize the AI ​​recognition model deployed on the gateway or router to perform behavioral feature analysis on the collected data traffic in order to identify the video stream generated by the camera and distinguish its service type; Step S3: Based on the service type, current network status, and preset user policies, generate a dynamic bandwidth allocation scheme through an AI scheduling decision model deployed on the gateway or router; Step S4: The gateway or router sends a bitrate adjustment command to the corresponding camera through a private signaling channel according to the dynamic bandwidth allocation scheme. Step S5: The camera responds to the bitrate adjustment command and dynamically adjusts its own video encoding parameters to achieve coordinated scheduling of network traffic.

[0033] In step S2, the AI ​​recognition model is a lightweight deep learning model, which identifies the video stream generated by the camera and the service type by analyzing one or more temporal features of the data traffic, such as the five-tuple, packet length sequence and time interval, without parsing the encrypted content of the data traffic. The service types include at least one of real-time preview, video playback, and alarm upload.

[0034] The AI ​​recognition model employs a lightweight deep learning architecture, specifically a temporal feature analysis model based on LSTM (Long Short-Term Memory). This model identifies camera video streams without parsing encrypted content by analyzing temporal features such as the five-tuple (source IP, destination IP, source port, destination port, protocol type), packet length sequence, and time intervals. The model training utilizes a semi-supervised learning method, employing publicly available IoT traffic datasets (such as CIC-IDS2017) and labeled data collected from real-world home network environments.

[0035] In actual deployment, the model input is a continuous 10-second traffic feature sequence, and the output is the probability distribution of camera video streams and the service type classification results. To ensure efficient model operation on resource-constrained gateway devices, model pruning and quantization techniques are used to compress the model parameters to less than 1MB and control the inference latency to less than 50ms. For example, when the model detects that the traffic of a certain IP address exhibits the characteristic combination of "high frequency, small packet length, and fixed interval" (typical characteristics of camera video streams), it determines that it is a camera video stream and further distinguishes it into the service type of "real-time preview" (high frame rate, low latency requirement), "video playback" (low frame rate, large packet length), or "alarm upload" (bursting, small packet length).

[0036] In step S3, the AI ​​scheduling decision model is a reinforcement learning model, and its decision-making basis also includes the prediction results of short-term network traffic demand in order to achieve forward-looking scheduling. The user policy can be configured through a graphical interface, including setting device priority, time-based policy, maximum bandwidth threshold, minimum bandwidth threshold, and selecting one or more predefined scene modes; The scenario modes include security priority mode and game acceleration mode.

[0037] The AI ​​scheduling decision model adopts a reinforcement learning architecture based on DQN (Deep Q-Network). Its state space includes: current network bandwidth utilization, distribution of service types of each camera, real-time network latency, user policy configuration (such as security priority mode), and short-term traffic prediction results (predicting traffic demand in the next 5 minutes through LSTM).

[0038] The reward function is designed as follows: Basic incentive: The degree of improvement in Quality of Service (QoS) for critical services (such as security cameras) based on the bandwidth allocation plan; Penalty: The degree of QoS degradation for non-critical services (such as regular video playback); Scheduling costs: Penalties for adjusting bandwidth frequency and magnitude; In practical applications, when network bandwidth is strained, the system prioritizes real-time preview services for cameras in "security priority mode," limiting the bandwidth of non-critical services (such as gaming devices) to below a minimum threshold. For example, when network bandwidth utilization exceeds 80%, the system automatically reduces the video bitrate of ordinary cameras from 2Mbps to 1Mbps, while maintaining the bitrate of security cameras above 2Mbps to ensure that critical services are not affected.

[0039] The system also supports adaptive optimization, continuously optimizing the decision parameters of the reinforcement learning model by recording user feedback for each scheduling decision (such as users adjusting strategies through a graphical interface). For example, if users frequently increase the bandwidth priority of security cameras at night, the system will automatically learn this pattern and enable the security priority strategy by default during the 22:00-6:00 time period every day.

[0040] In step S4, the private signaling channel is constructed based on the CoAP protocol or the MQTT-SN protocol.

[0041] The private signaling channel is built on the MQTT-SN (MQTT for Sensor Networks) protocol, which is designed for resource-constrained devices and features low bandwidth consumption and low latency. The signaling channel establishment process is as follows: 1. When the gateway device starts up, it broadcasts a discovery request (MQTT-SN DISCONNECT message) to all cameras in the local area network. 2. After receiving the request, the camera responds with an MQTT-SN CONNACK to establish a secure connection; 3. The signaling channel uses TLS 1.2 encryption to ensure the security of command transmission; 4. The command format is JSON structure, which includes parameters such as camera ID, target resolution, target frame rate, and target bit rate; For example, when the system decides to adjust the bitrate of a security camera from 2Mbps to 1.5Mbps, the gateway sends the following command to the camera via the MQTT-SN protocol: { "cmd": "adjust_bitrate", "camera_id": "CAM-001", "target_bitrate": 1500, "valid_duration": 60 } The command can be transmitted within 100ms and occupies only about 50 bytes of bandwidth, which has a negligible impact on the overall traffic of the local area network.

[0042] In step S5, the video encoding parameters include one or more of resolution, frame rate, and bit rate; The method further includes step S6: The AI ​​scheduling decision model adaptively optimizes its scheduling strategy based on the user's historical scheduling records and feedback.

[0043] A preferred embodiment of the AI-based router, gateway, and camera traffic scheduling system of the present invention includes the following modules: The data traffic acquisition module is used by gateways or routers to collect data traffic from various network devices within the local area network in real time. The behavioral feature analysis module is used to perform behavioral feature analysis on the collected data traffic using an AI recognition model deployed on a gateway or router, in order to identify the video stream generated by the camera and distinguish its service type. The bandwidth allocation scheme generation module is used to generate a dynamic bandwidth allocation scheme based on the service type, current network status, and preset user policies, through an AI scheduling decision model deployed on the gateway or router. A bitrate adjustment instruction sending module is used by a gateway or router to send a bitrate adjustment instruction to the corresponding camera through a private signaling channel according to the dynamic bandwidth allocation scheme. The collaborative scheduling module is used by the camera to respond to the bitrate adjustment command and dynamically adjust its own video encoding parameters in order to achieve collaborative scheduling of network traffic.

[0044] In the behavioral feature analysis module, the AI ​​recognition model is a lightweight deep learning model, which identifies the video stream generated by the camera and the service type by analyzing one or more temporal features of the data traffic, such as the five-tuple, packet length sequence, and time interval, without parsing the encrypted content of the data traffic. The service types include at least one of real-time preview, video playback, and alarm upload.

[0045] The AI ​​recognition model employs a lightweight deep learning architecture, specifically a temporal feature analysis model based on LSTM (Long Short-Term Memory). This model identifies camera video streams without parsing encrypted content by analyzing temporal features such as the five-tuple (source IP, destination IP, source port, destination port, protocol type), packet length sequence, and time intervals. The model training utilizes a semi-supervised learning method, employing publicly available IoT traffic datasets (such as CIC-IDS2017) and labeled data collected from real-world home network environments.

[0046] In actual deployment, the model input is a continuous 10-second traffic feature sequence, and the output is the probability distribution of camera video streams and the service type classification results. To ensure efficient model operation on resource-constrained gateway devices, model pruning and quantization techniques are used to compress the model parameters to less than 1MB and control the inference latency to less than 50ms. For example, when the model detects that the traffic of a certain IP address exhibits the characteristic combination of "high frequency, small packet length, and fixed interval" (typical characteristics of camera video streams), it determines that it is a camera video stream and further distinguishes it into the service type of "real-time preview" (high frame rate, low latency requirement), "video playback" (low frame rate, large packet length), or "alarm upload" (bursting, small packet length).

[0047] In the bandwidth allocation scheme generation module, the AI ​​scheduling decision model is a reinforcement learning model, and its decision-making basis also includes the prediction results of short-term network traffic demand in order to achieve forward-looking scheduling. The user policy can be configured through a graphical interface, including setting device priority, time-based policy, maximum bandwidth threshold, minimum bandwidth threshold, and selecting one or more predefined scene modes; The scenario modes include security priority mode and game acceleration mode.

[0048] The AI ​​scheduling decision model adopts a reinforcement learning architecture based on DQN (Deep Q-Network). Its state space includes: current network bandwidth utilization, distribution of service types of each camera, real-time network latency, user policy configuration (such as security priority mode), and short-term traffic prediction results (predicting traffic demand in the next 5 minutes through LSTM).

[0049] The reward function is designed as follows: Basic incentive: The degree of improvement in Quality of Service (QoS) for critical services (such as security cameras) based on the bandwidth allocation plan; Penalty: The degree of QoS degradation for non-critical services (such as regular video playback); Scheduling costs: Penalties for adjusting bandwidth frequency and magnitude; In practical applications, when network bandwidth is strained, the system prioritizes real-time preview services for cameras in "security priority mode," limiting the bandwidth of non-critical services (such as gaming devices) to below a minimum threshold. For example, when network bandwidth utilization exceeds 80%, the system automatically reduces the video bitrate of ordinary cameras from 2Mbps to 1Mbps, while maintaining the bitrate of security cameras above 2Mbps to ensure that critical services are not affected.

[0050] The system also supports adaptive optimization, continuously optimizing the decision parameters of the reinforcement learning model by recording user feedback for each scheduling decision (such as users adjusting strategies through a graphical interface). For example, if users frequently increase the bandwidth priority of security cameras at night, the system will automatically learn this pattern and enable the security priority strategy by default during the 22:00-6:00 time period every day.

[0051] In the rate adjustment instruction sending module, the private signaling channel is constructed based on the CoAP protocol or the MQTT-SN protocol.

[0052] The private signaling channel is built on the MQTT-SN (MQTT for Sensor Networks) protocol, which is designed for resource-constrained devices and features low bandwidth consumption and low latency. The signaling channel establishment process is as follows: 1. When the gateway device starts up, it broadcasts a discovery request (MQTT-SN DISCONNECT message) to all cameras in the local area network. 2. After receiving the request, the camera responds with an MQTT-SN CONNACK to establish a secure connection; 3. The signaling channel uses TLS 1.2 encryption to ensure the security of command transmission; 4. The command format is JSON structure, which includes parameters such as camera ID, target resolution, target frame rate, and target bit rate; For example, when the system decides to adjust the bitrate of a security camera from 2Mbps to 1.5Mbps, the gateway sends the following command to the camera via the MQTT-SN protocol: { "cmd": "adjust_bitrate", "camera_id": "CAM-001", "target_bitrate": 1500, "valid_duration": 60 } The command can be transmitted within 100ms and occupies only about 50 bytes of bandwidth, which has a negligible impact on the overall traffic of the local area network.

[0053] In the collaborative scheduling module, the video encoding parameters include one or more of resolution, frame rate, and bit rate; The method further includes a model optimization module, used for: The AI ​​scheduling decision model adaptively optimizes its scheduling strategy based on the user's historical scheduling records and feedback.

[0054] To facilitate understanding of the present invention, a specific scenario is provided for further illustration: 1. The gateway collects traffic from all devices in real time. The AI ​​recognition model identifies the real-time preview service of 3 security cameras (high frame rate, low latency requirements), the video playback service of 2 ordinary cameras (low frame rate, large packet length), and the high bandwidth requirement service of the game console. 2. The AI ​​scheduling decision model analyzes that the current network bandwidth utilization rate is 75%, and predicts that the bandwidth demand will increase by 30% in the next 10 minutes (because the user is about to start a video conference). 3. Generate a dynamic bandwidth allocation scheme: security cameras maintain 2Mbps, ordinary cameras are reduced to 0.8Mbps, gaming consoles maintain 10Mbps, and video conferencing equipment is allocated 5Mbps; 4. Send a bitrate adjustment command to the ordinary camera through the MQTT-SN signaling channel. The camera responds to the command and reduces the video bitrate from 1Mbps to 0.8Mbps. 5. The system continuously monitors the network status. When it detects that the video conference has ended and the bandwidth demand has decreased, the system automatically restores the bitrate of the ordinary camera to 1Mbps.

[0055] Implementation results: In this scenario, the system successfully ensured the real-time preview service of security cameras (latency <100ms), while reducing the bandwidth usage of ordinary cameras by 20%, freeing up sufficient bandwidth for gaming and video conferencing services, and improving the overall network resource utilization efficiency by about 25%.

[0056] This invention operates in a typical home network environment, and its performance metrics are as follows: AI recognition model recognition accuracy: 95.2% (on the test set); Bandwidth allocation decision response time: <50ms; Resource usage: CPU utilization <15%, memory usage <50MB; Bandwidth utilization improvement: 25-35% on average; QoS guarantee rate for critical services: 98.5% (real-time preview service for security cameras).

[0057] In summary, the advantages of this invention are as follows: 1. Real-time data traffic from various network devices within the local area network is collected via gateways or routers. AI recognition models are used to analyze the behavioral characteristics of the collected data traffic to identify video streams generated by cameras and distinguish their service types. Then, based on the service type, current network status, and preset user policies, an AI scheduling decision model generates a dynamic bandwidth allocation scheme. According to the dynamic bandwidth allocation scheme, the gateway or router sends a bitrate adjustment command to the corresponding camera via a private signaling channel. The camera responds to the bitrate adjustment command and dynamically adjusts its own video encoding parameters to achieve coordinated network traffic scheduling. In other words, by deploying an AI model locally on the gateway or router, a lightweight deep learning model is first used to analyze the external behaviors of encrypted data traffic, such as the 5-tuple and packet length sequence. The system accurately identifies the video stream and service type of cameras without cracking the content. Then, a reinforcement learning model combines this identification result with real-time network status and user policies to predict short-term traffic and generate a forward-looking dynamic bandwidth allocation scheme. Subsequently, through a private signaling channel based on CoAP / MQTT-SN, a bitrate adjustment command is sent to a specific camera to dynamically adjust its encoding parameters such as resolution and frame rate. This forms a closed-loop control system within the local area network without cloud intervention, enabling accurate service perception and network status prediction of encrypted traffic locally. Based on this, dynamic and differentiated bandwidth scheduling is performed, prioritizing the quality of service for critical audio and video services in complex multi-device environments and improving the overall network resource utilization efficiency.

[0058] 2. By deploying AI recognition models and AI scheduling decision models on gateways or routers, real-time behavioral characteristic analysis and dynamic bandwidth allocation of data traffic within the local area network are achieved. This AI-based scheduling method can accurately identify the service type of camera video streams (such as real-time preview and video playback) and generate optimization schemes based on the current network status and user policies, thereby effectively reducing network congestion, improving bandwidth utilization, and significantly enhancing the automation level and efficiency of network management, meeting the needs of modern smart homes or enterprise networks for efficient resource scheduling.

[0059] 3. The AI ​​recognition model adopts a lightweight deep learning model, which identifies video streams by analyzing time-series features such as the five-tuple and packet length sequence of data traffic. It does not require parsing encrypted content. This design avoids direct access to sensitive user data, reduces the risk of privacy leakage, and complies with increasingly stringent data protection regulations (such as GDPR), improving the security and credibility of the system. It is particularly suitable for privacy-sensitive scenarios such as security cameras.

[0060] 4. User policies can be configured through a graphical interface, including setting device priorities, time-based policies, maximum / minimum bandwidth thresholds, and predefined scenario modes (such as security priority mode). This highly customizable design allows non-professional users to easily adjust scheduling policies to meet personalized needs in different scenarios, enhancing the product's ease of use and market adaptability, thereby improving user satisfaction and stickiness.

[0061] 5. Sending rate adjustment commands to the camera via a private signaling channel based on the CoAP or MQTT-SN protocol. This dedicated channel design ensures the confidentiality and integrity of communication and prevents external interference or eavesdropping. This not only improves the reliability of scheduling command transmission but also reduces the risk of network attacks, making it suitable for IoT environments with high security requirements.

[0062] 6. The AI ​​scheduling decision model adopts reinforcement learning technology, which can adaptively optimize the scheduling strategy based on the user's historical scheduling records and feedback. This self-learning ability enables the system to dynamically adapt to changes in the network environment and user behavior patterns, achieve long-term performance improvement, reduce the need for manual intervention, and demonstrate the technology's forward-looking nature and potential for sustainable improvement.

[0063] 7. The deployment of lightweight deep learning and reinforcement learning models optimizes the use of computing resources, enabling the solution to run efficiently on resource-constrained gateways or routers. This low-overhead design facilitates large-scale deployment and expansion, and is suitable for networks of different sizes, from home to enterprise, with high industrial applicability and cost-effectiveness.

[0064] 8. By collecting and analyzing LAN data traffic in real time, a lightweight AI model is used to identify the service type of camera video streams, and reinforcement learning is combined for dynamic bandwidth allocation, significantly improving the intelligence and efficiency of network scheduling. Meanwhile, its non-intrusive traffic analysis avoids the parsing of encrypted content, enhancing user privacy and compliance. The graphical interface supports personalized policy configurations, such as time-based priorities and scene modes, ensuring a flexible user experience. Furthermore, a private signaling channel ensures secure and reliable command transmission, and the adaptive learning mechanism continuously optimizes scheduling strategies based on historical feedback. Combined with a lightweight design that reduces resource overhead, the solution possesses good scalability and practicality, achieving efficient, secure, and user-friendly traffic management overall.

[0065] While specific embodiments of the present invention have been described above, those skilled in the art should understand that the specific embodiments described are merely illustrative and not intended to limit the scope of the present invention. Equivalent modifications and variations made by those skilled in the art in accordance with the spirit of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. An AI-based method for scheduling traffic between routers, gateways, and cameras, characterized in that: Includes the following steps: Step S1: The gateway or router collects the data traffic of each network device in the local area network in real time; Step S2: Utilize the AI ​​recognition model deployed on the gateway or router to perform behavioral feature analysis on the collected data traffic in order to identify the video stream generated by the camera and distinguish its service type; Step S3: Based on the service type, current network status, and preset user policies, generate a dynamic bandwidth allocation scheme through an AI scheduling decision model deployed on the gateway or router; Step S4: The gateway or router sends a bitrate adjustment command to the corresponding camera through a private signaling channel according to the dynamic bandwidth allocation scheme. Step S5: The camera responds to the bitrate adjustment command and dynamically adjusts its own video encoding parameters to achieve coordinated scheduling of network traffic.

2. The AI-based router, gateway, and camera traffic scheduling method as described in claim 1, characterized in that: In step S2, the AI ​​recognition model is a lightweight deep learning model, which identifies the video stream generated by the camera and the service type by analyzing one or more temporal features of the data traffic, such as the five-tuple, packet length sequence and time interval, without parsing the encrypted content of the data traffic. The service types include at least one of real-time preview, video playback, and alarm upload.

3. The AI-based router, gateway, and camera traffic scheduling method as described in claim 1, characterized in that: In step S3, the AI ​​scheduling decision model is a reinforcement learning model, and its decision-making basis also includes the prediction results of short-term network traffic demand in order to achieve forward-looking scheduling. The user policy can be configured through a graphical interface, including setting device priority, time-based policy, maximum bandwidth threshold, minimum bandwidth threshold, and selecting one or more predefined scene modes; The scenario modes include security priority mode and game acceleration mode.

4. The AI-based router, gateway, and camera traffic scheduling method as described in claim 1, characterized in that: In step S4, the private signaling channel is constructed based on the CoAP protocol or the MQTT-SN protocol.

5. The AI-based router, gateway, and camera traffic scheduling method as described in claim 1, characterized in that: In step S5, the video encoding parameters include one or more of resolution, frame rate, and bit rate; The method further includes step S6: The AI ​​scheduling decision model adaptively optimizes its scheduling strategy based on the user's historical scheduling records and feedback.

6. An AI-based router, gateway, and camera traffic scheduling system, characterized in that: Includes the following modules: The data traffic acquisition module is used by gateways or routers to collect data traffic from various network devices within the local area network in real time. The behavioral feature analysis module is used to perform behavioral feature analysis on the collected data traffic using an AI recognition model deployed on a gateway or router, in order to identify the video stream generated by the camera and distinguish its service type. The bandwidth allocation scheme generation module is used to generate a dynamic bandwidth allocation scheme based on the service type, current network status, and preset user policies, through an AI scheduling decision model deployed on the gateway or router. A bitrate adjustment instruction sending module is used by a gateway or router to send a bitrate adjustment instruction to the corresponding camera through a private signaling channel according to the dynamic bandwidth allocation scheme. The collaborative scheduling module is used by the camera to respond to the bitrate adjustment command and dynamically adjust its own video encoding parameters in order to achieve collaborative scheduling of network traffic.

7. The AI-based router, gateway, and camera traffic scheduling system as described in claim 6, characterized in that: In the behavioral feature analysis module, the AI ​​recognition model is a lightweight deep learning model, which identifies the video stream generated by the camera and the service type by analyzing one or more temporal features of the data traffic, such as the five-tuple, packet length sequence, and time interval, without parsing the encrypted content of the data traffic. The service types include at least one of real-time preview, video playback, and alarm upload.

8. The AI-based router, gateway, and camera traffic scheduling system as described in claim 6, characterized in that: In the bandwidth allocation scheme generation module, the AI ​​scheduling decision model is a reinforcement learning model, and its decision-making basis also includes the prediction results of short-term network traffic demand in order to achieve forward-looking scheduling. The user policy can be configured through a graphical interface, including setting device priority, time-based policy, maximum bandwidth threshold, minimum bandwidth threshold, and selecting one or more predefined scene modes; The scenario modes include security priority mode and game acceleration mode.

9. The AI-based router, gateway, and camera traffic scheduling system as described in claim 6, characterized in that: In the rate adjustment instruction sending module, the private signaling channel is constructed based on the CoAP protocol or the MQTT-SN protocol.

10. The AI-based router, gateway, and camera traffic scheduling system as described in claim 6, characterized in that: In the collaborative scheduling module, the video encoding parameters include one or more of resolution, frame rate, and bit rate; The method further includes a model optimization module, used for: The AI ​​scheduling decision model adaptively optimizes its scheduling strategy based on the user's historical scheduling records and feedback.

Images