Traffic incident secondary identification method based on video cloud platform pull flow detection

Abstract

The invention relates to the technical field of intelligent traffic, and discloses a traffic event secondary identification method based on video cloud platform pull stream detection, which comprises the following steps: S1, pulling low-code stream video data at a cloud end, carrying out low-code stream polling event detection through a traffic event identification system, and outputting a preliminary detection result; s2, if the preliminary detection result is a traffic event, pulling high-code-stream video data corresponding to the low-code-stream video data, performing high-code-stream event recheck detection through a traffic event recognition system, and outputting a final recognition result; according to the operation of the traffic incident recognition system, cloud resources are shared through dynamic scheduling of cloud computing power. According to the invention, through low code stream polling event detection and high code stream event recheck detection of the traffic event, the bandwidth resource consumption is reduced, the investment of network resources and computing resources can be effectively saved, and the recognition accuracy can be ensured under the support of an event detection algorithm based on a video analysis technology.

Description

Technical Field

[0001] This application relates to the field of intelligent transportation technology, and specifically to a traffic event secondary recognition method based on video cloud platform pull stream detection. Background Technique

[0002] Currently, most provinces build video cloud platforms using the architecture of "road section - provincial cloud platform - ministerial cloud platform", that is, build road section video monitoring systems through local deployment, and build provincial cloud platforms through cloud deployment.

[0003] In addition, to meet the requirements of traffic event recognition accuracy, the traffic event recognition system uses a single recognition mode for event detection, that is, usually directly pulls videos with a bitrate of 1Mbps or above for detection, which requires a large amount of computing resources and network resources to support.

[0004] However, the single recognition mode pulls full - volume and real - time video data with a bitrate of no less than 1Mbps from the system or platform where the data source is located to the traffic event recognition system for detection through protocols such as RTMP, HTTP - HLS, and FLV. Usually, most of the monitoring video detection results show that there are no traffic events, and at the same time, a large amount of resource overhead is generated, that is, a large amount of indispensable but not completely necessary resources are consumed, and the cloud network resources are not utilized efficiently. Summary of the Invention

[0005] The purpose of this application is to provide a traffic event secondary recognition method based on video cloud platform pull stream detection. Based on the secondary recognition technology of low - bitrate round - robin event detection and high - bitrate event review detection, it can improve the event recognition accuracy while making full use of resources.

[0006] To achieve the above - mentioned purpose, this application discloses the following technical solutions: A traffic event secondary recognition method based on video cloud platform pull stream detection, the method includes the following steps: S1 - Pull low - bitrate video data in the cloud, perform low - bitrate round - robin event detection through the traffic event recognition system, and output preliminary detection results; S2 - If the preliminary detection result is a traffic event, pull high - bitrate video data corresponding to the low - bitrate video data, perform high - bitrate event review detection through the traffic event recognition system, and output the final recognition result; Among them, the operation of the traffic event recognition system shares cloud resources through dynamic scheduling of cloud computing power.

[0007] Preferably, the low bitrate is less than 1Mbps; the high bitrate is higher than 1Mbps.

[0008] Preferably, the network framework of the traffic event recognition system includes: The provincial video cloud platform deployed in the cloud is configured to communicate with the local business system through IPsec VPN or SSL VPN; The network exit is configured to use a firewall and a router as the exit devices, with security policies and routing policies; The core layer is configured to achieve high-speed data transmission through a core switch; The access layer is configured to connect application servers, data servers, and distributed storage devices through access switches.

[0009] Preferably, the network framework of the traffic event recognition system further includes: A traffic load balancing module is provided between the network exit and the core layer, and the traffic load balancing module is configured to dynamically adjust the data transmission path according to the real-time traffic.

[0010] Preferably, the dynamic adjustment of the data transmission path according to the real-time traffic specifically includes: D1 - Real-time monitoring of the traffic load rate of the network exit and the traffic load rate of the core layer; D2 - Calculating the difference between the traffic load rate of the network exit and the traffic load rate of the core layer; D3 - Dynamically allocating the path weight of the transmission path according to the difference and the network maximum theoretical load rate.

[0011] Preferably, the network maximum theoretical load rate is calculated through the following formula: Wherein, is the total bandwidth capacity of the network exit and the core layer; is the network peak traffic demand; is the network transmission delay time; is the task execution cycle; The difference is specifically: ; Wherein, is the traffic load rate of the network exit, is the traffic load rate of the core layer; The path weight is specifically: ; Wherein, is the sensitivity coefficient, is the threshold parameter.

[0012] Preferably, the dynamic scheduling of the cloud computing power specifically includes: Reducing network bandwidth consumption through cloud data docking, and dynamically allocating computing resources according to the computing power requirements of the low-bitstream round-robin event detection and the high-bitstream event review detection; Adjust the allocation ratio of computing resources based on the urgency of traffic events.

[0013] Preferably, the dynamic allocation of computing resources specifically includes: According to the number of video channels detected by the low-bitstream round-robin event and the corresponding bitstream , and the number of video channels detected by the high-bitstream event review and the corresponding bitstream , dynamically calculate the total computing power requirement ; The total computing power requirement is calculated by the following formula: where, is the computing power coefficient per unit bitstream of the low-bitstream video; is the computing power coefficient per unit bitstream of the high-bitstream video; is the actual bitstream value of the i-th low-bitstream video; is the actual bitstream value of the j-th high-bitstream video; and are non-linear adjustment exponents; is the priority weighting factor of the j-th high-bitstream video.

[0014] Preferably, the computing power coefficient per unit bitstream of the low-bitstream video and the computing power coefficient per unit bitstream of the high-bitstream video are based on the data of historical detection tasks, and the average detection time of the low-bitstream video and the average detection time of the high-bitstream video are obtained, specifically: where, is the average bitstream value of the low-bitstream video; is the average bitstream value of the high-bitstream video; and are correction coefficients; is the dynamic compensation factor.

[0015] Preferably, the value range of the correction coefficient is: , , and in one dynamic allocation of computing resources, the value satisfies .

[0016] Beneficial effects: The traffic event secondary recognition method based on pull stream detection of a video cloud platform according to the present application pulls low-bitrate video data in the cloud for low-bitrate round-robin event detection. When the preliminary detection result is a traffic event, the corresponding high-bitrate video data is pulled for high-bitrate event review detection, and the final recognition result is output, reducing the consumption of bandwidth resources, effectively saving the investment in network resources and computing resources. Moreover, with the support of the event detection algorithm based on video analysis technology, the recognition accuracy can be ensured. Brief Description of the Drawings

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present application. For those skilled in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0018] Figure 1 It is a flowchart of the traffic event secondary recognition method based on pull stream detection of a video cloud platform provided by an embodiment of the present application. Detailed Embodiments

[0019] The following will clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present application.

[0020] In this article, the term "including" is intended to cover non-exclusive inclusion, so that a process, method, article, or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such a process, method, article, or device. Without more limitations, the elements defined by the statement "including..." do not exclude the existence of additional identical elements in the process, method, article, or device including the said elements.

[0021] Taking the traffic event recognition system launched on the provincial cloud platform of a certain province as an example, a total of 10,096 monitoring videos were accessed. According to the single recognition mode of the existing technology, if event recognition of full-scale and real-time videos is carried out with a 1 Mbps bitrate, the total bandwidth requirement for the traffic event recognition system to pull the stream is approximately 9.86 G; calculated with a 70% bandwidth utilization rate, a 14.09 G bandwidth needs to be built. According to the secondary recognition technology, if 1,200 videos with a 128 kbps bitrate are polled once and 60 videos with a 1 Mbps bitrate are rechecked once, the total bandwidth requirement for the traffic event recognition system to pull the stream is approximately 0.21 G; calculated with a 70% bandwidth utilization rate, a 0.30 G bandwidth needs to be built.

[0022] Secondly, the recognition effect of the event detection algorithm based on video analysis technology depends to a great extent on the clarity of the picture. At the same time, the clarity of the video picture is directly affected by factors such as video bitrate and resolution. Therefore, the recognition effect of the event detection algorithm will be affected by video bitrate and resolution.

[0023] Video bitrate refers to the data traffic used by a video file per unit time, usually measured in kbps or Mbps. The size of the bitrate directly determines the clarity of the video and the size of the video file. Video resolution represents the number of pixels in the video picture, which affects the clarity and detail performance of the video. At the same video resolution, the larger the bitrate of the video file, the smaller the compression ratio, the closer the decoded video file is to the source file, the better the image quality, and the clearer the picture. Therefore, the recognition accuracy of the event detection algorithm is positively correlated with the bitrate size.

[0024] To obtain a clear picture and improve the recognition accuracy of the algorithm, the existing technology usually uses videos with a high bitrate (1 Mbps or above) for event detection, that is, full-scale and real-time high-bitrate videos are used for event recognition, resulting in a large amount of resource overhead. To maximize the utilization of resources, the present invention has studied the event recognition accuracy under different video bitrate conditions to explore an event recognition mode that can balance event recognition accuracy and resource consumption. By performing traffic event recognition on 3,000 videos with different bitrates (including 13,000 event samples), it is found that the recognition accuracy of 32 K bitrate videos is 60%, the recognition accuracy of 128 K bitrate videos is 85%, and the recognition accuracy of 1 M bitrate videos is 95%, indicating that the event recognition accuracy increases with the increase of video bitrate.

[0025] Therefore, to meet the requirement of traffic event recognition accuracy ≥ 85% and maximize the utilization of resources at the same time, this embodiment proposes a traffic event secondary recognition technology of "low-bitrate polling event detection + high-bitrate event recheck detection". Specifically, this embodiment discloses as Figure 1A traffic event secondary recognition method based on pulling stream detection of a video cloud platform, the method comprising the following steps: S1 - Pull video data with a low bitrate (below 1 Mbps) from the cloud, and perform low bitrate round-robin event detection through a traffic event recognition system to output a preliminary detection result; S2 - If the preliminary detection result is a traffic event, pull video data with a high bitrate (greater than or equal to 1 Mbps) corresponding to the low bitrate video data, and perform high bitrate event review detection through a traffic event recognition system to output a final recognition result; Pulling of low bitrate / high bitrate video can be implemented based on existing video streaming protocols (such as RTSP, RTMP), and the bandwidth occupancy can be controlled by adjusting the bitrate resolution (such as switching from 480p to 1080p).

[0026] Among them, the operation of the traffic event recognition system dynamically schedules and shares cloud resources through cloud computing power. The traffic event recognition system is built in a cloud deployment manner and is deployed on the same cloud as the cloud platform. For the traffic event detection requirements of a video cloud platform with video sources in the cloud, a traffic event recognition system is adopted, combined with traffic event secondary recognition technology, and through the dynamic scheduling of cloud computing power, the effective utilization of computing resources is achieved, and through cloud data docking, the network resource investment is reduced and the construction cost is saved.

[0027] The results of a specific application show that in terms of network resources, the bandwidth resources required by the secondary recognition technology are approximately 2.13% of that of the single recognition mode. In terms of computing resources, the computing power resources are calculated according to the single-channel 128 kbps bitrate video detection computing power of 3 TFLOPS (FP16) and the single-channel 1 Mbps bitrate video detection computing power of 9 TFLOPS (FP16). The computing power required for a single round of detection in the single recognition mode is 90864 TFLOPS; the maximum computing power required for a single round of the secondary recognition technology is 4140 TFLOPS; the computing resources required by the secondary recognition technology are approximately 4.56% of that of the single recognition mode. Compared with the single recognition mode, the secondary recognition technology can effectively save the investment in network resources and computing resources.

[0028] Therefore, based on the above, the traffic event secondary recognition method based on pulling stream detection of a video cloud platform in this embodiment pulls low bitrate video data from the cloud for low bitrate round-robin event detection, filters videos without events, and when the preliminary detection result is a traffic event, pulls the corresponding high bitrate video data for high bitrate event review detection and outputs a final recognition result, realizing high bitrate review only for suspected events, reducing the consumption of bandwidth resources, effectively saving the investment in network resources and computing resources, and moreover, with the support of an event detection algorithm based on video analysis technology, the recognition accuracy can be ensured.

[0029] In one embodiment, the network framework of the traffic event recognition system includes: A provincial video cloud platform deployed in the cloud, configured to communicate with the local business system through methods such as IPsec VPN or SSL VPN; A network exit, mainly meeting the north-south communication of the internal and external networks of the local area network, configured to: use a firewall and a router as exit devices, and configure security policies, routing policies, etc. as needed to ensure network security; A core layer, responsible for high-speed and efficient data and traffic transmission between various parts of the network, configured to: achieve high-speed data transmission through a core switch, that is, provide high-speed forwarding for traffic packets entering and leaving the local area network; An access layer, responsible for connecting devices such as application servers, data servers, and distributed storage to the local area network, configured to: connect application servers, data servers, and distributed storage devices through an access switch, that is, business systems such as operation monitoring and video monitoring are deployed on servers in the local computer room and are connected to the local area network through the access switch.

[0030] Furthermore, in this embodiment, the network framework of the traffic event recognition system further includes: A traffic load balancing module is set between the network exit and the core layer, and the traffic load balancing module is configured to: dynamically adjust the data transmission path according to the real-time traffic. Specifically, the dynamically adjusting the data transmission path according to the real-time traffic includes: D1 - Real-time monitoring of the traffic load rate of the network exit and the traffic load rate of the core layer; D2 - Calculating the difference between the traffic load rate of the network exit and the traffic load rate of the core layer; D3 - Dynamically allocating the path weight of the transmission path according to the difference and the network maximum theoretical load rate.

[0031] Among them, the difference is specifically: ; is the traffic load rate of the network exit, is the traffic load rate of the core layer.

[0032] The network maximum theoretical load rate is used to quantify the upper limit of network load and is calculated by the following formula: , is the total bandwidth capacity of the network exit and the core layer, in Gbps, obtained from the network device specification, such as 10 Gbps; is the network peak traffic demand, in Gbps; is the network transmission delay time, in seconds, obtained through Ping test or network monitoring tools, such as 50 ms; is the task execution period, in seconds, set according to the detection task frequency, such as 10 seconds. This formula quantifies the impact of delay on network load capacity by introducing network transmission delay time, providing a more accurate theoretical basis for load balancing.

[0033] The specific path weight is as follows: ; is the sensitivity coefficient, used to control the steepness of weight adjustment, adjusted through experiments to control the steepness of weight adjustment, for example ; is the threshold parameter, used to set the reference value of network load difference, such as set to 50% of Lmax, that is ; The value range of is 0 to 1, used to represent the path priority.

[0034] This formula realizes a smooth transition of weight adjustment, avoids the path switching jitter caused by traditional linear adjustment, and improves the stability of network transmission. And combined with the upper limit constraint of the network maximum theoretical load rate, dynamically adjusts the path weight to reduce the risk of network congestion.

[0035] Feasibly, in this embodiment, the network peak traffic demand is calculated by the following formula: Wherein, is the total number of video streams; is the bitstream value of the k-th video, in Mbps; is the activity coefficient of the k-th video, with a value range of 0 to 1, indicating the activity probability of this video stream within a unit time; is the fluctuation amplitude coefficient, used to characterize the periodic fluctuation of network load, fitted through historical traffic data analysis; is the fluctuation frequency, in rad / s, fitted through historical traffic data analysis; is the attenuation coefficient, used to characterize the attenuation characteristic of the fluctuation over time, fitted through historical traffic data analysis; is the current time, in seconds. This formula dynamically predicts the network peak traffic demand by introducing a periodic fluctuation function and an attenuation term, improving the adaptability to the periodic changes of network load.

[0036] Further, the activity coefficient of the k-th video is calculated by the following formula: Wherein, is the active duration of the k-th video within a unit time, in seconds; is the total duration of a unit time, in seconds; is the activity fluctuation coefficient, which is used to characterize the periodic change of the active state of the video stream and is fitted through the historical active data of the video stream; is the phase angle of the k-th video, with the unit of rad, and is fitted through the historical active data of the video stream. This formula further optimizes the dynamic scheduling strategy of network resources and improves the adaptability to the periodic change of the active state of the video stream by introducing a periodic fluctuation function.

[0037] In one implementation, the dynamic scheduling of the cloud computing power specifically includes: Reducing network bandwidth consumption through cloud data docking, and dynamically allocating computing resources according to the computing power requirements of the low-bitrate round-robin event detection and the high-bitrate event review detection; Adjusting the allocation ratio of computing resources based on the urgency of traffic events.

[0038] Specifically, adjusting the allocation ratio according to the urgency includes the following steps: P1 - Priority division: Defining the low-bitrate round-robin event detection as a low-priority task and the high-bitrate event review detection as a high-priority task. Among them, the goal of the low-priority task is to perform preliminary event detection through low-bitrate (<1Mbps) videos, covering all video channels, but only marking suspected events; the computing power requirement is that the computing power requirement for single-channel low-bitrate video detection is relatively low; the characteristics are that the task volume is large (covering all video channels), but the resource consumption of a single-channel task is small and high-precision identification is not required. The goal of the high-priority task is to pull high-bitrate (≥1Mbps) videos for review of the videos with suspected events in the low-bitrate round-robin event detection and output the final identification result; the computing power requirement is that the computing power requirement for single-channel high-bitrate video detection is relatively high; the characteristics are that the task volume is small (only for suspected events), but the resource consumption of a single-channel task is large and high-precision identification is required.

[0039] P2 - Priority-driven allocation: High-priority tasks preferentially occupy computing power resources to ensure that their computing power requirements are met; low-priority tasks are allocated as needed from the remaining computing power.

[0040] Furthermore, the dynamic allocation of computing resources specifically includes: According to the number of video channels of the low-bitrate round-robin event detection and the corresponding bitrate , and the number of video channels of the high-bitrate event review detection and the corresponding bitrate , dynamically calculate the total computing power requirement ; the total computing power requirement is calculated through the following formula: Wherein, is the computing power coefficient per unit bitrate for low-bitrate video, with the unit of TFLOPS / (Mbps); is the computing power coefficient per unit bitrate for high-bitrate video, with the unit of TFLOPS / (Mbps); is the actual bitrate value of the i-th low-bitrate video, with the unit of Mbps; is the actual bitrate value of the j-th high-bitrate video, with the unit of Mbps; and are the non-linear adjustment exponents, which are respectively used for correcting the computing power requirements of low-bitrate and high-bitrate videos, and fitting the computing power requirement curves under different bitrate scenarios through experiments; is the priority weighting factor of the j-th high-bitrate video, with the range of 0 to 1, which is dynamically adjusted according to the urgency of the task. For example, it is set to 0.9 for accident detection and 0.5 for illegal parking detection. By introducing the non-linear adjustment exponent and , this formula realizes the dynamic non-linear modeling of the computing power requirements. For example, when the computing power requirements of low-bitrate tasks increase non-linearly with the increase of bitrate, it improves the adaptability to extreme bitrate scenarios, while the priority weighting factor ensures that high-urgency tasks obtain computing power first and avoids resource waste.

[0041] When adjusting the allocation ratio of computing resources, the computing power allocation ratio is dynamically adjusted according to the real-time task queue (the number of tasks of low-bitrate round-robin event detection and high-bitrate event review detection). For example, when the number of high-bitrate event review detection tasks surges, the system automatically increases the computing power allocation ratio of high-bitrate tasks (specifically, by increasing the value of ), and reduces the concurrent number of low-bitrate round-robin event detections.

[0042] Furthermore, the computing power coefficient per unit bitrate of the low-bitrate video and the computing power coefficient per unit bitrate of the high-bitrate video are obtained by statistically calculating the average detection time of the low-bitrate video and the average detection time of the high-bitrate video based on the data of historical detection tasks. Specifically: Among them, is the average bitrate value of the low-bitrate video, with the unit of Mbps; is the average bitrate value of the high-bitrate video, with the unit of Mbps; and are correction coefficients, which are used to compensate for the algorithm complexity differences of different bitrate videos. Generally, they are obtained through experimental verification. It is feasible that the value range of the correction coefficient is: , , and in one dynamic allocation of computing resources, the value satisfies ; is a dynamic compensation factor, which is used to balance the relationship between detection time consumption and computing power allocation, and is adjusted according to the change of algorithm complexity, such as . Through the introduction of the dynamic compensation factor, this formula realizes the adaptive correction of computing power allocation, balances the detection time consumption and computing power allocation, and improves the adaptability to the change of algorithm complexity. Secondly, the correction coefficient is verified through experiments to ensure that the computing power allocation of low-bitrate videos will not be underestimated due to algorithm complexity differences, and at the same time avoid overestimating the computing power requirements of high-bitrate videos.

[0043] In summary, the traffic event secondary recognition method based on pull stream detection of the video cloud platform in this embodiment only needs to perform high-bitrate event review detection on videos with the initial inspection result of traffic events, which can greatly reduce the computing resource overhead. Moreover, through a series of matching designs such as dynamic computing power allocation, traffic load balancing, and priority-driven scheduling, the reasonable allocation, effective utilization, high-efficiency response, and smooth network are ensured.

[0044] In the embodiments provided in this application, it should be understood that the embodiments described here can be implemented in hardware, software, firmware, middleware, code, or any appropriate combination thereof. For hardware implementation, the processor can be implemented in one or more of the following units: application specific integrated circuit (ASIC), digital signal processor (DSP), digital signal processing device (DSPD), programmable logic device (PLD), field programmable gate array (FPGA), processor, controller, microcontroller, microprocessor, other electronic units designed to implement the functions described here, or a combination thereof. For software implementation, part or all of the processes of the embodiments can be completed by a computer program instructing the relevant hardware. When implemented, the above program can be stored in a computer-readable storage medium or transmitted as one or more instructions or codes on the computer-readable storage medium. The computer-readable storage medium includes computer storage media and communication media, where the communication media includes any medium that facilitates the transmission of a computer program from one place to another. The storage media can be any available medium that can be accessed by a computer. The computer-readable storage medium can include, but is not limited to, RAM, ROM, EEPROM, CD-ROM or other optical disc storage, magnetic disk storage media or other magnetic storage devices, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer.

[0045] Finally, it should be noted that the above are only the preferred embodiments of the present application and are not intended to limit the present application. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for secondary recognition of traffic events based on pulling stream detection of a video cloud platform, characterized in that, The method includes the following steps: S1 - Pull low-bitrate video data in the cloud, perform low-bitrate round-robin event detection through a traffic event recognition system, and output a preliminary detection result; S2 - If the preliminary detection result is a traffic event, pull the high-bitrate video data corresponding to the low-bitrate video data, perform high-bitrate event review detection through the traffic event recognition system, and output a final recognition result; Among them, the operation of the traffic event recognition system dynamically schedules and shares cloud resources through cloud computing power.

2. The traffic event secondary recognition method based on pull stream detection of a video cloud platform according to claim 1, wherein The low bitrate is lower than 1 Mbps; the high bitrate is higher than 1 Mbps.

3. The traffic event secondary recognition method based on pull stream detection of a video cloud platform according to claim 1, characterized in that The network framework of the traffic event recognition system includes: The provincial video cloud platform deployed in the cloud, configured to communicate with the local business system through IPsec VPN or SSL VPN; The network exit, configured to use a firewall and a router as the exit devices, with security policies and routing policies; The core layer, configured to achieve high-speed data transmission through a core switch; The access layer, configured to connect to application servers, data servers, and distributed storage devices through access switches.

4. The traffic event secondary recognition method based on pull stream detection of a video cloud platform according to claim 3, wherein The network framework of the traffic event recognition system further includes: A traffic load balancing module is set between the network exit and the core layer, and the traffic load balancing module is configured to: dynamically adjust the data transmission path according to the real-time traffic.

5. The traffic event secondary recognition method based on pull stream detection of a video cloud platform according to claim 4, wherein The dynamically adjusting the data transmission path according to the real-time traffic specifically includes: D1 - Real-time monitor the traffic load rate of the network exit and the traffic load rate of the core layer; D2 - Calculate the difference between the traffic load rate of the network exit and the traffic load rate of the core layer; D3 - Dynamically allocate the path weight of the transmission path according to the difference and the network maximum theoretical load rate.

6. The traffic event secondary recognition method based on pull stream detection of a video cloud platform according to claim 5, wherein The network maximum theoretical load rate is calculated through the following formula: Among them, is the total bandwidth capacity of the network exit and the core layer; is the network peak traffic demand; is the network transmission delay time; is the task execution cycle; The specific difference is as follows: ; Among them, is the traffic load rate of the network exit, is the traffic load rate of the core layer; The specific path weight is as follows: ; Among them, is the sensitivity coefficient; is the threshold parameter.

7. The traffic event secondary recognition method based on pull stream detection of a video cloud platform according to claim 1, wherein The dynamic scheduling of the cloud computing power specifically includes: Reduce network bandwidth consumption through cloud data docking, and dynamically allocate computing resources according to the computing power requirements of the low-bitrate round-robin event detection and the high-bitrate event review detection; Adjust the allocation ratio of computing resources based on the urgency of traffic events.

8. The traffic event secondary recognition method based on pull stream detection of a video cloud platform according to claim 7, characterized in that The dynamically allocating computing resources specifically includes: The number of video channels for low-bitrate round-robin event detection and the corresponding bitrate , and the number of video channels for high-bitrate event review and detection and the corresponding bitrate , dynamically calculate the total computing power requirement ; The total computing power requirement is calculated by the following formula: Among them, is the unit bitstream computing power coefficient of the low-bitrate video; is the unit bitstream computing power coefficient of the high-bitrate video; is the actual bitstream value of the i-th low-bitrate video; is the actual bitstream value of the j-th high-bitrate video; and is the non-linear adjustment exponent; is the priority weighting factor of the j-th high-bitrate video.

9. The traffic event secondary recognition method based on pull stream detection of a video cloud platform according to claim 8, characterized in that The unit bitstream computing power coefficient of the low-bitstream video and the unit bitstream computing power coefficient of the high-bitstream video are obtained by statistically calculating the average detection time of the low-bitstream video and the average detection time of the high-bitstream video based on the data of historical detection tasks. and the average detection time of the high-bitstream video Specifically, it is as follows: Among them, is the average bitrate value of the low-bitrate video; is the average bitrate value of the high-bitrate video; and are correction coefficients; is the dynamic compensation factor.

10. The traffic event secondary recognition method based on pull stream detection of a video cloud platform according to claim 9, characterized in that, The value range of the correction coefficient is as follows: , , and in one dynamic allocation of computing resources, the value satisfies .

Images