System and method for supporting computing power resource scheduling management and data cross-network desensitization exchange

Abstract

The invention discloses a system supporting computing power resource scheduling management and data cross-network desensitization exchange, which comprises a video access layer, a computing power scheduling layer, an analysis processing layer, a data desensitization layer and a storage exchange layer, and is characterized in that the computing power scheduling layer comprises a task scheduling component, a computing power management component and a container management component; and the task scheduling component is used for controlling the video access layer, the computing power management component and the container management component to respectively schedule data to be analyzed, computing power resources and algorithm containers according to an operation task issued by a user terminal. According to the system, computing power resources are dynamically allocated to save the computing power resources, meanwhile, a container management technology is combined with a multi-algorithm collaborative framework to realize collaborative work among different algorithms, and meanwhile, deep desensitization and encrypted storage exchange are performed on analysis and processing data. The invention further discloses a method which is used for the system to carry out computing power resource scheduling management and data cross-network desensitization exchange.

Description

Technical Field

[0001] The present invention relates to computing power resource scheduling and various computing power optimization technologies, and in particular to a system and method that supports computing power resource scheduling management and cross-network desensitized data exchange. Background Art

[0002] The video surveillance analysis system consists of three parts: a surveillance camera, a cloud computing platform, and a user terminal. The cloud computing platform obtains the video stream data collected by the surveillance camera through the network, analyzes and processes the data to be parsed output by the video access layer in the cloud, and then sends the analysis results and alarm information to the user terminal through the network.

[0003] With the explosive growth of AI technology, the security industry is also experiencing a surge in intelligent business operations. However, a common challenge in building intelligent business operations is the extremely high costs of computing resources, algorithm software, and supporting computer rooms. The procurement cost of computing power card servers is often several or even dozens of times that of conventional CPU servers. Furthermore, security operations involve a large amount of video streaming data, such as surveillance video and police case video evidence. Compared to text-based data, intelligent analysis of video streaming data places higher demands on hardware. Therefore, compared to traditional information technology, building intelligent business operations presents a more significant demand-cost conflict, and the high construction costs can even make the work impossible.

[0004] The existing video surveillance and analysis system uses a fixed computing power allocation model and cannot be dynamically adjusted according to the task load, resulting in a waste of computing power resources. Different algorithms cannot work together and have difficulty coping with complex scenarios. As a result, when large model analysis requires a large amount of computing power, the system cannot be elastically expanded. In addition, cross-network data exchange only uses basic encryption and lacks deep desensitization processing. Moreover, video analysis, data processing, storage exchange and other links are independent of each other and inefficient. Summary of the Invention

[0005] In order to address the deficiencies of the above-mentioned prior art, the present invention provides a system and method that supports computing resource scheduling management and cross-network desensitized data exchange. It adopts dynamic allocation of computing resources to save computing resources, and adopts container management technology combined with a multi-algorithm collaborative framework to achieve collaborative work between different algorithms. At the same time, it performs deep desensitization and encrypted storage and exchange of analysis and processing data.

[0006] The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0007] A system that supports computing resource scheduling and management and cross-network data desensitization exchange, including:

[0008] The video access layer is used to access the surveillance camera to obtain the video stream data collected by the surveillance camera and output the corresponding data to be parsed;

[0009] The computing power scheduling layer is used to access user terminals and includes a task scheduling component, a computing power management component, and a container management component. The computing power management component is used to manage various computing power resources, and the container management component is used to manage various algorithm containers. The task scheduling component is used to control the video access layer, computing power management component, and container management component to respectively schedule the data to be parsed, computing power resources, and algorithm containers according to the computing tasks issued by the user terminal;

[0010] The analysis and processing layer is used to analyze and process the data to be parsed and output the analysis and processing data, including a computing resource component composed of multiple computing resources and an algorithm resource component composed of multiple algorithm containers, where different algorithm containers encapsulate different algorithm models;

[0011] A data desensitization layer, configured to perform desensitization processing on the analysis and processing data output by the analysis and processing layer, and output the desensitized data;

[0012] The storage exchange layer is used to exchange and store the desensitized data output by the data desensitization layer across networks based on blockchain technology.

[0013] Furthermore, the data to be parsed is video stream data or image frame data, and the video access layer includes:

[0014] A video request component, configured to request and obtain the video stream data from the surveillance camera;

[0015] The protocol conversion component is used to convert the obtained video stream data of different transmission protocols into the same transmission protocol;

[0016] A video output component, configured to output the video stream data after protocol conversion as the data to be parsed;

[0017] An image output component is used to perform encoding, decoding and frame extraction processing on the video stream data after protocol conversion at a preset time interval to form the image frame data, and output the image frame data as the data to be parsed;

[0018] The output selection component is used to select the video output component or the image output component for output according to the real-time nature of the computing task.

[0019] Furthermore, the video request component includes:

[0020] A secondary stream request module, configured to request the surveillance camera to output the secondary stream and output the obtained secondary stream;

[0021] A main stream request module, configured to request the surveillance camera to output a main stream of the video stream data;

[0022] The bit rate conversion module is used to downcode the acquired main bit stream to form a downcoded bit stream for output;

[0023] The stream code switching module is used to preferentially start the auxiliary stream request module to request the surveillance camera to output the auxiliary code stream. When the auxiliary stream request module does not receive the auxiliary code stream within a specified time or receives an error instruction fed back by the surveillance camera, the main code stream request module is switched to start to request the surveillance camera to output the main code stream.

[0024] Furthermore, a computing power model is provided in the task scheduling component for calculating the computing power required for the computing task. The computing power model is as follows:

[0025]

[0026] Among them, F is the required computing power, Di is the data density of the i-th data to be parsed among the required n data to be parsed, and Ai is the algorithm coefficient of the required algorithm model under the data density of the i-th data to be parsed.

[0027] Furthermore, the data desensitization layer includes:

[0028] A data grading component, configured to grade the analyzed and processed data according to its content, and provided with a graded lookup table between data content and sensitivity level;

[0029] A data association component for associating the analysis and processing data based on the same subject, the same time or the same location;

[0030] The data desensitization component is used to desensitize the analysis and processing data according to the data type and sensitivity level, and is provided with a desensitization query table between the data type, sensitivity level and desensitization method.

[0031] A method for supporting computing resource scheduling management and cross-network desensitized data exchange is used in the above system; the method comprises the following steps:

[0032] Step 100: The task scheduling component receives a computing task;

[0033] Step 200: The task scheduling component controls the video access layer, computing power management component, and container management component to respectively schedule data to be parsed, computing power resources, and algorithm containers according to the computing task;

[0034] Step 300: The analysis and processing layer uses the computing resources scheduled by the computing power management component and the algorithm container scheduled by the container management component to analyze and process the data to be parsed output by the video access layer, and outputs the analyzed and processed data;

[0035] Step 400: After the analysis and processing layer completes the computing task, the task scheduling component first sends a stop instruction to the container management component to cause the container management component to stop computing on the corresponding computing container, and then sends a reclaim instruction to the computing power management component to cause the computing power management component to reclaim the corresponding computing power resources and re-mark them as idle computing power.

[0036] Step 500: The data desensitization layer desensitizes the analysis and processing data output by the analysis and processing layer, and outputs the desensitized data;

[0037] Step 600: The storage exchange layer performs cross-network exchange and storage on the desensitized data output by the data desensitization layer.

[0038] Furthermore, the computing task includes a task type and data source information. In step 200, the task scheduling component controls the video access layer, computing power management component, and container management component to respectively schedule the data to be parsed, computing power resources, and algorithm containers according to the computing task, including the following steps:

[0039] Step 210: The task scheduling component determines the required algorithm model and video source data according to the task type and data source information;

[0040] Step 220: The task scheduling component sends an algorithm scheduling instruction and a data request instruction to the container management component and the video access layer respectively according to the required algorithm model and video source data;

[0041] Step 230: After receiving the data request instruction, the video access layer requests the surveillance camera to output corresponding video source data, and pre-processes the video source data to generate the data to be parsed and sends it to the task scheduling component;

[0042] Step 240: The task scheduling component calculates the computing power required for the computing task based on the required algorithm model and the data density of the data to be parsed;

[0043] Step 250: The task scheduling component sends a computing power scheduling instruction to the computing power management component according to the required computing power;

[0044] Step 260: After receiving the computing power scheduling instruction, the computing power management component determines a computing power address corresponding to the computing power size from the idle computing power of the analysis and processing layer, and sends the determined computing power address to the container management component;

[0045] Step 270: After receiving the algorithm scheduling instruction and computing power address, the container management component calls the corresponding algorithm container in the analysis and processing layer, and deploys the called algorithm container on the computing power resources corresponding to the computing power address.

[0046] Furthermore, the computing power model is as follows:

[0047]

[0048] Among them, F is the required computing power, Di is the data density of the i-th data to be parsed among the required n data to be parsed, and Ai is the algorithm coefficient of the required algorithm model under the data density of the i-th data to be parsed.

[0049] Furthermore, the data to be parsed is video stream data or image frame data, and the computing task also includes task real-time performance. In step 220, when the task scheduling component sends a data request instruction to the video access layer, it also informs the video access layer of the task real-time performance. In step 230, after receiving the data request instruction, the video access layer requests the surveillance camera to output corresponding video source data, and pre-processes the video source data to form the data to be parsed and send it to the task scheduling component, which includes the following steps:

[0050] Step 231: After receiving the data request instruction, the video access layer sends a video request instruction to the surveillance camera to obtain the video stream data collected by the surveillance camera;

[0051] Step 232: The video access layer converts video stream data of different transmission protocols into the same transmission protocol;

[0052] Step 233: If the real-time requirement of the computing task is high, the video access layer executes step 234; if the real-time requirement of the computing task is low, the video access layer executes step 235;

[0053] Step 234: The video access layer outputs the video stream data after the protocol conversion as the data to be parsed;

[0054] Step 235: The video access layer performs encoding, decoding, and frame extraction processing on the video stream data after protocol conversion at a preset time interval to form the image frame data, and outputs the image frame data as the data to be parsed.

[0055] Furthermore, the video request instruction includes a secondary code request instruction and a primary code request instruction. In step 231, after receiving the data request instruction, the video access layer sends a video request instruction to the surveillance camera to obtain the video stream data collected by the surveillance camera, including the following steps:

[0056] Step 231a: After receiving the data request instruction, the video access layer sends the auxiliary code request instruction to the surveillance camera;

[0057] Step 231b: If the video access layer does not receive the auxiliary code stream of the video stream data within the specified time or receives an error instruction fed back by the surveillance camera, it sends the main code request instruction to the surveillance camera and executes step 231c;

[0058] Step 231c: After receiving the main code stream of the video stream data, the video access layer performs downcoding on the main code stream of the video stream data to form a downcoded stream.

[0059] The present invention has the following beneficial effects: the present invention realizes dynamic allocation of computing power through containerization technology, computing power management and scheduling technology, solves the problem of idle computing power caused by unreasonable computing power planning, and establishes an automated algorithm task scheduling mechanism, and sets up a task scheduling module. The module can automatically start and stop tasks in staggered task configuration, thereby reducing computing power waste caused by invalid calculations and improving the effective utilization rate of computing power. Based on the characteristics of auxiliary code streams, video transcoding technology, and video frame extraction technology, the data is pre-processed to greatly reduce the computing power consumption of tasks of the same type and scale without affecting the results of algorithm analysis. The computing power conversion model is established to realize the businessization and visualization of the computing power concept. On this basis, users can clearly grasp the existing computing power scale and plan computing power as expected. At the same time, the pixel-level precise desensitization and adaptive rule discovery mechanism are improved, and multimodal joint desensitization (video + audio + metadata) are combined to finally construct a "transmission-scheduling-desensitization-transmission" full-link security framework. With dynamic resource scheduling and intelligent privacy protection technology, the efficient and secure exchange of algorithms and algorithm analysis results is realized in a cross-network environment for the first time. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is an application architecture diagram of the system provided by the present invention that supports computing resource scheduling management and cross-network desensitized data exchange.

[0061] Figure 2 This is a schematic diagram of the principle architecture of the system provided by the present invention that supports computing resource scheduling management and cross-network desensitized data exchange.

[0062] Figure 3A flowchart of the steps of the method provided by the present invention for supporting computing resource scheduling management and cross-network desensitized data exchange.

[0063] Figure 4 This is a block diagram of step 200 in the method for supporting computing resource scheduling management and cross-network desensitized data exchange provided by the present invention.

[0064] Figure 5 This is a flowchart of step 230 in the method for supporting computing resource scheduling management and cross-network desensitized data exchange provided by the present invention.

[0065] Figure 6 This is a flowchart of step 231 in the method for supporting computing resource scheduling management and cross-network desensitized data exchange provided by the present invention. DETAILED DESCRIPTION

[0066] The present invention will be described in detail below with reference to the accompanying drawings and embodiments. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and are not to be construed as limiting the present invention.

[0067] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention.

[0068] Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature identified as "first," "second," or "third" may explicitly or implicitly include one or more of such features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0069] In the present invention, unless otherwise expressly specified or limited, terms such as "installed," "connected," "connect," "fixed," and "disposed" should be understood in a broad sense. For example, they may refer to fixed or detachable connections, or integration; mechanical or electrical connections; direct or indirect connections through an intermediate medium; or internal communication between two components or interaction between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0070] Example 1

[0071] like Figure 1 As shown, a system that supports computing resource scheduling management and cross-network desensitized data exchange is connected between a surveillance camera and a user terminal through a network. The video stream data collected by the surveillance camera is output to the system through the network, and the user terminal sends computing tasks to the system through the network; wherein the number of the surveillance cameras can be multiple, and the number of the user terminals can also be multiple.

[0072] Specifically, such as Figure 2 As shown, this system includes:

[0073] The video access layer is used to access the surveillance camera to obtain the video stream data collected by the surveillance camera and output the corresponding data to be parsed;

[0074] The computing power scheduling layer is used to access user terminals and includes a task scheduling component, a computing power management component, and a container management component. The computing power management component is used to manage various computing power resources, and the container management component is used to manage various algorithm containers. The task scheduling component is used to control the video access layer, computing power management component, and container management component to respectively schedule the data to be parsed, computing power resources, and algorithm containers according to the computing tasks issued by the user terminal;

[0075] The analysis and processing layer is used to analyze and process the data to be parsed and output the analysis and processing data, including a computing resource component composed of multiple computing resources and an algorithm resource component composed of multiple algorithm containers, where different algorithm containers encapsulate different algorithm models;

[0076] A data desensitization layer, configured to perform desensitization processing on the analysis and processing data output by the analysis and processing layer, and output the desensitized data;

[0077] The storage exchange layer is used to exchange and store the desensitized data output by the data desensitization layer across networks based on blockchain technology.

[0078] The system provided by the present invention adopts containerization technology to encapsulate each large algorithm model in a corresponding container to form each algorithm container. While reducing computing power overhead, it can achieve unified scheduling across the network and dynamically allocate the computing power resources and algorithm containers deployed in the analysis and processing layer according to the computing tasks. After completing the computing task, the computing power management component can promptly recycle the computing power resources occupied by the algorithm container to allocate them to the next computing task, greatly improving the utilization rate of computing power resources. At the same time, the analysis and processing data output by the analysis and processing layer is desensitized, and the desensitized data is exchanged and stored across the network using blockchain technology, thereby ensuring data security and avoiding sensitive data leakage and data tampering.

[0079] The computing power resources in this patent refer to the computing power provided by computing hardware with computing capabilities such as CPU, GPU, FPGA and server.

[0080] The data to be parsed is video stream data or image frame data, and the video access layer includes:

[0081] A video request component, configured to request and obtain the video stream data from the surveillance camera;

[0082] The protocol conversion component is used to convert the obtained video stream data of different transmission protocols into the same transmission protocol;

[0083] A video output component, configured to output the video stream data after protocol conversion as the data to be parsed;

[0084] An image output component is used to perform encoding, decoding and frame extraction processing on the video stream data after protocol conversion at a preset time interval to form the image frame data, and output the image frame data as the data to be parsed;

[0085] The output selection component is used to select the video output component or the image output component for output according to the real-time nature of the computing task.

[0086] Since surveillance cameras from different manufacturers may use different video transmission protocols, and surveillance cameras deployed in different years may also use different video transmission protocols, the video access layer will convert the video stream data obtained from each surveillance camera using different transmission protocols (such as RTMP, SFTP, GB28181 and other protocols) into the same transmission protocol (such as RTSP protocol) through the protocol conversion component, thereby achieving protocol standardization and unified management, which can simplify the scheduling and analysis processing of the video stream data by the computing power scheduling layer and the analysis and processing layer.

[0087] In addition, when the video access layer is outputting, if the real-time requirement of the computing task is low, the image output component is selected to encode, decode and extract frames of the video stream data after the protocol conversion at a preset time interval to form the image frame data, and the image frame data is used as the data to be parsed and output to the analysis and processing layer for analysis and processing. If the real-time requirement of the computing task is high, the video output component is selected to use the video stream data after the protocol conversion as the data to be parsed and output to the analysis and processing layer for analysis and processing, which can greatly save the computing power resources of the analysis and processing layer.

[0088] The video request component includes:

[0089] A secondary stream request module, configured to request the surveillance camera to output the secondary stream and output the obtained secondary stream;

[0090] A main stream request module, configured to request the surveillance camera to output a main stream of the video stream data;

[0091] The bit rate conversion module is used to downcode the acquired main bit stream to form a downcoded bit stream for output;

[0092] The stream code switching module is used to preferentially start the auxiliary stream request module to request the surveillance camera to output the auxiliary code stream. When the auxiliary stream request module does not receive the auxiliary code stream within a specified time or receives an error instruction fed back by the surveillance camera, the main code stream request module is switched to start to request the surveillance camera to output the main code stream.

[0093] The main stream is the primary video stream captured by the surveillance camera. It uses high resolution, high frame rate, and high bit rate to ensure video clarity and smoothness. For example, in a 4K UHD surveillance system, the main stream can have a resolution of up to 3840×2160, a frame rate of 25 / 30fps, and a bit rate of 8-16Mbps.

[0094] The auxiliary stream is a low-bitrate version of the main stream, with lower resolution (such as 720p or lower), frame rate and bit rate. It usually adopts a high-compression encoding method such as H.264 Baseline Profile, and the bit rate is generally between 0.5-2Mbps.

[0095] Since the bit rate of the auxiliary code stream is only half of that of the main code stream or even lower, the video access layer preferentially requests and outputs the auxiliary code stream, which can save the computing power resources of the analysis and processing layer; and since some surveillance cameras may have been deployed earlier or have simpler functions and do not support dual-stream output, the video access layer can first request the main code stream for these surveillance cameras, and then use the code reduction function to reduce the bit rate of the main code stream to form the reduced code stream and output it, which can also save the computing power resources of the analysis and processing layer.

[0096] For surveillance cameras that support dual-stream output, the RTSP addresses of the main stream and the auxiliary stream are different. Therefore, the video access layer can use the relevant instructions of the RTSP transmission protocol for detection, and write the RTSP address of the auxiliary stream in the auxiliary code request instruction for requesting the auxiliary stream. If the surveillance camera supports the auxiliary stream output, it will output the auxiliary stream through the RTSP address written in the instruction after receiving the auxiliary code request instruction. If the surveillance camera does not support the auxiliary stream output, it will not respond or feedback an error instruction to the video access layer after receiving the auxiliary code request instruction because it cannot find the RTSP address written in the instruction.

[0097] The task scheduling component is provided with a computing power model for calculating the computing power required for the computing task. The computing power model is as follows:

[0098]

[0099] Among them, F is the required computing power, Di is the data density of the i-th data to be parsed among the required n data to be parsed, and Ai is the algorithm coefficient of the required algorithm model under the data density of the i-th data to be parsed.

[0100] The algorithm coefficient refers to the standard computing power required by a specific algorithm model to process a single video stream or image frame. Different algorithm models require different standard computing power. For example, a human fight algorithm might require several times more computing power than a human run algorithm. The same algorithm model also requires different standard computing power when processing video streams with different bitrates or image frame rates. For example, a human fight algorithm requires more computing power when analyzing and processing a 2Mbps video stream than a 3Mbps video stream.

[0101] The algorithm coefficient is calculated by providing a benchmark computing power card and testing how many video streams at a benchmark bitrate (e.g., 2Mbps) or image frames at a benchmark frame rate (e.g., 1 frame per 5 seconds) a specific algorithm model can simultaneously analyze when the benchmark computing power card is fully loaded. The inverse of the measured result, multiplied by a safety factor, is the standard computing power of the algorithm model at the benchmark bitrate or frame rate. The algorithm coefficient for this standard computing power is set to 1, and the algorithm coefficient for the algorithm model at other bitrates or frame rates is then calculated using a quasi-linear model. The algorithm coefficients of other algorithm models at different bitrates or frame rates are calculated in the same manner.

[0102] Data density (data-density): the data density of the video stream data is its bit rate, and the data density of the image frame data is its frame frequency.

[0103] The data desensitization layer includes:

[0104] A data grading component, configured to grade the analyzed and processed data according to its content, and provided with a graded lookup table between data content and sensitivity level;

[0105] A data association component for associating the analysis and processing data based on the same subject, the same time or the same location;

[0106] The data desensitization component is used to desensitize the analysis and processing data according to the data type and sensitivity level, and is provided with a desensitization query table between the data type, sensitivity level and desensitization method.

[0107] The analysis and processing data output by the analysis and processing layer includes audio and video data (decomposed into video frames, audio streams, and metadata), image data (parsed EXIF ​​information and image content), and structured data (identified field types and semantics). The data classification component uses the following hierarchical query table to perform hierarchical processing on the analysis and processing data:

[0108]

[0109]

[0110] The data association component can perform cross-modal association of data of different modalities in the analysis and processing data based on the same subject, such as associating facial data and voice data of the same person; it can also perform cross-temporal and spatial association of data of different subjects in the analysis and processing data based on the same time or the same place, such as associating different people and different vehicles that appear at the same time or the same place.

[0111] The data desensitization component first uses an adaptive sampling algorithm (dynamically adjusting the sampling rate based on content complexity) to extract key frames from the analyzed and processed data, and then performs multimodal sensitive information detection on the extracted key frames. The detection targets include but are not limited to: biometrics (such as face, iris, fingerprint, etc.), identity features (such as work badges, certificates, license plates, etc.), and scene features (confidential areas, special equipment, etc.). The following desensitization query table is used to desensitize the analyzed and processed data:

[0112]

[0113] EXIF (Exchangeable Image File Format) is a metadata standard embedded in image files (such as JPEG, TIFF, RAW, etc.) and is used to record data such as shooting device parameters, shooting conditions, and copyright information.

[0114] After the desensitization process is completed, the desensitization effect of the data desensitization component can also be verified. The verification methods mainly include manual spot checks, usability evaluation, quantitative indicators, etc.

[0115] Manual spot check: Manually randomly select desensitized data samples and compare them with the original data to verify whether the desensitization rules are correctly implemented and check the rationality of the desensitized data.

[0116] Usability evaluation: Verify whether the desensitized data supports business scenarios (such as development, testing, and analysis) to avoid data invalidation caused by excessive desensitization.

[0117] Quantitative indicators: Measure the masking effect through objective indicators and provide comparable evaluation results through statistical analysis (common indicators: masking coverage, the ratio of sensitive fields to be masked; information entropy, measuring the unpredictability of masked data; re-identification risk, calculating the probability of data restoration through statistical methods; business impact indicators, such as test case pass rate, data analysis error rate, etc.).

[0118] The storage exchange layer ultimately outputs data for secure exchange, using the following technologies:

[0119] Audit log records, trusted storage module, business partitioned storage, data encryption, tamper-proof design and log tracking audit monitoring.

[0120] Example 2

[0121] like Figure 3 As shown, a method for supporting computing resource scheduling management and cross-network desensitized data exchange is used in the system described in Example 1; the method includes the following steps:

[0122] Step 100: The task scheduling component receives a computing task.

[0123] In step 100, the computing task is sent by relevant personnel to the task scheduling component of the computing power scheduling layer by operating the user terminal.

[0124] Step 200: The task scheduling component controls the video access layer, computing power management component and container management component to schedule the data to be parsed, computing power resources and algorithm containers respectively according to the computing task.

[0125] In step 200, the task scheduling component parses the computing task to determine the data to be parsed, computing resources and algorithm containers required for the computing task, and sends corresponding scheduling instructions to the video access layer, computing power management component and container management component respectively, so as to deploy the required algorithm containers on the required computing resources in the analysis and processing layer, thereby analyzing and processing the required data to be parsed.

[0126] Specifically, the computing task includes a task type and data source information. The task type can be one or more, and the data source information can also be one or more. That is, the computing task can be different analysis and processing tasks for the same data to be parsed, the same analysis and processing task for multiple data to be parsed, or multiple analysis and processing tasks for multiple data to be parsed.

[0127] The task type corresponds to the algorithm model, for example, a face recognition task corresponds to a face recognition algorithm, a running recognition task corresponds to a running recognition algorithm, a movement trajectory prediction task corresponds to a movement trajectory prediction model, and the like.

[0128] The data source information corresponds to the source and time of the data to be parsed, including a camera identifier and a shooting time period. The camera identifier is used to identify the surveillance camera that captured the data to be parsed, and can be at least one of a camera name, a camera address, and a camera number. The shooting time period is used to identify the time when the data to be parsed was captured.

[0129] In step 200, if Figure 4 As shown, when the task scheduling component controls the video access layer, computing power management component, and container management component according to the computing task to respectively schedule the data to be parsed, computing power resources, and algorithm containers, the following steps are included:

[0130] Step 210: The task scheduling component determines the required algorithm model and video source data according to the task type and data source information.

[0131] In step 210, each computing task may include only one task type, such as performing a face recognition task, or multiple task types, such as simultaneously performing a face recognition task and a motion trajectory prediction task. Therefore, the number of algorithmic large models determined may be one or more. Similarly, each computing task may include data source information for only one video source data or multiple video source data. Therefore, the number of video source data determined may be one or more.

[0132] Step 220: The task scheduling component sends an algorithm scheduling instruction and a data request instruction to the container management component and the video access layer respectively according to the required algorithm large model and video source data.

[0133] Step 230: After receiving the data request instruction, the video access layer requests the surveillance camera to output corresponding video source data, and pre-processes the video source data to form the data to be parsed and sends it to the task scheduling component.

[0134] In step 230, the original data obtained by the video access layer from the surveillance camera is video stream data, and the data to be parsed sent to the task scheduling component is video stream data or image frame data formed by frame extraction of the video stream data.

[0135] Specifically, such as Figure 5 As shown, the computing task also includes task real-time performance. In step 220, when the task scheduling component sends a data request instruction to the video access layer, it also informs the video access layer of the task real-time performance. In step 230, after receiving the data request instruction, the video access layer requests the surveillance camera to output corresponding video source data, and pre-processes the video source data to form the data to be parsed and send it to the task scheduling component. The following steps are included:

[0136] Step 231: After receiving the data request instruction, the video access layer sends a video request instruction to the surveillance camera to obtain the video stream data collected by the surveillance camera.

[0137] In this step 231, the video request instruction includes an auxiliary code request instruction and a main code request instruction. The video request instruction is used to request the auxiliary code stream of the video stream data from the surveillance camera, and the main code request instruction is used to request the main code stream of the video stream data from the surveillance camera.

[0138] Specifically, such as Figure 6As shown, in step 231, after receiving the data request instruction, the video access layer sends a video request instruction to the surveillance camera to obtain the video stream data collected by the surveillance camera, including the following steps:

[0139] Step 231a: After receiving the data request instruction, the video access layer sends the auxiliary code request instruction to the surveillance camera;

[0140] Step 231b: If the video access layer does not receive the auxiliary code stream of the video stream data within the specified time or receives an error instruction fed back by the surveillance camera, it sends the main code request instruction to the surveillance camera and executes step 231c;

[0141] Step 231c: After receiving the main code stream of the video stream data, the video access layer performs downcoding on the main code stream of the video stream data to form a downcoded stream.

[0142] In step 231b, if the video access layer receives the auxiliary code stream of the video stream data within the specified time, step 232 is executed.

[0143] Step 232: The video access layer converts video stream data of different transmission protocols into the same transmission protocol.

[0144] In step 232, the video access layer converts the auxiliary streams or down-streams of video stream data using different transmission protocols (such as RTMP, SFTP, GB28181, etc.) to the same transmission protocol (such as RTSP protocol) for subsequent analysis and processing.

[0145] Step 233: If the real-time requirement of the computing task is high, the video access layer executes step 234; if the real-time requirement of the computing task is low, the video access layer executes step 235.

[0146] Step 234: The video access layer outputs the video stream data after protocol conversion as the data to be parsed.

[0147] In step 234, the data to be parsed output by the video access layer is the auxiliary stream or down-stream after the protocol is converted.

[0148] Step 235: The video access layer performs encoding, decoding, and frame extraction processing on the video stream data after protocol conversion at a preset time interval to form the image frame data, and outputs the image frame data as the data to be parsed.

[0149] In step 235, the data to be parsed output by the video access layer is the image frame data after the auxiliary stream or the down-stream is sequentially processed with protocol conversion, encoding and decoding, and frame extraction.

[0150] Step 240: The task scheduling component calculates the computing power required for the computing task based on the required algorithm model and the data density of the data to be parsed.

[0151] In step 240, the computing power model is as follows:

[0152]

[0153] Among them, F is the required computing power, Di is the data density of the i-th data to be parsed among the required n data to be parsed, and Ai is the algorithm coefficient of the required algorithm model under the data density of the i-th data to be parsed.

[0154] Step 250: The task scheduling component sends a computing power scheduling instruction to the computing power management component according to the required computing power.

[0155] Step 260: After receiving the computing power scheduling instruction, the computing power management component determines a computing power address corresponding to the computing power size in the idle computing power of the analysis and processing layer, and sends the determined computing power address to the container management component.

[0156] Step 270: After receiving the algorithm scheduling instruction and computing power address, the container management component calls the corresponding algorithm container in the analysis and processing layer, and deploys the called algorithm container on the computing power resources corresponding to the computing power address.

[0157] Step 300: The analysis and processing layer uses the computing resources scheduled by the computing power management component and the algorithm container scheduled by the container management component to analyze and process the data to be parsed output by the video access layer, and outputs the analyzed and processed data.

[0158] Step 400: After the analysis and processing layer completes the computing task, the task scheduling component first sends a stop instruction to the container management component to make the container management component stop the corresponding computing power container for computing, and then sends a recovery instruction to the computing power management component to make the computing power management component recover the corresponding computing power resources and re-mark them as idle computing power.

[0159] Step 500: The data desensitizing layer desensitizes the analysis and processing data output by the analysis and processing layer, and outputs the desensitized data.

[0160] Step 600: The storage exchange layer performs cross-network exchange and storage on the desensitized data output by the data desensitization layer.

[0161] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention rather than to limit them. Although the embodiments of the present invention are described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the embodiments of the present invention can still be modified or replaced by equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A system that supports computing resource scheduling management and cross-network data desensitization exchange, characterized by: include: The video access layer is used to access the surveillance camera to obtain the video stream data collected by the surveillance camera and output the corresponding data to be parsed; The computing power scheduling layer is used to access user terminals and includes a task scheduling component, a computing power management component, and a container management component. The computing power management component is used to manage various computing power resources, and the container management component is used to manage various algorithm containers. The task scheduling component is used to control the video access layer, computing power management component, and container management component to respectively schedule the data to be parsed, computing power resources, and algorithm containers according to the computing tasks issued by the user terminal; The analysis and processing layer is used to analyze and process the data to be parsed and output the analysis and processing data, including a computing resource component composed of multiple computing resources and an algorithm resource component composed of multiple algorithm containers, where different algorithm containers encapsulate different algorithm models; A data desensitization layer, configured to perform desensitization processing on the analysis and processing data output by the analysis and processing layer, and output the desensitized data; The storage exchange layer is used to exchange and store the desensitized data output by the data desensitization layer across networks based on blockchain technology.

2. The system according to claim 1, wherein: The data to be parsed is video stream data or image frame data, and the video access layer includes: A video request component, configured to request and obtain the video stream data from the surveillance camera; The protocol conversion component is used to convert the obtained video stream data of different transmission protocols into the same transmission protocol; A video output component, configured to output the video stream data after protocol conversion as the data to be parsed; An image output component is used to perform encoding, decoding and frame extraction processing on the video stream data after protocol conversion at a preset time interval to form the image frame data, and output the image frame data as the data to be parsed; The output selection component is used to select the video output component or the image output component for output according to the real-time nature of the computing task.

3. The system according to claim 2, characterized in that The video request component includes: A secondary stream request module, configured to request the surveillance camera to output the secondary stream and output the obtained secondary stream; A main stream request module, configured to request the surveillance camera to output a main stream of the video stream data; The bit rate conversion module is used to downcode the acquired main bit stream to form a downcoded bit stream for output; The stream code switching module is used to preferentially start the auxiliary stream request module to request the surveillance camera to output the auxiliary code stream. When the auxiliary stream request module does not receive the auxiliary code stream within a specified time or receives an error instruction fed back by the surveillance camera, the main code stream request module is switched to start to request the surveillance camera to output the main code stream.

4. The system according to claim 1, wherein: The task scheduling component is provided with a computing power model for calculating the computing power required for the computing task. The computing power model is as follows: Among them, F is the required computing power, Di is the data density of the i-th data to be parsed among the required n data to be parsed, and Ai is the algorithm coefficient of the required algorithm model under the data density of the i-th data to be parsed.

5. The system according to claim 1, wherein: The data desensitization layer includes: A data grading component, configured to grade the analyzed and processed data according to its content, and provided with a graded lookup table between data content and sensitivity level; A data association component for associating the analysis and processing data based on the same subject, the same time or the same location; The data desensitization component is used to desensitize the analysis and processing data according to the data type and sensitivity level, and is provided with a desensitization query table between the data type, sensitivity level and desensitization method.

6. A method for supporting computing resource scheduling management and cross-network data desensitization exchange, characterized in that: Used in the system of claim 1; the method comprises the following steps: Step 100: The task scheduling component receives a computing task; Step 200: The task scheduling component controls the video access layer, computing power management component, and container management component to respectively schedule data to be parsed, computing power resources, and algorithm containers according to the computing task; Step 300: The analysis and processing layer uses the computing resources scheduled by the computing power management component and the algorithm container scheduled by the container management component to analyze and process the data to be parsed output by the video access layer, and outputs the analyzed and processed data; Step 400: After the analysis and processing layer completes the computing task, the task scheduling component first sends a stop instruction to the container management component to cause the container management component to stop computing on the corresponding computing container, and then sends a reclaim instruction to the computing power management component to cause the computing power management component to reclaim the corresponding computing power resources and re-mark them as idle computing power. Step 500: The data desensitization layer desensitizes the analysis and processing data output by the analysis and processing layer, and outputs the desensitized data; Step 600: The storage exchange layer performs cross-network exchange and storage on the desensitized data output by the data desensitization layer.

7. The method according to claim 6, characterized in that The computing task includes a task type and data source information. In step 200, the task scheduling component controls the video access layer, computing power management component, and container management component to respectively schedule the data to be parsed, computing power resources, and algorithm containers according to the computing task, including the following steps: Step 210: The task scheduling component determines the required algorithm model and video source data according to the task type and data source information; Step 220: The task scheduling component sends an algorithm scheduling instruction and a data request instruction to the container management component and the video access layer respectively according to the required algorithm model and video source data; Step 230: After receiving the data request instruction, the video access layer requests the surveillance camera to output corresponding video source data, and pre-processes the video source data to generate the data to be parsed and sends it to the task scheduling component; Step 240: The task scheduling component calculates the computing power required for the computing task based on the required algorithm model and the data density of the data to be parsed; Step 250: The task scheduling component sends a computing power scheduling instruction to the computing power management component according to the required computing power; Step 260: After receiving the computing power scheduling instruction, the computing power management component determines a computing power address corresponding to the computing power size from the idle computing power of the analysis and processing layer, and sends the determined computing power address to the container management component; Step 270: After receiving the algorithm scheduling instruction and computing power address, the container management component calls the corresponding algorithm container in the analysis and processing layer, and deploys the called algorithm container on the computing power resources corresponding to the computing power address.

8. The method according to claim 7, characterized in that The computing power model is as follows: Among them, F is the required computing power, Di is the data density of the i-th data to be parsed among the required n data to be parsed, and Ai is the algorithm coefficient of the required algorithm model under the data density of the i-th data to be parsed.

9. The method according to claim 7, characterized in that The data to be parsed is video stream data or image frame data, and the computing task also includes task real-time performance. In step 220, when the task scheduling component sends a data request instruction to the video access layer, it also informs the video access layer of the task real-time performance. In step 230, after receiving the data request instruction, the video access layer requests the surveillance camera to output corresponding video source data, and pre-processes the video source data to generate the data to be parsed and send it to the task scheduling component. The following steps are included: Step 231: After receiving the data request instruction, the video access layer sends a video request instruction to the surveillance camera to obtain the video stream data collected by the surveillance camera; Step 232: The video access layer converts video stream data of different transmission protocols into the same transmission protocol; Step 233: If the real-time requirement of the computing task is high, the video access layer executes step 234; if the real-time requirement of the computing task is low, the video access layer executes step 235; Step 234: The video access layer outputs the video stream data after the protocol conversion as the data to be parsed; Step 235: The video access layer performs encoding, decoding, and frame extraction processing on the video stream data after protocol conversion at a preset time interval to form the image frame data, and outputs the image frame data as the data to be parsed.

10. The method according to claim 9, characterized in that The video request instruction includes a secondary code request instruction and a primary code request instruction. In step 231, after receiving the data request instruction, the video access layer sends a video request instruction to the surveillance camera to obtain the video stream data collected by the surveillance camera, including the following steps: Step 231a: After receiving the data request instruction, the video access layer sends the auxiliary code request instruction to the surveillance camera; Step 231b: If the video access layer does not receive the auxiliary code stream of the video stream data within the specified time or receives an error instruction fed back by the surveillance camera, it sends the main code request instruction to the surveillance camera and executes step 231c; Step 231c: After receiving the main code stream of the video stream data, the video access layer performs downcoding on the main code stream of the video stream data to form a downcoded stream.

Images