A computer vision data processing method and system for data stream collaboration
By using 3D resource modeling and dynamic load quantification assessment, the problem of unreasonable resource allocation in multi-source visual data stream processing is solved, achieving efficient collaborative optimization of resources and accurate load assessment, and adapting to dynamic resource adjustment in complex scenarios.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUAIAN COLLEGE OF INFORMATION TECH
- Filing Date
- 2026-03-03
- Publication Date
- 2026-06-12
AI Technical Summary
Existing multi-source visual data stream processing systems suffer from problems such as unreasonable resource allocation, inaccurate load assessment, and insufficient collaborative assessment, resulting in low resource utilization, processing latency, and poor adaptability, making it difficult to meet the application needs of complex industrial scenarios and large-scale intelligent terminals.
By employing a closed-loop logic of 3D resource modeling, dynamic load quantification, and collaborative relationship evaluation, and by constructing computing, storage, and transmission resource models, combined with dynamic data access degree and cluster cohesion evaluation, we can achieve refined resource allocation and dynamic optimization.
It achieves coordinated adaptation of computing, storage, and transmission resources, accurately quantifies load differences, reduces overload processing latency, improves resource utilization, and adapts to complex and ever-changing application scenarios.
Smart Images
Figure CN122195649A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of visual data processing technology, specifically to a computer vision data processing method and system for data flow collaboration. Background Technology
[0002] With the widespread application of computer vision technology in industrial production, intelligent monitoring, autonomous driving, and other fields, the demand for collaborative processing of multi-source visual data streams (such as industrial cameras, smart terminals, and cloud push) is becoming increasingly urgent. Currently, visual data processing systems generally face technical bottlenecks such as strong heterogeneity of multi-source data streams, large fluctuations in data volume, and unreasonable resource allocation, which seriously affect processing efficiency and stability.
[0003] In existing technologies, multi-source visual data stream processing often adopts a single-dimensional resource allocation model, focusing only on computing resource scheduling and neglecting the coordinated adaptation of storage and transmission resources. This leads to an imbalance between the supply and demand of 3D resources, with some scenarios exhibiting storage redundancy and insufficient transmission bandwidth, or idle computing resources and storage overflow, resulting in an overall resource utilization rate generally below 60%. Furthermore, load assessment often relies on static indicators, failing to consider the temporal continuity and dynamic fluctuation characteristics of the data stream. It judges the load status solely through preset thresholds, failing to accurately quantify the real-time load differences between different partitions, leading to load assessment errors exceeding 30%.
[0004] Furthermore, the existing system lacks an effective collaborative evaluation mechanism and has not established collaborative relationships between data flow nodes. It only processes data from individual partitions in isolation, making it difficult to cope with sudden data flow peaks. Overloaded partitions experience processing delays exceeding 500ms, while redundant partitions have idle resource rates exceeding 40%. Simultaneously, resource adjustments are mostly manual interventions or fixed-period adjustments, resulting in delayed responses and an inability to dynamically optimize resource allocation based on real-time load. This leads to processing congestion or resource waste in scenarios with concurrent multi-source data flows. Moreover, its poor adaptability makes it difficult to accommodate the differences in data flow characteristics across different data sources, limiting its adoption in complex industrial scenarios, large-scale intelligent terminal collaboration, and other high-end applications.
[0005] To address the aforementioned issues, there is an urgent need for a processing method and system capable of achieving collaborative modeling of 3D resources, precise quantification of dynamic load, intelligent evaluation of collaborative relationships, and real-time optimization of resources. This would overcome the limitations of existing technologies and meet the needs of efficient collaborative processing of multi-source visual data streams. Summary of the Invention
[0006] The purpose of this invention is to provide a computer vision data processing method and system for data stream collaboration, in order to solve the problems mentioned in the background art. In summary, the core principle of this invention is based on a closed-loop logic of "3D resource modeling - dynamic load quantification - collaborative relationship evaluation - intelligent resource optimization" to achieve efficient collaborative processing of multi-source visual data streams. First, by constructing a 3D resource model for computation, storage, and transmission, the multi-source data streams are cataloged by source partition, so that each data source corresponds to an independent collaborative partition and embedded partition units, achieving refined initial allocation of resources (Step 1). Second, based on the real-time data volume, the dynamic data call degree is calculated, and a sample set is generated in combination with time continuity to quantify the dynamic call gradient between partition units, accurately representing load differences (Step 2). Then, using the dynamic data call degree as the adaptive balance centroid, a flow collaboration set is constructed, and the gradient concentration degree is evaluated through cluster cohesion to reflect centroid stability (Step 3). Finally, the resource consumption status of partition units is determined by the cluster cohesion mean and standard deviation, realizing the cross-time-period dynamic transfer of redundant resources to overloaded partitions, forming a closed-loop collaborative mechanism of "allocation-quantification-evaluation-optimization" (Step 4).
[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:
[0008] A computer vision data processing system for data flow collaboration, comprising: a collaborative partitioning and unit division module, a data processing and parameter calculation module, a collaborative set construction and evaluation module, and a resource consumption evaluation and adjustment module;
[0009] The collaborative partitioning and unit division module is used to realize collaborative partitioning and cataloging of multi-source visual data streams, construction of resource dimension models, and partition unit division.
[0010] The data processing and parameter calculation module is used to configure the number of partition units, acquire real-time data, calculate dynamic call-related parameters, and quantify load differences.
[0011] The circulation collaboration set construction and evaluation module is used to determine the adaptive balance benchmark, construct the circulation collaboration set, and evaluate the degree of concentration of internal parameters;
[0012] The resource consumption assessment and adjustment module is used to collect relevant parameters, assess the resource consumption status of individual partitions, and process resource adjustments.
[0013] Preferably, the collaborative partitioning and unit division module includes:
[0014] Used for collaborative partitioning and cataloging of multi-source computer vision data streams, and recording the cataloging results into the collaborative partitioning and cataloging unit of the collaborative processing center;
[0015] This is a resource dimension model building unit used to build a model that includes computing resources, storage resources, and transmission resources, and to equally divide the total resources of each type to determine the basis for model scaling.
[0016] It is used to map the three-dimensional scale combination of the model into the smallest co-processing unit, and to serve as the partition unit of the embedded partition unit within the co-partition.
[0017] Preferably, the data processing and parameter calculation module includes:
[0018] Used to configure the number of partition units embedded in each collaborative partition, and to specify the partition unit configuration unit that a single partition unit handles the corresponding visual data stream node;
[0019] Used to obtain the real-time data volume of each partition unit and its corresponding collaborative partition, calculate the dynamic data access degree of the partition unit, and upload it to the data acquisition and access degree calculation unit of the collaborative processing center.
[0020] A sample set generation unit for generating a sample set containing dynamic data call degree within a continuous time period based on the temporal continuity of visual data streams;
[0021] The load difference quantization unit is used to select different partition units within the same collaborative partition to form flow redirection and quantize the dynamic load differences corresponding to the vectors.
[0022] Preferably, the transfer collaboration set construction and evaluation module includes:
[0023] The equilibrium centroid determination unit is used to determine the adaptive equilibrium centroid based on the dynamic data call degree, using a sample set as a benchmark.
[0024] The collaborative set building unit is used to collect all dynamic call gradients corresponding to the adaptive equilibrium centroid and construct the flow collaborative set of the partition unit.
[0025] Cluster cohesion evaluation unit used to calculate the cluster cohesion of the flow collaboration set and to evaluate the degree of cluster cohesion of dynamic call gradient.
[0026] Preferably, the resource consumption assessment and adjustment module includes:
[0027] Used to change the selection object of the adaptive equilibrium centroid within a continuous time period, and to obtain the centroid adjustment unit of the flow cooperation set and cluster cohesion corresponding to different centroids;
[0028] A parametric statistical unit used to calculate the mean and standard deviation of cluster cohesion within a cohesive partition;
[0029] A resource consumption assessment unit used to evaluate the resource consumption status of each partition's individual units and label their status based on the cluster cohesion mean and cluster cohesion standard deviation.
[0030] Used to transfer the processing resource space of redundant state partition units to the processing resource adjustment unit of overloaded state partition units in subsequent consecutive time periods.
[0031] A computer vision data processing method for data stream collaboration, comprising the following steps:
[0032] Step S1: Perform collaborative partitioning, cataloging, and resource allocation on the multi-source computer vision data stream, establish a model that includes dimensions of computing resources, storage resources, and transmission resources, and divide the data into individual partitions;
[0033] Step S2: Configure the number of partition units in each collaborative partition, obtain the real-time data volume of each partition unit and the corresponding collaborative partition, and calculate the dynamic data access degree;
[0034] Step S3: Based on the dynamic invocation gradient, construct the flow collaboration set of the partitioned unit, and evaluate the cluster cohesion of the flow collaboration set;
[0035] Step S4: Based on the cluster mean and cluster standard deviation, assess the resource consumption of individual partitions to adjust the processing resources for collaborative partitioning.
[0036] Preferably, the specific implementation process of step S1 includes:
[0037] The visual data streams from multiple computer sources (including industrial cameras, smart terminals, cloud push, etc.) are collaboratively partitioned and cataloged, and recorded in the collaborative processing center. Each data source corresponds to a collaborative partition, and each collaborative partition is allocated a processing resource.
[0038] A three-dimensional model of the collaborative processing center is established. The x, y, and z axes of the three-dimensional model are respectively composed of the dimensions of computing resources, storage resources, and transmission resources. The total computing resources, total storage resources, and total transmission resources of the collaborative processing center are equally divided in turn, and used as the scale basis on the x, y, and z axes of the three-dimensional model.
[0039] Each scale point on the x, y, and z axes of the 3D model is combined and mapped to the smallest collaborative processing unit of the collaborative processing center, and each smallest collaborative processing unit is used as a partition unit embedded within the collaborative partition.
[0040] Preferably, the specific implementation process of step S2 includes:
[0041] The collaborative partition corresponding to the i-th processing resource is cataloged as follows: Let the nth partition unit be denoted as And partitioned individual Embedded in collaborative partition In this configuration, the minimum number of collaborative processing units embedded within the collaborative partition is set, and each partition unit is instructed to process one visual data stream node.
[0042] Obtain the real-time data volume of the corresponding visual data stream node within each partition unit, and the sum of the real-time data volumes of all partition units embedded in the collaborative partition. Record the sum of these real-time data volumes as the real-time processing resource volume of the collaborative partition. Real-time data volume and collaborative partitioning The ratio of real-time processing resources is denoted as partition unit. Dynamic data retrieval degree And upload it to the collaborative processing center;
[0043] Based on the temporal continuity of visual data streams, a partitioned sample set is generated within the t-th consecutive time period, denoted as . ,in, Indicates collaborative partitioning The number of embedded partition units (i.e., the total number of visual data stream nodes corresponding to the data source).
[0044] Select the nth and mth partition units within the same collaborative partition, and The flow redirection of the visual data stream nodes and quantify the flow direction. Dynamic gradient invocation , used to characterize the dynamic load difference between two individual partitions within the same collaborative partition, where max{} is the maximum value function. Let m be the dynamic data access degree corresponding to the m-th partition unit, and .
[0045] Preferably, the specific implementation process of step S3 includes:
[0046] Within the same collaborative partition, based on the partition sample set With dynamic data retrieval degree For partitioned sample sets The adaptive equilibrium centroid is obtained by collecting all dynamic call gradients of the adaptive equilibrium centroid during the t-th consecutive time interval to form a partitioned unit. Flowing Collaborative Set ;
[0047] Evaluation of the flow of collaborative sets Cluster cohesion In the formula, Represents a flow of collaborative sets The total number of gradients dynamically invoked in the process. For the t-th consecutive time interval, the flow of the cooperative set The mean of the dynamic call gradient (the closer the clustering is to 1, the more concentrated the dynamic call gradient is within the same flow collaboration set, and the more stable the adaptive balance centroid is).
[0048] It should be noted that during the flow of image data stream nodes (such as switching surveillance camera images or handing over tasks at industrial quality inspection nodes), the load will dynamically shift between nodes. A fixed centroid will lead to load assessment bias. The dynamic data call degree is used as the core benchmark for adaptive balance centroid because the dynamic data call degree reflects the load ratio of a single data stream node in real time, and its numerical distribution directly corresponds to the load balance of the nodes. In the t-th consecutive time period, for the sample set of the same collaborative partition, the dynamic data call degree corresponding to each partition is used as a "potential adaptive balance centroid". Dynamic switching is achieved by traversing and selecting. That is, the load state of each data stream node may become a reference benchmark for load balance within the partition, rather than fixing a certain preset node. The selection process needs to be combined with the construction of the flow collaborative set. Only when a certain DDi,n can collect the dynamic call gradient (f(nm)) of all other nodes in the same partition is it determined as an effective centroid, ensuring that the centroid has "global load coverage" to clarify the load adaptation direction of node flow (such as the load distribution path from high-load nodes to centroid nodes) and adapt to the dynamic correlation characteristics of image data stream node flow.
[0049] Preferably, the specific implementation process of step S4 includes:
[0050] Within the same collaborative partition, based on the partition sample set During the t-th consecutive time period, the selection object of the adaptive equilibrium centroid is changed to obtain the flow coordination set when different adaptive equilibrium centroids are selected, and the cluster cohesion of the flow coordination set.
[0051] Quantify the cluster mean of the i-th co-occurrence partition within the t-th consecutive time period. ,and and cluster cohesion standard deviation ,and ;
[0052] The evaluation of dynamic data retrieval degree is based on the cluster cohesion mean and cluster cohesion standard deviation. For partitioned sample sets When the adaptive equilibrium centroid is reached, the nth partition unit resource consumption level ;
[0053] If the level of resource consumption If the value is less than 0, then the partition will be a single unit. Marked as redundant, if resource consumption level If the value is greater than 0, then the partition will be a single unit. Marked as overloaded, if resource consumption is high If the value is 0, then the partition will be a single unit. Marked as a state of equilibrium;
[0054] During the (t+1)th consecutive time period, the processing resource space of the redundant partition unit is transferred to the processing resource space of the overloaded partition unit.
[0055] It should be noted that by quantifying the statistical differences within cluster cohesion, resource consumption is bound to load synergy. In visual data processing, resource consumption depends not only on the amount of data but also on the efficiency of inter-node collaboration (e.g., overloaded nodes are often accompanied by excessively high collaboration concentration). By using the statistical characteristics of continuous time periods to adapt to the long-term fluctuation characteristics of visual data streams (e.g., the difference in daytime / nighttime data volume in intelligent monitoring), resource reallocation becomes "predictive".
[0056] Compared with the prior art, the beneficial effects achieved by the present invention are:
[0057] The present invention, with its three-dimensional resource collaborative modeling and partitioned single-unit embedded design, differs from the existing single-dimensional resource allocation, and realizes the linkage and adaptation of computing, storage, and transmission resources, which helps to solve the problem of resource supply and demand imbalance caused by the heterogeneity of multi-source data streams.
[0058] By adopting a dual-parameter evaluation system of dynamic data access degree and cluster cohesion, it breaks through the limitations of existing single load indicators. It can not only accurately quantify real-time load differences, but also reflect load stability through cluster cohesion, realizing the transformation from "passive response" to "proactive prediction" and reducing overload processing latency.
[0059] Based on the state determination of cluster mean and standard deviation and cross-period resource transfer, unlike the existing fixed-period adjustment, it realizes the "dynamic flow of resources on demand", which helps to solve the resource adaptation problem under the peak of sudden data flow, improves the utilization rate of redundant resources, and does not require manual intervention, adapting to complex and ever-changing application scenarios. Attached Figure Description
[0060] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.
[0061] Figure 1 This is a schematic diagram illustrating the steps of a computer vision data processing method for data flow collaboration according to the present invention. Detailed Implementation
[0062] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0063] In this first embodiment: a computer vision data processing system for data flow collaboration is provided. The system includes: a collaborative partitioning and unit division module, a data processing and parameter calculation module, a flow collaboration set construction and evaluation module, and a resource consumption evaluation and adjustment module.
[0064] The collaborative partitioning and unit division module is used to realize collaborative partitioning and cataloging of multi-source visual data streams, resource dimension model construction, and partition unit division.
[0065] Specifically, the collaborative partitioning and cell division module includes:
[0066] Used for collaborative partitioning and cataloging of multi-source computer vision data streams, and recording the cataloging results into the collaborative partitioning and cataloging unit of the collaborative processing center;
[0067] This is a resource dimension model building unit used to build a model that includes computing resources, storage resources, and transmission resources, and to equally divide the total resources of each type to determine the basis for model scaling.
[0068] Used to map the three-dimensional scale combination of the model into the smallest co-processing unit, and as the partition unit partitioning unit of the embedded partition unit within the co-partition;
[0069] The data processing and parameter calculation module is used to configure the number of partition units, acquire real-time data, calculate relevant parameters for dynamic calls, and quantify load differences.
[0070] Specifically, the data processing and parameter calculation module includes:
[0071] Used to configure the number of partition units embedded in each collaborative partition, and to specify the partition unit configuration unit that a single partition unit handles the corresponding visual data stream node;
[0072] Used to obtain the real-time data volume of each partition unit and its corresponding collaborative partition, calculate the dynamic data access degree of the partition unit, and upload it to the data acquisition and access degree calculation unit of the collaborative processing center.
[0073] A sample set generation unit for generating a sample set containing dynamic data call degree within a continuous time period based on the temporal continuity of visual data streams;
[0074] Load difference quantization unit is used to select different partition units within the same collaborative partition to form flow redirection and quantize the dynamic load difference corresponding to the vector.
[0075] The module for constructing and evaluating the flow collaboration set is used to determine the adaptive equilibrium benchmark, construct the flow collaboration set, and evaluate the degree of concentration of internal parameters.
[0076] Specifically, the collaborative set construction and evaluation module includes:
[0077] The equilibrium centroid determination unit is used to determine the adaptive equilibrium centroid based on the dynamic data call degree, using a sample set as a benchmark.
[0078] The collaborative set building unit is used to collect all dynamic call gradients corresponding to the adaptive equilibrium centroid and construct the flow collaborative set of the partition unit.
[0079] Cluster cohesion evaluation unit used to calculate the cluster cohesion of the flow collaboration set and to evaluate the degree of dynamic call gradient concentration;
[0080] The resource consumption assessment and adjustment module is used to collect relevant parameters, assess the resource consumption status of individual partitions, and process resource adjustments.
[0081] Specifically, the resource consumption assessment and adjustment module includes:
[0082] Used to change the selection object of the adaptive equilibrium centroid within a continuous time period, and to obtain the centroid adjustment unit of the flow cooperation set and cluster cohesion corresponding to different centroids;
[0083] Parametric statistical units used to calculate the mean and standard deviation of cluster cohesion within a cohesive partition;
[0084] A resource consumption assessment unit used to evaluate the resource consumption status of each partition's individual units and label their status based on the cluster cohesion mean and cluster cohesion standard deviation.
[0085] Used to transfer the processing resource space of redundant state partition units to the processing resource adjustment unit of overloaded state partition units in subsequent consecutive time periods.
[0086] Please see Figure 1 In this second embodiment: a computer vision data processing method for data flow collaboration is provided to be applicable to the first embodiment above. This embodiment is applied to the visual quality inspection scenario of an automotive parts production line. The scenario includes 10 industrial cameras (data sources) that collect defect image data of different parts of the engine block, and the image data stream needs to be processed in real time.
[0087] The method includes the following steps:
[0088] Step S1: Perform collaborative partitioning, cataloging, and resource allocation on the multi-source computer vision data stream, establish a model that includes dimensions of computing resources, storage resources, and transmission resources, and divide the data into individual partitions;
[0089] For example, visual data streams from multiple computer sources are collaboratively partitioned and cataloged, and recorded in a collaborative processing center, where one data source corresponds to one collaborative partition, and one collaborative partition is allocated a processing resource.
[0090] A three-dimensional model of the collaborative processing center is established. The x, y, and z axes of the three-dimensional model are respectively composed of the dimensions of computing resources, storage resources, and transmission resources. The total computing resources, total storage resources, and total transmission resources of the collaborative processing center are equally divided in turn, and used as the scale basis on the x, y, and z axes of the three-dimensional model.
[0091] Each scale point on the x, y, and z axes of the 3D model is combined and mapped to the smallest collaborative processing unit of the collaborative processing center, and each smallest collaborative processing unit is used as a partition unit embedded within the collaborative partition.
[0092] For example, 10 data sources correspond to 10 collaborative partitions (CP1-CP). 10 The total computing resources are 1000 GFLOPS, storage resources are 5TB, and transmission resources are 10Gbps. The 3D model is divided into 20 equal scales for each axis, and each partition is a single unit (DP). n ) is the smallest unit of the three-dimensional scale combination. Each collaborative partition contains 5 partition units (Nᵢ=5), corresponding to 5 image data stream nodes.
[0093] Step S2: Configure the number of partition units in each collaborative partition, obtain the real-time data volume of each partition unit and the corresponding collaborative partition, and calculate the dynamic data access degree;
[0094] For example, the collaborative partition catalog corresponding to the i-th processing resource is denoted as Let the nth partition unit be denoted as And partitioned individual Embedded in collaborative partition In this configuration, the minimum number of collaborative processing units embedded within the collaborative partition is set, and each partition unit is instructed to process one visual data stream node.
[0095] Obtain the real-time data volume of the corresponding visual data stream node within each partition unit, and the sum of the real-time data volumes of all partition units embedded in the collaborative partition. Record the sum of the real-time data volumes as the real-time processing resource volume of the collaborative partition. Real-time data volume and collaborative partitioning The ratio of real-time processing resources is denoted as partition unit. Dynamic data retrieval degree And upload it to the collaborative processing center;
[0096] Based on the temporal continuity of visual data streams, a partitioned sample set is generated within the t-th consecutive time period, denoted as . ,in, Indicates collaborative partitioning Number of embedded partition units;
[0097] Select the nth and mth partition units within the same collaborative partition, and The flow redirection of the visual data stream nodes and quantify the flow direction. Dynamic gradient invocation , used to characterize the dynamic load difference between two individual partitions within the same collaborative partition, where max{} is the maximum value function. Let m be the dynamic data access degree corresponding to the m-th partition unit, and ;
[0098] For example, within time period t (10s), the real-time data volume of each partition unit is 10-50MB / s, and the real-time processing resource volume of the collaborative partition is the sum of the data volumes of each partition unit (50-250MB / s). The dynamic data access frequency DDᵢ n =Individual partition data volume / Collaborative partition data volume.
[0099] Step S3: Based on the dynamic invocation gradient, construct the flow collaboration set of the partitioned unit, and evaluate the cluster cohesion of the flow collaboration set;
[0100] For example, within the same collaborative partition, based on the partition sample set With dynamic data retrieval degree For partitioned sample sets The adaptive equilibrium centroid is obtained by collecting all dynamic call gradients of the adaptive equilibrium centroid during the t-th consecutive time interval to form a partitioned unit. Flowing Collaborative Set ;
[0101] Evaluation of the flow of collaborative sets Cluster cohesion In the formula, Represents a flow of collaborative sets The total number of gradients dynamically invoked in the process. For the t-th consecutive time interval, the flow of the cooperative set Dynamically calling the gradient mean;
[0102] For example, with DDᵢ n To adaptively balance the centroids, four dynamic call gradients are collected for each centroid (five individuals in the same partition, four groups of combinations for n≠m), and the cluster cohesion CC is calculated.t (i, n), CC t The values of (i, n) range from 0.82 to 0.91, with a mean of 0.85, indicating that the dynamic call gradient within the flow collaboration set is highly concentrated and the adaptive balance centroid is stable.
[0103] Step S4: Based on the cluster mean and cluster standard deviation, assess the resource consumption of individual partitions to adjust the processing resources for collaborative partitioning;
[0104] For example, within the same collaborative partition, based on the partition sample set During the t-th consecutive time period, the selection object of the adaptive equilibrium centroid is changed to obtain the flow coordination set when different adaptive equilibrium centroids are selected, and the cluster cohesion of the flow coordination set.
[0105] Quantify the cluster mean of the i-th co-occurrence partition within the t-th consecutive time period. ,and and cluster cohesion standard deviation ,and ;
[0106] The evaluation of dynamic data retrieval degree is based on the cluster cohesion mean and cluster cohesion standard deviation. For partitioned sample sets When the adaptive equilibrium centroid is reached, the nth partition unit resource consumption level ;
[0107] If the level of resource consumption If the value is less than 0, then the partition will be a single unit. Marked as redundant, if resource consumption level If the value is greater than 0, then the partition will be a single unit. Marked as overloaded, if resource consumption is high If the value is 0, then the partition will be a single unit. Marked as a state of equilibrium;
[0108] During the (t+1)th consecutive time period, the processing resource space of the redundant partition unit is transferred to the processing resource space of the overloaded partition unit.
[0109] For example, DP2 (CL=0.3) of CP3 and DP4 (CL=0.25) of CP7 are in an overload state, while DP5 (CL=-0.2) of CP2 and DP1 (CL=-0.18) of CP8 are in a redundant state; during the t+1 period (the next 10s), 20% of the resources of each redundant unit will be transferred to the overloaded unit.
[0110] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0111] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention 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 make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A computer vision data processing method for data stream collaboration, characterized in that, The method includes the following steps: Step S1: Perform collaborative partitioning, cataloging, and resource allocation on the multi-source computer vision data stream, establish a model that includes dimensions of computing resources, storage resources, and transmission resources, and divide the data into individual partitions; Step S2: Configure the number of partition units in each collaborative partition, obtain the real-time data volume of each partition unit and the corresponding collaborative partition, and calculate the dynamic data access degree; Step S3: Based on the dynamic invocation gradient, construct the flow collaboration set of the partitioned unit, and evaluate the cluster cohesion of the flow collaboration set; Step S4: Based on the cluster mean and cluster standard deviation, assess the resource consumption of individual partitions to adjust the processing resources for collaborative partitioning.
2. The computer vision data processing method for data stream collaboration according to claim 1, characterized in that, The specific implementation process of step S1 includes: The visual data streams from multiple computer sources are collaboratively partitioned and cataloged, and recorded in the collaborative processing center. Each data source corresponds to one collaborative partition, and each collaborative partition is allocated a processing resource. A three-dimensional model of the collaborative processing center is established. The x, y, and z axes of the three-dimensional model are respectively composed of the dimensions of computing resources, storage resources, and transmission resources. The total computing resources, total storage resources, and total transmission resources of the collaborative processing center are equally divided in turn, and used as the scale basis on the x, y, and z axes of the three-dimensional model. Each scale point on the x, y, and z axes of the 3D model is combined and mapped to the smallest collaborative processing unit of the collaborative processing center, and each smallest collaborative processing unit is used as a partition unit embedded within the collaborative partition.
3. The computer vision data processing method for data stream collaboration according to claim 2, characterized in that, The specific implementation process of step S2 includes: The collaborative partition corresponding to the i-th processing resource is cataloged as follows: Let the nth partition unit be denoted as And partitioned individual Embedded in collaborative partition In this configuration, the minimum number of collaborative processing units embedded within the collaborative partition is set, and each partition unit is instructed to process one visual data stream node. Obtain the real-time data volume of the corresponding visual data stream node within each partition unit, and the sum of the real-time data volumes of all partition units embedded in the collaborative partition. Record the sum of these real-time data volumes as the real-time processing resource volume of the collaborative partition. Real-time data volume and collaborative partitioning The ratio of real-time processing resources is denoted as partition unit. Dynamic data retrieval degree And upload it to the collaborative processing center; Based on the temporal continuity of visual data streams, a partitioned sample set is generated within the t-th consecutive time period, denoted as . ,in, Indicates collaborative partitioning Number of embedded partition units; Select the nth and mth partition units within the same collaborative partition, and The flow redirection of the visual data stream nodes and quantify the flow direction. Dynamic gradient invocation , used to characterize the dynamic load difference between two individual partitions within the same collaborative partition, where max{} is the maximum value function. Let m be the dynamic data access degree corresponding to the m-th partition unit, and .
4. The computer vision data processing method for data stream collaboration according to claim 3, characterized in that, The specific implementation process of step S3 includes: Within the same collaborative partition, based on the partition sample set With dynamic data retrieval degree For partitioned sample sets The adaptive equilibrium centroid is obtained by collecting all dynamic call gradients of the adaptive equilibrium centroid during the t-th consecutive time interval to form a partitioned unit. Flowing Collaborative Set ; Evaluation of the flow of collaborative sets Cluster cohesion In the formula, Represents a flow of collaborative sets The total number of gradients dynamically invoked in the process. For the t-th consecutive time interval, the flow of the cooperative set Dynamically call the gradient mean.
5. A computer vision data processing method for data stream collaboration according to claim 4, characterized in that, The specific implementation process of step S4 includes: Within the same collaborative partition, based on the partition sample set During the t-th consecutive time period, the selection object of the adaptive equilibrium centroid is changed to obtain the flow coordination set when different adaptive equilibrium centroids are selected, and the cluster cohesion of the flow coordination set. Quantify the cluster mean of the i-th co-occurrence partition within the t-th consecutive time period. ,and and cluster cohesion standard deviation ,and ; The evaluation of dynamic data retrieval degree is based on the cluster cohesion mean and cluster cohesion standard deviation. For partitioned sample sets When the adaptive equilibrium centroid is reached, the nth partition unit resource consumption level ; If the level of resource consumption If the value is less than 0, then the partition will be a single unit. Marked as redundant, if resource consumption level If the value is greater than 0, then the partition will be a single unit. Marked as overloaded, if resource consumption is high If the value is 0, then the partition will be a single unit. Marked as a state of equilibrium; During the (t+1)th consecutive time period, the processing resource space of the redundant partition unit is transferred to the processing resource space of the overloaded partition unit.
6. A computer vision data processing system for data flow collaboration, executing a computer vision data processing method for data flow collaboration as described in any one of claims 1-5, characterized in that, The system includes: a collaborative partitioning and unit division module, a data processing and parameter calculation module, a flow collaborative set construction and evaluation module, and a resource consumption evaluation and adjustment module; The collaborative partitioning and unit division module is used to realize collaborative partitioning and cataloging of multi-source visual data streams, construction of resource dimension models, and partition unit division. The data processing and parameter calculation module is used to configure the number of partition units, acquire real-time data, calculate dynamic call-related parameters, and quantify load differences. The circulation collaboration set construction and evaluation module is used to determine the adaptive balance benchmark, construct the circulation collaboration set, and evaluate the degree of concentration of internal parameters; The resource consumption assessment and adjustment module is used to collect relevant parameters, assess the resource consumption status of individual partitions, and process resource adjustments.
7. A computer vision data processing system for data stream collaboration according to claim 6, characterized in that, The collaborative partitioning and unit division module includes: Used for collaborative partitioning and cataloging of multi-source computer vision data streams, and recording the cataloging results into the collaborative partitioning and cataloging unit of the collaborative processing center; This is a resource dimension model building unit used to build a model that includes computing resources, storage resources, and transmission resources, and to equally divide the total resources of each type to determine the basis for model scaling. It is used to map the three-dimensional scale combination of the model into the smallest co-processing unit, and to serve as the partition unit of the embedded partition unit within the co-partition.
8. A computer vision data processing system for data stream collaboration according to claim 6, characterized in that, The data processing and parameter calculation module includes: Used to configure the number of partition units embedded in each collaborative partition, and to specify the partition unit configuration unit that a single partition unit processes the corresponding visual data stream node; Used to obtain the real-time data volume of each partition unit and its corresponding collaborative partition, calculate the dynamic data access degree of the partition unit, and upload it to the data acquisition and access degree calculation unit of the collaborative processing center. A sample set generation unit for generating a sample set containing dynamic data call degree within a continuous time period based on the temporal continuity of visual data streams; The load difference quantization unit is used to select different partition units within the same collaborative partition to form flow redirection and quantize the dynamic load differences corresponding to the vectors.
9. A computer vision data processing system for data stream collaboration according to claim 6, characterized in that, The module for constructing and evaluating the transfer collaboration set includes: The equilibrium centroid determination unit is used to determine the adaptive equilibrium centroid based on the dynamic data call degree, using a sample set as a benchmark. The collaborative set building unit is used to collect all dynamic call gradients corresponding to the adaptive equilibrium centroid and construct the flow collaborative set of the partition unit. Cluster cohesion evaluation unit used to calculate the cluster cohesion of the flow cohesion set and to evaluate the degree of dynamic call gradient concentration.
10. A computer vision data processing system for data stream collaboration according to claim 6, characterized in that, The resource consumption assessment and adjustment module includes: Used to change the selection object of the adaptive equilibrium centroid within a continuous time period, and to obtain the centroid adjustment unit of the flow cooperation set and cluster cohesion corresponding to different centroids; Parametric statistical units used to calculate the mean and standard deviation of cluster cohesion within a cohesive partition; A resource consumption assessment unit used to evaluate the resource consumption status of each partition's individual units and label their status based on the cluster cohesion mean and cluster cohesion standard deviation. Used to transfer the processing resource space of redundant state partition units to the processing resource adjustment unit of overloaded state partition units in subsequent consecutive time periods.