LED display screen task allocation matrix generation and resource scheduling system

Abstract

The invention discloses an LED display screen task allocation matrix generation and resource scheduling system, relates to the technical field of edge calculation and LED display control, and is used for solving the problems of high video transmission redundancy, low task scheduling efficiency and inaccurate flicker identification. An image frame and state data are extracted, an inter-frame difference and optical flow fusion algorithm is adopted to identify a high dynamic region, a periodic flicker region is extracted in combination with phase consistency and short-time Fourier transform, and a flicker mask and a compressed data packet are generated. A templating mechanism and a meta-learning strategy are introduced into the system, and efficient coding and recognition of a low-change area are achieved. And meanwhile, efficient synchronization and fault-tolerant processing among distributed nodes is supported. The scheduling module dynamically constructs a task allocation matrix based on the node resource state and the flicker burden, and gives consideration to both the transmission efficiency and the task suitability. According to the system, the real-time performance, the accuracy and the resource utilization rate of data transmission and instruction execution of the LED display screen can be remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of edge computing and LED display control, more particularly, the present application relates to a LED display screen task allocation matrix generation and resource scheduling system. BACKGROUND

[0002] With the wide application of LED display screens in the fields of advertising media, traffic guidance, stage performance, etc., how to improve the data transmission efficiency and resource scheduling capability of the system while ensuring the display effect has become a key problem to be solved. The existing LED display control system relies on full-frame refresh and static scheduling strategy, ignores the uneven characteristics of regional changes in video content, resulting in a large amount of redundant data being repeatedly transmitted, and the system bandwidth resources being inefficiently occupied. At the same time, in the face of high-density node deployment of distributed LED array system, the traditional scheduling mechanism is difficult to dynamically perceive the load state of each node, cannot flexibly adapt to high-change tasks and resource bottlenecks, and is prone to problems such as delay increase and display stuttering.

[0003] In addition, in actual use scenarios, LED pictures exist periodic flickering, slight jitter and other phenomena. Such slight brightness disturbance in static background is often misjudged as a large area of update region by traditional frame difference or optical flow algorithm, resulting in unbalanced allocation of coding resources. The existing method lacks identification and processing mechanism for such regions, and cannot compress low-value data while ensuring content consistency.

[0004] In view of the above problems, the present application provides a solution. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a LED display screen task allocation matrix generation and resource scheduling system to solve the problems raised in the background art.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical scheme: In one preferred embodiment, it comprises: an edge intelligent perception and processing module, a distributed protocol fusion and instruction synchronization module, a dynamic resource scheduling and task allocation matrix generation module and a data storage module, the modules are signal connected; The edge intelligent perception and processing module is mainly used for identifying fault state by edge perception of LED video stream and device state; The distributed protocol fusion and instruction synchronization module is mainly used for converting the protocol of compressed and fault data packets, synchronizing the instruction timeline, and dynamically adjusting the scheduling weight of the device node according to the same; The dynamic resource scheduling and task allocation matrix generation module is mainly used for evaluating the available resource capability of each node and the performance-price ratio of the adaptive task, and then constructing a task allocation matrix; The data storage module is mainly used for storing all data in the processing process.

[0007] In a preferred embodiment, in the edge intelligent perception and processing module, image data and state variables of the LED display screen are collected, regions with obvious displacement or brightness change in the image data are captured, region screening is performed, the image data is divided into full-frame grids, and a unique identification code is generated for the image region not participating in dynamic change.

[0008] In a preferred embodiment, in the edge intelligent perception and processing module, the pixel difference region is classified into several closed regions, the quantization bit width Qi and the compression code rate μi of each region are determined, and the fault probability value is determined.

[0009] In a preferred embodiment, in the edge intelligent perception and processing module, a fine-grained grid of the dynamic region and a coarse-grained grid of the static region are generated, and brightness extreme normalization is performed; a phase consistency gating factor is constructed, and the optical flow tensor is multiplied by the gating factor; A sliding window pixel difference sequence with a length of L is maintained for each grid, and a short-time Fourier transform is performed on the time sequence to calculate the energy concentration degree and the phase drift degree.

[0010] In a preferred embodiment, in the edge intelligent perception and processing module, the mode that the amplitude energy is concentrated in a single thin strip and the phase hardly drifts is detected, a corresponding binary mask is generated, and real motion and static flicker are distinguished; The mask generated in the above step is applied to the frame difference tensor and the optical flow tensor, difference screening is performed in the template library, and an evolvable flicker knowledge base is constructed.

[0011] In a preferred embodiment, in the distributed protocol fusion and instruction synchronization module, a key field is added to the header of each differential data packet, and the corresponding intermediate instruction is automatically reduced to a low-priority sub-queue, only one signature confirmation is performed, and repeated retransmission is skipped; the fast path or the slow path of PBFT is dynamically selected.

[0012] In a preferred embodiment, in the dynamic resource scheduling and task allocation matrix generation module, the number of flicker mask regions in the current cache is counted, the flicker burden coefficient is defined, and the scheduling weight is corrected.

[0013] In a preferred embodiment, in the dynamic resource scheduling and task allocation matrix generation module, the appearance period and the matching hit rate of each newly detected flicker template are recorded, a mutual information distance matrix is calculated, a hit rate threshold is set, and if the mask matching hit rate decreases by more than the hit rate threshold in a continuous period, the meta-learning process is triggered.

[0014] The technical effects and advantages of the LED display screen task allocation matrix generation and resource scheduling system of the application are: The present application can not only accurately identify high dynamic regions and periodic flickering regions in LED display screens by deeply integrating image change detection and device state perception, but also realize differential coding based on compression weights and phase characteristics to effectively reduce data redundancy and transmission pressure. Compared with traditional scheduling methods, the present application introduces a flickering mask and a template matching mechanism to make task allocation more targeted and improve the scheduling tolerance of low change regions and the overall load balancing ability of the system. In addition, through the optimized distributed protocol fusion and lightweight PBFT synchronization path switching strategy, the number of redundant confirmations and broadcasts is greatly reduced, and the instruction synchronization efficiency and fault tolerance are improved. The system has stronger adaptability and evolvability, and can maintain stable and reliable display control performance in a multi-source heterogeneous device environment. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 The embodiment of the LED display screen task allocation matrix generation and resource scheduling system is shown in the flowchart.

[0016] Figure 2 The timing diagram of the LED display screen task allocation matrix generation and resource scheduling system is shown in the flowchart. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0018] EMBODIMENT The present application discloses an LED display screen task allocation matrix generation and resource scheduling system, as shown in Figure 1 The system includes an edge intelligent perception and processing module, a distributed protocol fusion and instruction synchronization module, a dynamic resource scheduling and task allocation matrix generation module, and a data storage module, which are signal connected.

[0019] The data storage module is mainly used for storing all data in the processing process. The edge intelligent perception and processing module is mainly used for identifying high dynamic regions and fault states through edge perception of LED video streams and device states, and then structuring the identification results into compression data packets and fault data packets. First, deploy the collector on the edge controller of the LED display screen, set to capture a frame Ft of image data every fixed period, and keep the image data cache of the last frame F(t-1), while collecting the state variables of the LED display screen from the underlying sensor interface: St=[Tt, Bt, Pt, Lt], where Tt represents temperature, Vt represents power supply voltage, Pt represents power load, and Lt represents data packet loss rate, and then output the image frame pair (F(t-1), Ft) and the state vector St.

[0020] Then, use the inter-frame difference + optical flow fusion algorithm to capture the area with obvious displacement or brightness change in the image data, and the specific steps are as follows: Calculate the inter-frame difference map, specifically according to the formula: ; Use the gray-scale pixels for operation, which is strong in anti-interference; Calculate the optical flow vector map, specifically according to the formula: ; Get the motion vector of each pixel point, which represents the local displacement direction and speed; Calculate the change fusion intensity map, specifically according to the formula: ; Where α, β ∈ [0, 1], which are obtained by cross-validation training.

[0021] Then, fuse the change intensity map Ct(x, y) and set the threshold Yc to perform region screening, specifically according to the following: ; At the same time, in the actual scene, in order to avoid false detection caused by screen shaking, brightness disturbance or edge blurring, etc., multi-scale filtering and median noise reduction processing need to be performed on the original intensity map to further improve the robustness of region screening. Therefore, perform two-dimensional connected component extraction on Rt, merge it into a rectangular region Bk, and filter out the pseudo-change block with a pixel number less than the threshold, extract the boundary coordinates of the high dynamic region from the change intensity map, mark it as the key update unit, and output the high dynamic region set B={B1,...,Bk} as the target region for difference compression.

[0022] Further, construct a full-frame grid division for the acquired image data of the LED display screen, filter out all the grids located in the dynamic region Bk, and for the remaining static grid region, perform the perceptual hash function pHash to generate a unique identification code for the image region that does not participate in dynamic change, while keeping the current time hash table for instruction abstraction and data structure compression.

[0023] After the completion of the inter-frame difference and optical flow fusion, mask compensation, the pixel coordinate set of the salient region difference is obtained, and the grid space clustering is applied to the set. The pixel difference region is classified into several closed regions Ri using a density clustering algorithm such as DBSCAN or a connected domain marking algorithm, and a unique region ID RIDi is assigned to each region.

[0024] Then the gray level histogram features, contrast level, texture complexity and other indicators of each region are extracted, and the compression weight vector qi is constructed. Specifically according to the formula: qi = [Ci, Vi, Ti]; Among them, Ci represents the contrast factor, Vi represents the brightness variance, and Ti represents the high-frequency texture distribution amount. Combined with the region characteristics and the current bandwidth load B(t), the quantization bit width Qi and the compression code rate μi of the region are adaptively determined, wherein the calculation of the quantization bit width Qi is specifically according to the formula: ; The calculation of the compression code rate μi is specifically according to the formula: ; Among them, β1, β2, β3, γ represent the compression strategy parameters, and Δt represents the encoding time window.

[0025] It should be noted that the compression strategy not only focuses on the data complexity of a single region, but also considers the current transmission state and computing load of the device. Under a specific bandwidth bottleneck, the encoding quality of high-priority regions is prioritized, and stronger quantization strategies are adopted for secondary change regions, thereby achieving a dynamic balance between global compression efficiency and content recognition importance.

[0026] Then, the quantization bit width Qi is used to perform nonlinear gray mapping and variable-length coding on the pixels in the region to construct a structured compression data packet.

[0027] Further, three types of device operation data, temperature, voltage fluctuation and packet loss rate, are collected; and the data stream is divided into fixed time windows Tw, and a multivariate sequence is generated for each window. Then, the window sequence of each node is input into the trained LSTM model, and the fault probability value of the node in the current window is output. It is worth noting that the LSTM model used has been trained through historical large-scale device operation logs and has strong sequence prediction and context modeling capabilities, which can identify the gradual deterioration trend of the device at an early stage, rather than only reacting to sudden failures, thereby improving the stability and fault warning lead time of the entire system.

[0028] Further, a fault threshold Yf is set, when the fault probability of a certain device is greater than the preset value, a fault data packet containing the device ID and the fault type is generated, and is broadcast in a structured packet form, supporting subsequent scheduling reconstruction and resource migration.

[0029] It should be noted that when the overall brightness region of the collected input video stream data is almost unchanged, but the local pixels periodically flicker, and the content position is fixed and the change rhythm is consistent in different frames, it will cause the optical flow direction to be unstable and the frame difference to change frequently but the region to be unchanged when the frame difference combined with the optical flow fusion algorithm extracts the change region, resulting in the inability to accurately reflect the jump of the small area and the easy misidentification as a large area update region. Therefore, in the embodiment, when calculating the frame difference and the optical flow vector pixel by pixel, the following steps are performed: Step A1: the density distribution of the optical flow tensor in space within the past time window is counted; and a fast clustering algorithm K-means based on optical flow density or DBSCAN based on density is performed on the picture to generate a fine-grained grid of dynamic regions and a coarse-grained grid of static regions. Then, for each grid, the brightness extreme value normalization is performed, which is specifically based on the formula: ; Where Igrid represents the set of all pixel brightness in the grid. Step A2: After performing frame difference and TV-L1 optical flow calculation, the original tensor Ft(x,y) is obtained, and a phase consistency gating factor Gt(x,y) is constructed for measuring the brightness phase stability of the pixel position in multiple frames, which is defined as Gt(x,y), which is specifically based on the following formula: ; Where ∠I represents the phase of the local brightness signal. At the same time, in order to prevent it from misleading subsequent dynamic detection, it is necessary to multiply the optical flow tensor by the gating factor , which is specifically based on the formula: ; To ensure that the optical flow vector of the pixel that "reverses the brightness but does not move in space" is de-weighted in advance, and is suppressed to near zero. Step A3: maintain a sliding window pixel difference sequence with a length of L for each grid, and perform short-time Fourier transform STFT on the time sequence, which is specifically based on the following formula: ; Further, the amplitude spectrum At(f) and the phase spectrum ϕt(f) are extracted respectively, and the energy concentration degree and the phase drift degree are calculated, wherein the energy concentration degree calculation formula is specifically based on the following formula: ; The phase dispersion metric calculation formula is specifically as follows: ; Wherein, f* represents the main frequency; After that, if the mode of "amplitude energy concentrated in a single narrow band, and phase almost does not drift" is detected, that is, CA>θ1 and Dϕ<θ2, the grid is marked as "flicker template", and the corresponding binary mask Mflash is generated, so as to accurately identify the pixel change area with periodicity, narrow frequency band and stable phase, and to distinguish between real motion and "static flicker".

[0030] Step A4: the mask Mflash generated in the last step is applied to the frame difference tensor and the optical flow tensor, Specifically: ; At the same time, a "low confidence area" flag bit is added in the metadata, the flicker template is compressed into a frequency domain feature vector, and stored in the metadata memory pool; After that, the mutual information index I(Ti, Tj) is used to make difference screening in the template library, and the most distinctive template is reserved to build an evolvable flicker knowledge base, which is used for fast matching in subsequent video stream processing, so as to realize the fast convergence and long-term adaptive optimization of flicker recognition in the cold start stage.

[0031] The distributed protocol fusion and instruction synchronization module first analyzes the protocol fields in the data packet, such as brightness, refresh rate and instruction type; using the protocol intelligent mapping library, different manufacturer command fields are mapped into a unified intermediate operation code structure, such as Figure 2 As shown in the figure; Specifically, the distributed protocol fusion and instruction synchronization module adds two key fields flash_mask_flag and template_id in the header of each differential data packet, wherein the flash_mask_flag field represents a Boolean bit, marking whether the packet contains a flicker mask; the template_id field represents a unique identifier pointing to an existing template in the local flicker template pool.

[0032] Further, when the protocol fusion parses flash_mask_flag=1 and template_id already exists in the opposite end cache, the instruction is marked as "confirmation-free repeated instruction", and the instruction only needs to be signed once in the Byzantine Fault Tolerance (PBFT) process, without broadcasting full data, skipping redundant retransmission; After that, based on the network node delay prediction model T^latency(i,j), the "fast path" or "slow path" of PBFT is dynamically selected, which is specifically as follows: ; Wherein, in the "slow path" execution, the invalid or non-responsive node is allowed to insert a null placeholder value, ensuring that the master synchronization window continues to advance and is not delayed, reducing the communication cost of frequent repeated synchronization and improving processing throughput and response sensitivity.

[0033] The dynamic resource scheduling and task allocation matrix generation module is mainly used for modeling the state space, evaluating the available resource capacity and cost performance of each node, and constructing a task allocation matrix according to the weight, energy consumption, delay and other limiting factors, and finally outputting a real-time scheduling decision.

[0034] First, in the dynamic resource scheduling and task allocation matrix generation module, for each edge node n, the number of "flicker mask regions" Mn in its current cache is counted, and the flicker burden coefficient of the node is defined: ; Where Tn represents the total number of tasks carried by the node.

[0035] Then, the scheduling weight correction is performed on the original task scheduling score function Wn, and a flicker weighting item is introduced: ; Where α represents the scheduling sensitivity coefficient. It should be noted that after the scheduling weight correction, the nodes with more "update-free regions" are preferentially used to carry low-change-rate tasks.

[0036] At the same time, all tasks containing "flicker masks" are marked as "locally visually composable" type, and the delay constraint function Dmax of the scheduler is automatically relaxed by one level: Record the appearance period and matching hit rate Hi(t) of each newly detected flicker template Ti in the event log, and at low load periods, perform K-Medoids clustering on all historical templates, calculate the mutual information distance matrix between each other, and retain the center templates that cover most of the samples, delete redundant templates, and compress the space occupation.

[0037] Further, set the hit rate threshold ε, if the mask matching hit rate drops by more than the hit rate threshold ε in the last N periods, trigger the meta-learning process: Extract the "difficult example" fragments that fail to match from the event log; Re-train the parameter weights of the phase gating function Gt(x, y) to optimize its time window and phase perturbation tolerance; Integrate the newly generated template into the memory pool, and reduce the weight or eliminate the old template.

[0038] The above formulas are all de-dimensioned to calculate the numerical values, the formulas are obtained by collecting a large amount of data to simulate a formula of the most recent real situation, and preset parameters in the formulas are set by a person skilled in the art according to actual conditions.

[0039] The above embodiments can be realized wholly or partially by software, hardware, firmware or any other combination. When realized by software, the above embodiments can be realized wholly or partially in the form of a computer program product.

[0040] Those skilled in the art can realize that the modules and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and the constraints of the technical solution. A person skilled in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0041] In addition, each functional module in each embodiment of the present application can be integrated in one processing module, or each module can exist physically alone, or two or more modules can be integrated in one module.

[0042] The above is merely a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0043] Finally, the above is merely a preferred embodiment of the present application, and is not used to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. LED display task allocation matrix generation and resource scheduling system, characterized by ; Includes: edge intelligent perception and processing module, distributed protocol fusion and instruction synchronization module, dynamic resource scheduling and task allocation matrix generation module and data storage module, and signal connection between modules; The edge intelligent perception and processing module is mainly used to identify fault conditions through edge perception of LED video streams and device status; The distributed protocol fusion and instruction synchronization module is mainly used to synchronize instruction timelines through protocol abstract conversion of compressed and faulty data packets, and dynamically adjust the scheduling weights of device nodes accordingly; The dynamic resource scheduling and task allocation matrix generation module is mainly used to evaluate the available resource capacity of each node and the cost-effectiveness of the adapted tasks, and then construct the task allocation matrix; The data storage module is mainly used to store all data in the processing process.

2. The LED display task allocation matrix generation and resource scheduling system according to claim 1, characterized in that: In the edge intelligent perception and processing module, the image data and state variables of the LED display are collected, areas with obvious displacement or brightness changes in the image data are captured, regional screening is performed, the full frame of the image data is gridded, and a unique identification code is generated for the image areas that do not participate in dynamic changes.

3. The LED display task allocation matrix generation and resource scheduling system according to claim 2, characterized in that: In the edge intelligent perception and processing module, the pixel difference area is divided into several closed areas, the quantization bit width Qi and compression code rate μi of each area are determined, and the fault probability value is determined.

4. The LED display screen task allocation matrix generation and resource scheduling system according to claim 3, characterized in that; In the edge intelligent perception and processing module, a fine-grained grid is generated in the dynamic area and a coarse-grained grid is generated in the static area, and brightness extreme value normalization is performed; a phase consistency gating factor is constructed and the optical flow tensor is multiplied by the gating factor; A sliding window pixel difference sequence of length L is maintained for each grid, and a short-time Fourier transform is performed on the time series to calculate the energy concentration and phase drift.

5. The LED display task allocation matrix generation and resource scheduling system according to claim 4, characterized in that: In the edge intelligent perception and processing module, a pattern with amplitude energy concentrated in a single thin band and almost no phase drift is detected, and a corresponding binary mask is generated to distinguish between real motion and static flicker; Apply the mask generated in the previous step to the frame difference tensor and the optical flow tensor, perform differential screening in the template library, and build an evolvable flicker knowledge base.

6. The LED display task allocation matrix generation and resource scheduling system according to claim 5, characterized in that: In the distributed protocol fusion and instruction synchronization module, a key field is added to the header of each differential data packet, and the corresponding intermediate instructions are automatically relegated to a low-priority subqueue, with only one signature confirmation performed, skipping repeated retransmissions; and the PBFT fast path or slow path is dynamically selected.

7. The LED display screen task allocation matrix generation and resource scheduling system according to claim 6, characterized in that: In the dynamic resource scheduling and task allocation matrix generation module, the number of flicker mask areas in the current cache is counted, the flicker burden coefficient is defined, and the scheduling weight is corrected.

8. The LED display screen task allocation matrix generation and resource scheduling system according to claim 7, characterized in that; In the dynamic resource scheduling and task allocation matrix generation module, the occurrence period and matching hit rate of each newly detected flicker template are recorded, the pairwise mutual information distance matrix is ​​calculated, and the hit rate threshold is set. If the mask matching hit rate drops below the hit rate threshold in consecutive cycles, the meta-learning process is triggered.

Images