Distributed LED transverse fine tearing immediate alignment method and device and storage medium

Abstract

A distributed LED transverse fine tearing instant alignment method and device and a storage medium are applied to the field of artificial intelligence, and the method comprises the following steps: obtaining the optimal offset of each control card and the local timestamp of each scanning line, and generating a global alignment time matrix; generating a time-space-environment three-dimensional coupling data cube according to the global alignment time matrix, the bidirectional mapping function and the temperature and the power supply voltage of each control card; performing digital twinborn prediction according to the three-dimensional coupling data cube to obtain a predicted phase field; carrying out weight fusion on a virtual tearing saliency map obtained according to the predicted phase field and an optical flow dislocation intensity map obtained according to the camera shooting screenshot, generating a tearing probability map, and obtaining phase deviation corresponding to each control card; obtaining a discrete phase compensation vector according to the phase deviation, and packaging the discrete phase compensation vector into a phase compensation instruction frame; and in the inter-frame security window, issuing phase compensation instruction frames to all the control cards in parallel. And fine transverse fine tearing of the distributed LED is eliminated on line.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to a distributed LED horizontal fine tearing instant alignment method and device and storage medium. BACKGROUND

[0002] The existing distributed LED display system is prone to horizontal fine tearing when playing high-speed pictures on a large-size screen driven by multiple control cards. The existing public technology usually classifies this problem as "frame synchronization" or "line synchronization" mismatch, and gives solutions in hardware, software and visual feedback, forming several industry practices that can be implemented. Among them, multi-clock distribution, network timestamp correction, visual feedback compensation and environmental monitoring correction constitute the main horizontal tearing (HT) suppression means in the current industry.

[0003] Although the current hardware-centric timing scheme provides a unified reference at the physical layer, the installation process requires sub-microsecond length and delay calibration for each distribution cable, and cannot make online corrections to dynamic phase changes caused by temperature drift, power fluctuations or board card aging in actual operation. Once the screen size expands or the wiring path changes, the cost of re-calibration increases significantly. The synchronization method based on network timestamp relies on the statistical average of link round-trip delay, and in the face of sudden jitter or exchange node re-routing, transient errors can easily return to the line start counter. Microsecond-level residual errors are still enough to appear as 1-2 pixel horizontal tearing in high-speed pictures. The visual closed loop introduced by the camera can sense the final display effect, but it is very sensitive to moving pictures, high brightness flicker and environmental light interference, and its correction period is limited by the camera frame rate, making it difficult to complete closed adjustment within each frame refresh window. The practice of mapping temperature and power information into a fixed compensation curve lacks a detailed modeling of the timing, thermal and electrical coupling characteristics, and cannot accurately depict the phase shift caused by local hot spots or transient load changes, which is prone to overcompensation or insufficient compensation at the boundary position.

[0004] In summary, the existing technical routes are still in a scattered state between clock synchronization, visual detection and environmental adaptation, lacking a real-time closed-loop mechanism that can integrate multi-source information within a millisecond time scale and output a directly executable compensation amount, resulting in limited reliability and accuracy of online elimination of fine tearing. SUMMARY

[0005] At least one embodiment of the present application provides a distributed LED horizontal fine tearing instant alignment method, device and storage medium, which solves the problem of limited reliability and accuracy of online elimination of fine horizontal tearing in the prior art.

[0006] To solve the above technical problems, the present application is implemented as follows:

[0007] In a first aspect, the embodiments of the present application provide a distributed LED horizontal fine tearing instant alignment method, comprising:

[0008] obtaining optimal offset of each control card relative to a unified reference clock and local timestamps of each scanning line of the control card, and generating a globally aligned time matrix with double indexes of line and card; wherein the optimal offset is determined according to network round-trip delay measurement and time difference of common boundary lines between control cards; the local timestamp is obtained by triggering a board-mounted nanosecond counter through a line start hardware interrupt of each control card;

[0009] generating a time-space-environment three-dimensional coupled data cube according to the globally aligned time matrix, a preset bidirectional mapping function, and temperature and power supply voltage of each control card through downsampling to a pixel grid, wherein the bidirectional mapping function is a mapping function from physical pixel coordinates to control card number-line sequence index established according to installation drawing and pixel row-column sequence;

[0010] performing digital twin prediction according to the three-dimensional coupled data cube to obtain a predicted phase field;

[0011] performing weight fusion on a virtual tearing saliency map obtained by calculating a horizontal gradient of the predicted phase field and an optical flow misplacement intensity map obtained by performing optical flow difference processing on a camera screenshot, generating a tearing probability map and obtaining a phase deviation corresponding to each control card;

[0012] performing cross-card phase compensation solving and mapping according to the phase deviation, obtaining a discrete phase compensation vector, and packaging as a phase compensation instruction frame; wherein the phase compensation instruction frame is used to make each control card write the discrete phase compensation vector in a firmware programmable clock interface, and lock the effective time at the next frame line sequence counter reset point;

[0013] After entering an inter-frame safety window, the phase compensation instruction frame is issued to all the control cards in parallel.

[0014] Specifically, the method as described above, the obtaining of the optimal offset of each control card relative to the unified reference clock and the local timestamps of each scanning line of the control card, and the generation of the globally aligned time matrix with double indexes of line and card, comprises:

[0015] obtaining clock offset of each control card relative to the unified reference clock according to the network round-trip delay measurement;

[0016] According to the clock offset, the local timestamp, a row index set located at the control card boundary, and a first preset algorithm, iterative optimization of gradient descent is performed to obtain the optimal offset;

[0017] According to the optimal offset, the local timestamp matrix formed by the local timestamp is corrected to obtain the global alignment time matrix.

[0018] Specifically, according to the above method, the time-space-environment three-dimensional coupled data cube is generated according to the global alignment time matrix, a preset bidirectional mapping function, and the temperature and power supply voltage of each control card by downsampling to a pixel grid, including:

[0019] According to the bidirectional mapping function, the global alignment time matrix is accessed element by element to generate a pixel phase matrix, wherein the control card number in the bidirectional mapping function is used to read the column in the global alignment time matrix, and the row sequence index in the bidirectional mapping function is used to read the row in the global alignment time matrix.

[0020] According to the control card number, the amplitude of the temperature and the power supply voltage is obtained for each point of the two-dimensional pixel plane to obtain a temperature matrix and a voltage matrix.

[0021] The temperature matrix is subjected to range normalization processing to obtain a corresponding target temperature matrix.

[0022] The voltage matrix is subjected to range normalization processing to obtain a corresponding target voltage matrix.

[0023] The pixel phase matrix, the target temperature matrix, and the target voltage matrix are channel spliced to obtain the three-dimensional coupled data cube.

[0024] Specifically, according to the above method, the digital twin prediction is performed according to the three-dimensional coupled data cube to obtain a predicted phase field, including:

[0025] According to the control card number list, the module physical block size, and the pixel row and column sequence, a two-dimensional driving grid is constructed, and a three-dimensional state vector of each node in the two-dimensional driving grid is initialized, the three-dimensional state vector including a phase value, a temperature value, and a voltage value.

[0026] The three-dimensional state vector is forward multi-physical step predicted on the two-dimensional driving grid to obtain an a priori state vector.

[0027] According to the node, the three-dimensional coupled data cube is parsed into an observation vector and an observation mapping operator is constructed.

[0028] performing dynamic assimilation according to the prior state vector, the observation vector, the observation mapping operator, and a gain matrix, the gain matrix being obtained according to a current neighborhood variance, an observation precision, and a scanning direction;

[0029] extracting a phase channel from the optimal state set and rearranging according to node coordinates to obtain the predicted phase field.

[0030] Further, the method as described above, the forward multi-physical step prediction of the three-dimensional state vector on the two-dimensional driving grid to obtain a prior state vector, comprises:

[0031] calling a line sequence step operator to perform single-frame recursion on the phase value to obtain a prior phase;

[0032] calculating the temperature value based on a two-dimensional heat conduction operator to obtain a prior temperature;

[0033] calculating the voltage value based on a preset node-edge resistance network to obtain a prior voltage;

[0034] concatenating the prior phase, the prior temperature, and the prior voltage into the prior state vector.

[0035] Preferably, the method as described above, after obtaining the predicted phase field, the method further comprises:

[0036] performing temporal consistency verification and spatial smoothness verification on the predicted phase field;

[0037] marking an area passing the temporal consistency verification and the spatial smoothness verification as an ideal synchronous state.

[0038] Preferably, the method as described above, obtaining the optical flow misplacement intensity map obtained by performing optical flow difference processing on the camera screenshot, comprises:

[0039] calculating an optical flow vector field of the camera screenshot by a depth separable pyramid algorithm; taking an absolute value and squaring a transverse component of the optical flow vector field to visualize and output an actual misplacement intensity map;

[0040] performing weight suppression on the actual misplacement intensity map according to a normalized environment confidence of the environment channel of the three-dimensional coupled data cube at a current time to obtain the optical flow misplacement intensity map.

[0041] Specifically, the method as described above, the weight fusion of the virtual tearing saliency map obtained according to the transverse gradient calculation on the predicted phase field and the optical flow misplacement intensity map obtained by performing optical flow difference processing on the camera screenshot to generate a tearing probability map and obtain the phase deviation corresponding to each control card, comprises:

[0042] According to the virtual tearing saliency map, the optical flow misplacement intensity map and the environment confidence, weight fusion is performed to generate the tearing probability map, wherein the weights corresponding to the virtual tearing saliency map, the optical flow misplacement intensity map and the environment confidence are all calibrated offline according to the Bayesian minimum risk criterion;

[0043] According to the bidirectional mapping function, a public boundary set of the control card is obtained;

[0044] Taking the public boundary set as an index, a product of a tearing probability of a cumulative pixel of each boundary and a pixel phase difference is calculated to obtain the phase deviation.

[0045] Preferably, the method as described above, before the weight fusion of the virtual tearing saliency map obtained by calculating the transverse gradient of the predicted phase field and the optical flow misplacement intensity map obtained by performing optical flow difference processing on the camera screenshot, the generation of the tearing probability map and the acquisition of the phase deviation corresponding to each control card, the method further comprises:

[0046] According to the frame sequence number returned by the control card and the camera time stamp of the camera image captured by the camera system, the average time delay of the camera system to the display system is estimated;

[0047] According to the difference between the camera time stamp and the average time delay, the predicted phase field and the camera screenshot corresponding to the output difference value are obtained;

[0048] According to the bidirectional mapping function and the preset quadratic polynomial correction term, the camera screenshot is projected into the coordinate system corresponding to the predicted phase field;

[0049] According to the sub-pixel level bilinear interpolation, the camera screenshot and the predicted phase field are coordinate-aligned to obtain the aligned camera screenshot.

[0050] Specifically, according to the method as described above, the cross-card phase compensation solving and mapping according to the phase deviation, the acquisition of the discrete phase compensation vector and the encapsulation into the phase compensation instruction frame, comprise:

[0051] Taking the tearing probability in the tearing probability map as a weight, a target function of cross-card boundary phase consistency is constructed;

[0052] Taking the phase deviation as an initial value, a first-order gradient descent solving of the target function is performed based on a constrained least squares iteration to obtain a continuous compensation solution;

[0053] The continuous compensation solution is projected into the adjustable step set corresponding to each control card to obtain the discrete phase compensation value corresponding to each control card;

[0054] recombine all the discrete phase compensation values into the discrete phase compensation vector, and encapsulate the discrete phase compensation vector into the phase compensation instruction frame.

[0055] Preferably, the method as described above, before the encapsulating the discrete phase compensation vector into the phase compensation instruction frame, further comprises:

[0056] substituting the discrete phase compensation vector into the objective function to obtain an objective value;

[0057] performing global consistency verification on the objective value, if the verification is passed, encapsulating the discrete phase compensation vector into the phase compensation instruction frame; otherwise, adjusting an iteration step length, and returning to execute the step of using a constrained least square iteration to perform a first-order gradient descent solving on the objective function with the phase deviation as an initial value to obtain a continuous compensation solution.

[0058] Preferably, the method as described above, further comprises:

[0059] receiving a new row commutation timestamp sequence returned by each of the control cards after the compensation takes effect;

[0060] performing fast incremental link verification according to the row commutation timestamp sequence to obtain a verification result;

[0061] if the verification result is that a preset time difference threshold is not met, returning to execute the step of performing cross-card phase compensation solving and mapping according to the phase deviation, obtaining a discrete phase compensation vector, and encapsulating the discrete phase compensation vector into a phase compensation instruction frame.

[0062] In a second aspect, an embodiment of the present application provides a control device, comprising:

[0063] a first processing module configured to obtain optimal offset amounts of control cards relative to a unified reference clock and local timestamps of scanning rows of the control cards, and generate a globally aligned time matrix with row-card double indexes; wherein the optimal offset amounts are determined according to network round-trip delay measurement and time difference of common boundary rows between the control cards; and the local timestamps are obtained by triggering a board-mounted nanosecond counter through row start hardware interrupts of the control cards;

[0064] a second processing module configured to generate a time-space-environment three-dimensional coupled data cube according to the globally aligned time matrix, a preset bidirectional mapping function, and temperature and power supply voltage of the control cards after down-sampling to a pixel grid; wherein the bidirectional mapping function is a mapping function from physical pixel coordinates to control card number-row sequence indexes, which is established according to installation drawings and pixel row and column sequences;

[0065] The third processing module is used to perform digital twin prediction based on the three-dimensional coupled data cube and obtain the predicted phase field.

[0066] The fourth processing module is used to perform weighted fusion based on the virtual tear saliency map obtained by calculating the lateral gradient of the predicted phase field and the optical flow misalignment intensity map obtained by optical flow differential processing of the camera screenshot, to generate a tear probability map and obtain the phase deviation corresponding to each of the control cards.

[0067] The fifth processing module is used to perform cross-card phase compensation solution and mapping based on the phase deviation, obtain discrete phase compensation vector, and encapsulate it into a phase compensation instruction frame; wherein, the phase compensation instruction frame is used to enable each of the control cards to write the discrete phase compensation vector in the firmware programmable clock interface, and lock the effective time at the next frame row sequence counter reset point;

[0068] The sixth processing module is used to send the phase compensation command frame in parallel to all the control cards after entering the inter-frame security window.

[0069] Thirdly, embodiments of this application provide an electronic device, including: a processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, it implements the steps of the distributed LED lateral fine tear instant alignment method as described above.

[0070] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the distributed LED lateral fine tear instantaneous alignment method as described above.

[0071] Fifthly, embodiments of this application provide a computer program product, including computer instructions that, when executed by a processor, implement the steps of the distributed LED lateral fine tear instantaneous alignment method as described above.

[0072] Compared with existing technologies, the distributed LED lateral fine tear real-time alignment method, device, and storage medium provided in this application achieve a closed loop from data acquisition and state estimation to command issuance within the same frame refresh cycle through the synchronous fusion of three information channels: time interception, digital twin prediction, and visual observation. This enables sub-microsecond uniformity of cross-card scanning phase without the need for cable re-lengthening or waiting for network latency statistical convergence. Furthermore, during digital twin prediction, a three-dimensional coupled data cube of time, space, and environment is constructed, and predictions are performed in real time based on this cube, maintaining temporal consistency of phase field prediction even with rapid changes in ambient temperature or power supply fluctuations. It can provide pixel-resolution compensation for local hotspots or transient loads, avoiding compensation imbalances at boundaries. In the visual branch, a tear probability map is generated by weighted fusion of the lateral gradient saliency map and the optical flow misalignment intensity map. This utilizes the prior constraint of the predicted phase field on the theoretical tear position while maintaining the sensitivity of camera observation to the actual misalignment amplitude, reducing the risk of false triggering of high-brightness flicker and moving content compared to traditional pure visual closed-loop systems. Finally, by issuing discretized phase compensation commands through an inter-frame safety window and delaying their effect until the next frame reset point, the fine lateral tearing of distributed LEDs is eliminated online, ensuring that the adjustment process is transparent to the playing image and improving the flexibility and scalability of maintenance operations. Attached Figure Description

[0073] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0074] Figure 1 This is one of the flowcharts illustrating the instant alignment method for lateral fine tearing of distributed LEDs in this application embodiment;

[0075] Figure 2 This is the third flowchart illustrating the distributed LED lateral fine tear instant alignment method in this application embodiment;

[0076] Figure 3 This is the third flowchart illustrating the distributed LED lateral fine tear instant alignment method in this application embodiment;

[0077] Figure 4 This is the fourth flowchart illustrating the distributed LED lateral fine tear instant alignment method in the embodiments of this application;

[0078] Figure 5 This is the fifth flowchart illustrating the distributed LED lateral fine tear instant alignment method in the embodiments of this application;

[0079] Figure 6 This is the sixth flowchart illustrating the distributed LED lateral fine tear instant alignment method in this application embodiment;

[0080] Figure 7 This is the seventh flowchart illustrating the distributed LED lateral fine tear instant alignment method in the embodiments of this application;

[0081] Figure 8 This is the eighth flowchart illustrating the distributed LED lateral fine tear instant alignment method in the embodiments of this application;

[0082] Figure 9 This is the ninth flowchart illustrating the distributed LED lateral fine tear instant alignment method in the embodiments of this application;

[0083] Figure 10 This is a schematic diagram of the network architecture that can obtain phase compensation vectors in the embodiments of this application;

[0084] Figure 11 This is a schematic diagram of the control device in the embodiments of this application;

[0085] Figure 12 This is a schematic diagram of the structure of the electronic device in the embodiments of this application. Detailed Implementation

[0086] Exemplary embodiments of the present application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thorough and complete, and will fully convey the scope of the present application to those skilled in the art.

[0087] The terms “first,” “second,” etc., used in the specification and claims of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus. The terms “and / or” in the specification and claims indicate at least one of the connected objects.

[0088] The following description provides examples and is not intended to limit the scope, applicability, or configuration set forth in the claims. Changes may be made to the function and arrangement of the elements discussed without departing from the spirit and scope of this disclosure. Various procedures or components may be appropriately omitted, substituted, or added to the examples. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Furthermore, features described with reference to certain examples may be combined in other examples.

[0089] To address the aforementioned technical issues, this application provides a method, apparatus, and storage medium for real-time alignment of distributed LED lateral fine tears, applicable to a large-size screen including a central synchronization server and multiple control cards working together, wherein the central synchronization server is connected to each control card respectively. The following example uses the method applied to the central synchronization server.

[0090] See Figure 1 This application provides a method for real-time alignment of lateral fine tears in distributed LEDs, comprising:

[0091] Step S101: Obtain the optimal offset of each control card relative to a unified reference clock and the local timestamp of each scan line of the control card, and generate a global alignment time matrix with row-card dual index; wherein, the optimal offset is determined based on the network round-trip delay measurement and the time difference of the shared boundary line between control cards; the local timestamp is obtained by triggering the onboard nanosecond counter through the row start hardware interrupt of each control card.

[0092] Step S102: Based on the global alignment time matrix, the preset bidirectional mapping function, and the temperature and power supply voltage of each control card downsampled to the pixel grid, a time-space-environment three-dimensional coupled data cube is generated. The bidirectional mapping function is a mapping function from physical pixel coordinates to control card number-row index established according to the installation drawing and pixel row and column order.

[0093] Step S103: Perform digital twin prediction based on the three-dimensional coupled data cube to obtain the predicted phase field;

[0094] Step S104: Based on the virtual tear saliency map obtained by calculating the lateral gradient of the predicted phase field and the optical flow misalignment intensity map obtained by optical flow differential processing of the camera screenshot, a weighted fusion is performed to generate a tear probability map and obtain the phase deviation corresponding to each of the control cards.

[0095] Step S105: Perform cross-card phase compensation solution and mapping based on the phase deviation to obtain a discrete phase compensation vector and encapsulate it into a phase compensation instruction frame; wherein, the phase compensation instruction frame is used to enable each of the control cards to write the discrete phase compensation vector in the firmware programmable clock interface and lock the effective time at the next frame row sequence counter reset point.

[0096] Step S106: After entering the inter-frame security window, the phase compensation command frame is sent in parallel to all the control cards.

[0097] In this embodiment, the central synchronization server first obtains the optimal offset of each control card relative to a unified reference clock and the local timestamp of each scan line of the control card, and generates a globally aligned time matrix with row-card dual indices. The optimal offset is determined by the central synchronization server based on dynamic network round-trip delay measurements and the time difference of the shared boundary lines of each control card. Network round-trip delay measurements can dynamically compensate for network transmission delay, while the shared boundary lines and the provision of physical reference points allow the optimal offset to be obtained dynamically in real time, more closely reflecting the actual operating state, which is beneficial for maintaining synchronization accuracy. It is also applicable to heterogeneous control cards, requires no unified clock hardware, has a wider range of applications, and eliminates the need for manual calibration, thus reducing costs. The local timestamp of each scan line is recorded by an onboard nanosecond counter (such as an FPGA or dedicated timing chip) directly triggered by the row start hardware interrupt of the control card, rather than relying on operating system software interrupts. This helps avoid jitter caused by operating system scheduling, interrupt delays, or task switching, achieving near-zero latency timestamps, eliminating software latency, and ensuring strict alignment with physical events (such as the start of row scanning), providing a high-fidelity data source. Simultaneously, based on the aforementioned optimal offset and local timestamps, a globally aligned time matrix for row-card dual indices is generated, directly supporting spatiotemporal management of cross-control card data without requiring additional time conversion calculations, thus saving costs. Furthermore, it can be updated in real-time, exhibiting good dynamic applicability. It also retains the high-resolution characteristics of hardware triggering and the overall consistency of centralized network estimation, unlike traditional synchronization methods that rely on a central time base or simple round-trip delay averaging, providing a time base input with precise row-level positioning for subsequent compensation.

[0098] After generating the global alignment time matrix, a time-space-environment three-dimensional coupled data cube is generated by combining the global alignment time matrix with a preset bidirectional mapping function and the temperature and voltage downsampled to the pixel grid. The spatial information can be obtained by mapping the global alignment time matrix and the preset bidirectional mapping function, achieving a bidirectional mapping between the physical pixel space and the card number-row order, facilitating subsequent coupling. Since the formation of lateral fine tears is affected not only by clock phase difference but also by timing drift caused by control card temperature rise and power supply ripple, coupling the temperature and power supply voltage downsampled to the pixel grid as environmental information into the three-dimensional coupled data cube ensures that the three-dimensional coupled data cube can reflect the characteristics of lateral fine tears from multiple angles, thus facilitating high-precision, real-time identification and correction of lateral fine tears (especially those 1-2 pixels wide). It should be noted that the aforementioned bidirectional mapping function is a mapping function established based on the installation drawings and pixel row and column order, from physical pixel coordinates to control card number-row order index. Once established, this bidirectional mapping function uniquely determines the affiliation of each pixel in the distributed drive grid. Based on this bidirectional mapping function, the time data at the control card level can be accurately mapped to the pixel level, ensuring that there is no deviation in the timing description of the tear location.

[0099] Furthermore, the three-dimensional coupled data cube is predicted using a digital twin (DT) to obtain a predicted phase field representing the downscan sequence under ideal conditions. This ensures the temporal consistency of the phase field prediction even when environmental temperature rises or power supply fluctuations change rapidly. Compared to adjusting the blanking time using empirical curves, this mechanism can provide pixel-resolution compensation for local hotspots or transient loads, avoiding compensation imbalances at boundaries. Further, a lateral gradient calculation is performed on the predicted phase field to obtain a virtual tear saliency map representing the virtual tear location. Simultaneously, optical flow differential processing is performed on the camera images or screenshots captured by the camera system to obtain an optical flow misalignment intensity map representing the actual misalignment intensity. By weighting and fusing these two maps, a tear probability map that retains both the theoretical tear location and the measured misalignment intensity can be generated. This facilitates the acquisition of the phase deviations that need to be applied to each control card. Preferably, the weighting and fusing is implemented using the Sigmoid function. By integrating multi-source alignment, virtual-real verification, and weighted adaptive fusion, the industry problems of easy false detection by relying solely on vision and easy missed detection by relying solely on timing are solved, and a criterion system is established for real-time alignment of lateral fine tears in distributed LEDs.

[0100] After obtaining the phase deviation, cross-card phase compensation can be solved and mapped based on the phase deviation to obtain the discrete phase compensation vector (PCV), which is then encapsulated into a phase compensation instruction frame. The frame synchronization monitoring thread continuously running in the central synchronization server calculates the minimum remaining scan time between two adjacent frames based on the global alignment time matrix and the latest row commutation timestamp returned by each control card. When it is determined that the inter-frame safety window (IFSW) has been entered, the instruction distribution process is started, and the obtained phase compensation instruction frame is broadcast in parallel to all control cards through the message bus. This allows the hot-plugging (HP) thread of each control card to write the discrete phase compensation vector into the on-chip adjustable phase-locked loop (PLL) / delay line register through the application programming interface (API) of the firmware programmable clock after receiving the corresponding phase compensation instruction frame, and lock the effective time at the next frame row sequence counter reset point to achieve instant alignment of the lateral fine tear. This closed-loop process closely integrates theoretical timing, visual observation, and execution feasibility. Unlike correction modes that rely solely on vision or clocks, it can achieve sub-microsecond synchronization without interrupting the rendering of the current frame and maintain strict alignment between the compensation action and the next frame's line sequence reset point.

[0101] In summary, this application achieves a closed loop from data acquisition and state estimation to command issuance within the same frame refresh cycle by synchronously fusing information from time interception, digital twin prediction, and visual observation. This enables sub-microsecond uniformity of cross-card scanning phase without requiring cable relengthening or waiting for network latency statistical convergence. Furthermore, during digital twin prediction, a three-dimensional coupled data cube of time, space, and environment is constructed, and predictions are performed in real-time based on this cube, maintaining temporal consistency of phase field prediction even with rapid changes in ambient temperature or power supply fluctuations. It can provide pixel-resolution compensation for local hotspots or transient loads, avoiding compensation imbalances at boundaries. In the visual branch, a tear probability map is generated by weighted fusion of the lateral gradient saliency map and the optical flow misalignment intensity map. This utilizes the prior constraint of the predicted phase field on the theoretical tear location while maintaining the sensitivity of camera observation to the actual misalignment amplitude, reducing the risk of false triggering by high-brightness flickering and moving content compared to traditional pure visual closed-loop systems. Finally, the discretized phase compensation command is issued through the inter-frame safety window and delayed until the next frame reset point to take effect, ensuring that the adjustment process is transparent to the picture being played, thus improving the flexibility and scalability of maintenance operations.

[0102] In one specific embodiment, when acquiring the local scan time, an additional time interception service routine (ISR) is pre-registered in the real-time display driver of each control card for the row start hardware interrupt. The control card in the... When the horizontal scan level edge arrives, the ISR immediately calls the onboard nanosecond counter to read the current count value and generate a local timestamp. ,in For the control card serial number, , This refers to the number of lines scanned per frame. Immediately after ISR completion... Write to the buffer to ensure that line-level timing precision is not interfered with by other threads. By only adding time-capturing logic to the existing interrupt chain, the playback flow remains continuous throughout. After completing the line-level timing capture, the data in the buffer is processed line by line in sequence. Rearrange to form a column vector: Furthermore, the control card invokes a reliable transmission protocol to... The data is packaged with the sending time and uploaded to the central synchronization server, enabling the central synchronization server to centrally acquire the local scan times of all control cards.

[0103] In one specific embodiment, when obtaining the virtual tear saliency map, the Sobel-X derivative of the predicted phase field is performed to calculate the lateral phase gradient field. Then, the phase bias caused by global brightness is suppressed by normalization, thus forming the virtual tear saliency map V. ,in, Represents pixel coordinates The corresponding tear significance; Represents pixel coordinates The corresponding predicted phase; A three-point difference operator is used to maintain sub-pixel gradient accuracy. When Greater than the dynamic threshold ,in, When the preset reference value is used, the pixel is determined to be in the potential phase step region; at the same time, its gradient sign is recorded to indicate the two tearing directions, namely left fast and right slow or left slow and right fast, providing a basis for determining the deviation sign.

[0104] In one specific embodiment, an example is given of the specific application of the phase compensation vector on the control card side, wherein the actual phase compensation vector can be determined according to the following formula:

[0105]

[0106] in: This is the actual phase compensation vector of control card k at time t; The existing compensation amount is in effect; This refers to the discrete phase compensation vector in the phase compensation command frame. This is the start time of the current frame; This is the reset time for the next frame's line sequence counter; The duration is for a single frame. By employing a segmented control strategy for the phase compensation vector, it is ensured that the scanning timing remains unchanged within the same frame, avoiding new tearing caused by half-frame activation; moreover, phase adjustment and pixel refresh are completely decoupled, achieving transparent, lossless transmission to the currently playing video stream.

[0107] This application achieves sub-microsecond cross-card phase unification on distributed LED video walls, providing manufacturers with a real-time correction framework combining hardware and software. This significantly reduces maintenance costs caused by video wall size expansion, wiring differences, or component aging. Replacing traditional equal-length cables or pure visual correction with row-level time interception and digital twin assimilation means that adaptive leveling can be completed on-site without downtime. This improves the availability and maintenance efficiency of rental and fixed-installation display systems, making it highly attractive for time-sensitive scenarios such as stage performances, stadiums, and advertising media. Incorporating temperature and voltage into the assimilation model also allows the supply chain to continue using lower-cost power supplies and heat sinks without sacrificing image integrity, providing OEMs with room for Bill of Materials (BOM) optimization.

[0108] The multi-source temporal assimilation method proposed in this application is universal and can be transferred to other applications requiring distributed high-speed synchronization. For example, large-scale Micro-LED or Quantum Dot Light-Emitting Diode (QLED) video walls also rely on row-level scanning, where minute phase differences are amplified in high dynamic range (HDR) motion images; high-speed laser projection matrices and naked-eye 3D screen rendering also require strict row-order consistency across multiple channels, where the temporal truncation and dynamic compensation strategies can be directly reused; in industrial inspection or scientific research experiments, if multiple cameras or sensor arrays sample in parallel at the row or column level, the fusion algorithm can help smooth out spatial distortions caused by trigger delays. Furthermore, for multi-beam additive manufacturing, synchrotron radiation shutter control, and even automotive-grade electronic control unit (ECU) video stitching, synchronization accuracy and real-time self-correction are often key quality parameters. The framework of this application can provide a reference for these fields, achieving higher throughput and lower maintenance burden.

[0109] See Figure 2Specifically, as described above, obtaining the optimal offset of each control card relative to a unified reference clock and the local timestamp of each scan line of the control card, and generating a globally aligned time matrix with row-card dual indices, includes:

[0110] Step S201: Based on the network round-trip delay measurement, obtain the clock offset of each control card relative to the unified reference clock;

[0111] Step S202: Based on the clock offset, the local timestamp, the row index set located at the control card boundary, and the first preset algorithm, perform gradient descent iterative optimization to obtain the optimal offset;

[0112] Step S203: Correct the local timestamp matrix formed by the local timestamps according to the optimal offset to obtain the global aligned time matrix.

[0113] The steps for obtaining local timestamps have been explained above. Therefore, this embodiment mainly illustrates the steps for obtaining the optimal offset and the global alignment time matrix. First, based on network round-trip delay measurements, the clock offset of each control card relative to the unified reference clock is obtained. This clock offset can be recorded as the initial clock offset. Then, the set of row indices located at the boundary of the control card is obtained. Based on the clock offset, local timestamp, row index set, and the first preset algorithm, gradient descent iterative optimization is performed to obtain the desired optimal offset. The first preset algorithm can be:

[0114]

[0115] in, For an ordered set of control cards with physical splicing boundaries;

[0116] Let be the global clock offset vector to be determined;

[0117] The offset of control card k relative to a unified reference clock;

[0118] K represents the total number of control cards in the distributed system;

[0119] For control card k in line Local timestamps collected at the location;

[0120] The control card k and control card p share the same row index set.

[0121] Specifically, the first preset algorithm uses the boundary row scan time difference as the objective to ensure that the solution corresponds to the location where the lateral tear occurs. The preferred termination condition for the first preset algorithm is that the change in the objective function between two consecutive iterations is less than a preset threshold, for example: The square of microseconds enables convergence within a two-frame interval.

[0122] Obtain the optimal offset Then, based on the optimal offset The local timestamp matrix, composed of local timestamps from all control cards, is corrected to generate a globally aligned time matrix for the row-card dual index. : ,in, A two-dimensional globally aligned time matrix for both row index and control card index; It is an L-dimensional column vector with all elements being 1; It is a matrix composed of local timestamp vectors of each control card. .

[0123] Among them, the global alignment time matrix serves as the sole temporal input for constructing the three-dimensional coupled data cube in subsequent steps, achieving absolute temporal uniformity of cross-card scanning events and laying a time-domain foundation for real-time alignment of lateral fine tears in distributed LEDs.

[0124] See Figure 3 Specifically, the method described above, wherein generating a time-space-environment three-dimensional coupled data cube based on the global alignment time matrix, a preset bidirectional mapping function, and the temperature and power supply voltage of each control card downsampled to the pixel grid, includes:

[0125] Step S301: Element-wise access is performed on the global alignment time matrix according to the bidirectional mapping function to generate a pixel phase matrix, wherein the control card number in the bidirectional mapping function is used to read the columns in the global alignment time matrix, and the row index in the bidirectional mapping function is used to read the rows in the global alignment time matrix;

[0126] Step S302: According to the control card number, perform point-by-point calculations on the two-dimensional pixel plane with respect to the amplitude of the temperature and the power supply voltage to obtain a temperature matrix and a voltage matrix;

[0127] Step S303: Perform range normalization on the temperature matrix to obtain the corresponding target temperature matrix;

[0128] Step S304: Perform range normalization on the voltage matrix to obtain the corresponding target voltage matrix;

[0129] Step S305: Channel-based stitching is performed on the pixel phase matrix, the target temperature matrix, and the target voltage matrix to obtain the three-dimensional coupled data cube.

[0130] This embodiment illustrates the steps for generating a time-space-environment three-dimensional coupled data cube. First, the global alignment time matrix is ​​accessed element-by-element using a bidirectional mapping function to generate a pixel phase matrix. The control card number in the bidirectional mapping function is used to read columns in the global alignment time matrix, and the row index is used to read rows. Specifically, for any pixel coordinate... Its actual scanning completion time This can be determined using the bidirectional mapping function:

[0131]

[0132] in, For pixels Scan completion time;

[0133] For control card Global alignment time for line l;

[0134] To control the total number of cards;

[0135] This represents the number of scan lines per frame.

[0136] This is the Kronecker Delta Function (KDF). The value is 1 if the condition is met, otherwise it is 0.

[0137] For pixels The control card number to which it belongs;

[0138] x-coordinate of pixels The corresponding scan row order index.

[0139] Using the aforementioned bidirectional mapping function, row-card time elements perfectly aligned with pixel affiliation can be selected, ensuring that each element of the pixel phase matrix corresponds to its physical scanning process. This is particularly useful at the boundary rows between two cards, directly quantifying the timing difference of potential tear gaps. Furthermore, based on the control card number, the two-dimensional pixel plane is analyzed point-by-point with respect to temperature. and the amplitude of the supply voltage To obtain the temperature matrix and voltage matrix ,in, , Furthermore, the temperature matrix is ​​normalized by range to obtain the corresponding target temperature matrix, which can be expressed as: , and The temperature represents the historical extreme values. By performing range normalization on the temperature matrix, the corresponding target temperature matrix is ​​obtained, which can be expressed as: , and The voltage historical extreme values ​​are represented. By normalizing the ranges of the temperature and voltage matrices, environmental disturbances are transformed into continuous tensors of the same dimension as the pixel coordinates, providing a pixel-by-pixel physical basis for predicting phase drift.

[0140] Furthermore, by performing channel-based stitching of the pixel matrix and the normalized environment matrix (i.e., the target temperature matrix and the target voltage matrix) in the third dimension, the required three-dimensional coupled data cube can be obtained; the elements corresponding to each spatial coordinate and channel index in this three-dimensional coupled data cube It can be represented as:

[0141]

[0142] in, This is the phase channel, used to store the globally aligned time matrix. The scan completion time corresponding to each coordinate in the middle; The temperature channel (z=3) stores the temperature values ​​corresponding to each coordinate in the normalized target temperature matrix; the voltage channel (z=3) stores the voltage values ​​corresponding to each coordinate in the normalized target voltage matrix. Through pixel-by-pixel stitching, phase, temperature, and voltage are fused under the same indexing system, forming a coupled feature describing time-space-environment. Digital twin prediction based on this three-dimensional coupled data cube enables tear localization and phase prediction to simultaneously consider temporal differences and environmental drift effects, thus achieving high-precision, real-time identification and correction of 1-2 pixel lateral fine tears.

[0143] See Figure 4 Specifically, in the method described above, the step of performing digital twin prediction based on the three-dimensional coupled data cube to obtain the predicted phase field includes:

[0144] Step S401: Construct a two-dimensional driving mesh according to the control card number list, module physical block size and pixel row and column order, and initialize the three-dimensional state vector of each node in the two-dimensional driving mesh. The three-dimensional state vector includes phase value, temperature value and voltage value.

[0145] Step S402: Perform forward multi-physics step prediction on the three-dimensional state vector on the two-dimensional driving mesh to obtain the prior state vector;

[0146] Step S403: Based on the nodes, the three-dimensional coupled data cube is parsed into observation vectors and an observation mapping operator is constructed;

[0147] Step S404: Dynamic assimilation is performed based on the prior state vector, the observation vector, the observation mapping operator, and the gain matrix to obtain the optimal state set. The gain matrix is ​​obtained based on the current neighborhood variance, observation accuracy, and scanning direction.

[0148] Step S405: Extract phase channels from the optimal state set and rearrange them according to the node coordinates to obtain the predicted phase field.

[0149] In this embodiment, a specific example is given of the steps described above for performing digital twin prediction based on the three-dimensional coupled data cube to obtain the predicted phase field. Specifically, to ensure the model can accurately distinguish the hardware layout and guarantee the reliability of the digital twin, a list of control card numbers is first obtained by reading the installation topology file. Module physical block size And the pixel row and column order, etc., to construct the corresponding two-dimensional driving mesh. ,in, Indicates the node number. Represents the total number of nodes, making the grid nodes This achieves precise mapping between the digital twin model and physical hardware, corresponding to the physical pixels of the screen. It eliminates the risk of errors from manual mesh generation and supports modular expansion (e.g., no model reconstruction is needed when adding a new control card), significantly improving deployment efficiency. Furthermore, it initializes each node in the 2D driven mesh based on the control card number and row number. The corresponding three-dimensional state vector is: ,in, Indicates the initial value of the scan phase. Indicates the initial temperature value. The initial value of the supply voltage is represented by three values, which are inherited from the historical frame state or the factory calibration value, respectively. The grid and state vector together constitute the digital twin space and the basis of variables. Multiphysics integrated modeling helps to improve the physical consistency of the twin model prediction, and the initialization of the three-dimensional state vector also helps to provide accurate initial values ​​for subsequent step predictions, which helps to speed up the prediction process.

[0150] Furthermore, based on the three-dimensional state vector obtained above, multi-physics step prediction is performed to obtain the prior state vector. The overall set of prior states is represented as: It simulates continuous evolution over time, predicts transient behavior, provides a critical time window for preventative maintenance, and emphasizes multi-physics steps, making the prediction results more consistent with physical laws than single-physics models, thus improving prediction accuracy. Furthermore, computation through a two-dimensional grid simplifies the computational workload in three-dimensional space and supports parallel processing, which helps to accelerate response speed while maintaining accuracy.

[0151] Furthermore, based on the node indices, the obtained three-dimensional coupled data cube will be parsed into two-dimensional observation vectors. This analytical process rigorously maps discrete 3D data to the nodes of a 2D driving mesh, eliminating data spatial misalignment and filtering out non-core noise (such as sensor drift), retaining only high-reliability observations. Simultaneously, it significantly reduces subsequent computational load through dimensionality reduction. Furthermore, based on node topology analysis, it ensures consistency between the observation vector and the 2D driving mesh, avoiding error accumulation caused by coordinate transformation. In one specific embodiment, after obtaining the observation vector, an observation mapping operator H is constructed: This facilitates the mapping between the observation vector and the three-dimensional state vector based on the mapping operator, thereby enabling subsequent dynamic assimilation steps.

[0152] Furthermore, by performing dynamic assimilation (DA) based on the prior state vector, observation vector, and observation mapping operator obtained above, and combining this with weight adjustments using the gain matrix obtained based on the current neighborhood variance, observation accuracy, and scan direction, adaptive error correction can be achieved. Specifically, the gain matrix... Based on neighborhood variance, spatial correlation (such as temperature fluctuations of neighboring nodes) is considered to avoid amplifying local noise. The gain matrix is ​​obtained based on observation accuracy and can be weighted according to sensor confidence to ensure accuracy. The gain matrix is ​​also obtained based on scanning direction, which can adapt to scanning path characteristics and ensure that the assimilation directionality matches the actual measurement. Through real-time calculation of the gain matrix, it can dynamically adapt to changes in system operating conditions, ensuring real-time robustness. Specifically, Kalman assimilation is used to achieve the above dynamic assimilation, and the assimilation kernel of this dynamic assimilation can be expressed as: The set of optimal states after assimilation correction can be represented as: It should be noted that the following characteristics need to be ensured during the assimilation process: weights are increased at the control card splicing boundary to prioritize the elimination of lateral phase steps; suppression coefficients are added to high-temperature and high-pressure nodes to avoid introducing transient thermal noise into the phase plane; and assimilation iterations occur within a single frame period. Forced convergence within the internal space ensures real-time performance.

[0153] After obtaining the optimal state set, phase channels can be extracted from the optimal state set and rearranged according to the node coordinates to form the predicted phase field. The acquisition of the predicted phase field extracts only the phase channel (the core indicator of user concern), eliminating irrelevant dimensions (temperature / voltage) to generate a pure phase prediction field, ensuring the accuracy of the prediction results and simplifying subsequent applications. Furthermore, based on node coordinate rearrangement, it can ensure that the output phase field is completely consistent with the physical module layout (such as matching the pixel row and column order to the display device), avoiding distortion caused by coordinate transformation, and allowing the prediction results to be directly used for hardware verification. Even better, the predicted phase field is output in a standard phase field format (such as PNG / CSV), supporting real-time interface with display systems and control software, accelerating the "prediction-decision-execution" closed loop, and improving system response speed.

[0154] In summary, this embodiment, by achieving a closed loop of "structured modeling, multi-physics prediction, intelligent assimilation, and accurate output," helps improve the accuracy of digital twin predictions and reduce deployment costs. It also realizes a paradigm shift from "post-event diagnosis" to "real-time prediction and proactive intervention," making it particularly suitable for high-reliability scenarios. Furthermore, real-time processing ensures the effectiveness of the process in dynamic environments.

[0155] Furthermore, in the method described above, the step of performing forward multi-physics step prediction on the three-dimensional state vector on the two-dimensional driving mesh to obtain the prior state vector includes:

[0156] The phase value is recursively calculated in a single frame by calling the row sequence step operator to obtain the prior phase;

[0157] The temperature value is calculated based on a two-dimensional thermal conductivity operator to obtain the prior temperature;

[0158] The voltage value is calculated based on a preset node-edge resistor network to obtain the prior voltage;

[0159] The prior phase, the prior temperature, and the prior voltage are combined to form the prior state vector.

[0160] In this embodiment, the above-mentioned forward multiphysics step prediction is illustrated, specifically, three coupled calculations are performed sequentially on a two-dimensional driven mesh, namely scan-driven prediction, thermal diffusion prediction, and power supply voltage prediction.

[0161] Specifically, scan-driven prediction involves calling the row-order step operator. For phase value Perform single-frame recursion to obtain the prior phase. The use of single-frame recursion helps avoid full-frame calculations and ensures the real-time performance of line scanning.

[0162] Thermal diffusion prediction is specifically based on a two-dimensional thermal conductivity operator. For temperature value Calculations were performed to obtain the prior temperature. Since temperature diffusion is a two-dimensional heat conduction problem, the heat equations can be discretized using the finite difference method, for example: ,in, Indicates thermal conductivity. This represents the two-dimensional Laplace operator calculated using the five-point format: , Indicates the grid step size. Indicates as a node ( The corresponding temperature value is used to ensure spatial coupling.

[0163] Supply voltage prediction is specifically based on node-edge resistance networks. For the voltage value Calculations are performed to obtain the prior voltage. In this node-edge resistor network, a node represents each control card, and an edge represents the resistance of the wire connecting the control cards. The input is a current source, which is generally derived from the control signal of the driver card. Specifically, the node voltages are solved using Kirchhoff's equations, for example: ,in, This represents the admittance matrix constructed based on a pre-defined resistor network. This represents the voltage vector output by the node. This represents the current source vector obtained from the drive command. The Alchian equations are solved using the faster-converging Gauss-Seidel iteration to ensure hardware accuracy. Furthermore, this node-edge resistor network only depends on the design document, can run stably after deployment, and requires no training.

[0164] Finally, by concatenating the aforementioned prior phase, prior temperature, and prior voltage, the desired prior state vector can be obtained. Specifically, this is achieved by concatenating the prior state vector node by node according to the node index list of the two-dimensional driving mesh. .

[0165] Preferably, in the method described above, after obtaining the predicted phase field, the method further includes:

[0166] The predicted phase field is subjected to temporal consistency verification and spatial smoothness verification;

[0167] Regions that pass the timing consistency check and the spatial smoothness check are marked as ideally synchronized.

[0168] In a preferred embodiment of this application, after obtaining the predicted phase field, a two-level consistency check is applied to the predicted phase field: a temporal consistency check and a spatial smoothness check. The temporal consistency check is used to determine whether the phase difference between any two adjacent nodes is less than a first threshold. ,Right now If satisfied, the adjacent node is determined to have passed the temporal consistency check; the spatial smoothness check is used to determine whether the sum of the second-order differences within a preset size (e.g., 3×3) pixel window is less than the second threshold. If the conditions are met, then the window region is determined to have passed the spatial smoothness check. Furthermore, regions that have passed both levels of consistency checks can be marked as being in an ideal synchronization state. At this point, the ideal synchronization state and the predicted phase field obtained above can be output together as a reference surface in the subsequent virtual-real comparison step.

[0169] Through the above steps, the digital twin model can complete a closed-loop evolution from prior prediction to assimilation correction, and then to ideal synchronization state extraction within each frame period; the output predicted phase field Ideal synchronization state It provides highly reliable virtual evidence containing temporal information and thermo-electric coupling for real-time detection of transverse fine tears.

[0170] See Figure 5 Preferably, the method described above, obtaining the optical flow misalignment intensity map obtained by optical flow differential processing of the camera screenshot, includes:

[0171] Step S501: Calculate the optical flow vector field of the camera screenshot using the depth-separable pyramid algorithm;

[0172] Step S502: Take the absolute value of the transverse component of the optical flow vector field and square it, and visualize the output as the actual misalignment intensity map.

[0173] Step S503: Based on the normalized environmental confidence of the environmental channel of the three-dimensional coupled data cube at the current moment, weight suppression is applied to the actual misalignment intensity map to obtain the optical flow misalignment intensity map.

[0174] In this embodiment, the steps for obtaining the optical flow misalignment intensity map are illustrated. Specifically, after obtaining a screenshot from the camera system, the optical flow vector field of the screenshot is calculated using a depth-separable pyramid algorithm. in, For the horizontal component, The longitudinal component is then used; subsequently, the absolute value of the transverse component of the optical flow vector field is taken and squared, and the visualization output (projection) is the actual misalignment intensity map. This involves extracting the intensity of the lateral motion (i.e., the lateral component) from the optical flow vector field, and enhancing its significance by squaring it. The visualization output is a map of the actual misalignment intensity that directly shows the magnitude of the horizontal displacement, making the actual misalignment more visually striking.

[0175] To eliminate localized brightness flicker caused by thermal noise and power supply ripple, the normalized environmental confidence scores of the environmental channels (i.e., voltage and temperature channels) of the aforementioned three-dimensional coupled data cube at the current moment will be obtained first. Based on this, weighted suppression is performed on the actual misalignment intensity map to obtain the desired optical flow misalignment intensity map. Weight suppression can be performed in the following ways: Its purpose is to reduce the actual dislocation strength in high-temperature and high-pressure regions. The weights in the image are adjusted to prevent environmental noise from being mistaken for actual tearing, thus improving the optical flow misalignment intensity map. High-value areas are considered strong evidence of tearing in actual measurements. This embodiment uses only low-frequency screenshots to perform differential optical flow on adjacent frames, effectively eliminating high-frequency motion components and making the residual information highly correlated with the line scan, facilitating accurate identification and correction in the future.

[0176] See Figure 6 Specifically, in the method described above, the step of weighting and fusing the virtual tear saliency map obtained by calculating the lateral gradient of the predicted phase field and the optical flow misalignment intensity map obtained by optical flow differential processing of the camera screenshot to generate a tear probability map and obtain the phase deviation corresponding to each of the control cards includes:

[0177] Step S601: Based on the virtual tear saliency map, the optical flow misalignment intensity map, and the environmental confidence level, perform weight fusion to generate the tear probability map. The weights corresponding to the virtual tear saliency map, the optical flow misalignment intensity map, and the environmental confidence level are all calibrated offline according to the Bayesian minimum risk criterion.

[0178] Step S602: Obtain the common boundary set of the control card according to the bidirectional mapping function;

[0179] Step S603: Using the common boundary set as an index, calculate the product of the tearing probability of the cumulative pixels of each boundary and the pixel phase difference to obtain the phase deviation.

[0180] In this embodiment, the steps for obtaining phase deviation are illustrated. After obtaining the aforementioned virtual tear saliency map, optical flow misalignment intensity map, and environmental confidence score, the three are weighted and fused to generate a tear probability map at the pixel level. Specifically, the fusion method can be expressed as follows: ,in, For the Sigmoid function; , and The coefficients are offline calibrations based on the Bayesian minimum risk criterion, and The resulting tear probability map retains the theoretical tear zone location while incorporating measured misalignment strength and environmental confidence level, thus achieving dynamic confidence adjustment.

[0181] Furthermore, the common boundary set of the control card will be obtained based on the aforementioned bidirectional mapping function. By using the common boundaries in this set of common boundaries as indices, and accumulating the product of the pixel-level tearing probability and the pixel phase difference for each boundary, the phase deviation at the control card level can be obtained. ,in, For pixels ( The probability of tearing; The predicted phase difference between the pixels on both sides of the boundary; Let k be the set of common boundaries between control card k and its right-adjacent control card. This indicates the number of pixels in the common boundary set; This is the phase deviation that needs to be applied to control card k. The probability value is obtained from the above equation. Integrating the phase difference as a confidence weight improves the robustness of boundary bias estimation; the sign of the bias is changed from... Automatically provided, enabling directional closure. The final output probability value provides weights for the optimization objective function in subsequent discrete phase solving. This provides quantitative initial values ​​for subsequent discrete phase solutions.

[0182] Based on the above, this application solves the industry problems of easy false detection by relying solely on vision and easy missed detection by relying solely on timing by using multi-source alignment, virtual and real mutual verification and weight adaptive fusion, and establishes a criterion system for real-time alignment of lateral fine tears in distributed LEDs.

[0183] See Figure 7 Preferably, in the method described above, before performing weighted fusion of the virtual tear saliency map obtained by calculating the lateral gradient of the predicted phase field and the optical flow misalignment intensity map obtained by optical flow differential processing of the camera screenshot to generate a tear probability map and obtain the phase deviation corresponding to each of the control cards, the method further includes:

[0184] Step S701: Estimate the average latency from the camera system to the display system based on the frame sequence number returned by the control card and the timestamp of the camera image captured by the camera system;

[0185] Step S702: Based on the difference between the camera timestamp and the average delay, obtain the predicted phase field and camera screenshot corresponding to the output difference;

[0186] Step S703: Based on the bidirectional mapping function and the preset quadratic polynomial correction term, project the camera screenshot onto the coordinate system corresponding to the predicted phase field;

[0187] Step S704: Align the camera screenshot and the predicted phase field with coordinates based on sub-pixel level bilinear interpolation to obtain the aligned camera screenshot.

[0188] In this embodiment, since the predicted phase field and the camera screenshot are generated from different clock domains, without temporal synchronization and spatial registration, subsequent tear alignment will lose its reference base. Therefore, temporal synchronization and spatial registration are performed before obtaining the phase deviation. Specifically, the temporal synchronization step can be based on the frame sequence number returned by the control card. The timestamp of the video images captured by the camera system Estimate the average latency from the camera system to the display system. When estimating the average time delay, it is preferable to estimate it based on a linear regression equation; after obtaining the average time delay, the difference between the camera timestamp and the average time delay is taken ( The corresponding predicted phase field and screenshots This ensures that the predicted phase field and the camera screenshot point to the same single-frame period or rendering period.

[0189] For spatial registration, the bidirectional mapping function described above and the preset quadratic polynomial correction term are used. The camera screenshot is projected onto the coordinate system corresponding to the predicted phase field, i.e., the two-dimensional driving grid. The quadratic polynomial correction term is used to correct the deviation between the actual and ideal positions of image points caused by the optical physical characteristics of the camera system. By fitting the distortion with the quadratic polynomial correction term, the lens distortion is accurately compensated with only one physical parameter under computational complexity. This enables the digital twin system to achieve 100% alignment between physical coordinates and image coordinates within millisecond latency, ensuring the real-time performance and reliability of the camera screenshot. It also helps to avoid image stretching or tearing caused by fitting with high-order polynomials.

[0190] Furthermore, the camera screenshot and the predicted phase field are aligned using sub-pixel-level bilinear interpolation to ensure the accuracy, real-time performance, and system reliability of the aligned camera screenshot.

[0191] See Figure 8 Specifically, as described above, the step of performing cross-card phase compensation solution and mapping based on the phase deviation to obtain a discrete phase compensation vector, and encapsulating it into a phase compensation command frame, includes:

[0192] Step S801: Using the tear probability in the tear probability map as weights, construct an objective function for cross-card boundary phase consistency;

[0193] Step S802: Using the phase deviation as the initial value, the objective function is solved by first-order gradient descent based on constrained least squares iteration to obtain a continuous compensation solution;

[0194] Step S803: Project the continuous compensation solution onto the adjustable step set corresponding to each of the control cards to obtain the discrete phase compensation value corresponding to each of the control cards;

[0195] Step S804: Recombine all the discrete phase compensation values ​​into the discrete phase compensation vector, and encapsulate the discrete phase compensation vector into the phase compensation command frame.

[0196] In this embodiment, the steps for obtaining the discrete phase compensation vector are illustrated. To further accurately measure the residual between the theoretical value and the expected synchronization, the boundary set of all control cards is obtained according to the bidirectional mapping function. This boundary set includes, but is not limited to, the common boundary set between the control cards. Then, using the tear probability in the obtained tear probability map as weights, an objective function for cross-card boundary phase consistency is constructed. Specifically, this objective function can be expressed as: The smaller the value of the objective function, the smoother the boundary. Let be the continuous phase compensation vector to be found, where the value is the corresponding continuous compensation solution; For pixels ( The probability of tearing; To predict the phase field at pixel ( The phase value of ) Let K be the set of pixels sharing the common boundary between control card k and control card k+1 (located to the right of control card k); K is the total number of control cards. This objective function allows pixels with high tearing probabilities to be given higher weights, ensuring the optimization process focuses on truly tear-prone locations rather than global averaging, thus guaranteeing the accuracy of the obtained continuous compensation solution.

[0197] Furthermore, using the phase deviation obtained above as the initial value, and solving the objective function using first-order gradient descent based on a preset constrained least squares iteration method, the converged continuous compensation solution can be obtained. This fine-grained approximation of the ideal synchronization state obtained above. The constrained least squares iteration in the digital twin system, through mathematical constraints and multiple iterations, quickly finds the solution that best matches the actual observations while ensuring physical plausibility. The mathematical constraints include, but are not limited to, constraints on temperature and voltage. In a specific embodiment, the simplified model of this constrained least squares iteration can be expressed as the product of the obtained observation mapping operator and the state vector, minus the observation vector.

[0198] After obtaining the continuous compensation solution Then, the continuous compensation solution By projecting the values ​​onto the adjustable step sets corresponding to each control card, the final discrete phase compensation values ​​executed by each control card can be obtained. The adjustable step sets of the control cards can be represented as follows: The continuous demapping method can be expressed as: ,in, To write to the control card Discrete phase compensation values ​​in firmware; For control card The public walk into the assembly area; For control card The minimum adjustable step.

[0199] Furthermore, while ensuring executability, the discretization error is minimized to avoid over-adjustment that introduces new phase jitter. (The rest of the text appears to be incomplete and requires further context.) Recombined into discrete phase compensation vector The data is encapsulated into a phase compensation command frame, which is used to control each control card to perform corresponding phase compensation in order to solve the lateral fine tearing.

[0200] Preferably, the method described above further includes, before encapsulating the discrete phase compensation vector into the phase compensation command frame:

[0201] Substitute the discrete phase compensation vector into the objective function to calculate the target value;

[0202] The target value is subjected to global consistency verification. If the verification passes, the discrete phase compensation vector is encapsulated into the phase compensation instruction frame. Otherwise, the iteration step size is adjusted, and the process returns to the step of using the phase deviation as the initial value and constrained least squares iteration to solve the objective function using first-order gradient descent to obtain a continuous compensation solution.

[0203] In this embodiment, before encapsulating the phase compensation command frame, a global consistency verification is performed. Specifically, the discrete phase compensation vector obtained above is substituted into the objective function to calculate the corresponding target value. And perform global consistency verification on the target value, where the target value is less than or equal to the phase adjustment threshold. If the global consistency is satisfied (specifically, global boundary consistency), the discrete phase compensation vector is encapsulated into a phase compensation instruction frame and written into the instruction buffer of the control system. It waits for the next refresh cycle and is executed by the firmware. Otherwise, if the global consistency is not satisfied, the iteration step size is adjusted, and the process returns to the steps of solving the continuous phase compensation solution and restarts.

[0204] In the above embodiments, a complete closed loop is achieved from obtaining the continuous phase compensation solution, discrete phase compensation vector, verification, and issuing phase compensation command frames. The objective function focuses on boundary pixels, improving the accuracy of the correction; discrete projection ensures that the compensation amount can be directly executed by the firmware; global verification ensures monotonic synchronization of multi-card boundaries, avoiding secondary tearing. The final output discrete phase compensation vector... It provides precise and operable instructions for real-time adjustment of the control card phase, laying the execution layer foundation for the instant alignment closed loop of lateral fine tearing.

[0205] See Figure 9 Preferably, the method described above further includes:

[0206] Step S901: Receive a new frame of row reversal timestamp sequence transmitted back by each of the control cards after the compensation takes effect;

[0207] Step S902: Perform fast incremental link verification based on the row reversal timestamp sequence to obtain the verification result;

[0208] Step S903: If the verification result does not meet the preset time difference threshold, then return to the step of performing cross-card phase compensation solution and mapping based on the phase deviation, obtaining discrete phase compensation vector, and encapsulating it into a phase compensation instruction frame.

[0209] In this embodiment, after sending out the phase compensation command frame obtained above, the system also receives the commutation timestamp sequence of the new frame returned by each control card after the compensation takes effect. Based on this commutation timestamp sequence, a fast incremental link verification is performed to obtain the verification result. Specifically, this verification is performed by appending the timestamp to the global log and triggering the acquisition of the time difference between the local timestamp and the determined phase deviation. If the time difference is less than or equal to a preset time difference threshold, the requirement is met; otherwise, the verification result is determined not to meet the preset time difference threshold, and the system returns to the step of performing cross-card phase compensation solution and mapping based on the phase deviation to obtain a discrete phase compensation vector, which is then encapsulated into a phase compensation command frame. This ensures the accuracy of real-time alignment of lateral fine tears.

[0210] See Figure 10An embodiment of this application also provides a simplified schematic diagram of a network architecture that can achieve the acquisition of phase compensation vector in the above method. The input vector in the input module is composed of four channels: the scanning phase of the previous frame, the junction temperature, the power supply voltage, and the pixel row and column index. It first enters a 64-unit fully connected layer and is activated by the Rectified Linear Unit (ReLU) function. Then, it passes through two 64-dimensional graph convolutional layers (including: Graph Convolutional Neural Network (GraphConv) and ReLU function) along the pixel adjacency matrix to capture spatial coupling. After frame splicing, it is sent to a 64-dimensional gated recurrent unit (GRU) in the time dimension to model row order dynamics. The hidden state passes through a 32-unit fully connected layer + ReLU, random deactivation (Dropout) (0.1) and a 1-dimensional linear output layer to directly give the phase compensation vector of the pixel in the current frame.

[0211] See Figure 11 Another embodiment of this application provides a control device, including:

[0212] The first processing module 1101 is used to obtain the optimal offset of each control card relative to a unified reference clock and the local timestamp of each scan line of the control card, and generate a global alignment time matrix with row-card dual index; wherein, the optimal offset is determined based on the network round-trip delay measurement and the time difference of the shared boundary line between the control cards; the local timestamp is obtained by triggering an onboard nanosecond counter through the row start hardware interrupt of each control card.

[0213] The second processing module 1102 is used to generate a time-space-environment three-dimensional coupled data cube based on the global alignment time matrix, a preset bidirectional mapping function, and the temperature and power supply voltage of each control card downsampled to the pixel grid. The bidirectional mapping function is a mapping function from physical pixel coordinates to control card number-row index established according to the installation drawing and pixel row and column order.

[0214] The third processing module 1103 is used to perform digital twin prediction based on the three-dimensional coupled data cube and obtain the predicted phase field.

[0215] The fourth processing module 1104 is used to perform weighted fusion based on the virtual tear saliency map obtained by calculating the lateral gradient of the predicted phase field and the optical flow misalignment intensity map obtained by optical flow differential processing of the camera screenshot, generate a tear probability map and obtain the phase deviation corresponding to each of the control cards.

[0216] The fifth processing module 1105 is used to perform cross-card phase compensation solution and mapping based on the phase deviation, obtain discrete phase compensation vector, and encapsulate it into a phase compensation instruction frame; wherein, the phase compensation instruction frame is used to enable each of the control cards to write the discrete phase compensation vector in the firmware programmable clock interface, and lock the effective time at the next frame row sequence counter reset point.

[0217] The sixth processing module 1106 is used to send the phase compensation command frame in parallel to all the control cards after entering the inter-frame security window.

[0218] Specifically, in the control device described above, the first processing module includes:

[0219] The first processing unit is used to obtain the clock offset of each control card relative to the unified reference clock based on the network round-trip delay measurement.

[0220] The second processing unit is used to perform gradient descent iterative optimization based on the clock offset, the local timestamp, the row index set located at the control card boundary, and the first preset algorithm to obtain the optimal offset.

[0221] The third processing unit is used to correct the local timestamp matrix composed of the local timestamps according to the optimal offset to obtain the global aligned time matrix.

[0222] Specifically, in the control device described above, the second processing module includes:

[0223] The fourth processing unit is used to perform element-wise access to the global alignment time matrix according to the bidirectional mapping function to generate a pixel phase matrix, wherein the control card number in the bidirectional mapping function is used to read the column in the global alignment time matrix, and the row index in the bidirectional mapping function is used to read the row in the global alignment time matrix;

[0224] The fifth processing unit is used to perform point-by-point calculations on the two-dimensional pixel plane based on the control card number, with respect to the amplitude of the temperature and the power supply voltage, to obtain a temperature matrix and a voltage matrix.

[0225] The sixth processing unit is used to perform range normalization on the temperature matrix to obtain the corresponding target temperature matrix;

[0226] The seventh processing unit is used to perform range normalization processing on the voltage matrix to obtain the corresponding target voltage matrix;

[0227] The eighth processing unit is used to perform channel-based stitching of the pixel phase matrix, the target temperature matrix, and the target voltage matrix to obtain the three-dimensional coupled data cube.

[0228] Specifically, in the control device described above, the third processing module includes:

[0229] The ninth processing unit is used to construct a two-dimensional driving mesh according to the number list of the control card, the physical block size of the module and the pixel row and column order, and initialize the three-dimensional state vector of each node in the two-dimensional driving mesh, wherein the three-dimensional state vector includes phase value, temperature value and voltage value.

[0230] The tenth processing unit is used to perform forward multi-physics step prediction on the three-dimensional state vector on the two-dimensional driving mesh to obtain the prior state vector.

[0231] The eleventh processing unit is used to parse the three-dimensional coupled data cube into observation vectors and construct an observation mapping operator based on the nodes.

[0232] The twelfth processing unit is used to perform dynamic assimilation based on the prior state vector, the observation vector, the observation mapping operator, and the gain matrix to obtain the optimal state set. The gain matrix is ​​obtained based on the current neighborhood variance, observation accuracy, and scanning direction.

[0233] The thirteenth processing unit is used to extract phase channels from the set of optimal states and rearrange them according to the node coordinates to obtain the predicted phase field.

[0234] Furthermore, in the control device described above, the tenth processing unit includes:

[0235] The first sub-processing unit is used to call the row sequence stepping operator to perform single-frame recursion on the phase value to obtain the prior phase;

[0236] The second sub-processing unit is used to calculate the temperature value based on the two-dimensional thermal conductivity operator to obtain the prior temperature;

[0237] The third sub-processing unit is used to calculate the voltage value based on a preset node-edge resistance network to obtain the prior voltage;

[0238] The fourth sub-processing unit is used to combine the prior phase, the prior temperature and the prior voltage into the prior state vector.

[0239] Preferably, in the control device described above, after obtaining the predicted phase field, the control device further includes:

[0240] The fourteenth processing unit is used to perform temporal consistency verification and spatial smoothness verification on the predicted phase field.

[0241] The fifteenth processing unit is used to mark the regions that pass the timing consistency check and the spatial smoothness check as ideal synchronization states.

[0242] Preferably, the control device described above, the fourth processing module, includes:

[0243] The sixteenth processing unit is used to calculate the optical flow vector field of the camera screenshot using the depth-separable pyramid algorithm; take the absolute value of the lateral component of the optical flow vector field and square it, and visualize the output as the actual misalignment intensity map.

[0244] The seventeenth processing unit is used to perform weighted suppression on the actual misalignment intensity map based on the normalized environmental confidence of the environmental channel of the three-dimensional coupled data cube at the current time, so as to obtain the optical flow misalignment intensity map.

[0245] Specifically, in the control device described above, the fourth processing module includes:

[0246] The eighteenth processing unit is used to perform weight fusion based on the virtual tear saliency map, the optical flow misalignment intensity map, and the environmental confidence level to generate the tear probability map. The weights corresponding to the virtual tear saliency map, the optical flow misalignment intensity map, and the environmental confidence level are all calibrated offline according to the Bayesian minimum risk criterion.

[0247] The nineteenth processing unit is used to obtain the common boundary set of the control card according to the bidirectional mapping function;

[0248] The twentieth processing unit is used to calculate the product of the tearing probability of the cumulative pixels of each boundary and the pixel phase difference, using the common boundary set as an index, to obtain the phase deviation.

[0249] Preferably, in the control device described above, the control device further includes:

[0250] The seventh processing module is used to estimate the average latency from the camera system to the display system based on the frame sequence number returned by the control card and the timestamp of the camera image captured by the camera system.

[0251] The eighth processing module is used to obtain the predicted phase field and the camera screenshot corresponding to the output difference based on the difference between the camera timestamp and the average delay.

[0252] The ninth processing module is used to project the camera screenshot onto the coordinate system corresponding to the predicted phase field according to the bidirectional mapping function and the preset quadratic polynomial correction term;

[0253] The tenth processing module is used to perform coordinate alignment between the camera screenshot and the predicted phase field based on sub-pixel level bilinear interpolation to obtain the aligned camera screenshot.

[0254] Specifically, in the control device described above, the fifth processing module includes:

[0255] The twenty-first processing unit is used to construct an objective function for cross-card boundary phase consistency using the tear probability in the tear probability map as weights.

[0256] The twenty-second processing unit is used to solve the objective function by first-order gradient descent based on constrained least squares iteration, using the phase deviation as the initial value, to obtain a continuous compensation solution.

[0257] The twenty-third processing unit is used to project the continuous compensation solution onto the adjustable step set corresponding to each of the control cards to obtain the discrete phase compensation value corresponding to each control card.

[0258] The twenty-fourth processing unit is used to recombine all the discrete phase compensation values ​​into the discrete phase compensation vector, and encapsulate the discrete phase compensation vector into the phase compensation instruction frame.

[0259] Preferably, the control device described above further includes:

[0260] The twenty-fifth processing unit is used to substitute the discrete phase compensation vector into the objective function to calculate the target value;

[0261] The 26th processing unit is used to perform global consistency verification on the target value. If the verification is successful, the discrete phase compensation vector is encapsulated into the phase compensation instruction frame; otherwise, the iteration step size is adjusted and the process of using the phase deviation as the initial value and constrained least squares iteration to solve the objective function with first-order gradient descent to obtain a continuous compensation solution is returned.

[0262] Preferably, the control device described above further includes:

[0263] The eleventh processing module is used to receive a new frame of row reversal timestamp sequence transmitted back by each of the control cards after the compensation takes effect;

[0264] The twelfth processing module is used to perform fast incremental link verification based on the row reversal timestamp sequence to obtain the verification result.

[0265] The thirteenth processing module is used to return to the step of performing cross-card phase compensation solution and mapping based on the phase deviation, obtaining discrete phase compensation vector, and encapsulating it into a phase compensation instruction frame if the verification result does not meet the preset time difference threshold.

[0266] It should be noted that the control device in this embodiment is the same as the device described above. The implementation methods in each of the above embodiments are applicable to the embodiments of this device and can achieve the same technical effect. The device provided in this application embodiment can implement all the method steps implemented in the above method embodiments and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiments and the beneficial effects will not be described in detail.

[0267] See Figure 12 Another embodiment of this application provides an electronic device 12, including: a processor 1201, a memory 1202, and a program stored in the memory 1202 and executable on the processor 1201. When the program is executed by the processor 1201, it implements the steps of the distributed LED lateral fine tear instant alignment method as described above and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0268] Another embodiment of this application provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the steps of the distributed LED lateral fine tear instantaneous alignment method described above, achieving the same technical effect. To avoid repetition, it will not be described again here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0269] Another embodiment of this application provides a computer program product, including computer instructions that, when executed by a processor, implement the steps of the distributed LED lateral fine tear instant alignment method as described above, and achieve the same technical effect. To avoid repetition, it will not be described again here.

[0270] It should be noted that, in this document, 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 a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0271] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0272] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A distributed LED lateral scribe instant alignment method, characterized in that, The method comprises the following steps: acquiring optimal offsets of each control card relative to a unified reference clock and local timestamps of each scanning line of the control card, and generating a globally aligned time matrix with double indexes of line and card; wherein the optimal offset is determined according to network round-trip delay measurement and time difference of common boundary lines between control cards; the local timestamp is acquired by triggering a board-mounted nanosecond counter through a line start hardware interrupt of each control card; generating a time-space-environment three-dimensional coupled data cube according to the globally aligned time matrix, a preset bidirectional mapping function and temperature and power supply voltage of each control card down-sampled to a pixel grid, wherein the bidirectional mapping function is a mapping function from physical pixel coordinates to control card number-line sequence index established according to installation drawing and pixel row and column sequence; carrying out digital twin prediction according to the three-dimensional coupled data cube to acquire a predicted phase field; performing weight fusion on a virtual tearing saliency map obtained by calculating a transverse gradient of the predicted phase field and an optical flow misplacement intensity map obtained by optical flow difference processing on a camera screenshot to generate a tearing probability map and acquire a phase deviation corresponding to each control card; performing cross-card phase compensation solving and mapping according to the phase deviation to acquire a discrete phase compensation vector and package it into a phase compensation instruction frame; wherein the phase compensation instruction frame is used for enabling each control card to write the discrete phase compensation vector into a firmware programmable clock interface and lock an effective time instant at a next frame line sequence counter reset point; after entering an inter-frame safety window, the phase compensation instruction frame is issued to all the control cards in parallel.

2. The method of claim 1, wherein, The method of acquiring optimal offsets of each control card relative to a unified reference clock and local timestamps of each scanning line of the control card, and generating a globally aligned time matrix with double indexes of line and card, comprises the following steps: acquiring clock offsets of each control card relative to the unified reference clock according to the network round-trip delay measurement; performing gradient descent iterative optimization according to the clock offset, the local timestamp, a line index set located at a control card boundary and a first preset algorithm to obtain the optimal offset; correcting a local timestamp matrix composed of the local timestamp according to the optimal offset to obtain the globally aligned time matrix.

3. The method of claim 1, wherein, The method of generating a time-space-environment three-dimensional coupled data cube according to the globally aligned time matrix, a preset bidirectional mapping function and temperature and power supply voltage of each control card down-sampled to a pixel grid, comprises the following steps: performing element-by-element access on the globally aligned time matrix according to the bidirectional mapping function to generate a pixel phase matrix, wherein the control card number in the bidirectional mapping function is used for reading a column in the globally aligned time matrix, and the line sequence index in the bidirectional mapping function is used for reading a row in the globally aligned time matrix; performing amplitude point-by-point on a two-dimensional pixel plane with respect to the temperature and the power supply voltage to obtain a temperature matrix and a voltage matrix according to the control card number; performing range normalization processing on the temperature matrix to obtain a corresponding target temperature matrix; Range normalization is performed on the voltage matrix to obtain a corresponding target voltage matrix; The pixel phase matrix, the target temperature matrix, and the target voltage matrix are channel spliced to obtain the three-dimensional coupling data cube.

4. The method of claim 1, wherein, The digital twin prediction is performed according to the three-dimensional coupling data cube to obtain a predicted phase field, including: According to the number list of the control card, the module physical block size, and the pixel row and column sequence, a two-dimensional driving grid is constructed, and a three-dimensional state vector of each node in the two-dimensional driving grid is initialized, the three-dimensional state vector including a phase value, a temperature value, and a voltage value; A forward multi-physical step prediction is performed on the three-dimensional state vector on the two-dimensional driving grid to obtain a prior state vector; According to the node, the three-dimensional coupling data cube is parsed into an observation vector and an observation mapping operator is constructed; A dynamic assimilation is performed according to the prior state vector, the observation vector, the observation mapping operator, and a gain matrix to obtain an optimal state set, the gain matrix being obtained according to a current neighborhood variance, an observation accuracy, and a scanning direction; A phase channel is extracted from the optimal state set, and rearranged according to a node coordinate to obtain the predicted phase field.

5. The method of claim 4, wherein, The forward multi-physical step prediction is performed on the three-dimensional state vector on the two-dimensional driving grid to obtain a prior state vector, including: A single-frame recursion is performed on the phase value by calling a row sequence step operator to obtain a prior phase; The temperature value is calculated based on a two-dimensional heat conduction operator to obtain a prior temperature; The voltage value is calculated according to a preset node-edge resistance network to obtain a prior voltage; The prior phase, the prior temperature, and the prior voltage are spliced into the prior state vector.

6. The method of claim 4, wherein, After obtaining the predicted phase field, the method further includes: A temporal consistency check and a spatial smoothness check are performed on the predicted phase field; An area passing the temporal consistency check and the spatial smoothness check is marked as an ideal synchronization state.

7. The method of claim 1, wherein, The light flow dislocation intensity map obtained by performing light flow difference processing on the camera screenshot includes: A light flow vector field of the camera screenshot is calculated by a depth separable pyramid algorithm; an absolute value of a transverse component of the light flow vector field is taken and squared to be visualized and output as an actual dislocation intensity map; According to a normalized environment confidence of an environment channel of the three-dimensional coupling data cube at a current time, the actual dislocation intensity map is subjected to weight suppression to obtain the light flow dislocation intensity map.

8. The method of claim 7, wherein, The weight fusion is performed on the virtual tearing saliency map obtained by performing a transverse gradient calculation on the predicted phase field and the light flow dislocation intensity map obtained by performing light flow difference processing on the camera screenshot to generate a tearing probability map and obtain a phase deviation corresponding to each control card, including: The weight fusion is performed according to the virtual tearing saliency map, the light flow dislocation intensity map, and the environment confidence to generate the tearing probability map, wherein the weights corresponding to the virtual tearing saliency map, the light flow dislocation intensity map, and the environment confidence are all calibrated offline according to a Bayesian minimum risk criterion. According to the bidirectional mapping function, a public boundary set of the control cards is obtained; Taking the public boundary set as an index, a product of a tearing probability and a pixel phase difference of each boundary accumulated pixel is calculated to obtain the phase deviation.

9. The method of claim 4, wherein, Before the step of generating a tearing probability map by weight fusion of a virtual tearing saliency map obtained according to a transverse gradient calculation of the predicted phase field and an optical flow misplacement intensity map obtained by optical flow difference processing of the camera screenshot and obtaining the phase deviation corresponding to each control card, the method further comprises: According to the frame sequence number returned by the control card and a camera time stamp of the camera image captured by the camera system, an average time delay of the camera system to the display system is estimated; According to a difference between the camera time stamp and the average time delay, the predicted phase field and the camera screenshot corresponding to the output difference value are obtained; According to the bidirectional mapping function and a preset quadratic polynomial correction term, the camera screenshot is projected into a coordinate system corresponding to the predicted phase field; According to sub-pixel level bilinear interpolation, the camera screenshot and the predicted phase field are aligned in coordinates to obtain the aligned camera screenshot.

10. The method of claim 8, wherein, The step of solving and mapping the phase deviation to obtain a discrete phase compensation vector and packaging the phase compensation instruction frame comprises: Taking the tearing probability in the tearing probability map as a weight, a target function of cross-card boundary phase consistency is constructed; Taking the phase deviation as an initial value, a continuous compensation solution is obtained by solving the target function by first-order gradient descent based on constrained least squares iteration; The continuous compensation solution is projected into an adjustable step set corresponding to each control card to obtain a discrete phase compensation value corresponding to each control card; All the discrete phase compensation values are recombined into the discrete phase compensation vector, and the discrete phase compensation vector is packaged into the phase compensation instruction frame.

11. The method of claim 10, wherein, Before the step of packaging the discrete phase compensation vector into the phase compensation instruction frame, the method further comprises: The discrete phase compensation vector is substituted into the target function to obtain a target value; The target value is verified for global consistency, if the verification is passed, the discrete phase compensation vector is packaged into the phase compensation instruction frame; otherwise, an iteration step is adjusted, and the step of taking the phase deviation as an initial value and solving the target function by first-order gradient descent based on constrained least squares iteration to obtain a continuous compensation solution is executed.

12. The method of claim 1, wherein, The method further comprises: A new sequence of line commutation time stamps returned by each control card after the compensation takes effect is received; A verification result is obtained by performing fast incremental link verification according to the sequence of line commutation time stamps; If the verification result does not satisfy a preset time difference threshold, the step of solving and mapping the phase deviation to obtain a discrete phase compensation vector and packaging the phase compensation instruction frame is executed.

13. A control device characterized by comprising: The method comprises: The first processing module is configured to acquire optimal offsets of each control card relative to a unified reference clock and local time stamps of each scanning line of the control card, and generate a global alignment time matrix with double indexes of line and card; wherein the optimal offset is determined according to network round-trip delay measurement and time difference of common boundary lines between control cards; the local time stamp is acquired by triggering a board-mounted nanosecond counter through a line start hardware interrupt of each control card; The second processing module is configured to generate a time-space-environment three-dimensional coupled data cube according to the global alignment time matrix, a preset bidirectional mapping function and temperature and power supply voltage of each control card after down-sampling to a pixel grid, wherein the bidirectional mapping function is a mapping function from physical pixel coordinates to card number-line sequence index, which is established according to an installation drawing and pixel row and column sequence; The third processing module is configured to perform digital twin prediction according to the three-dimensional coupled data cube to acquire a predicted phase field; The fourth processing module is configured to perform weight fusion on a virtual tearing saliency map obtained by calculating a transverse gradient of the predicted phase field and an optical flow misplacement intensity map obtained by optical flow difference processing on a camera screenshot, generate a tearing probability map and acquire a phase deviation corresponding to each control card; The fifth processing module is configured to perform cross-card phase compensation solving and mapping according to the phase deviation, acquire a discrete phase compensation vector and package as a phase compensation instruction frame; wherein the phase compensation instruction frame is used to make each control card write the discrete phase compensation vector in a firmware programmable clock interface and lock an effective time to a next frame line sequence counter reset point; The sixth processing module is configured to parallelly issue the phase compensation instruction frame to all the control cards after entering an inter-frame safety window.

14. An electronic device, comprising: A processor, a memory and a program stored on the memory and executable on the processor, wherein the program is executed by the processor to implement steps of the distributed LED horizontal fine tearing real-time alignment method according to any one of claims 1 to 12. A computer program is stored on the computer readable storage medium, and the computer program is executed by the processor to implement steps of the distributed LED horizontal fine tearing real-time alignment method according to any one of claims 1 to 12.

15. A computer-readable storage medium, characterized in that, Computer instructions are included, and the computer instructions are executed by the processor to implement steps of the distributed LED horizontal fine tearing real-time alignment method according to any one of claims 1 to 12.

16. A computer program product, characterised in that, ​

Images