Method and device for positioning, arranging and calibrating LED (light-emitting diode) modules

Abstract

The invention relates to the technical field of LED modules, and discloses an LED module positioning arrangement and calibration method and device. The method comprises the following steps: acquiring an LED module mounting image and detecting pose data of a plurality of first LED modules; calculating a target position coordinate of each first LED module based on the pose data; the multiple first LED modules are assembled according to the target position coordinates, and a second LED module is obtained; according to the second LED module, position light intensity coupling analysis is executed, and target calibration parameters are generated. Image information of a plurality of modules can be processed at the same time, the detection efficiency is improved, and then rapid positioning assembly and test calibration of the LED modules are achieved.

Description

Technical Field

[0001] This invention relates to the field of LED module technology, and in particular to a method and apparatus for positioning, arranging and calibrating LED modules. Background Technology

[0002] Traditional LED module manufacturing relies on mechanical positioning systems and manual calibration. While mechanical positioning methods can meet basic installation requirements in high-precision assembly and mass production, they often involve individual detection and processing when coordinating the positioning of multiple LED modules. This makes it difficult to guarantee overall positioning accuracy and easily leads to cumulative errors. When dealing with flicker caused by 100Hz dual-line frequency ripple, existing technologies often only passively suppress it by adding hardware filtering circuits, failing to fundamentally address the mechanism of ripple generation. This results in limited calibration effectiveness and high costs. Summary of the Invention

[0003] This invention provides a method and apparatus for positioning, arranging, and calibrating LED modules. This invention can process image information from multiple modules simultaneously, improving detection efficiency and thus enabling rapid positioning, assembly, testing, and calibration of LED modules.

[0004] In a first aspect, the present invention provides a method for positioning, arranging, and calibrating LED modules, the method comprising: Acquire LED module mounting images and detect the pose data of multiple first LED modules; Calculate the target position coordinates of each first LED module based on the pose data; The multiple first LED modules are assembled according to the target position coordinates to obtain a second LED module; Based on the position light intensity coupling analysis performed by the second LED module, target calibration parameters are generated.

[0005] In conjunction with the first aspect, in a first implementation of the first aspect of the present invention, the step of acquiring LED module mounting images and detecting pose data of multiple first LED modules includes: The mounting status of multiple first LED modules is simultaneously acquired by a machine vision system to obtain the original mounting image; The original mounting image is subjected to resolution normalization processing to obtain a standard mounting image; Pixel physical size calibration is performed based on the standard mounting image to obtain pixel coordinate calibration parameters; The pixel coordinate calibration parameters are applied to the standard mounting image for coordinate system unification processing to obtain the LED module mounting image. Parallel pose detection is performed on the LED module mounting image to obtain pose data of multiple first LED modules.

[0006] In conjunction with the first aspect, in a second implementation of the first aspect of the present invention, the step of performing parallel pose detection on the LED module mounting image to obtain pose data of multiple first LED modules includes: The LED module mounting image is input into a multi-channel blind deconvolutional neural network for image region decomposition and feature extraction to obtain the edge texture feature map of each first LED module. Based on the edge texture feature map, pose regression calculation is performed to obtain the pose vector of each first LED module. The pose vectors are output and integrated in parallel to obtain pose data of multiple first LED modules.

[0007] In conjunction with the first aspect, in a third implementation of the first aspect of the present invention, the step of inputting the LED module mounting image into a multi-channel blind deconvolutional neural network for image region decomposition and feature extraction to obtain the edge texture feature map of each first LED module includes: The LED module mounting image is input into the multi-channel signal separation layer of a multi-channel blind deconvolutional neural network to perform image channel decomposition, thereby obtaining the channel image data of each first LED module. The channel image data is input into the adaptive FIR filter bank layer of a multi-channel blind deconvolutional neural network for channel filtering to obtain multiple channel feature data. The channel feature data is input into the neural blind deconvolution feature extraction layer of the multi-channel neural blind deconvolution network for blind source separation and feature extraction, resulting in multiple separated feature maps. The separated feature maps are input into the channel orthogonality constraint processing layer of the multi-channel neural blind deconvolution network to perform inter-channel interference suppression processing, resulting in multiple channel feature maps. The channel feature maps are input into the multi-channel pose fusion output layer of the multi-channel neural blind deconvolution network for parallel feature integration to obtain the edge texture feature maps of each first LED module.

[0008] In conjunction with the first aspect, in a fourth implementation of the first aspect of the present invention, the step of calculating the target position coordinates of each first LED module based on the pose data includes: Calculate the spatial pulse characteristics comprehensive index of each first LED module based on the pose data; Based on the aforementioned spatial impulsivity comprehensive index, the spatial impulsivity index value is gradually maximized through gradient descent algorithm and momentum term adjustment to obtain the target position parameters of each first LED module; Multi-scale feature fusion is performed on the target position parameters to obtain the target position coordinates of each first LED module.

[0009] In conjunction with the first aspect, in a fifth implementation of the first aspect of the present invention, the step of calculating the spatial impulsiveness comprehensive index of each first LED module based on the pose data includes: Based on the pose data, the fourth-order cumulant kurtosis function value is calculated to detect the spatial concentration of the light-emitting area of ​​each first LED module, the peak detection function value is used to identify the maximum light intensity response value of each first LED module, and the sharpness function value is used to evaluate the edge sharpness of each first LED module. The spatial concentration, the maximum light intensity response value, and the edge sharpness are weighted to obtain the weighted concentration component, weighted peak component, and weighted sharpness component of each first LED module. The weighted concentration component, weighted peak component, and weighted sharpness component are summed to obtain the single-module spatial pulseness index of each first LED module; The spatial pulse characteristics of each first LED module are globally summed to obtain the comprehensive spatial pulse characteristics of each first LED module.

[0010] In conjunction with the first aspect, in a sixth implementation of the first aspect of the present invention, the step of assembling the plurality of first LED modules according to the target position coordinates to obtain a second LED module includes: Based on the target location coordinates, LED module spatial layout planning is performed to determine the spatial layout parameters of each first LED module in the display screen frame. The spatial layout parameters are used to calculate the alignment accuracy between modules and the splicing gap constraint to obtain a global arrangement scheme. Based on the global arrangement scheme, the installation sequence and positioning reference of each first LED module are determined, and an LED module assembly process scheme is generated. According to the LED module assembly process, multiple first LED modules are precisely positioned and installed one by one, and each first LED module is fixed into the corresponding slot in the display frame to obtain the second LED module.

[0011] In conjunction with the first aspect, in the seventh implementation of the first aspect of the present invention, the step of generating target calibration parameters based on the position light intensity coupling analysis performed by the second LED module includes: The second LED module was tested for power-on drive. The voltage fluctuation of each LED module was monitored in real time by an adaptive peak voltage tracking controller to obtain voltage fluctuation data of each LED module. Based on the voltage fluctuation data, dual-line frequency ripple detection and suppression processing are performed to obtain the driving voltage parameters; Based on the driving voltage parameters, the light-emitting characteristics of the LED module are analyzed and the correspondence between the physical position of the second LED module and the light intensity distribution is established to obtain position-light intensity coupling relationship data. Global calibration is performed based on the location-intensity coupling relationship data to obtain the target calibration parameters.

[0012] In conjunction with the first aspect, in the eighth implementation of the first aspect of the present invention, the step of performing global calibration based on the position-intensity coupling relationship data to obtain target calibration parameters includes: Based on the positional light intensity coupling relationship data, a modified constitutive relation error function is constructed to calculate the calibration deviation and obtain the error evaluation index. The error evaluation index is used as the optimization target input Gaussian process to establish a probabilistic model of calibration parameters. The correlation between parameters is described by the Matérn kernel function and the prior distribution of the objective function is established to obtain the probability distribution model. Based on the probability distribution model, the calibration parameters are optimized iteratively to obtain the initial calibration parameters. The initial calibration parameters are then weighted by the Fisher information matrix to obtain the target calibration parameters.

[0013] Secondly, the present invention provides a positioning, arranging, and calibration device for LED modules, the positioning, arranging, and calibration device for LED modules comprising: The acquisition module is used to acquire LED module mounting images and detect the pose data of multiple first LED modules; The calculation module is used to calculate the target position coordinates of each first LED module based on the pose data; An assembly module is used to assemble the plurality of first LED modules according to the target position coordinates to obtain a second LED module; The coupling analysis module is used to perform position light intensity coupling analysis based on the second LED module and generate target calibration parameters.

[0014] The technical solution provided by this invention employs a multi-channel blind deconvolutional neural network to achieve parallel pose detection of N LED modules. Compared to the traditional method of detecting each module individually, this approach can process image information from multiple modules simultaneously, improving detection efficiency. Channel orthogonal constraint processing layers eliminate inter-channel interference, ensuring the accuracy and reliability of parallel multi-module detection. A spatial impulsivity index designed based on the optical characteristics of LED modules comprehensively evaluates three dimensions: luminous concentration, peak response, and edge sharpness, specifically addressing the unique optical problems in LED module positioning. Compared to general computer vision algorithms, this index is more suitable for the physical characteristics of LED modules, enabling more accurate position coordinate calculations. By constructing a global optimization model including position accuracy, alignment accuracy, gap control, and flatness control terms, collaborative positioning and arrangement of multiple modules is achieved. Compared to the traditional method of processing each module independently, global constraint optimization ensures overall positioning accuracy at the system level, avoiding the generation of accumulated errors. An adaptive peak voltage tracking controller replaces the traditional passive filtering method, actively tracking voltage peak changes and dynamically adjusting the PWM duty cycle. The 100Hz dual-line frequency ripple problem was solved at the voltage control source, achieving truly flicker-free drive. A mathematical coupling model of the physical position and light intensity distribution of the LED module was established, realizing the organic integration of position calibration and optical calibration. Compared with traditional simple geometric calibration methods, position-light intensity coupling analysis can comprehensively consider the influence of module position on optical performance, achieving optoelectronic integrated calibration. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart illustrating the positioning, arrangement, and calibration method for LED modules provided in this application embodiment; Figure 2 A schematic block diagram of the structure of the LED module positioning, arrangement and calibration device provided in the embodiments of this application. Detailed Implementation

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

[0018] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the described order. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change based on the actual situation.

[0019] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0020] It should also be further understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0021] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features described herein can be combined with each other.

[0022] Please see Figure 1 , Figure 1 This is a flowchart illustrating the positioning, arrangement, and calibration method for LED modules provided in the embodiments of this application, as shown below. Figure 1 As shown, the LED module positioning, arrangement, and calibration method provided in this application includes: Step S100: Obtain LED module mounting images and detect the pose data of multiple first LED modules; Specifically, an industrial-grade machine vision system synchronously acquires the status of multiple LED modules on the mounting platform. This system, with its high-resolution and high-speed image acquisition capabilities, completes a full-coverage scan before any module position drift occurs, generating original mounting images covering multiple module areas. The original mounting images undergo resolution standardization, employing a resampling algorithm to interpolate and adjust the images to a preset standard resolution, for example, unifying all images to a 4096×2160 pixel format. This eliminates inconsistencies in sampling scale caused by different camera acquisition channels, resulting in standard mounting images. Based on these standard mounting images, pixel physical size calibration is performed. Using a calibration board pattern or high-precision ruler, the mapping relationship between multiple pixel positions and their corresponding physical positions is extracted, constructing a pixel coordinate calibration parameter matrix. This matrix describes the conversion ratio between each pixel in the image and actual physical space units (e.g., mm) and the nonlinear distortion correction model. The pixel coordinate calibration parameters are applied to the standard mounting images, performing coordinate system unification processing. All image content is mapped to a two-dimensional coordinate system with a unified physical scale and coordinate axis orientation, resulting in LED module mounting images. Parallel detection algorithms based on deep neural networks (such as multi-channel convolutional neural network structure MCNBD) are used to quickly identify all visible first LED modules in the processed mounting image. Combining the image texture features, boundary contours and arrangement rules of each module, the corresponding three-dimensional pose data is extracted, including the lateral offset, longitudinal offset and rotation angle of each LED module in the image coordinate system, forming a set of pose data vectors of multiple first LED modules.

[0023] Step S200: Calculate the target position coordinates of each first LED module based on the pose data; Specifically, based on the initial pose data of multiple first LED modules, a spatial impulsivity comprehensive index is constructed for each module. This index is quantitatively evaluated by combining three key dimensions: luminous concentration, local response peak, and edge sharpness in the image space. Luminous concentration measures the focusing characteristics of the module's light intensity distribution; local response peak captures the response intensity of the strongest luminous region; and edge sharpness evaluates the image clarity of the module's boundaries. These three together constitute a composite index reflecting the module's spatial response characteristics. Based on the spatial impulsivity index, target position parameters are optimized for each module. During optimization, a gradient descent algorithm with momentum is used to iteratively adjust the module's position parameters, enabling the optimization process to overcome local extrema traps and gradually converge towards the global optimum. Each parameter update is determined by the superposition of the current index's improvement trend and the previous update direction. Through continuous iteration, the spatial impulsivity response of each module gradually approaches its strongest point, determining the optimal spatial parameter combination for the module in the current coordinate system. The target position parameters obtained from each module are introduced into the multi-scale feature fusion process. That is, the spatial response features of the modules are re-evaluated at multiple different spatial scales. After eliminating the differences between scales through normalization, the fusion weights are dynamically allocated according to the feature response intensity or signal-to-noise ratio at each scale. The position estimation results at all scales are then weighted and fused to generate the target position coordinates.

[0024] Step S300: Assemble multiple first LED modules according to the target position coordinates to obtain a second LED module; Specifically, based on the target position coordinates of all first LED modules, the spatial layout planning of the LED modules is performed. By meshing the effective installation area within the display frame, analyzing boundary constraints, and mapping the actual dimensions of the modules, the spatial layout parameters of each first LED module in the overall splicing structure are determined. These parameters include its positional offset relative to the edge of the display screen, the center point, or other modules, its rotation angle, and its row and column number, while also considering the mechanical interference between the modules and structural components. The spatial layout parameters are input into the module arrangement engine to perform alignment accuracy control and splicing gap constraint calculations. By establishing error evaluation models in the vertical and horizontal directions between modules, and combining factors such as module dimensional tolerances, structural assembly boundaries, and target splicing gaps, a global arrangement scheme that meets the consistency requirements of the entire screen structure is solved. The arrangement scheme aims to minimize the pose offset between modules, gap deviations, and planar height differences, and outputs the positioning matrix and adjacent mating relationships of each module in the physical frame. After completing the arrangement planning, the specific installation sequence and positioning benchmarks for each first LED module are determined based on the global arrangement scheme. This includes priority installation positions, alignment reference surfaces, pre-tightening sequence, and temperature compensation strategies. A repeatable LED module assembly process plan suitable for multiple batches is generated, clarifying the workstation operation flow, module handling path, positioning fixture settings, and clamping and release sequence. Following the assembly process plan, multiple first LED modules are precisely positioned and installed one by one. Using a high-precision displacement platform or automated assembly robot arm, the modules are installed into the corresponding slots in the display frame according to the process coordinates, completing the mechanical fixation and initial electrical contact of the modules, thus constructing the second LED module.

[0025] Step S400: Perform position light intensity coupling analysis on the second LED module to generate target calibration parameters.

[0026] Specifically, after the second LED module is assembled, a power-on drive test is performed on the entire screen. Each module is connected to a voltage control system with dynamic response capabilities, and an adaptive peak voltage tracking controller is used to monitor and record the power supply voltage of the module in real time during operation. The voltage tracking controller dynamically extracts the voltage peak change trend of each module within a specific time window through sliding window analysis, weighted calculation, and edge tracking algorithms, obtaining voltage fluctuation data that reflects the stability and response characteristics of the LED power input. Based on the voltage fluctuation data, possible dual-frequency ripple signals are detected and filtered. By introducing an active notch filter mechanism and a linear voltage regulation stage in the control loop, deep suppression of 100Hz and its higher-order harmonic ripple components is achieved, minimizing the impact of voltage disturbances and outputting drive voltage parameters within a controllable range, including the basic duty cycle and average voltage value, as well as the dynamically compensated voltage waveform response characteristics. Based on the processed driving voltage parameters, multi-dimensional modeling of the luminous characteristics of the LED module under different voltage response conditions is performed, including luminous intensity distribution, brightness stability, color consistency, and directional attenuation. Through large-scale image acquisition and analysis, a mapping relationship between the physical location of the second LED module and the spatial light intensity distribution is established. The position-light intensity coupling data can describe the contribution characteristics of each module to the overall light field at a specific spatial location. Based on the position-light intensity coupling data, global calibration and optimization processing is performed. Taking into account the light intensity overlap area of ​​all modules, the module edge fusion area, and the overall screen brightness consistency requirements, by adjusting the driving voltage duty cycle, compensation parameters, or PWM control signals of each module, a target calibration parameter set that can achieve brightness balance, color temperature consistency, and minimum ripple across the entire screen is constructed.

[0027] In one specific embodiment, the process of performing step S100 may specifically include the following steps: The mounting status of multiple first LED modules is simultaneously acquired by a machine vision system to obtain the original mounting image; The original mounting image is normalized to obtain a standard mounting image; Pixel physical size calibration is performed based on standard mounting images to obtain pixel coordinate calibration parameters; The pixel coordinate calibration parameters are applied to the standard mounting image for coordinate system unification to obtain the LED module mounting image. Parallel pose detection is performed on the LED module mounting images to obtain pose data for multiple first LED modules.

[0028] Specifically, an industrial machine vision system is deployed at the intermediate workstation where LED modules are mounted but not yet cured or electrically packaged. This system includes multiple synchronously triggered area scan cameras, line scan cameras, or high-frame-rate CMOS imaging modules, connected to a programmable vision processing controller for synchronous acquisition and control. During actual operation, by setting a trigger signal, an image scan of the entire mounting area is completed instantaneously before the LED modules experience positional drift, generating original mounting images containing multiple first LED modules. Resolution standardization processing is performed on the original mounting images, using a resampling interpolation algorithm to uniformly adjust the image size, pixel density, and sampling alignment, mapping all image content to a specified standard, such as a 4096×2160 pixel format. Simultaneously, image illumination intensity and grayscale dynamic range are normalized to enhance edge detection accuracy. Pixel physical size calibration is performed based on the standard mounting image. Using a high-precision reference calibration template or pattern pre-placed in the mounting area, the distribution of calibration points in the image coordinates is detected. Combined with the actual physical dimensions, the physical spatial length, width, and coordinate distortion correction factor corresponding to each pixel in the image are calculated, yielding pixel coordinate calibration parameters, including pixel scaling factor, rotation deviation, principal point position, and distortion correction model. After applying these pixel coordinate calibration parameters to the standard mounting image, coordinate system unification processing is performed. This involves mapping the image coordinate system to a unified physical reference coordinate system using a coordinate transformation model. The physical reference coordinate system is based on the display screen's structural frame, ensuring that image data from different image acquisition batches and different camera acquisition channels maintain consistent spatial positioning, thus generating the LED module mounting image. A high-performance parallel processing framework is used to perform pose recognition and state calculation on LED module mounting images. A parallel pose detection model based on deep neural networks is introduced, such as a neural network structure that integrates convolutional feature extraction, region localization, shape reconstruction and inverse mapping. This model can quickly identify and calculate the position distribution, rotation posture and boundary integrity of multiple first LED modules in the image. The output of each module includes its two-dimensional position offset and rotation angle information in a unified coordinate system, forming a pose dataset.

[0029] In one specific embodiment, the process of performing parallel pose detection on the LED module mounting image to obtain pose data of multiple first LED modules can specifically include the following steps: The LED module mounting image is input into a multi-channel blind deconvolutional neural network for image region decomposition and feature extraction to obtain the edge texture feature map of each first LED module. Pose regression calculation is performed based on edge texture feature map to obtain the pose vector of each first LED module; The pose vectors are output in parallel and integrated to obtain pose data of multiple first LED modules.

[0030] Specifically, the LED module mounting image is input into a multi-channel blind deconvolutional neural network. The network employs a multi-input channel design, with each channel processing information from different regions of the image or specific local regions of the module. A shared parameter mechanism within the network ensures a uniform response standard for the overall structure. In the pre-structure of the neural network, multiple convolutional layers perform local receptive field feature acquisition on the input image, extracting low-level spatial information such as edges, contours, array patterns, and pad shapes of the LED module. A non-linear activation function is then used to compress and sparsify the feature map, enhancing the network's sensitivity to subtle structural features in the image. The core blind deconvolution structure of the network reconstructs images without explicit degradation function constraints. Combined with multi-scale feature mapping and signal reconstruction strategies, it effectively removes spatial blur and local distortion caused by illumination interference, focus deviation, and lens distortion in the mounting image, gradually restoring clear edge responses and local texture morphology, generating an edge texture feature map. This map features high local contrast, high orientation preservation, and low background noise, describing the spatial contour structure and geometric pose of each first LED module in the mounting image. The edge texture feature map is then input into the pose regression module. Multiple fully connected layers or residual mapping structures are used to deeply encode the geometric features of each module region in the map. A spatial parameter learning mechanism establishes the mapping relationship between feature responses and physical pose, enabling regression calculation of module displacement and rotation angle. This outputs a pose vector for each module, containing translational displacement and rotational orientation, reflecting the precise geometric state of the module relative to a unified coordinate reference. After the regression calculation is completed, the output of the neural network performs parallel integration processing, and outputs the pose vector set of all modules in a unified format, including fields such as module number, center coordinates, attitude angle, rotation reference axis, and offset error evaluation value, forming pose data of multiple first LED modules.

[0031] In one specific embodiment, the process of inputting the LED module mounting image into a multi-channel blind deconvolutional neural network for image region decomposition and feature extraction to obtain the edge texture feature map of each first LED module can specifically include the following steps: The LED module mounting image is input into the multi-channel signal separation layer of a multi-channel neural blind deconvolution network to perform image channel decomposition, thereby obtaining the channel image data of each first LED module. The channel image data is input into the adaptive FIR filter bank layer of the multi-channel neural blind deconvolutional network for channel filtering to obtain multiple channel feature data. The channel feature data is input into the neural blind deconvolution feature extraction layer of the multi-channel neural blind deconvolution network to perform blind source separation and feature extraction, resulting in multiple separated feature maps. The separated feature maps are input into the channel orthogonality constraint processing layer of the multi-channel neural blind deconvolutional network to perform inter-channel interference suppression processing, resulting in multiple channel feature maps. The channel feature maps are input into the multi-channel pose fusion output layer of the multi-channel neural blind deconvolution network for parallel feature integration, resulting in edge texture feature maps for each first LED module.

[0032] Specifically, the LED module mounting image is input into a multi-channel blind deconvolutional neural network structure. This network features a highly modular design, with each layer corresponding to a key sub-function in the mounting image analysis. The LED module mounting image enters a multi-channel signal separation layer. In this layer, based on the initial spatial distribution characteristics of the modules in the mounting image, image segmentation, region clustering, and channel mapping strategies are used to divide the entire mounting image into multiple independent module channel image data. These channel images correspond to local region images or physical channel inputs of multiple modules, forming a channel image data set. The channel image data is then input into the network's adaptive FIR filter bank layer. This layer configures a set of dynamically adjustable finite impulse response filters for each module channel. The filters perform one-dimensional or two-dimensional filtering operations on the input image based on features such as noise distribution, edge blurring, and texture directionality. Through a learning mechanism, the filter coefficients are automatically optimized to enhance module edge response, suppress background interference, and improve texture clarity. During this process, the image of each module is converted into channel feature data, including local geometric features, gray-level gradient distribution, and spatial frequency response information. The channel feature data is input into the neural blind deconvolution feature extraction layer. This layer works collaboratively through depthwise convolution, deconvolution mapping, and blind source separation mechanisms. Without a predefined convolution kernel, it adaptively decouples the image superposition signals in the module region, effectively separating mixed features introduced by illumination changes, material reflections, and mounting errors. It also extracts feature maps that accurately reflect the module's geometric contours, solder joint edges, and array structure. These feature maps are highly focused on the physical structure region of the module, eliminating irrelevant background and interfering structures from the image. To prevent inter-channel interference caused by light spot diffusion, edge proximity, or optical ghosting, the extracted feature maps are input into the channel orthogonal constraint processing layer for inter-channel interference suppression. Based on the frequency domain response orthogonality principle and feature space overlap evaluation index, the channel orthogonal constraint processing layer performs resonance suppression, phase offsetting, and signal reconstruction operations on potential feature coupling regions between adjacent module channels. This reduces cross-interference between channel features, ensuring high independence and regional consistency of the feature data for each module, and outputting a set of channel feature maps. The channel feature maps are input into the multi-channel pose fusion output layer of the network. The fusion output layer performs cross-channel parallel feature integration operations. Through pose parameter association, edge direction unification, scale normalization and semantic completion mechanisms, the local maps of each module are fused into edge texture feature maps for output, forming a feature map set.

[0033] The process involves inputting channel feature data into the blind deconvolution feature extraction layer of a multi-channel neural blind deconvolution network for blind source separation and feature extraction, resulting in multiple separated feature maps. This includes: performing LED module feature spectrum analysis on the channel feature data; extracting the frequency domain feature distribution of LED module edges, textures, and geometry in each channel using Fast Fourier Transform to obtain the spectral feature vector for each channel; performing real-time adaptive filter coefficient optimization based on the spectral feature vector; dynamically adjusting the coefficient weights of the 512th-order FIR filter using the minimum mean square error algorithm and recursive least squares method to automatically match the filter frequency response to the optical characteristics of the LED module, resulting in an adaptively optimized filter coefficient matrix; and applying the filter coefficient matrix to inter-channel blind source separation. The algorithm demixes the LED module signal, separating the mixed LED module image signal into independent source signals through independent component analysis and the principle of maximizing non-Gaussianity, thus obtaining independent feature components after source signal separation. LED module-specific feature enhancement is then applied to these independent feature components, highlighting the regular arrangement of the LED chip array, the rectangular boundary of the PCB substrate, and the standardized interface of the connectors, while suppressing background noise and irrelevant information interference, resulting in enhanced LED module feature signals. Finally, deconvolution reconstruction is performed based on the LED module feature signals, reconstructing the abstract feature signals into feature maps with clear spatial location information while maintaining the geometric integrity of the LED module, resulting in multiple separated feature maps.

[0034] In one specific embodiment, the process of performing step S200 may specifically include the following steps: Calculate the comprehensive spatial pulse characteristics of each first LED module based on pose data; Based on the spatial impulsivity comprehensive index, the spatial impulsivity index value is gradually maximized through gradient descent algorithm and momentum term adjustment to obtain the target position parameters of each first LED module; Multi-scale feature fusion is performed on the target position parameters to obtain the target position coordinates of each first LED module.

[0035] Specifically, based on the geometric pose, edge response, and light intensity distribution characteristics of each module in space, a spatial impulsivity comprehensive index is constructed to characterize the spatial imaging features of the modules. This index comprehensively evaluates the integrity and structural completeness of the modules in the mounting image from multiple dimensions, reflecting their spatial performance quality at the current pose. The spatial impulsivity comprehensive index integrates three core aspects: the concentration of local responses in the module image, the prominence of light intensity peaks, and the clarity of edge contours. Concentration reflects whether the module's luminous area is highly focused; light intensity peaks reflect whether there are abnormal brightness distributions or blurry spots; and edge clarity reflects the positioning accuracy of the mounting structure. A composite index function is constructed by extracting signals such as the light intensity histogram, grayscale gradient distribution, and local edge operator response from the feature map of the module image. A larger composite index function value indicates that the current module is closer to the ideal state at the current pose. After constructing the index, to optimize the spatial position of the modules and improve the alignment accuracy of the entire system, the spatial parameters of each module are iteratively adjusted based on the spatial impulse comprehensive index. The gradient descent optimization algorithm with momentum term is used to learn the target position parameters of the modules. The optimization objective is to maximize the spatial impulse comprehensive index value of each module under the adjusted posture. During the operation, the optimal adjustment direction is gradually determined by calculating the partial derivative of the index function with respect to the module pose parameters. The historical gradient update trend is introduced in combination with the momentum factor to suppress gradient oscillation and accelerate the convergence speed. In each iteration, the position coordinates and attitude angle of the module are finely adjusted, and after several rounds of optimization, they are stably converged to the target position parameters that are locally optimal or globally optimal. After completing the iteration process, the target position parameters of each module determined based on the index optimization results are obtained. To improve the robustness and scale adaptability of target location parameters, a multi-scale feature fusion framework is introduced for processing. Considering the performance characteristics of the module under different image resolutions or different spatial receptive domains, the original mounting image is resampled at multiple scales to construct multiple scale versions. The module's features are re-extracted at different spatial resolutions such as 1x, 2x, and 4x. The scale weighting factor is constructed by combining the position response, feature sharpness, and relative signal-to-noise ratio presented at these scales. After completing feature alignment and scale response normalization between scales, a weighted fusion method is used to comprehensively summarize the target location parameters at multiple scales to form the final target location coordinates with scale invariance, edge enhancement, and resistance to local disturbances.

[0036] In one specific embodiment, the process of calculating the spatial impulsivity comprehensive index of each first LED module based on pose data may specifically include the following steps: Based on the pose data, the fourth-order cumulant kurtosis function value is calculated to detect the spatial concentration of the light-emitting area of ​​each first LED module, the peak detection function value is used to identify the maximum light intensity response value of each first LED module, and the sharpness function value is used to evaluate the edge sharpness of each first LED module. The spatial concentration, maximum light intensity response value, and edge sharpness are weighted to obtain the weighted concentration component, weighted peak component, and weighted sharpness component of each first LED module. The single-module spatial pulse index of each first LED module is obtained by summing the weighted concentration component, weighted peak component, and weighted sharpness component. The spatial pulse characteristics of each first LED module are summed globally to obtain the comprehensive spatial pulse characteristics of each first LED module.

[0037] Specifically, based on the pose data of each first LED module, its corresponding region in the image is located, and the luminous image data within that region is extracted. After region extraction, a fourth-order cumulant kurtosis function is calculated on the grayscale or brightness distribution within the module image region to determine the light intensity concentration of the module image region. By analyzing the sharpness of the pixel brightness distribution, it is determined whether the luminous region is highly concentrated. If the kurtosis value is high, it indicates that the intensity of the module's luminous center is significantly higher than that of the edges, reflecting the correlation between mounting and light source centering, and is used to evaluate the module's spatial focusing capability under the current pose. Next, peak detection processing is performed within the module image region. By searching for the maximum brightness value in a sliding window in the local region, and calculating the ratio of the maximum brightness value to the global average brightness or regional average, the maximum response capability of each module in the current state is extracted, reflecting the photoelectric excitation efficiency, central pixel stability, and degree of local saturation characteristics of the LED module. Sharpness function analysis is performed on the boundary regions of the module images. By applying methods such as Laplacian operator, Sobel operator, or second-order gradient operator to the module boundary regions, the response to changes in the edge contours in the image is calculated to obtain a sharpness factor characterizing the edge clarity, reflecting the geometric features of the image such as whether the module arrangement is regular, whether the mounting is neat, and whether the transition between adjacent pixels is abrupt. Spatial concentration, maximum light intensity response value, and edge sharpness are weighted. The weighting is combined with the statistical analysis results of the luminous physical characteristics, structural design rules, and arrangement of the LED modules. Spatial concentration is assigned a higher weight to emphasize the geometric focusing characteristics of the luminous area, peak response is used as a secondary weight to enhance the intensity of the optical center, and edge sharpness is used to assist in judging the reliability of positioning. Therefore, after setting appropriate weighting coefficients according to the different requirements of accuracy and structural integrity for the task, the kurtosis value, peak response value, and sharpness function value are multiplied by the corresponding weights to obtain the weighted concentration component, weighted peak component, and weighted sharpness component of each module. The three weighted components are directly added together to form a single-module spatial impulsivity index, which represents the composite evaluation result of the optical response intensity and geometric imaging quality exhibited by the current module in the image space under the current pose. The single-module spatial impulsivity indices of all first LED modules are uniformly accumulated and summed to form a comprehensive spatial impulsivity index covering the entire mounted image, which is used to measure the overall image sharpness, luminous concentration, and structural regularity of the entire module array in the current arrangement and assembly state.

[0038] Specifically, the spatial impulsivity indices of each first LED module are globally summed to obtain a comprehensive spatial impulsivity index for each first LED module. This includes: modeling a multi-objective optimization problem for the spatial impulsivity indices of each first LED module, setting four optimization objectives: maximizing LED module positioning accuracy, minimizing the gap between adjacent modules, optimizing overall arrangement consistency, and minimizing computational complexity, resulting in a multi-objective optimization mathematical model; performing a Pareto optimal solution set search based on the multi-objective optimization mathematical model, using a non-dominated sorting genetic algorithm to find Pareto front solutions in the four-dimensional objective space that cannot be simultaneously improved by other solutions, avoiding local optima problems caused by single-objective optimization, and obtaining the Pareto optimal solution set for the LED module arrangement; and then... The optimal solution set is input into a multi-attribute decision analysis process for optimal solution selection. The weight coefficients of each optimization objective are determined using the analytic hierarchy process (AHP), and the relative proximity of each candidate solution to the ideal solution is calculated using the TOPSIS method, yielding the optimal arrangement scheme with the highest comprehensive evaluation. Based on the optimal arrangement scheme, the LED module position coordinates are finely adjusted. A quadratic programming algorithm is used to fine-tune the initially determined LED module positions while satisfying physical constraints and optical performance requirements, resulting in finely adjusted LED module position parameters. Global consistency verification of the LED module position parameters is then performed. By calculating the statistical distribution of positional deviations, angular deviations, and spacing deviations among the LED modules, the consistency and stability of the overall arrangement scheme are ensured, yielding a spatial impulsivity comprehensive index verified by global consistency.

[0039] In one specific embodiment, the process of performing step S300 may specifically include the following steps: Based on the target location coordinates, LED module spatial layout planning is performed to determine the spatial layout parameters of each first LED module in the display frame. The spatial layout parameters are calculated to determine the alignment accuracy between modules and the splicing gap constraint, thus obtaining a global arrangement scheme. Based on the global arrangement scheme, the installation sequence and positioning reference of each first LED module are determined, and an LED module assembly process scheme is generated. According to the LED module assembly process, multiple first LED modules are precisely positioned and installed one by one, and each first LED module is fixed into the corresponding slot in the display frame to obtain the second LED module.

[0040] Specifically, based on the target location coordinate data, a spatial layout planning model matching the size of the display installation area is constructed. This model uses the target coordinate system as a reference, mapping each first LED module to the physical frame coordinate system of the screen. Combining the module's structural dimensions, tolerance specifications, and border constraints, a module layout sketch is generated through geometric mapping, spatial compression analysis, and array planning algorithms. This clarifies the placement coordinates and rotation angle of each module and preliminarily defines the module's row and column, adjacency relationship, and boundary distance, generating spatial layout parameters. Constraint analysis is then performed on these parameters, with alignment accuracy control and splicing gap control between modules as the core optimization objectives. A multi-objective error function is constructed to jointly constrain the X-axis alignment, Y-axis alignment, rotational parallelism, and module edge splicing gaps between modules. The calculation fully considers the module shell manufacturing tolerances, the thermal expansion effect of the structural border, and the guiding tolerance of electrical interfaces. Combined with the overall assembly requirements of the display screen, a constraint solving algorithm or layout optimization algorithm is selected to output a global arrangement scheme that satisfies maximum spatial consistency, minimum misalignment error, and assemblability. Based on the overall arrangement scheme, an LED module assembly process scheme suitable for batch assembly operations is developed. The process scheme clearly specifies the installation sequence of the modules, such as whether to prioritize by column or by row, and whether to prioritize the installation of edge modules or reference modules. The positioning reference for each module is set according to the arrangement sequence, such as using the lower left corner slot as the reference or the center hole as the positioning reference. At the same time, combined with the structure of the fixture platform, the arrangement of positioning pins and the size of the adsorption device, the key parameters such as the initial position, transportation path, descent posture and placement speed of the module during the assembly process are clarified. Corresponding detection and compensation mechanisms are developed according to the assembly error risk points to ensure that the positional deviation caused by environmental disturbances or the accumulation of module tolerances can be dynamically corrected during the placement process. Based on the above assembly process, the module installation operation is carried out step by step. The system or operators use a high-precision robotic arm, positioning platform or guide rail system to transport each first LED module to the designated installation slot in the display frame. Positioning pins, adsorption surfaces or locking mechanisms are used to ensure that the translation and rotation of the module in three-dimensional space meet the design requirements. After the module is positioned, it is firmly fixed in the frame by means of structural clamping, screw fixing or magnetic buckle. After several modules are assembled, machine vision or laser alignment system is used to perform local consistency verification to ensure that the positional error is not accumulated and spread during the entire assembly process, so as to achieve precise splicing of multiple first LED modules in the physical frame to form a second LED module.

[0041] The process, after obtaining the second LED module, includes multi-functional connection quality monitoring of the second LED module. This includes: real-time contact resistance detection using the conductivity characteristics of the connectors between LED modules in the second LED module; obtaining interface resistance data reflecting the contact quality of each connection point by applying a standard test current to the connection interface of each LED module and measuring the corresponding voltage drop change; establishing a correlation model between LED module connection strength and contact resistance based on the interface resistance data; obtaining resistance-strength correlation parameters for non-destructive quality prediction by statistically analyzing the correspondence between the resistance range and mechanical connection strength under normal connection conditions; and applying the resistance-strength correlation parameters to real-time monitoring of the LED module connection status, while simultaneously monitoring the drive of each LED module. Dynamic monitoring data of LED module connections is obtained by dynamically tracking and analyzing the stability of the connection interface through power-resistance dual-parameter analysis of power changes and connection interface resistance changes. Mechanical stress and resistance change correlation analysis is performed on the dynamic monitoring data of LED module connections. By comparing the stress-strain curves of the LED module under different mechanical loads with the corresponding resistance change curves, the stress distribution and deformation patterns of the connection interface are identified, and stress-resistance correlation characteristic parameters are obtained. Based on the stress-resistance correlation characteristic parameters, LED module connection failure early warning processing is performed. By establishing resistance change thresholds corresponding to different damage stages and realizing graded early warning of connection status, early fault detection signals are provided when connection strength decreases or contact is poor, resulting in LED module connection quality verification results.

[0042] In one specific embodiment, the process of performing step S400 may specifically include the following steps: The second LED module was powered on and driven. The voltage fluctuation of each LED module was monitored in real time by an adaptive peak voltage tracking controller to obtain voltage fluctuation data of each LED module. Dual-line frequency ripple detection and suppression processing are performed based on voltage fluctuation data to obtain driving voltage parameters; The light-emitting characteristics of the LED module are analyzed based on the driving voltage parameters, and the correspondence between the physical position of the second LED module and the light intensity distribution is established to obtain position-light intensity coupling relationship data. Global calibration is performed based on position-intensity coupling data to obtain target calibration parameters.

[0043] Specifically, after the second LED module is assembled and physically connected, the power drive system performs step-by-step power-on initialization of the entire screen and starts the voltage monitoring system composed of a high-precision ADC sampling unit and a signal conditioning module. At the same time, the adaptive peak voltage tracking controller is activated. During the actual operation of the module, the adaptive peak voltage tracking controller continuously scans the input voltage signal using a sliding window weighted analysis method, extracts the peak voltage data in each cycle or frame of the drive signal in real time, compares the peak voltage data with the reference voltage, records the voltage fluctuation amplitude and waveform change trend of each module in a specific time period, and outputs a voltage fluctuation data set including parameters such as timestamp, voltage rise edge, fall edge, peak value, and average value. Voltage fluctuation data is input into the ripple detection module. The ripple detection module performs Fourier transform or spectral expansion on the voltage signal using frequency domain analysis methods, and identifies the presence of ripple components with a dominant dual-line frequency of 50Hz or 100Hz. At the same time, it combines amplitude threshold and phase stability analysis to determine whether the ripple will interfere with the uniformity of light emission. If a significant ripple signal is detected, a notch filter mechanism or a voltage compensation algorithm is used to inject an inverse waveform into the control path to achieve real-time reduction or suppression of the ripple signal. A set of driving voltage parameters, including basic voltage parameters, dynamic compensation terms, and response time constants, is generated to describe the stable operating state of the module under ripple-free disturbance. The luminous characteristics of LED modules are analyzed based on driving voltage parameters. By conducting area-by-area lighting tests on the second LED module under stable control conditions, a high-sensitivity image acquisition device is used to sample the light intensity distribution of each module at high resolution. A mathematical mapping model is established based on the physical spatial location of the module, driving voltage parameters, and brightness values ​​of the corresponding pixel areas to form a one-to-one correspondence between physical location and light intensity response. This constructs LED position light intensity coupling relationship data, which reflects the optical uniformity of the module under rated drive and reveals the brightness reduction phenomenon in certain areas due to position deviation or voltage instability. The positional light intensity coupling data is input into the global calibration module. In the global calibration module, the brightness output, relative position, and orientation angle of all modules are jointly optimized through error distribution analysis, gradient compensation strategy, and brightness reconstruction algorithm. By constructing an error model between the target brightness field and the actual brightness field, iterative fitting or Bayesian optimization methods are used to dynamically adjust the driving voltage compensation value, PWM duty cycle, or reference brightness level of each module. In the process of multiple rounds of optimization, the brightness distribution of uneven areas is continuously corrected so that the final calibrated screen achieves high-quality optical performance such as consistent color temperature, uniform brightness, and no flicker or jitter on a macroscopic scale. The target calibration parameter set is output, including the driving correction factor, voltage feedback threshold, ripple suppression status, and corresponding spatial number in the display system for each module.

[0044] In one specific embodiment, the process of performing global calibration based on position-intensity coupling relationship data to obtain target calibration parameters can specifically include the following steps: Based on the position-light intensity coupling relationship data, a modified constitutive relation error function is constructed to calculate the calibration deviation and obtain the error evaluation index. By using the error evaluation index as the optimization target input, a probabilistic model of the calibration parameters is established through a Gaussian process. The correlation between the parameters is described by the Matérn kernel function, and the prior distribution of the objective function is established to obtain the probability distribution model. The calibration parameters are optimized iteratively based on the probability distribution model to obtain the initial calibration parameters. The initial calibration parameters are then weighted by the Fisher information matrix to obtain the target calibration parameters.

[0045] Specifically, using position-intensity coupling data as the basic input, a corrected constitutive error function is constructed to reflect the correspondence between calibration deviation and the degree of light intensity response deviation. In the process of constructing the corrected constitutive error function, the nonlinear response characteristics between the LED module's driving input, voltage modulation, position offset, and optical output are included in the modeling scope. The spatial response error is then calculated by weighting and accumulating the difference function, weighted smoothing term, and gradient regularization term to obtain an error evaluation index that reflects the impact of each module on the overall light field consistency. This index includes the root mean square deviation between the actual light intensity value and the expected light intensity distribution, the sensitivity coefficient of the driving voltage disturbance to brightness fluctuations, and the degree of spatial brightness gradient anomaly caused by uneven module response. The error evaluation index is used as the optimization objective function and input into the Gaussian process modeling framework to construct a probabilistic model for estimating the relationship between driving calibration parameters and spatial light intensity error. In the Gaussian process, a kernel function in the parameter space is defined to describe the correlation between any two combinations of calibration parameters. The Matérn kernel function is used as the structural kernel, which can characterize the influence of the distance between parameters on the output change of the response function in the feature space. This allows the Gaussian process to establish a prior probability distribution model of the influence of calibration parameters on light intensity error using limited sampled data, forming a probability distribution model jointly defined by the mean function and the covariance structure. Based on the probabilistic model, an optimal solution search for the objective function is performed. The next parameter combination most likely to improve calibration performance is selected by maximizing the expected improvement criterion, the upper confidence limit criterion, or other uncertainty-based acquisition functions. By iteratively updating the parameter sampling points, updating the model's predicted mean and confidence interval boundaries, the driving voltage, PWM duty cycle, or other adjustable control quantities are continuously optimized until the predicted minimum or maximum a posteriori point of the objective function converges to the global optimum or near-global optimum. This optimum is then used as the initial calibration parameter set. Meanwhile, to improve the stability and information effectiveness of the initial parameters, the initial calibration parameters are weighted. The sensitivity of each parameter is analyzed by introducing the Fisher information matrix as a weight adjustment factor. The Fisher information matrix is ​​constructed by calculating the expected square value of the first derivative of the objective function with respect to the control parameters, which characterizes the uncertainty density and information intensity of the model in each parameter direction. Based on this, the update step size in each parameter direction is adjusted differently, so that the parameters in the high confidence direction are corrected more significantly, while the parameters in the low confidence direction are fine-tuned. In this way, the parameter optimization process is guided to tend to stable convergence on the path with the minimum information entropy, resulting in a target calibration parameter set that includes dynamic learning rate, adaptive parameter weights, and accuracy enhancement mechanism.

[0046] Please see Figure 2 , Figure 2 A schematic block diagram of the structure of the LED module positioning, arrangement and calibration device 200 provided in the embodiments of this application, as shown below. Figure 2 As shown, the LED module positioning, arrangement, and calibration device 200 includes: The acquisition module 210 is used to acquire LED module mounting images and detect the pose data of multiple first LED modules; Calculation module 220 is used to calculate the target position coordinates of each first LED module based on the pose data; Assembly module 230 is used to assemble multiple first LED modules according to the target position coordinates to obtain a second LED module; The coupling analysis module 240 is used to perform position light intensity coupling analysis based on the second LED module and generate target calibration parameters.

[0047] Through the collaborative efforts of the aforementioned components, a multi-channel blind deconvolutional neural network was used to achieve parallel pose detection for N LED modules. Compared to the traditional method of detecting each module individually, this approach can process image information from multiple modules simultaneously, improving detection efficiency. Channel orthogonal constraint processing layers eliminate inter-channel interference, ensuring the accuracy and reliability of parallel multi-module detection. A spatial impulsivity index designed based on the optical characteristics of LED modules comprehensively evaluates three dimensions: luminous concentration, peak response, and edge sharpness, specifically addressing the unique optical problems in LED module positioning. Compared to general computer vision algorithms, this index is more suitable for the physical characteristics of LED modules, enabling more accurate position coordinate calculations. By constructing a global optimization model including position accuracy, alignment accuracy, gap control, and flatness control terms, collaborative positioning and arrangement of multiple modules was achieved. Compared to the traditional method of processing each module independently, global constraint optimization ensures overall positioning accuracy at the system level, avoiding the generation of accumulated errors. An adaptive peak voltage tracking controller was used instead of the traditional passive filtering method, actively tracking voltage peak changes and dynamically adjusting the PWM duty cycle. The 100Hz dual-line frequency ripple problem was solved at the voltage control source, achieving truly flicker-free drive. A mathematical coupling model of the physical position and light intensity distribution of the LED module was established, realizing the organic integration of position calibration and optical calibration. Compared with traditional simple geometric calibration methods, position-light intensity coupling analysis can comprehensively consider the influence of module position on optical performance, achieving optoelectronic integrated calibration.

[0048] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0049] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0050] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for positioning, arranging, and calibrating LED modules, characterized in that, include: Acquire LED module mounting images and detect the pose data of multiple first LED modules; Calculate the target position coordinates of each first LED module based on the pose data; The multiple first LED modules are assembled according to the target position coordinates to obtain a second LED module; Based on the position light intensity coupling analysis performed by the second LED module, target calibration parameters are generated.

2. The method for positioning, arranging, and calibrating LED modules according to claim 1, characterized in that, The process of acquiring LED module mounting images and detecting the pose data of multiple first LED modules includes: The mounting status of multiple first LED modules is simultaneously acquired by a machine vision system to obtain the original mounting image; The original mounting image is subjected to resolution normalization processing to obtain a standard mounting image; Pixel physical size calibration is performed based on the standard mounting image to obtain pixel coordinate calibration parameters; The pixel coordinate calibration parameters are applied to the standard mounting image for coordinate system unification processing to obtain the LED module mounting image. Parallel pose detection is performed on the LED module mounting image to obtain pose data of multiple first LED modules.

3. The method for positioning, arranging, and calibrating LED modules according to claim 2, characterized in that, The parallel pose detection of the LED module mounting image yields pose data for multiple first LED modules, including: The LED module mounting image is input into a multi-channel blind deconvolutional neural network for image region decomposition and feature extraction to obtain the edge texture feature map of each first LED module. Based on the edge texture feature map, pose regression calculation is performed to obtain the pose vector of each first LED module. The pose vectors are output and integrated in parallel to obtain pose data of multiple first LED modules.

4. The method for positioning, arranging, and calibrating LED modules according to claim 3, characterized in that, The step of inputting the LED module mounting image into a multi-channel blind deconvolutional neural network for image region decomposition and feature extraction to obtain the edge texture feature map of each first LED module includes: The LED module mounting image is input into the multi-channel signal separation layer of a multi-channel blind deconvolutional neural network to perform image channel decomposition, thereby obtaining the channel image data of each first LED module. The channel image data is input into the adaptive FIR filter bank layer of a multi-channel blind deconvolutional neural network for channel filtering to obtain multiple channel feature data. The channel feature data is input into the neural blind deconvolution feature extraction layer of the multi-channel neural blind deconvolution network for blind source separation and feature extraction, resulting in multiple separated feature maps. The separated feature maps are input into the channel orthogonality constraint processing layer of the multi-channel neural blind deconvolution network to perform inter-channel interference suppression processing, resulting in multiple channel feature maps. The channel feature maps are input into the multi-channel pose fusion output layer of the multi-channel neural blind deconvolution network for parallel feature integration to obtain the edge texture feature maps of each first LED module.

5. The method for positioning, arranging, and calibrating LED modules according to claim 1, characterized in that, The calculation of the target position coordinates of each first LED module based on the pose data includes: Calculate the spatial pulse characteristics comprehensive index of each first LED module based on the pose data; Based on the aforementioned spatial impulsivity comprehensive index, the spatial impulsivity index value is gradually maximized through gradient descent algorithm and momentum term adjustment to obtain the target position parameters of each first LED module; Multi-scale feature fusion is performed on the target position parameters to obtain the target position coordinates of each first LED module.

6. The method for positioning, arranging, and calibrating LED modules according to claim 5, characterized in that, The calculation of the spatial impulsivity comprehensive index of each first LED module based on the pose data includes: Based on the pose data, the fourth-order cumulant kurtosis function value is calculated to detect the spatial concentration of the light-emitting area of ​​each first LED module, the peak detection function value is used to identify the maximum light intensity response value of each first LED module, and the sharpness function value is used to evaluate the edge sharpness of each first LED module. The spatial concentration, the maximum light intensity response value, and the edge sharpness are weighted to obtain the weighted concentration component, weighted peak component, and weighted sharpness component of each first LED module. The weighted concentration component, weighted peak component, and weighted sharpness component are summed to obtain the single-module spatial pulseness index of each first LED module; The spatial pulse characteristics of each first LED module are globally summed to obtain the comprehensive spatial pulse characteristics of each first LED module.

7. The method for positioning, arranging, and calibrating LED modules according to claim 1, characterized in that, The assembly of the plurality of first LED modules according to the target position coordinates to obtain a second LED module includes: Based on the target location coordinates, LED module spatial layout planning is performed to determine the spatial layout parameters of each first LED module in the display screen frame. The spatial layout parameters are used to calculate the alignment accuracy between modules and the splicing gap constraint to obtain a global arrangement scheme. Based on the global arrangement scheme, the installation sequence and positioning reference of each first LED module are determined, and an LED module assembly process scheme is generated. According to the LED module assembly process, multiple first LED modules are precisely positioned and installed one by one, and each first LED module is fixed into the corresponding slot in the display frame to obtain the second LED module.

8. The method for positioning, arranging, and calibrating LED modules according to claim 1, characterized in that, The step of generating target calibration parameters based on the position light intensity coupling analysis performed on the second LED module includes: The second LED module was tested for power-on drive. The voltage fluctuation of each LED module was monitored in real time by an adaptive peak voltage tracking controller to obtain voltage fluctuation data of each LED module. Based on the voltage fluctuation data, dual-line frequency ripple detection and suppression processing are performed to obtain the driving voltage parameters; Based on the driving voltage parameters, the light-emitting characteristics of the LED module are analyzed and the correspondence between the physical position of the second LED module and the light intensity distribution is established to obtain position-light intensity coupling relationship data. Global calibration is performed based on the location-intensity coupling relationship data to obtain the target calibration parameters.

9. The method for positioning, arranging, and calibrating LED modules according to claim 8, characterized in that, The global calibration based on the position-intensity coupling relationship data, to obtain target calibration parameters, includes: Based on the positional light intensity coupling relationship data, a modified constitutive relation error function is constructed to calculate the calibration deviation and obtain the error evaluation index. The error evaluation index is used as the optimization target input Gaussian process to establish a probabilistic model of calibration parameters. The correlation between parameters is described by the Matérn kernel function and the prior distribution of the objective function is established to obtain the probability distribution model. Based on the probability distribution model, the calibration parameters are optimized iteratively to obtain the initial calibration parameters. The initial calibration parameters are then weighted by the Fisher information matrix to obtain the target calibration parameters.

10. A positioning, arranging, and calibration device for LED modules, characterized in that, A method for performing the positioning, arrangement, and calibration of LED modules as described in any one of claims 1-9, comprising: The acquisition module is used to acquire LED module mounting images and detect the pose data of multiple first LED modules; The calculation module is used to calculate the target position coordinates of each first LED module based on the pose data; An assembly module is used to assemble the plurality of first LED modules according to the target position coordinates to obtain a second LED module; The coupling analysis module is used to perform position light intensity coupling analysis based on the second LED module and generate target calibration parameters.

Images