Industrial vision detection and positioning guide integrated dynamic calibration adaptation system

Abstract

The application relates to the technical field of image recognition and understanding, and discloses a dynamic calibration and adaptation system for industrial visual detection and positioning guidance integration, which comprises an imaging and collecting module for collecting original digital images; a feature re-anchor module for extracting semantic anchor points and calculating a drift tensor; a dynamic calibration and adaptation module for generating a space modulation mask based on the drift tensor, according to which the original digital images are sharpened and their contrast is modulated, and a parameterized three-dimensional grid is driven to generate geometric topological deformation to reconstruct a virtual space positioning reference; and an execution guidance module for outputting a pose parameter according to the positioning reference. The application converges physical environment interference into high-frequency information repair in the digital domain of images, eliminates negative conduction of mechanical hysteresis on calibration accuracy, strengthens the supporting force of bottom image quality on high-level semantic understanding, and enhances the anti-shock stability of the system in a dynamic scene.

Description

Technical Field

[0001] This invention belongs to the field of image recognition and understanding technology, and in particular relates to a dynamic calibration and adaptation system that integrates industrial visual inspection and positioning guidance. Background Technology

[0002] Currently, in industrial automated production lines, vision inspection systems typically use hand-eye calibration to establish a linear mapping relationship between the pixel coordinate system and the spatial coordinate system of the actuator, and compensate for system drift by periodically updating the coordinate transformation matrix. Existing matrix compensation methods are based on rigid body kinematics models in physical space, assuming that the detection environment and material carrier maintain absolute geometric stability during movement. However, under the combined effects of high-frequency micro-vibrations, thermal expansion caused by temperature drift, and nonlinear displacement of the transmission mechanism, the detection environment produces local topological distortions. Since the discretized mathematical matrix is ​​difficult to approximate nonlinear physical deformations, and the analytical process of matrix parameters has a time-domain response delay, the calibration parameters fail, causing positioning guidance deviations.

[0003] Besides the physical constraints at the hardware level, there are also limitations at the software control level. For example, Chinese invention patent application CN120563630A discloses an online parameter calibration method and system for a dynamic measurement vision system in industrial settings. This method acquires multiple calibration plate images through a two-dimensional rotating mechanism and calculates the pose parameters of the mounting base based on the mapping relationship between pixel coordinates and physical coordinates. This scheme heavily relies on the premise of rigid body transformation, assuming that the relative displacement between the camera, turntable, and base follows linear rotation and translation. In actual working conditions, high-frequency excitation leads to minute mechanical deformation or non-uniform thermal stress causing creep in the supporting structure, which is a non-linear, non-rigid topological evolution. The LM iterative method used in the proposed solution only solves the macroscopic pose matrix, which cannot provide a refined representation of the topological folds generated in the local physical space. Furthermore, it does not consider the weakening effect of dynamic interference such as motion blur on the accuracy of low-level feature extraction. The degradation of the low-level image quality limits the reliability of high-level semantic understanding. In addition, the dynamic interference in the production line not only causes spatial displacement but also weakens the exposure features in the image acquisition process, resulting in motion blur and reduced edge contrast. Traditional visual calibration systems separate the low-level image quality processing from the physical space compensation logic, attempting to extract semantic features from degraded images, which reduces the recognition accuracy in dynamic scenes and may even trigger feature extraction lockout.

[0004] Therefore, how to establish an adaptive mechanism for synchronizing image features with spatial references in the face of dynamic environmental distortion, and to achieve deep collaborative linkage between detection and localization guidance, is the technical problem to be solved by this invention. Summary of the Invention

[0005] This invention provides a dynamic calibration and adaptation system integrating industrial visual inspection and positioning guidance, comprising: The imaging acquisition module is used to acquire raw digital images that characterize the operating state of physical objects; The feature re-anchoring module is used to identify geometrically constrained edge intersections, circle center feature points, and texture abrupt change points in the original digital image using a semantic prior model to determine multiple semantic anchors, and calculate the displacement vector of the semantic anchors relative to the preset reference position. By performing energy minimization fitting on the displacement vector, a drift tensor representing the nonlinear degradation trend of the image domain is generated. The dynamic calibration and adaptation module maps the drift tensor to a pixel spatial distribution probability map to identify local feature degradation regions affected by high-frequency motion features in dynamic scenes. Based on the drift tensor, it generates a spatial modulation mask corresponding to the local feature degradation region. Within the feature search domain defined by the spatial modulation mask, it performs pixel-level sharpening on the original digital image by dynamically adjusting the convolution kernel weight matrix of the Laplacian operator, and performs contrast gain modulation through local gray-level histogram reconstruction to reconstruct high-level semantic features in the original digital image. Simultaneously, the dynamic calibration and adaptation module uses the drift tensor to drive the parametric 3D mesh to generate geometric topological deformation. By aligning the deformed parametric 3D mesh with the sharpened semantic anchor points in the feature domain spatial domain, it reconstructs the virtual spatial positioning benchmark of the physical object. The execution guidance module is used to calculate the six-degree-of-freedom pose parameters of the physical object based on the reconstructed virtual space positioning reference, and encapsulate the six-degree-of-freedom pose parameters into a control data packet that can be recognized by the external execution terminal for output.

[0006] Preferably, when the feature re-anchoring module identifies semantic anchor points, it extracts high-frequency gradient change regions in the original digital image and establishes geometric constraints based on the response consistency of edge intersections under different illumination components.

[0007] Preferably, when the feature re-anchoring module calculates the drift tensor, it performs displacement vector fitting of each semantic anchor point between adjacent image frames to obtain the drift tensor. The calculation formula is: ,in, Here, n is the drift tensor; n is the total number of semantic anchors. Contribute weights to the features of the i-th semantic anchor point; The spatial coordinates of the semantic anchor point at the current moment; The preset reference coordinates for semantic anchor points stored during the static initialization phase of the system.

[0008] Preferably, when the dynamic calibration and adaptation module generates the spatial modulation mask, it establishes an inverse mapping relationship between the magnitude of the drift tensor and the pixel response intensity of the local feature degradation region, and uses the spatial modulation mask to perform weight gain allocation on the pixel region with a high degree of motion blur in the original digital image.

[0009] Preferably, when the dynamic calibration and adaptation module performs pixel-level sharpening using the spatial modulation mask, it real-time corrects the coefficients of the weight matrix based on the weight distribution in the spatial modulation mask to enhance the edge gradient in the local feature degradation region.

[0010] Preferably, when the dynamic calibration and adaptation module performs contrast gain modulation, it uses a spatial modulation mask to extract the brightness deviation component of the original digital image and performs adaptive histogram equalization on the original digital image.

[0011] Preferably, when the dynamic calibration and adaptation module generates geometric topological deformation, a linear offset model between the drift tensor and the vertex coordinates of the parameterized 3D mesh is established to keep the topological connectivity properties of the parameterized 3D mesh constant, and the displacement of the vertex coordinates is updated using the drift tensor.

[0012] Preferably, when the dynamic calibration and adaptation module reconstructs the virtual space positioning reference, it calculates the spatial overlap between the deformed parametric 3D mesh and the sharpened semantic anchor points, and determines the coordinate transformation matrix for real-time calibration through maximum likelihood estimation.

[0013] Preferably, the system also includes a closed-loop feedback module for calculating the control accuracy residual of the six-degree-of-freedom pose parameters and transmitting the control accuracy residual back to the feature re-anchoring module within a period of 10ms to 20ms.

[0014] Preferably, when the execution guidance module outputs control data packets, it converts the six-degree-of-freedom pose parameters into register control signals for the external execution terminal.

[0015] Compared with existing technologies, the dynamic calibration and adaptation system integrating industrial vision inspection and positioning guidance of this invention has the following advantages: 1. In industrial vision inspection and positioning, the system establishes a connection between a parameterized 3D mesh and the physical inspection environment, transforming the traditional physical coordinate transformation matrix update method into geometric topology reconstruction. When the production line environment produces nonlinear displacement or physical distortion, the system uses the drift vector of image features to drive the virtual mesh to generate synchronous deformation, eliminating the computational lag and accumulated error of matrix analysis when dealing with non-rigid deformation, and obtaining a stable spatial positioning reference.

[0016] 2. By generating a spatial modulation mask through semantic anchor point drift tensor, targeted pixel sharpening and contrast enhancement are achieved for local areas of the image affected by high-frequency micro-vibrations on the production line. This processing mechanism counteracts motion blur and grayscale attenuation at the low-level feature generation stage of the image, improves the image quality's support for high-level semantic understanding, and avoids the problem of semantic feature extraction lock-out due to physical environmental interference.

[0017] 3. By combining feature re-anchoring and energy minimization of grid deformation, a closed-loop feedback between the image feature domain and the spatial positioning domain is achieved. The image enhancement effect generated by local pixel reconstruction supports the stable extraction of semantic features. At the same time, the virtual grid after deformation reconstruction provides accurate coordinate mapping. This multi-dimensional collaborative mechanism enhances the seismic robustness of the system under complex working conditions and reduces the dependence on external high-frequency mechanical calibration. Attached Figure Description

[0018] Figure 1 This is the overall architecture and closed-loop data flow diagram of the dynamic calibration and adaptation system of this invention; Figure 2 This is a functional node decomposition diagram of each core module of the dynamic calibration and adaptation system of this invention. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0020] It should be noted that all directional and positional terms used in this invention, such as: up, down, left, right, front, back, vertical, horizontal, inner, outer, top, bottom, transverse, longitudinal, center, etc., are only used to explain the relative positional relationship and connection between components in a specific state (as shown in the accompanying drawings). They are only for the convenience of describing this invention and do not require that this invention be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention. In addition, the descriptions of "first," "second," etc., in this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated.

[0021] In the description of this invention, unless otherwise explicitly specified and limited, the terms installation, connection, and linking should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections; they can refer to direct connections or indirect connections through an intermediate medium; they can refer to the internal connection of two components. For those skilled in the art, the specific meaning of the above terms in this invention can be understood according to the specific circumstances.

[0022] In the description of this specification, references to the terms "an embodiment," "some embodiments," "illustrative embodiments," "examples," "specific examples," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example, and the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0023] A dynamic calibration and adaptation system integrating industrial vision inspection and positioning guidance, comprising: The imaging acquisition module is used to acquire raw digital images that characterize the operating state of physical objects; The feature re-anchoring module is used to identify geometrically constrained edge intersections, circle center feature points, and texture abrupt change points in the original digital image using a semantic prior model to determine multiple semantic anchors, and calculate the displacement vector of the semantic anchors relative to the preset reference position. By performing energy minimization fitting on the displacement vector, a drift tensor representing the nonlinear degradation trend of the image domain is generated. The dynamic calibration and adaptation module maps the drift tensor to a pixel spatial distribution probability map to identify local feature degradation regions affected by high-frequency motion features in dynamic scenes. Based on the drift tensor, it generates a spatial modulation mask corresponding to the local feature degradation region. Within the feature search domain defined by the spatial modulation mask, it performs pixel-level sharpening on the original digital image by dynamically adjusting the convolution kernel weight matrix of the Laplacian operator, and performs contrast gain modulation through local gray-level histogram reconstruction to reconstruct high-level semantic features in the original digital image. Simultaneously, the dynamic calibration and adaptation module uses the drift tensor to drive the parametric 3D mesh to generate geometric topological deformation. By aligning the deformed parametric 3D mesh with the sharpened semantic anchor points in the feature domain spatial domain, it reconstructs the virtual spatial positioning benchmark of the physical object. The execution guidance module is used to calculate the six-degree-of-freedom pose parameters of the physical object based on the reconstructed virtual space positioning reference, and encapsulate the six-degree-of-freedom pose parameters into a control data packet that can be recognized by the external execution terminal for output.

[0024] Preferably, when the feature re-anchoring module identifies semantic anchor points, it extracts high-frequency gradient change regions in the original digital image and establishes geometric constraints based on the response consistency of edge intersections under different illumination components.

[0025] Preferably, when the feature re-anchoring module calculates the drift tensor, it performs displacement vector fitting of each semantic anchor point between adjacent image frames to obtain the drift tensor. The calculation formula is: ,in, Here, n is the drift tensor; n is the total number of semantic anchors. Contribute weights to the features of the i-th semantic anchor point; The spatial coordinates of the semantic anchor point at the current moment; The preset reference coordinates for semantic anchor points stored during the static initialization phase of the system.

[0026] Preferably, when the dynamic calibration and adaptation module generates the spatial modulation mask, it establishes an inverse mapping relationship between the magnitude of the drift tensor and the pixel response intensity of the local feature degradation region, and uses the spatial modulation mask to perform weight gain allocation on the pixel region with a high degree of motion blur in the original digital image.

[0027] Preferably, when the dynamic calibration and adaptation module performs pixel-level sharpening using the spatial modulation mask, it real-time corrects the coefficients of the weight matrix based on the weight distribution in the spatial modulation mask to enhance the edge gradient in the local feature degradation region.

[0028] Preferably, when the dynamic calibration and adaptation module performs contrast gain modulation, it uses a spatial modulation mask to extract the brightness deviation component of the original digital image and performs adaptive histogram equalization on the original digital image.

[0029] Preferably, when the dynamic calibration and adaptation module generates geometric topological deformation, a linear offset model between the drift tensor and the vertex coordinates of the parameterized 3D mesh is established to keep the topological connectivity properties of the parameterized 3D mesh constant, and the displacement of the vertex coordinates is updated using the drift tensor.

[0030] Preferably, when the dynamic calibration and adaptation module reconstructs the virtual space positioning reference, it calculates the spatial overlap between the deformed parametric 3D mesh and the sharpened semantic anchor points, and determines the coordinate transformation matrix for real-time calibration through maximum likelihood estimation.

[0031] Preferably, the system also includes a closed-loop feedback module for calculating the control accuracy residual of the six-degree-of-freedom pose parameters and transmitting the control accuracy residual back to the feature re-anchoring module within a period of 10ms to 20ms.

[0032] Preferably, when the execution guidance module outputs control data packets, it converts the six-degree-of-freedom pose parameters into register control signals for the external execution terminal.

[0033] Example 1: When a dynamic inspection scenario on an industrial precision assembly line faces high-frequency mechanical micro-vibration and continuous thermodynamic temperature drift, the nonlinear topological distortion of the physical space causes nonlinear degradation of local exposure features in the original digital image. The imaging acquisition module acquires the original digital image representing the operating state of the physical object. The feature re-anchoring module establishes geometric constraints based on the response consistency of edge intersections under different illumination components, extracts high-frequency gradient change regions in the original digital image, and identifies edge intersections, center feature points, and texture abrupt change points as semantic anchors. The feature re-anchoring module fits the displacement vector of each semantic anchor between adjacent image frames to generate a drift tensor representing the nonlinear degradation trend of the image domain. The calculation formula for the drift tensor is as follows: ,in, Here, n is the drift tensor; n is the total number of semantic anchors. Contribute weights to the features of the i-th semantic anchor point; The spatial coordinates of the semantic anchor point at the current moment; The preset reference coordinates for semantic anchor points stored during the static initialization phase of the system.

[0034] The dynamic calibration and adaptation module establishes an inverse mapping relationship between the magnitude of the drift tensor and the pixel response intensity of local feature degradation regions affected by high-frequency motion features. The module maps the drift tensor to a pixel spatial distribution probability map, identifies local feature degradation regions, and generates spatial modulation masks corresponding to these regions based on the drift tensor. The module uses an anisotropic Gaussian kernel function to diffuse the drift tensor. Using the current spatial coordinates of each semantic anchor point Centered on the drift tensor The modulus is determined as the variance parameter of the Gaussian kernel function. A two-dimensional probability distribution field covering the original digital image is constructed, and the probability density of overlapping regions in the image is superimposed. The normalized result is a spatial distribution matrix with values ​​in the range of 0 to 1, which serves as a spatial modulation mask. The displacement trend of discrete anchor points is converted into pixel-level response intensity weights, providing a quantitative basis for the subsequent dynamic correction of the Laplacian operator weight matrix coefficients. Using the spatial modulation mask, the dynamic calibration and adaptation module assigns weight gain to pixel regions with high motion blur in the original digital image. Within the feature search domain defined by the spatial modulation mask, the dynamic calibration and adaptation module corrects the coefficients of the Laplacian operator's convolution kernel weight matrix in real time according to the weight distribution in the spatial modulation mask, sharpening the original digital image and enhancing the edge gradient in the local feature degradation region. At the same time, the dynamic calibration and adaptation module uses the spatial modulation mask to extract the brightness deviation component of the original digital image, modulates the contrast gain of the original digital image through adaptive histogram equalization, and reconstructs the high-level semantic features in the original digital image.

[0035] To address the mechanical lag in coordinate analysis of dynamic scenes and the physical constraints of underlying image degradation, the system transforms dynamic environmental distortion into a process of high-frequency information restoration in the digital domain and topological reconstruction in the virtual space. The dynamic calibration and adaptation module establishes a linear offset model between the drift tensor and the vertex coordinates of the parameterized 3D mesh. While maintaining the constant topological connectivity of the parameterized 3D mesh, the module uses the drift tensor to update the displacement of the vertex coordinates, driving geometric topological deformation of the parameterized 3D mesh. The module then uses an inverse distance weighting algorithm to adjust the drift tensor... The data is transferred to a parameterized 3D mesh, and the vertices adjacent to the semantic anchors are identified. The displacement offset of each vertex is calculated based on distance weights, driving the parameterized 3D mesh to undergo topological deformation synchronized with the physical environment distortion. The system uses the least squares method to fit the spatial alignment relationship between the deformed mesh topological nodes and the sharpened semantic anchors. A reprojection residual function is constructed and Levenberg-Marquardt iterative optimization is performed to calculate the real-time coordinate transformation matrix representing the physical object's pose. The system outputs six-DOF pose parameters to eliminate nonlinear measurement biases caused by high-frequency micro-vibrations. The dynamic calibration and adaptation module calculates the spatial overlap between the deformed parameterized 3D mesh and the sharpened and contrast-gain-modulated semantic anchors using maximum likelihood estimation, determining the real-time calibration... The system employs a precise coordinate transformation matrix. A dynamic calibration and adaptation module aligns the deformed parametric 3D mesh with semantic anchor points possessing high edge gradients, reconstructing a virtual spatial positioning reference for the physical object. The execution guidance module calculates the six-degree-of-freedom pose parameters of the physical object based on this reconstructed virtual spatial positioning reference. A closed-loop feedback module calculates the control accuracy residuals of the six-degree-of-freedom pose parameters and transmits these residuals back to the feature re-anchoring module within a 10ms to 20ms period. The execution guidance module converts the six-degree-of-freedom pose parameters into register control signals for the external execution terminal, encapsulating them into a control data packet recognizable by the external execution terminal. The system maintains the consistency of the continuous output of the physical object's pose parameters through the collaborative constraints of pixel-domain restoration and computational domain deformation.

[0036] Example 2: When dynamic inspection scenarios on industrial precision assembly lines face high-frequency mechanical micro-vibrations and continuous thermodynamic temperature drift, nonlinear topological distortion in the physical space causes nonlinear degradation of local exposure features in the original digital image. A visual calibration and verification platform is constructed, including an imaging acquisition module with a sampling frequency of 200 frames per second and a resolution of 5 megapixels, a three-axis mechanical vibration table providing a continuously adjustable excitation frequency from 10Hz to 60Hz, and an interference generator that injects Gaussian white noise with a signal-to-noise ratio of 30dB into the image transmission link. The sampling period of the feature re-anchoring module is established as 5ms. The technical considerations for establishing this sampling period are as follows: To balance feature tracking continuity with image processor computational load, the system allocates a 16-frame-deep sliding circular buffer in the memory control unit to store raw digital images output by the imaging acquisition module in real time. The feature re-anchoring module triggers a computation process every 2.5ms, extracting the most recent 8 frames from the buffer to form an analysis time window, ensuring a 50% data overlap in the time domain between adjacent computation tasks. Through this overlapping sampling mechanism, the system can utilize feature consistency between adjacent frames to eliminate pseudo-anchor point jumps caused by high-frequency electromagnetic noise, ensuring sub-pixel accuracy in the drift tensor calculation results. The smoothness of the image is improved; when the main frequency bandwidth of mechanical micro-vibration approaches half of the camera frame rate, the longer sampling period causes cross-pixel aliasing at the edge intersections of adjacent image frames; to avoid high-frequency signal aliasing under the Nyquist sampling theorem, the sampling period is set to the lower limit of its range of 5ms to capture the displacement vector of semantic anchor points, and four independent verification test groups are set; the first control group uses global coordinate transformation matrix update logic; the second control group uses drift tensor to generate spatial modulation mask and corrects the Laplacian operator convolution kernel weights, and removes the adaptive histogram equalization operation; the third control group sets the effective threshold of spatial modulation mask. The lower limit is below 0.1; the sample group of this invention uses a feature re-anchoring module and a dynamic calibration and adaptation module, setting three problem intensity gradients of 10Hz, 30Hz and 50Hz; a three-axis mechanical vibration table is started and electromagnetic noise is injected, and the imaging acquisition module acquires the original digital image; the feature re-anchoring module extracts the high-frequency gradient change region, establishes geometric constraints based on the response consistency of the edge intersection under different illumination components, and determines multiple semantic anchor points; the feature re-anchoring module fits the displacement vector of each semantic anchor point between adjacent image frames, and calculates the drift tensor representing the nonlinear degradation trend of the image domain according to the formula; the calculation formula of the drift tensor is as follows: ,in, Here, n is the drift tensor; n is the total number of semantic anchors. Contribute weights to the features of the i-th semantic anchor point; The spatial coordinates of the semantic anchor point at the current moment; The preset reference coordinates for semantic anchor points stored during the static initialization phase of the system.

[0037] The dynamic calibration and adaptation module maps the drift tensor to the pixel spatial distribution probability map; it extracts the mean edge gradient during the monitoring process as an intermediate analysis indicator; under 30Hz excitation conditions, the mean edge gradient of the original digital image input to the feature re-anchoring module decays to 15.4; the first control group did not identify local feature degradation areas, and its mean edge gradient remained at 16.2; the mean edge gradient of the second control group climbed to 42.7, and the brightness deviation component in the local feature degradation area masked the texture abrupt change point; the sample group of this invention uses a spatial modulation mask to assign weight gain to the original digital image, corrects the convolution kernel weight matrix of the Laplacian operator in real time, sharpens pixels, and modulates the contrast gain through adaptive histogram equalization, and its mean edge gradient reaches 78.6; the third control group, due to the low effective threshold setting, misidentifies and over-sharpenes white noise in non-degraded areas, causing the local image to exhibit oversaturation distortion and drastic fluctuations in the mean gradient.

[0038] The reprojection error of the vertex coordinates of the parameterized 3D mesh is extracted as the output evaluation data of the virtual space positioning reference. As the excitation frequency gradually increases from 10Hz to 30Hz and then to 50Hz, the measured reprojection errors of the sample group of this invention are 0.42 pixels, 0.58 pixels, and 0.85 pixels, respectively. The reprojection error of the first control group surges to 6.74 pixels at 30Hz and loses feature point tracking at 50Hz. The data evolution shows a nonlinear performance inflection point when the excitation frequency reaches 45Hz. When the excitation frequency exceeds 45Hz, the physical exposure time of the imaging acquisition module is limited, which reduces the number of photons integrated. The slope of the reprojection error of the sample group of this invention increases, reaching 1.42 pixels at 55Hz. In the frequency band that crosses this performance inflection point, the dynamic calibration and adaptation module adjusts according to spatial modulation. The real-time correction mechanism of the mask to the weight matrix of the Laplacian operator maintains the reprojection error at the sub-pixel level. The reprojection error output by the complete mechanism of the present invention is lower than the 2.86 pixels output by the second control group under the same working conditions after stripping the contrast gain modulation. The synergistic effect of pixel domain sharpening and contrast modulation in generating data mapping in geometric topological deformation reconstruction, based on the feature repair of drift tensor and spatial modulation mask and parameterized three-dimensional mesh deformation mechanism, changes the linear degradation correlation between forced vibration and calibration accuracy. When the system faces the superposition of high-frequency topological distortion and environmental noise in physical space, the dynamic calibration and adaptation module reconstructs the virtual space positioning reference based on the information of the image digital domain, controls the control accuracy residual of the six degrees of freedom pose parameters to converge to the threshold range, and cuts off the negative transmission chain generated by mechanical transmission lag to calibration accuracy.

[0039] Example 3: When the industrial vision inspection system is under conditions of light source flicker and non-uniform dynamic defocus, there are quantization deviations in the generation parameters of the spatial modulation mask and the weights of the Laplacian operator convolution kernel; the imaging acquisition module acquires the original digital image; the feature re-anchoring module calculates the feature contribution weight of the semantic anchor point; this feature contribution weight is directly proportional to the local gradient magnitude of the image neighborhood where the semantic anchor point is located; the feature re-anchoring module extracts a 5×5 pixel matrix centered on the semantic anchor point, calculates the average edge gradient of the pixel matrix using the Sobel operator, and sets the ratio of the average edge gradient to the maximum global gradient as the feature contribution weight; the dynamic calibration and adaptation module adjusts the feature contribution weight accordingly. The drift tensor fitted with the displacement vector is used to construct a pixel spatial distribution probability map. The dynamic calibration and adaptation module constructs a two-dimensional Gaussian probability density distribution function centered on the current spatial coordinates of each semantic anchor point, and sets the magnitude of the drift tensor as the variance parameter of this two-dimensional Gaussian probability density distribution function. The dynamic calibration and adaptation module superimposes the probability density distributions corresponding to all semantic anchor points to generate a normalized spatial modulation mask with values ​​between 0 and 1. Within the feature search domain defined by the spatial modulation mask, the dynamic calibration and adaptation module adjusts the convolution kernel weight matrix of the Laplacian operator point-by-point according to the local pixel weights of the spatial modulation mask. The specific formula for calculating the center weight coefficient of the Laplacian operator is as follows: ,in, These are the corrected central weight coefficients of the Laplace operator; The basic sharpening weight constant is determined during the static calibration phase of the system; k is the dimensionless contrast gain adjustment coefficient. This represents the normalized probability values ​​of the spatial modulation mask at pixel coordinates x and y.

[0040] The dynamic calibration and adaptation module updates the convolution kernel weight matrix of the Laplacian operator based on the center weight coefficients of the Laplacian operator, outputting a pixel-level sharpening matrix for the original digital image. The module also extracts pixel clusters with a probability value greater than 0.5 from the spatial modulation mask as brightness deviation components, calculates the local gray-level histogram of these pixel clusters, reconstructs the gray-level distribution range using a contrast-limited adaptive histogram equalization algorithm, and modulates the contrast gain of the original digital image. In the virtual space reconstruction stage, the module selects the grid vertex in the parametric 3D mesh that has the closest Euclidean distance to the semantic anchor point as the deformation control node. The calibration and adaptation module uses an inverse distance weighted interpolation algorithm to distribute the drift tensor to each deformation control node, driving the mesh vertex coordinates to generate corresponding topological displacements. The execution guidance module reconstructs the virtual space positioning reference based on the parameterized 3D mesh updated with topological displacements. The Laplacian operator correction method based on the spatial modulation mask probability distribution makes the image sharpening intensity and contrast gain controlled by the local nonlinear degradation parameters. The closed-loop feedback module extracts the control accuracy residual based on the geometric topological deformation features. The system maintains the continuous output of the six degrees of freedom pose parameters of the physical object based on the pixel domain repair and computational domain deformation linkage mechanism.

[0041] Example 4: When the system faces a new camera deployment scenario, the feature re-anchoring module establishes the initial reference domain for coordinate calculation according to the static extraction procedure; in a static state, it controls the imaging acquisition module to continuously acquire 100 frames of static raw digital images; in a static state, it continuously acquires M frames of static raw digital images, and the feature re-anchoring module extracts the candidate point set from the static raw digital images, calculates the variance of coordinate fluctuation of each candidate point in the sequence frames, eliminates unstable points with a variance greater than the preset sensor floor noise threshold, and stores the arithmetic mean of the coordinates of the remaining candidate points in the storage unit to determine the preset reference coordinates. M is a positive integer. The dynamic calibration and adaptation module extracts the global gradient maxima of the static original digital image and determines the basic sharpening weight constants based on the signal-to-noise ratio mapping model. To provide an initial benchmark for drift calculation and pixel-level sharpening correction in dynamic detection mode, the feature re-anchoring module identifies a set of candidate points containing edge intersections in the static original digital image and calculates the variance of pixel coordinate fluctuations of each candidate point in the image sequence. The feature re-anchoring module eliminates unstable points with variance fluctuations greater than the sensor's floor noise and determines the remaining intersections as semantic anchors. The feature re-anchoring module calculates the arithmetic mean of the spatial coordinates of the semantic anchor in the image sequence and uses it as the preset benchmark coordinates. Write to the storage unit; the dynamic calibration and adaptation module extracts the global gradient maximum value of the static original digital image and calculates the basic sharpening weight constant based on the signal-to-noise ratio mapping relationship. And store it in the underlying register.

[0042] After switching to the dynamic detection mode that includes high-frequency mechanical micro-vibrations, the system calls the preset reference coordinates stored in the storage unit. With the basic sharpening weight constant As a basic reference value, the feature re-anchoring module calculates the spatial coordinates of the semantic anchor point at the current moment. With preset reference coordinates Spatial displacement deviation is used to generate a drift tensor; the dynamic calibration and adaptation module constructs a spatial modulation mask based on the drift tensor; within the feature search domain defined by the spatial modulation mask, the dynamic calibration and adaptation module applies the basic sharpening weight constants. The system is superimposed with dynamically adjusted components to correct the convolution kernel weight matrix of the Laplacian operator; the execution guidance module receives a pixel-level sharpened image containing reconstructed semantic features and updates the virtual space positioning reference; the system uses the transmission link from static reference calibration to dynamic feedback correction to suppress the image coordinate divergence state and maintain the continuous closed-loop output of the six degrees of freedom pose parameters of the physical object.

[0043] Example 5: When the industrial vision inspection system initializes and constructs a parametric 3D mesh, the lack of quantified mesh vertex density setting criteria and environmental adaptability debugging procedures leads to deviations in the accuracy of reconstructing the virtual space positioning benchmark. During the pre-calibration phase of on-site deployment, the system initiates the initial density calibration process of the parametric 3D mesh. The control unit drives the imaging acquisition module to acquire a high-resolution benchmark image of the calibration target under static conditions without mechanical excitation. The feature re-anchoring module extracts all physical edge intersections in the benchmark image, calculates the average pixel distance between adjacent physical edge intersections, and determines it as the physical resolution limit. To ensure that the topological nodes of the parameterized 3D mesh can map the minimal deformation of physical space, while suppressing dimensional overload of matrix operations during geometric transformations, the initial mesh edge length of the parameterized 3D mesh is determined. According to the formula Set; where α is the grid resolution adaptation constant.

[0044] After establishing the initial parametric 3D mesh, the system executes an adaptive excitation debugging procedure to quantify the limit boundary of the dynamic distortion of the environment. The system starts a three-axis mechanical vibration table, linearly increasing the excitation frequency from 10Hz to the upper limit of the system's mechanical resonance frequency. During the frequency increase, the feature re-anchoring module tracks the displacement vector of the semantic anchor point between adjacent image frames and records the peak value sequence of the displacement vector. The control unit extracts the maximum displacement modulus from the peak value sequence and determines it as the extreme value of the dynamic distortion of the current deployment environment. The system will In the tolerance determination function of the input dynamic calibration and adaptation module; when the magnitude of the drift tensor calculated in real time exceeds the dynamic distortion extreme value. When the system determines that the nonlinear degradation of the image domain exceeds the confidence interval for pixel-level repair, the closed-loop feedback module outputs an interrupt protection signal indicating calibration failure to the external execution end. The interrupt protection signal triggers the system to lock the current virtual space positioning reference, preventing the output of distorted six-degree-of-freedom pose parameters and maintaining the stability of the detection process under boundary conditions.

[0045] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. 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 of this application and the scope of protection of this invention, and all of these forms are within the protection scope of this application.

Claims

1. A dynamic calibration and adaptation system integrating industrial visual inspection and positioning guidance, characterized in that, include: The imaging acquisition module is used to acquire raw digital images that characterize the operating state of physical objects; The feature re-anchoring module is used to identify geometrically constrained edge intersections, circle center feature points, and texture abrupt change points in the original digital image using a semantic prior model to determine multiple semantic anchors, and calculate the displacement vector of the semantic anchors relative to the preset reference position. By performing energy minimization fitting on the displacement vector, a drift tensor representing the nonlinear degradation trend of the image domain is generated. The dynamic calibration and adaptation module maps the drift tensor to a pixel spatial distribution probability map to identify local feature degradation regions affected by high-frequency motion features in dynamic scenes. Based on the drift tensor, it generates a spatial modulation mask corresponding to the local feature degradation region. Within the feature search domain defined by the spatial modulation mask, it performs pixel-level sharpening on the original digital image by dynamically adjusting the convolution kernel weight matrix of the Laplacian operator, and performs contrast gain modulation through local gray-level histogram reconstruction to reconstruct high-level semantic features in the original digital image. Simultaneously, the dynamic calibration and adaptation module uses the drift tensor to drive the parameterized 3D mesh to generate geometric topological deformation. By aligning the deformed parameterized 3D mesh with the sharpened semantic anchor points in the feature domain spatial domain, it reconstructs the virtual spatial positioning benchmark of the physical object. The execution guidance module is used to calculate the six-degree-of-freedom pose parameters of the physical object based on the reconstructed virtual space positioning reference, and encapsulate the six-degree-of-freedom pose parameters into a control data packet that can be recognized by the external execution terminal for output.

2. The dynamic calibration and adaptation system integrating industrial vision inspection and positioning guidance according to claim 1, characterized in that, When the feature re-anchoring module identifies semantic anchor points, it extracts high-frequency gradient change regions in the original digital image and establishes geometric constraints based on the consistency of the response of edge intersections under different illumination components.

3. The dynamic calibration and adaptation system integrating industrial vision inspection and positioning guidance according to claim 2, characterized in that, When the feature reanchoring module calculates the drift tensor, it performs displacement vector fitting of each semantic anchor point between adjacent image frames to obtain the drift tensor. The calculation formula is: ,in, Here, n is the drift tensor; n is the total number of semantic anchors. Contribute weights to the features of the i-th semantic anchor point; The spatial coordinates of the semantic anchor point at the current moment; The preset reference coordinates for semantic anchor points stored during the static initialization phase of the system.

4. The dynamic calibration and adaptation system integrating industrial vision inspection and positioning guidance according to claim 1, characterized in that, When the dynamic calibration and adaptation module generates the spatial modulation mask, it establishes an inverse mapping relationship between the magnitude of the drift tensor and the pixel response intensity of the local feature degradation region. The spatial modulation mask is then used to assign weight gain to pixel regions with high motion blur in the original digital image.

5. The dynamic calibration and adaptation system integrating industrial vision inspection and positioning guidance according to claim 4, characterized in that, When the dynamic calibration and adaptation module performs pixel-level sharpening using the spatial modulation mask, it adjusts the coefficients of the weight matrix in real time according to the weight distribution in the spatial modulation mask to enhance the edge gradient in the local feature degradation area.

6. The dynamic calibration and adaptation system integrating industrial vision inspection and positioning guidance according to claim 1, characterized in that, When the dynamic calibration and adaptation module performs contrast gain modulation, it uses a spatial modulation mask to extract the brightness deviation component of the original digital image and performs adaptive histogram equalization on the original digital image.

7. The dynamic calibration and adaptation system integrating industrial vision inspection and positioning guidance according to claim 1, characterized in that, When the dynamic calibration and adaptation module generates geometric topological deformation, it establishes a linear offset model between the drift tensor and the vertex coordinates of the parameterized 3D mesh, keeps the topological connectivity properties of the parameterized 3D mesh constant, and uses the drift tensor to update the displacement of the vertex coordinates.

8. The dynamic calibration and adaptation system integrating industrial vision inspection and positioning guidance according to claim 7, characterized in that, When the dynamic calibration and adaptation module reconstructs the virtual space positioning benchmark, it calculates the spatial overlap between the deformed parametric 3D mesh and the sharpened semantic anchor points, and determines the coordinate transformation matrix for real-time calibration through maximum likelihood estimation.

9. The dynamic calibration and adaptation system integrating industrial vision inspection and positioning guidance according to claim 8, characterized in that, The system also includes a closed-loop feedback module, which calculates the control accuracy residuals of the six-degree-of-freedom pose parameters and feeds the control accuracy residuals back to the feature re-anchoring module within a period of 10ms to 20ms.

10. The dynamic calibration and adaptation system integrating industrial vision inspection and positioning guidance according to claim 1, characterized in that, When the execution guidance module outputs control data packets, it converts the six-degree-of-freedom pose parameters into register control signals for the external execution terminal.

Images