Infrared lens assembly and adjustment accuracy calibration method based on industrial vision
By using an infrared lens assembly and adjustment method based on industrial vision, multi-dimensional parameters of the infrared lens are detected and adjusted simultaneously, overcoming the limitations of the single feature acquisition method in the existing technology. This achieves high-precision infrared lens assembly, quantifies assembly deviations, standardizes the adjustment process, and stabilizes the assembly accuracy of precision optical components.
Patent Information
- Application Number
- CN202610580450.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-04-29
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2046-04-29
AI Technical Summary
In existing technologies, infrared lens assembly inspection often adopts a single local feature acquisition method, which cannot simultaneously acquire multiple core assembly parameters and lacks a systematic multi-dimensional error quantification mechanism. This results in adjustment methods relying on manual experience, making it difficult to achieve high-precision assembly.
An industrial vision-based approach is adopted, which uses an industrial camera to acquire infrared lens images, extracts feature image sets of lens edge features, internal optical element markers and background reference benchmarks, executes feature matching algorithms to generate feature-corresponding point pairs, calculates measured parameters of optical center, optical axis pointing and lens spacing, uses a parameter error analysis model to generate a multi-dimensional assembly and adjustment error vector, and generates a calibration strategy through an assembly and adjustment decision model to drive the assembly and adjustment actuator to perform physical adjustments.
It enables synchronous detection and adjustment of multi-dimensional parameters of infrared lenses, quantifies assembly deviations, standardizes adjustment processes, reduces human intervention, stabilizes the assembly accuracy of precision optical components, and adapts to high-precision assembly requirements.
Smart Images

Figure CN122453940B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of lens precision calibration technology, and in particular to a method for calibrating the assembly and adjustment precision of infrared lenses based on industrial vision. Background Technology
[0002] Infrared lens assembly and debugging is a crucial step in the manufacturing of optical components. Currently, the accuracy inspection of infrared lens assembly and debugging mostly relies on single-local feature acquisition methods for parameter detection, focusing only on the local structural features of the lens for measurement and analysis. This singular feature acquisition mode cannot integrate multiple types of key structural information, making it difficult to simultaneously acquire multiple core assembly parameters of the lens. The parameter detection items are relatively limited, failing to achieve simultaneous detection of multi-dimensional assembly indicators.
[0003] Conventional assembly and adjustment error control methods often use single-index comparisons to determine deviations, simply comparing differences between independent parameters without a systematic, multi-dimensional error quantification mechanism. The deviation determination method is relatively simple and cannot fully reflect the overall deviation status of the lens assembly. Adjustments often rely on manual experience, lacking quantitative basis for the adjustments, and the adjustment actions cannot be precisely correlated with the source of the deviation.
[0004] The internal optical components of an infrared lens are precisely arranged, and even slight misalignments during assembly can cause a degradation in optical performance. Traditional inspection methods struggle to simultaneously acquire lens edges, internal markers, and external reference features, and cannot simultaneously calculate multiple key assembly parameters. Simple parameter comparison methods cannot separate deviation components across different dimensions, and adjustment operations lack quantitative guidance, making it difficult to meet the stringent requirements of high-precision assembly and debugging of infrared lenses. Therefore, a more sophisticated multi-feature collaborative inspection mechanism is needed, along with a systematic error analysis and quantitative adjustment model. Summary of the Invention
[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and propose an infrared lens assembly and adjustment accuracy calibration method based on industrial vision.
[0006] To achieve the above objectives, the present invention adopts the following technical solution: an infrared lens assembly and adjustment accuracy calibration method based on industrial vision, comprising:
[0007] An industrial camera is used to acquire optical image sequences of an infrared lens in the assembly and adjustment state. The optical image sequences are then preprocessed to extract a feature image set containing lens edge features, internal optical element markers, and background reference references.
[0008] The feature image set is processed by a feature matching algorithm to generate a set of feature-corresponding point pairs;
[0009] Based on the set of feature-corresponding point pairs, calculate the set of measured parameters for the optical center, optical axis direction, and lens spacing of the infrared lens;
[0010] Obtain the standard values of the design parameters of the infrared lens, including the standard position of the optical center, the standard direction of the optical axis, and the standard distance between each lens;
[0011] The measured parameter set and the standard values of the design parameters are input into the parameter error analysis model to calculate and generate a multi-dimensional assembly and adjustment error vector.
[0012] Based on the multi-dimensional assembly and adjustment error vector, a calibration strategy including adjustment components, adjustment direction, and adjustment amount is generated through the assembly and adjustment decision model;
[0013] The calibration strategy is converted into a sequence of control commands, which drives the assembly and adjustment mechanism to physically adjust the infrared lens.
[0014] As a further aspect of the present invention, a feature matching algorithm is performed on the feature image set to generate a set of feature-corresponding point pairs, including:
[0015] The feature image set is decomposed into a multi-scale Gaussian pyramid to generate image pyramid layers of multiple scales;
[0016] In each layer of the image pyramid, accelerated robust feature descriptors are extracted, and preliminary feature point matching is performed using the feature matching algorithm to form a set of candidate matching pairs.
[0017] When calculating the similarity of feature descriptors, the feature matching algorithm introduces consistency verification based on the local geometric context of feature points, and constructs a geometric consistency constraint function to filter candidate matching pairs based on the structural characteristics of the lens edge and marker points of the infrared lens image.
[0018] For the filtered candidate matching pair set, feature point trajectory tracking and bidirectional consistency verification are performed between image pyramid layers of adjacent scales, and finally the cross-scale stable set of feature corresponding point pairs is output.
[0019] As a further aspect of the present invention, the feature matching algorithm, when calculating the similarity of feature descriptors, introduces consistency verification based on the local geometric context of feature points, and constructs a geometric consistency constraint function to filter candidate matching pairs based on the structural characteristics of the lens edges and marker points of the infrared lens image, including:
[0020] Centered on each feature point in the candidate matching pair set, a local neighborhood window with a preset radius is selected in the image pyramid layer to which it belongs;
[0021] Within the local neighborhood window, the spatial relative positional relationship between the current feature point and other surrounding feature points is extracted, and a geometric context descriptor describing the local spatial distribution of the current feature point is calculated and generated.
[0022] For a pair of candidate matching feature points, calculate their geometric context descriptors respectively, and compare the difference between the two descriptors. If the difference exceeds a preset first threshold, the local geometric context is considered inconsistent, and the candidate matching pair is removed from the candidate matching pair set.
[0023] Based on the structural characteristics of the infrared lens image, where the lens edge appears as a smooth elliptical arc and the marker points are usually regular dots or crosses, the geometric consistency constraint function is constructed.
[0024] The geometric consistency constraint function is used to quantify whether the local image structure of two feature points in a pair of candidate matches conforms to the curvature continuity constraint of the lens edge or the neighborhood symmetry constraint of the marker point.
[0025] Calculate the geometric consistency constraint function value of each candidate pair in the candidate matching pair set. If the function value is lower than a preset second threshold, the corresponding candidate matching pair is considered to be inconsistent with the prior knowledge of the structure of the infrared lens, and the corresponding candidate matching pair is filtered out from the candidate matching pair set.
[0026] As a further aspect of the present invention, based on the set of feature-corresponding point pairs, a set of measured parameters for the optical center, optical axis direction, and lens spacing of the infrared lens is calculated, including:
[0027] From the set of feature corresponding point pairs, separate the feature point pairs belonging to the lens edge contour for the feature point pairs belonging to the internal optical element marker points;
[0028] Using feature point pairs belonging to the lens edge contour, the ellipse equation of the lens outer edge in the image coordinate system is calculated by an ellipse fitting algorithm. The coordinates of the center point of the ellipse equation are then transformed into the coordinate system and used as the measured position of the optical center.
[0029] Using feature point pairs belonging to the internal optical element markers, combined with the known three-dimensional design coordinates of the markers, the attitude of the infrared lens relative to the industrial camera is calculated by the perspective N-point positioning algorithm. The line of sight extracted from the attitude is inverted and used as the measured value of the optical axis pointing.
[0030] By utilizing feature point pairs belonging to markers on different lenses, and based on the calculated spatial position of each lens, the measured value of the lens spacing is obtained by calculating the spatial distance. All measured values together constitute the set of measured parameters.
[0031] As a further aspect of the present invention, the measured parameter set and the design parameter standard values are input into a parameter error analysis model to calculate and generate a multi-dimensional assembly and adjustment error vector, including:
[0032] The parameter error analysis model establishes a mapping relationship from the measured parameter space to the standard parameter space, and the mapping relationship is defined by the installation and adjustment parameter error model of the infrared lens.
[0033] By inputting the set of measured parameters into the mapping relationship, the equivalent measured parameters in the standard parameter space are calculated.
[0034] Calculate the difference between the equivalent measured parameters and the standard values of the design parameters in each parameter dimension;
[0035] The differences in each parameter dimension are normalized and dimensionally unified. The processed differences are arranged in a preset order to form the multi-dimensional assembly and adjustment error vector. Each element in the multi-dimensional assembly and adjustment error vector corresponds to a specific type of assembly and adjustment error.
[0036] As a further aspect of the present invention, the parameter error analysis model establishes a mapping relationship from the measured parameter space to the standard parameter space, including:
[0037] The mapping relationship is represented by a set of nonlinear equations, the independent variables of which are basic assembly and adjustment error variables including lens displacement, tilt, and lens spacing deviation.
[0038] The measured position of the optical center, the measured value of the optical axis direction, and the measured value of the lens spacing in the set of measured parameters are used as known quantities in the nonlinear equation system.
[0039] The nonlinear equations are solved by numerical iteration to obtain a set of numerical solutions for basic assembly and adjustment error variables. The numerical solutions for these basic assembly and adjustment error variables represent the equivalent representation of the measured parameter set in the standard parameter space. This equivalent representation is the equivalent measured parameter.
[0040] As a further aspect of the present invention, based on the multi-dimensional assembly and adjustment error vector, a calibration strategy including adjustment components, adjustment direction, and adjustment amount is generated through an assembly and adjustment decision model, including:
[0041] The assembly and adjustment decision model includes an error vector parsing module and a strategy mapping knowledge base;
[0042] The error vector parsing module receives the multi-dimensional assembly and adjustment error vector and decomposes it into sub-error components related to specific adjustable parts.
[0043] The strategy mapping knowledge base stores historical calibration data, which records the successful calibration strategies corresponding to different combinations of sub-error components.
[0044] In the strategy mapping knowledge base, the most similar historical calibration strategies are retrieved using the sub-error components as query conditions.
[0045] Based on the retrieved historical calibration strategies, a calibration strategy suitable for the current error state is generated through strategy fusion and optimization algorithms. The calibration strategy clearly indicates the names of the components that need to be adjusted, the adjustment direction of each component, and the physical adjustment amount.
[0046] As a further aspect of the present invention, the step of generating a calibration strategy suitable for the current error state based on the retrieved historical calibration strategies through a strategy fusion and optimization algorithm includes:
[0047] For each retrieved historical calibration strategy, calculate the Euclidean distance between its corresponding historical sub-error component and the current sub-error component;
[0048] Based on the Euclidean distance, a fusion weight is assigned to each historical calibration strategy, with a larger fusion weight assigned to the historical calibration strategy with a smaller distance.
[0049] The adjustment direction and adjustment amount in the historical calibration strategy are weighted and averaged according to their fusion weights to obtain the preliminary fused adjustment direction and adjustment amount;
[0050] The preliminary fusion of adjustment direction and adjustment amount is input into a physical model-based optimizer. The optimizer aims to minimize the multi-dimensional assembly and adjustment error vector, fine-tunes the adjustment direction and adjustment amount, outputs the final determined adjustment direction and adjustment amount, and integrates them to form the calibration strategy.
[0051] As a further aspect of the present invention, the calibration strategy is converted into a sequence of control commands to drive the assembly and adjustment mechanism to physically adjust the infrared lens, including:
[0052] The calibration strategy is analyzed to extract the names of the components that need to be adjusted, the adjustment direction of each component, and the physical adjustment amount.
[0053] Based on the motion characteristics and control interface protocol of each drive unit in the assembly and adjustment mechanism, the adjustment direction and physical adjustment amount of each component are converted into control commands for the corresponding drive unit.
[0054] Based on the physical dependencies and safety sequence between the adjustment actions of each component, all generated drive unit control commands are sorted to form an ordered sequence of control commands;
[0055] The control command sequence is sent one by one to the controller of the assembly and adjustment mechanism, and the controller drives the corresponding drive unit to complete the physical adjustment action.
[0056] As a further aspect of the present invention, based on the motion characteristics and control interface protocol of each drive unit in the assembly and adjustment mechanism, the adjustment direction and physical adjustment amount of each component are converted into control commands for the corresponding drive unit, including:
[0057] Search for the model and specifications of the drive unit corresponding to the name of the component that needs to be adjusted in the assembly and adjustment mechanism;
[0058] From the model and specification document, obtain the conversion coefficients for the corresponding drive unit to convert physical displacement into control pulse count, and the mapping relationship for converting adjustment direction into control signal level;
[0059] Using the conversion coefficient, the physical adjustment amount is calculated as the required number of control pulses;
[0060] Using the mapping relationship, the adjustment direction is converted into a specific control signal level value;
[0061] The number of control pulses and the level value of the control signal are encapsulated according to the data format specified in the control interface protocol to generate a single control instruction that can be recognized and executed by the corresponding drive unit.
[0062] Compared with the prior art, the advantages and positive effects of the present invention are as follows:
[0063] Simultaneously extracting lens edge features, internal optical component markers, and background reference points, a stable set of feature-corresponding point pairs is established based on feature matching algorithms. Based on multi-source feature correlations, the optical center, optical axis direction, and lens spacing are simultaneously calculated. Collaborative acquisition and joint computation of multiple feature types enrich the detection dimensions of lens assembly status, breaking the limitations of single-structure detection, unifying the calculation process for multiple core assembly parameters, and standardizing the acquisition logic of optical assembly parameters, resulting in a more comprehensive detection of the assembly position of internal lens components and the overall optical attitude.
[0064] A parametric error analysis model is introduced to perform component-by-component comparisons between measured parameters and standard design parameters, outputting a structured, multi-dimensional assembly error vector. A hierarchical error analysis method is employed to break down the deviation components of different structural positions and optical indicators, refining the quantitative expression of assembly deviations and clearly classifying the distribution types and attributes of various deviations.
[0065] By combining multi-dimensional assembly and adjustment error vectors and relying on the assembly and adjustment decision model, clear adjustment components, adjustment directions, and adjustment amounts are generated. The calibration strategy is transformed into a standardized sequence of control commands, directly driving the assembly and adjustment actuator to complete automated physical correction. Quantified adjustment rules can standardize the execution logic of assembly correction, unify the operation standards of assembly and adjustment correction, reduce operational fluctuations caused by human intervention, stabilize the execution process of precision optical component assembly correction, continuously maintain the precision control level of infrared lens assembly and adjustment operations, and adapt to the operational requirements of high-precision optical lens assembly. Attached Figure Description
[0066] Figure 1 This is a state diagram of the infrared lens assembly and adjustment accuracy calibration method based on industrial vision described in this invention.
[0067] Figure 2 A flowchart illustrating the process of generating a set of feature-corresponding point pairs by performing a feature matching algorithm on a set of feature images;
[0068] Figure 3 This is a flowchart for calculating the set of measured parameters of an infrared lens based on a set of feature correspondence point pairs. Detailed Implementation
[0069] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.
[0070] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.
[0071] See Figure 1This invention relates to a method for calibrating the assembly accuracy of an infrared lens based on industrial vision. The overall implementation scheme is as follows: An industrial camera is used to acquire images of an infrared lens in the assembly state, obtaining a series of optical images to form an optical image sequence. Preprocessing is performed on this optical image sequence to enhance image features and suppress noise, extracting a feature image set containing lens edge features, internal optical element markers, and a background reference. A feature matching algorithm is then applied to this feature image set. By comparing the feature information between different images, a series of two-dimensional coordinate correspondences describing the same physical points are generated, i.e., a set of feature-corresponding point pairs. Based on this set of feature-corresponding point pairs, a set of measured parameters for the optical center, optical axis direction, and lens spacing of the infrared lens in the assembly state is solved through mathematical models and geometric calculations. Simultaneously, the standard position of the optical center, the standard direction of the optical axis, and the standard distance between each lens are obtained from the infrared lens's design document, constituting the standard values of the design parameters. The set of measured parameters and the standard values of the design parameters are input into a preset parameter error analysis model. This model calculates and outputs a multi-dimensional assembly error vector, which quantifies the deviation between each assembly parameter and the design standard. Based on the multi-dimensional assembly and adjustment error vector, an assembly and adjustment decision model is used for analysis and mapping to generate a detailed calibration strategy. This strategy clearly indicates the names of the components that need adjustment, the adjustment direction of each component, and the specific physical adjustment amount. This calibration strategy is converted into a series of machine-readable control command sequences and sent to the controller of the assembly and adjustment actuator. This controller drives the internal physical execution units, such as motors and displacement stages, to precisely adjust the physical position and attitude of the infrared lens, thereby completing the calibration closed loop.
[0072] In one embodiment of the present invention, a feature matching algorithm is performed on the feature image set to generate a set of feature-corresponding point pairs. The process is as follows: (See reference) Figure 2 The feature image set is decomposed into a multi-scale Gaussian pyramid, generating image pyramid layers at multiple scales. At each layer of the image pyramid, accelerated robust feature descriptors are extracted, and a feature matching algorithm is used for preliminary feature point matching to form a candidate matching pair set. When calculating feature descriptor similarity, the feature matching algorithm introduces consistency verification based on the local geometric context of feature points, and constructs a geometric consistency constraint function based on the structural characteristics of the lens edges and marker points in the infrared lens image to filter the candidate matching pairs. For the filtered candidate matching pair set, feature point trajectory tracking and bidirectional consistency verification are performed between image pyramid layers at adjacent scales, outputting a cross-scale stable set of feature-corresponding point pairs.
[0073] When calculating the similarity of feature descriptors, the feature matching algorithm introduces consistency verification based on the local geometric context of feature points. Based on the structural characteristics of the lens edges and marker points in infrared lens images, a geometric consistency constraint function is constructed to filter candidate matching pairs. The specific steps are as follows: For each feature point in the candidate matching pair set, a local neighborhood window with a preset radius is selected within its corresponding image pyramid layer. Within the local neighborhood window, the spatial relative positional relationship between the current feature point and other surrounding feature points is extracted, and a geometric context descriptor describing the local spatial distribution of the current feature point is calculated. For a pair of candidate matching feature points, their geometric context descriptors are calculated separately, and the difference between the two descriptors is compared. If the difference exceeds a preset first threshold, their local geometric contexts are considered inconsistent, and this candidate matching pair is removed from the candidate matching pair set. Based on the structural characteristics of infrared lens images—where the lens edges appear as smooth elliptical arcs and the marker points are typically regular dots or crosses—a geometric consistency constraint function is constructed. The geometric consistency constraint function is used to quantify whether the local image structure of two feature points in a candidate match conforms to the curvature continuity constraint of the lens edge or the neighborhood symmetry constraint of the marker point. The geometric consistency constraint function value is calculated for each candidate match pair in the candidate match pair set. If the function value is lower than a preset second threshold, the corresponding candidate match pair is considered to not conform to the prior structural knowledge of the infrared lens, and is filtered out from the candidate match pair set.
[0074] In practical implementation, a feature matching algorithm is performed on the feature image set to generate a set of feature-corresponding point pairs. The process is as follows: The feature image set is decomposed into a multi-scale Gaussian pyramid, generating multiple image pyramid layers at various scales. The scale of each image pyramid layer is sequentially downsampled from the original resolution. At each layer of the image pyramid, accelerated robust feature descriptors are extracted, and preliminary feature point matching is performed using the feature matching algorithm to form a set of candidate matching pairs. When calculating the similarity of feature descriptors, the feature matching algorithm introduces consistency verification based on the local geometric context of feature points. Based on the structural characteristics of the lens edges and marker points in the infrared lens image, a geometric consistency constraint function is constructed to filter the candidate matching pairs. For the filtered set of candidate matching pairs, feature point trajectory tracking and bidirectional consistency verification are performed between image pyramid layers at adjacent scales, outputting a set of cross-scale stable feature-corresponding point pairs.
[0075] In its implementation, the feature matching algorithm introduces consistency verification based on the local geometric context of feature points when calculating the similarity of feature descriptors. Based on the structural characteristics of the lens edges and marker points in the infrared lens image, a geometric consistency constraint function is constructed to filter candidate matching pairs. The specific steps are as follows: For each feature point in the candidate matching pair set, a local neighborhood window with a preset radius is selected within its corresponding image pyramid layer. In some embodiments, the preset radius of the local neighborhood window can be set to a pixel region centered on the feature point.
[0076] Within a local neighborhood window, the spatial relative positions of the current feature point and its surrounding feature points are extracted, and a geometric context descriptor describing the local spatial distribution of the current feature point is calculated. For a pair of candidate matching feature points, the geometric context descriptors of the feature points are calculated separately, and the difference between the two geometric context descriptors is compared. If the difference between the geometric context descriptors of the feature points exceeds a preset first threshold, the local geometric context of the feature points is considered inconsistent, and this pair of candidate matches is removed from the candidate matching pair set. Based on the structural characteristics of infrared lens images, where the lens edge presents as a smooth elliptical arc and the marker points are usually regular dots or cross shapes, a geometric consistency constraint function is constructed. The geometric consistency constraint function is used to quantify whether the local image structure of each feature point in a pair of candidate matching features conforms to the curvature continuity constraint of the lens edge or the neighborhood symmetry constraint of the marker points. The geometric consistency constraint function value of each pair of candidate matches in the candidate matching pair set is calculated. If the geometric consistency constraint function value of the feature point pair is lower than a preset second threshold, the candidate matching pair corresponding to the feature point pair is considered not to conform to the prior knowledge of the infrared lens structure, and the candidate matching pair corresponding to the feature point pair is filtered out from the candidate matching pair set.
[0077] In some embodiments, the dissimilarity of the geometric context descriptor can be represented as the cosine distance between the spatial distribution vectors of feature points in the neighborhood of a feature point. In a specific implementation, the feature matching algorithm can use a bidirectional nearest neighbor matching method to generate preliminary feature point matches. Optionally, when performing multi-scale Gaussian pyramid decomposition on the feature image set, a Gaussian convolution kernel is used for continuous downsampling. It can be understood that the feature matching algorithm runs independently at each image pyramid layer, and the preliminary matching results output by the feature matching algorithm are correlated between layers. Optionally, during feature point trajectory tracking, the feature matching algorithm verifies the consistency of feature point responses in the scale space. It can be understood that the constructed geometric consistency constraint function combines the residuals of ellipse fitting to evaluate the local structure of feature points at the lens edge.
[0078] In practical implementation, a geometric consistency constraint function used to evaluate the consistency of feature points with respect to local geometry can be expressed as:
[0079]
[0080] in: This represents the similarity score of feature point pairs output by the geometric consistency constraint function. This indicates the number of corresponding point pairs used for sampling and evaluation within the local neighborhood window. In the local coordinate system of the first feature point, the... The azimuth angle of each sampling evaluation point relative to the central feature point This indicates that in the local coordinate system of the second feature point, it is related to the first... The azimuth angle of each sampled evaluation point relative to the central feature point. Similarity score. The value range is [0,1], and the similarity score is... The closer the value is to 1, the higher the consistency of the local geometric structure of the feature point pair, and the more the matched feature points pass the verification of the geometric consistency constraint function.
[0081] In one embodiment of the present invention, based on the set of feature-corresponding point pairs, the measured parameter set of the optical center, optical axis direction, and lens spacing of the infrared lens is calculated. The process is as follows: (See attached document). Figure 3 From the set of feature point pairs, feature point pairs belonging to the lens edge contour are separated from feature point pairs belonging to the internal optical element markers. Using the feature point pairs belonging to the lens edge contour, an ellipse equation for the lens outer edge in the image coordinate system is calculated using an ellipse fitting algorithm. The coordinates of the center point of this ellipse equation, after coordinate system transformation, are used as the measured position of the optical center. Using the feature point pairs belonging to the internal optical element markers, combined with the known 3D design coordinates of the markers, the attitude of the infrared lens relative to the industrial camera is calculated using a perspective N-point positioning algorithm. The line of sight extracted from this attitude is inverted and used as the measured value of the optical axis direction. Using the feature point pairs belonging to the markers on different lenses, based on the calculated spatial positions of each lens, the measured value of the lens spacing is obtained by calculating the spatial distance. All measured values together constitute the set of measured parameters.
[0082] In practical implementation, based on the set of feature-corresponding point pairs, the measured parameter set of the optical center, optical axis direction, and lens spacing of the infrared lens is calculated. The process is as follows: From the set of feature-corresponding point pairs, feature points belonging to the lens edge contour are separated from feature point pairs belonging to the internal optical element marker points. The separation operation is performed based on the positional distribution and descriptor attributes of the feature points. Using the feature point pairs belonging to the lens edge contour, the ellipse equation of the lens outer edge in the image coordinate system is calculated using an ellipse fitting algorithm. The ellipse fitting algorithm uses the least squares method to solve for the parameters of the general equation of the ellipse. The coordinates of the center point of the ellipse equation, after transformation from the image coordinate system to the world coordinate system, are used as the measured position of the optical center. Using the feature point pairs belonging to the internal optical element marker points, combined with the known three-dimensional design coordinates of the marker points, the orientation of the infrared lens relative to the industrial camera is calculated using a perspective N-point positioning algorithm. In practical implementation, the perspective N-point positioning algorithm uses the two-dimensional image coordinates and three-dimensional design coordinates of the marker points as input to solve for the extrinsic parameter matrix of the industrial camera. The viewing direction, represented by the rotation matrix extracted from the extrinsic matrix of the industrial camera, is transformed and inverted from the camera coordinate system to the lens's own coordinate system, and then used as the measured value of the optical axis pointing. Using feature point pairs belonging to markers on different lenses, and based on the spatial position of each lens calculated using a perspective N-point positioning algorithm, the measured value of the lens spacing is obtained by calculating the Euclidean distance in three-dimensional space between the center points of the corresponding markers on different lenses. The measured position of the optical center, the measured value of the optical axis pointing, and the measured value of the lens spacing together constitute the set of measured parameters.
[0083] In some embodiments, coordinate system transformation requires pre-calibrating the intrinsic and extrinsic parameters of the industrial camera. Optionally, the ellipse fitting algorithm can filter the feature points of the lens edge contour before processing to remove outliers caused by noise or discontinuous edges. When extracting the measured optical axis pointing value from the industrial camera's extrinsic parameter matrix, the rotation matrix needs to be normalized. It is understood that the measured value of the lens spacing is obtained by calculating the straight-line distance in space. In some embodiments, when there are multiple marker points on the same lens, the average value or centroid of the spatial positions of these marker points can be calculated as the representative position of the lens. Optionally, the perspective N-point positioning algorithm can introduce a random sampling consensus algorithm to improve the robustness of the solution and handle marker points that may have mismatched values. After extracting the rotation matrix from the industrial camera's extrinsic parameter matrix, it needs to be transformed into a coordinate system based on the mechanical axis of the infrared lens. It is understood that the calculated measured value of the lens spacing can be a scalar value or a vector containing multiple lens spacings.
[0084] In practical implementation, the equation of the lens outer edge ellipse can be expressed by the following general form of a quadratic curve equation:
[0085]
[0086] in: , , , , , These are the coefficients of the ellipse equation. These represent the pixel coordinates of feature points on the lens edge contour in the image coordinate system. The coefficients are obtained through least squares fitting. , , , , , Then, the image coordinates of the ellipse center It can be obtained through coefficient calculation. Image coordinates of the ellipse center. Coordinates transformed to world coordinate system This refers to the measured position of the optical center. The transformation relationship depends on the intrinsic parameter matrix of the industrial camera, the extrinsic parameter matrix of the industrial camera, and the pose relationship between the industrial camera coordinate system and the world coordinate system.
[0087] In one embodiment of the present invention, the measured parameter set and the standard values of the design parameters are input into a parameter error analysis model to calculate and generate a multi-dimensional assembly and adjustment error vector. The process is as follows: The parameter error analysis model establishes a mapping relationship from the measured parameter space to the standard parameter space, which is defined by the assembly and adjustment parameter error model of the infrared lens. The measured parameter set is input into this mapping relationship to calculate the equivalent measured parameters in the standard parameter space. The difference between the equivalent measured parameters and the standard values of the design parameters is calculated in each parameter dimension. The differences in each parameter dimension are normalized and dimensionally unified. The processed differences are arranged in a preset order to form a multi-dimensional assembly and adjustment error vector, where each element corresponds to a specific type of assembly and adjustment error.
[0088] The mapping relationship is represented by a set of nonlinear equations, where the independent variables are basic assembly error variables including lens displacement, tilt, and lens spacing deviation. The measured positions of the optical center, optical axis pointing, and lens spacing from the measured parameter set are taken as known quantities in the nonlinear equation set. The nonlinear equation set is solved using a numerical iteration method to obtain a set of numerical solutions for the basic assembly error variables. These numerical solutions represent the equivalent representation of the measured parameter set in the standard parameter space, and this equivalent representation is the equivalent measured parameter.
[0089] In practical implementation, the measured parameter set and the standard values of the design parameters are input into the parameter error analysis model to calculate and generate a multi-dimensional assembly error vector. The process is as follows: The parameter error analysis model establishes a mapping relationship from the measured parameter space to the standard parameter space, defined by the assembly parameter error model of the infrared lens. The measured parameter set is input into the mapping relationship to calculate the equivalent measured parameters in the standard parameter space. The differences between the equivalent measured parameters and the standard values of the design parameters are calculated in each parameter dimension. The differences in each parameter dimension are normalized and dimensionally unified. The processed differences are arranged in a preset order to form a multi-dimensional assembly error vector. Each element in the multi-dimensional assembly error vector corresponds to a specific type of assembly error. The mapping relationship is represented by a set of nonlinear equations, where the independent variables of the nonlinear equations are basic assembly error variables including lens displacement, tilt, and lens spacing deviation. The measured position of the optical center, the measured value of the optical axis direction, and the measured value of the lens spacing in the measured parameter set are used as known quantities in the nonlinear equations. By solving the nonlinear equations using a numerical iteration method, a set of numerical solutions for the basic assembly and adjustment error variables is obtained. The numerical solutions for the basic assembly and adjustment error variables represent the equivalent representation of the measured parameter set in the standard parameter space, and the equivalent representation is the equivalent measured parameter.
[0090] In practical implementation, solving the nonlinear equation system can employ the Newton-Raphson method or the Levenberg-Marquardt method. In some embodiments, normalization involves dividing the difference by the absolute value of the corresponding design parameter standard value. Optionally, dimensional unification converts all differences involving angles and lengths to a unified unit, such as unifying angles to radians and lengths to millimeters. The nonlinear equation system of mapping relationships contains equations describing the geometric imaging relationships of the optical system. It can be understood that the multi-dimensional alignment error vector is a column vector or a row vector. In some embodiments, the preset order can be arranged according to the order of optical center error, optical axis pointing error, and lens spacing error. Optionally, the basic alignment error variables can include the translation error of the lens along the X, Y, and Z axes and the rotation error around the X, Y, and Z axes. It can be understood that numerical iterative methods require setting initial values and convergence tolerances.
[0091] In practical implementation, the set of nonlinear equations defining the mapping relationship in the parameter error analysis model can be expressed as:
[0092]
[0093] in: This represents a vector consisting of basic assembly and adjustment error variables, including lens displacement, tilt, and lens spacing deviation. Indicates the measured position of the optical center. This indicates the measured value of the optical axis pointing to the optical axis. Represents the measured value of lens spacing, function This represents the nonlinear functional relationship mapping from the basic assembly and adjustment error variable to specific measured values. Solving the nonlinear equation system is equivalent to finding the basic assembly and adjustment error variable. The numerical solution makes the function The outputs are respectively equal to the known For the calculation of parameter dimension differences and the construction of the error vector, please refer to Table 1:
[0094] Table 1: Parameter Error Calculation and Error Vector Construction Table
[0095] Parameter type Equivalent measured parameters Standard values of design parameters Original difference Error vector elements after processing Optical center X (mm) 0.12 0.00 0.12 0.12 Optical center Y (mm) -0.08 0.00 -0.08 -0.08 Optical axis pointing in pitch (mrad) 1.5 0.0 1.5 0.0015 Optical axis orientation (mrad) -0.7 0.0 -0.7 -0.0007 Lens spacing 1-2 mm 5.15 5.10 0.05 0.05
[0096] This demonstrates the processing flow from equivalent measured parameters to error vector elements. In practical implementation, the equivalent measured parameters are the basic assembly error variables obtained by solving the nonlinear equation system. The representation in standard parameter space calculated using a specific geometric model. The standard values of design parameters are usually zero or theoretical design values. The original difference is the direct arithmetic difference between the equivalent measured parameter and the standard value of the design parameter. The processed error vector elements are the result of dimension unification (e.g., converting angles from milliradians to radians) and possible normalization. These elements, arranged in rows, constitute a multi-dimensional assembly and adjustment error vector.
[0097] In one embodiment of the present invention, a calibration strategy including adjustment components, adjustment directions, and adjustment amounts is generated through an assembly and adjustment decision model based on a multi-dimensional assembly and adjustment error vector. The process is as follows: The assembly and adjustment decision model includes an error vector parsing module and a strategy mapping knowledge base. The error vector parsing module receives the multi-dimensional assembly and adjustment error vector and decomposes it into sub-error components related to specific adjustable components. The strategy mapping knowledge base stores historical assembly and adjustment data, which records successful calibration strategies corresponding to different combinations of sub-error components. In the strategy mapping knowledge base, using the current sub-error component as a query condition, the most similar historical calibration strategies are retrieved. Based on the retrieved historical calibration strategies, a calibration strategy suitable for the current error state is generated through a strategy fusion and optimization algorithm. This calibration strategy explicitly indicates the names of the components that need adjustment, the adjustment direction of each component, and the physical adjustment amount.
[0098] Based on the retrieved historical calibration strategies, a calibration strategy suitable for the current error state is generated through strategy fusion and optimization algorithms. The specific steps are as follows: For each retrieved historical calibration strategy, the Euclidean distance between its corresponding historical sub-error component and the current sub-error component is calculated. Based on this Euclidean distance, a fusion weight is assigned to each historical calibration strategy; the smaller the distance, the larger the fusion weight. The adjustment direction and adjustment amount in the historical calibration strategies are weighted and averaged according to their fusion weights to obtain the preliminary fused adjustment direction and adjustment amount. The preliminary fused adjustment direction and adjustment amount are input into a physical model-based optimizer. This optimizer aims to minimize the multi-dimensional assembly and adjustment error vector, fine-tunes the adjustment direction and adjustment amount, and outputs the final determined adjustment direction and adjustment amount, which are then integrated to form the calibration strategy.
[0099] In practical implementation, based on the multi-dimensional assembly and adjustment error vector, a calibration strategy including the adjustment components, adjustment direction, and adjustment amount is generated through the assembly and adjustment decision model. The process is as follows: The assembly and adjustment decision model includes an error vector parsing module and a strategy mapping knowledge base. The error vector parsing module receives the multi-dimensional assembly and adjustment error vector and decomposes it into sub-error components related to specific adjustable components. The strategy mapping knowledge base stores historical assembly and adjustment data, which records successful calibration strategies corresponding to different combinations of sub-error components. In the strategy mapping knowledge base, using the current sub-error component as the query condition, the most similar historical calibration strategies are retrieved. Based on the retrieved historical calibration strategies, a calibration strategy suitable for the current error state is generated through strategy fusion and optimization algorithms. The calibration strategy clearly indicates the names of the components that need adjustment, the adjustment direction of each component, and the physical adjustment amount.
[0100] Based on the retrieved historical calibration strategies, a calibration strategy suitable for the current error state is generated through strategy fusion and optimization algorithms. The specific steps are as follows: For each retrieved historical calibration strategy, calculate the Euclidean distance between the historical sub-error component corresponding to the historical calibration strategy and the current sub-error component. Based on the Euclidean distance, assign a fusion weight to each historical calibration strategy; the smaller the distance, the larger the fusion weight. Calculate the weighted average of the adjustment direction and adjustment amount in the historical calibration strategies according to their corresponding fusion weights to obtain the preliminary fused adjustment direction and adjustment amount. Input the preliminary fused adjustment direction and adjustment amount into a physics-based optimizer. The physics-based optimizer aims to minimize the multi-dimensional assembly and adjustment error vector, fine-tunes the adjustment direction and adjustment amount, outputs the final determined adjustment direction and adjustment amount, and integrates them to form the calibration strategy.
[0101] In specific implementations, the policy mapping knowledge base can be stored using a relational database or a NoSQL database. In some embodiments, retrieving the most similar historical calibration policy can be achieved using the k-nearest neighbor algorithm. Optionally, the fusion weights are proportional to the inverse or negative exponential relationship of the Euclidean distance. The error vector parsing module can decompose the multi-dimensional assembly error vector into sub-error components based on a predefined mapping matrix. It is understood that the association between the sub-error components and the adjustable parts is determined during the model training phase. In some embodiments, the physical model-based optimizer can employ gradient descent or sequential quadratic programming algorithms. Optionally, the weighted averaging operation can be performed separately for the scalar adjustment amount and the vector adjustment direction. It is understood that the optimizer fine-tuning needs to satisfy the physical travel and resolution constraints of the assembly actuator.
[0102] In practical implementation, the formula for calculating the Euclidean distance between the historical sub-error component and the current sub-error component can be expressed as:
[0103]
[0104] in: Indicates the first The Euclidean distance between the historical sub-error components and the current sub-error components corresponding to each historical calibration strategy. This represents the total number of dimensions of the sub-error components. Represents the first sub-error component. Each component value Indicates the first The first historical sub-error component corresponding to the historical calibration strategy Each component value. Euclidean distance. The smaller the value, the higher the value. The more similar the historical error state to the current error state of a given historical calibration strategy, the better. For historical calibration strategy retrieval and weight allocation, please refer to Table 2.
[0105] Table 2: Historical Strategy Similarity Assessment and Fusion Weight Allocation Table
[0106] Historical Strategy Number Historical sub-error component vector Euclidean distance from the current error Fusion weights H001 [0.10,-0.05,0.0012] 0.051 0.45 H002 [0.08,-0.10,0.0015] 0.103 0.22 H003 [0.12,0.00,0.0010] 0.091 0.28 H004 [0.20,0.05,0.0020] 0.152 0.05
[0107] This demonstrates the process of retrieving similar strategies from historical assembly and adjustment data and assigning weights. In the specific implementation, the current sub-error component is assumed to be [0.11, -0.07, 0.0013]. Historical strategy numbers identify different historical records stored in the strategy mapping knowledge base. The historical sub-error component vector records the error state before the success of a historical assembly and adjustment case. The Euclidean distance to the current error is calculated using a formula. The fusion weights are calculated based on the distances, and the total weight is 1. For example, historical strategy H001 has the closest distance and is therefore assigned the largest fusion weight of 0.45.
[0108] The above embodiment converts the calibration strategy into a sequence of control commands to drive the assembly and adjustment actuator to physically adjust the infrared lens. The process is as follows: The calibration strategy is parsed to extract the names of the components requiring adjustment, the adjustment direction of each component, and the physical adjustment amount. Based on the motion characteristics and control interface protocol of each drive unit in the assembly and adjustment actuator, the adjustment direction and physical adjustment amount of each component are converted into control commands for the corresponding drive unit. According to the physical dependencies and safety order between the adjustment actions of each component, all generated drive unit control commands are sorted to form an ordered sequence of control commands. This sequence of control commands is sent one by one to the controller of the assembly and adjustment actuator, which then drives the corresponding drive unit to complete the physical adjustment action.
[0109] Based on the motion characteristics and control interface protocol of each drive unit in the assembly and adjustment mechanism, the adjustment direction and physical adjustment amount of each component are converted into control commands for the corresponding drive unit. The specific steps are as follows: Query the model and specifications of the drive unit corresponding to the name of the component to be adjusted in the assembly and adjustment mechanism. From the model and specifications, obtain the conversion coefficient for converting physical displacement to control pulse count, and the mapping relationship for converting adjustment direction to control signal level. Using the conversion coefficient, calculate the required number of control pulses for the physical adjustment amount. Using the mapping relationship, convert the adjustment direction into a specific control signal level value. Encapsulate the number of control pulses and the control signal level value according to the data format specified in the control interface protocol to generate a single control command that the corresponding drive unit can recognize and execute.
[0110] In practical implementation, the calibration strategy is converted into a sequence of control commands to drive the assembly and adjustment actuator to physically adjust the infrared lens. The process is as follows: The calibration strategy is parsed to extract the names of the components requiring adjustment, the adjustment direction of each component, and the physical adjustment amount. Based on the motion characteristics and control interface protocol of each drive unit in the assembly and adjustment actuator, the adjustment direction and physical adjustment amount of each component are converted into control commands for the corresponding drive unit. According to the physical dependencies and safety sequence between the adjustment actions of each component, all generated drive unit control commands are sorted to form an ordered sequence of control commands. The control command sequence is sent one by one to the controller of the assembly and adjustment actuator, which then drives the corresponding drive unit to complete the physical adjustment action.
[0111] Based on the motion characteristics and control interface protocol of each drive unit in the assembly and adjustment mechanism, the adjustment direction and physical adjustment amount of each component are converted into control commands for the corresponding drive unit. The specific steps are as follows: 1. Locate the model and specification manual of the drive unit in the assembly and adjustment mechanism corresponding to the name of the component to be adjusted. 2. Obtain the conversion coefficient for converting physical displacement to control pulse count, and the mapping relationship for converting adjustment direction to control signal level, from the model and specification manual. 3. Calculate the required number of control pulses for the physical adjustment amount using the conversion coefficient. 4. Convert the adjustment direction into a specific control signal level value using the mapping relationship. 5. Encapsulate the number of control pulses and the control signal level value according to the data format specified in the control interface protocol to generate a single control command that the corresponding drive unit can recognize and execute.
[0112] In practical implementation, physical dependency refers to the requirement that certain components must be adjusted before adjustment can begin. Optionally, the control interface protocol can be Modbus, CANopen, or a custom serial communication protocol. The ordering of the control command sequence can be determined based on the action dependencies represented by a directed acyclic graph. In some embodiments, the safety sequence requires unlocking the locking mechanism before position adjustment. Querying the model and specifications of the drive unit can be done by accessing a local or network device database. Control signal level values can be high and low levels representing forward and reverse rotation. The conversion factor obtained from the model and specifications is in pulses per millimeter. It is understood that after the control command sequence is sent, a response from the controller confirming or indicating completion is required. In some embodiments, the mapping relationship may be a lookup table mapping directional descriptions such as "clockwise" and "counterclockwise" to specific level combinations. Optionally, the encapsulated control commands typically include a start bit, address code, function code, data field, and check bit. In practical implementation, the formula for calculating the number of control pulses from the physical adjustment amount can be expressed as:
[0113]
[0114] in: This represents the calculated number of control pulses required, and is a unitless positive integer. This represents the conversion factor obtained from the model and specifications of the drive unit, and its unit is pulses per millimeter (pulse / mm). This represents the physical adjustment amount extracted from the calibration strategy, and its unit is millimeters (mm). The symbol is... This indicates rounding to the nearest integer. (Physical adjustment amount) Multiply by conversion factor The theoretical pulse count is obtained, and after rounding, the final integer number of pulses used to generate control commands is obtained. Control the number of pulses. It is encapsulated together with the control signal level value into an instruction.
[0115] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.
Claims
1. A method for calibrating the assembly accuracy of infrared lenses based on industrial vision, characterized in that, The method includes: An industrial camera is used to acquire optical image sequences of an infrared lens in the assembly and adjustment state. The optical image sequences are then preprocessed to extract a feature image set containing lens edge features, internal optical element markers, and background reference references. The feature image set is processed by a feature matching algorithm to generate a set of feature-corresponding point pairs; Based on the set of feature-corresponding point pairs, calculate the set of measured parameters for the optical center, optical axis direction, and lens spacing of the infrared lens; Obtain the standard values of the design parameters of the infrared lens, including the standard position of the optical center, the standard direction of the optical axis, and the standard distance between each lens; The measured parameter set and the standard values of the design parameters are input into the parameter error analysis model to calculate and generate a multi-dimensional assembly and adjustment error vector, including: The parameter error analysis model establishes a mapping relationship from the measured parameter space to the standard parameter space, which is defined by the assembly and adjustment parameter error model of the infrared lens. The measured parameter set is input into the mapping relationship to calculate the equivalent measured parameters in the standard parameter space. The difference between the equivalent measured parameters and the standard values of the design parameters is calculated in each parameter dimension. The differences in each parameter dimension are normalized and dimensionally unified. The processed differences are arranged in a preset order to form the multi-dimensional assembly and adjustment error vector. Each element in the multi-dimensional assembly and adjustment error vector corresponds to a specific type of assembly and adjustment error. The parameter error analysis model establishes a mapping relationship from the measured parameter space to the standard parameter space, including: The mapping relationship is represented by a set of nonlinear equations, the independent variables of which are basic assembly error variables including lens displacement, tilt, and lens spacing deviation. The measured position of the optical center, the measured value of the optical axis direction, and the measured value of the lens spacing in the measured parameter set are taken as known quantities of the nonlinear equation set. The nonlinear equation set is solved by numerical iteration method to obtain a set of numerical solutions for the basic assembly error variables. The numerical solutions for the basic assembly error variables represent the equivalent representation of the measured parameter set in the standard parameter space. The equivalent representation is the equivalent measured parameters. Based on the multi-dimensional assembly and adjustment error vector, a calibration strategy including adjustment components, adjustment direction, and adjustment amount is generated through the assembly and adjustment decision model; The calibration strategy is converted into a sequence of control commands, which drives the assembly and adjustment mechanism to physically adjust the infrared lens.
2. The infrared lens assembly and adjustment accuracy calibration method based on industrial vision according to claim 1, characterized in that, The feature image set is processed by a feature matching algorithm to generate a set of feature-corresponding point pairs, including: The feature image set is decomposed into a multi-scale Gaussian pyramid to generate image pyramid layers of multiple scales; In each layer of the image pyramid, accelerated robust feature descriptors are extracted, and preliminary feature point matching is performed using the feature matching algorithm to form a set of candidate matching pairs. When calculating the similarity of feature descriptors, the feature matching algorithm introduces consistency verification based on the local geometric context of feature points, and constructs a geometric consistency constraint function to filter candidate matching pairs based on the structural characteristics of the lens edge and marker points of the infrared lens image. For the filtered candidate matching pair set, feature point trajectory tracking and bidirectional consistency verification are performed between image pyramid layers at adjacent scales, and finally, the cross-scale stable set of feature corresponding point pairs is output.
3. The infrared lens assembly and adjustment accuracy calibration method based on industrial vision according to claim 2, characterized in that, The feature matching algorithm, when calculating the similarity of feature descriptors, introduces consistency verification based on the local geometric context of feature points, and constructs a geometric consistency constraint function to filter candidate matching pairs based on the structural characteristics of the lens edges and marker points in the infrared lens image, including: Centered on each feature point in the candidate matching pair set, a local neighborhood window with a preset radius is selected in the image pyramid layer to which it belongs; Within the local neighborhood window, the spatial relative positional relationship between the current feature point and other surrounding feature points is extracted, and a geometric context descriptor describing the local spatial distribution of the current feature point is calculated and generated. For a pair of candidate matching feature points, calculate their geometric context descriptors respectively, and compare the difference between the two descriptors. If the difference exceeds a preset first threshold, the local geometric context is considered inconsistent, and the candidate matching pair is removed from the candidate matching pair set. Based on the structural characteristics of the infrared lens image, where the lens edge appears as a smooth elliptical arc and the marker points are usually regular dots or crosses, the geometric consistency constraint function is constructed. The geometric consistency constraint function is used to quantify whether the local image structure of two feature points in a pair of candidate matches conforms to the curvature continuity constraint of the lens edge or the neighborhood symmetry constraint of the marker point. Calculate the geometric consistency constraint function value of each candidate pair in the candidate matching pair set. If the function value is lower than a preset second threshold, the corresponding candidate matching pair is considered to be inconsistent with the prior knowledge of the structure of the infrared lens, and the corresponding candidate matching pair is filtered out from the candidate matching pair set.
4. The infrared lens assembly and adjustment accuracy calibration method based on industrial vision according to claim 1, characterized in that, Based on the set of feature-corresponding point pairs, the measured parameter set of the optical center, optical axis direction, and lens spacing of the infrared lens is calculated, including: From the set of feature corresponding point pairs, separate the feature point pairs belonging to the lens edge contour for the feature point pairs belonging to the internal optical element marker points; Using feature point pairs belonging to the lens edge contour, the ellipse equation of the lens outer edge in the image coordinate system is calculated by an ellipse fitting algorithm. The coordinates of the center point of the ellipse equation are then transformed into the coordinate system and used as the measured position of the optical center. Using feature point pairs belonging to the internal optical element markers, combined with the known three-dimensional design coordinates of the markers, the attitude of the infrared lens relative to the industrial camera is calculated by the perspective N-point positioning algorithm. The line of sight extracted from the attitude is inverted and used as the measured value of the optical axis pointing. By utilizing feature point pairs belonging to markers on different lenses, and based on the calculated spatial position of each lens, the measured value of the lens spacing is obtained by calculating the spatial distance. All measured values together constitute the set of measured parameters.
5. The infrared lens assembly and adjustment accuracy calibration method based on industrial vision according to claim 1, characterized in that, Based on the multi-dimensional assembly and adjustment error vector, a calibration strategy including adjustment components, adjustment direction, and adjustment amount is generated through the assembly and adjustment decision model, including: The assembly and adjustment decision model includes an error vector parsing module and a strategy mapping knowledge base; The error vector parsing module receives the multi-dimensional assembly and adjustment error vector and decomposes it into sub-error components related to specific adjustable parts. The strategy mapping knowledge base stores historical calibration data, which records the successful calibration strategies corresponding to different combinations of sub-error components. In the strategy mapping knowledge base, the most similar historical calibration strategies are retrieved using the sub-error components as query conditions. Based on the retrieved historical calibration strategies, a calibration strategy suitable for the current error state is generated through strategy fusion and optimization algorithms. The calibration strategy clearly indicates the names of the components that need to be adjusted, the adjustment direction of each component, and the physical adjustment amount.
6. The infrared lens assembly and adjustment accuracy calibration method based on industrial vision according to claim 5, characterized in that, The calibration strategy based on the retrieved historical calibration strategies is generated through a strategy fusion and optimization algorithm to be suitable for the current error state, including: For each retrieved historical calibration strategy, calculate the Euclidean distance between its corresponding historical sub-error component and the current sub-error component; Based on the Euclidean distance, a fusion weight is assigned to each historical calibration strategy, with a larger fusion weight assigned to the historical calibration strategy with a smaller distance. The adjustment direction and adjustment amount in the historical calibration strategy are weighted and averaged according to their fusion weights to obtain the preliminary fused adjustment direction and adjustment amount; The initially fused adjustment direction and adjustment amount are input into a physical model-based optimizer. The optimizer aims to minimize the multi-dimensional assembly and adjustment error vector, fine-tunes the adjustment direction and adjustment amount, outputs the final determined adjustment direction and adjustment amount, and integrates them to form the calibration strategy.
7. The infrared lens assembly and adjustment accuracy calibration method based on industrial vision according to claim 1, characterized in that, The calibration strategy is converted into a sequence of control commands to drive the assembly and adjustment mechanism to physically adjust the infrared lens, including: The calibration strategy is analyzed to extract the names of the components that need to be adjusted, the adjustment direction of each component, and the physical adjustment amount. Based on the motion characteristics and control interface protocol of each drive unit in the assembly and adjustment mechanism, the adjustment direction and physical adjustment amount of each component are converted into control commands for the corresponding drive unit. Based on the physical dependencies and safety sequence between the adjustment actions of each component, all generated drive unit control commands are sorted to form an ordered sequence of control commands; The control command sequence is sent one by one to the controller of the assembly and adjustment mechanism, and the controller drives the corresponding drive unit to complete the physical adjustment action.
8. The infrared lens assembly and adjustment accuracy calibration method based on industrial vision according to claim 7, characterized in that, Based on the motion characteristics and control interface protocol of each drive unit in the assembly and adjustment mechanism, the adjustment direction and physical adjustment amount of each component are converted into control commands for the corresponding drive unit, including: Search for the model and specifications of the drive unit corresponding to the name of the component that needs to be adjusted in the assembly and adjustment mechanism; From the model and specification document, obtain the conversion coefficients for the corresponding drive unit to convert physical displacement into control pulse count, and the mapping relationship for converting adjustment direction into control signal level; Using the conversion coefficient, the physical adjustment amount is calculated as the required number of control pulses; Using the mapping relationship, the adjustment direction is converted into a specific control signal level value; The number of control pulses and the level value of the control signal are encapsulated according to the data format specified in the control interface protocol to generate a single control instruction that can be recognized and executed by the corresponding drive unit.
Citation Information
Patent Citations
Lens assembly coupling method and system
CN118426130A
Wide-angle lens transformation method based on 3D space reconstruction
CN121563851A
