A stereovision method and system for detecting internal defects in a multilayer transparent material

Abstract

This invention discloses a stereoscopic vision detection method and system for internal defects in multilayer transparent materials, relating to the fields of machine vision and optical inspection technology. The method includes system calibration, initial localization, refraction correction, and layer number determination. System calibration acquires the camera's intrinsic and extrinsic parameters; initial localization obtains the apparent three-dimensional coordinates of the defect through feature matching and triangulation; refraction correction uses the apparent coordinates as initial values, performs forward ray tracing, calculates the refraction direction according to Snell's law, constructs an optimization objective function, and uses an iterative optimization algorithm to solve for the true three-dimensional coordinates; layer number determination identifies the layer where the defect is located. This invention, by considering the light refraction effect, achieves high-precision three-dimensional localization and layer number determination of defects, possessing advantages such as non-destructiveness and wide applicability, and can meet the quality control requirements of high-end manufacturing industries.

Description

Technical Field

[0001] This invention relates to the fields of machine vision and optical inspection technology, particularly for the precise three-dimensional localization of internal defects in industrial ultrathin glass, multilayer composite glass, and other transparent layered materials. This method combines image processing, computer vision, artificial intelligence, and industrial automation technologies with stereo vision and optical refraction correction to provide a high-precision, non-contact solution for detecting internal defects in transparent materials. Background Technology

[0002] In high-end manufacturing industries, such as display panels, specialty glass, and semiconductor packaging, the application of multilayer transparent materials is becoming increasingly widespread. During the production process, these materials (such as multilayer glass and laminates) inevitably develop micro-defects, such as bubbles, impurities, and cracks. The presence of these defects can severely affect the optical performance, mechanical strength, and ultimate lifespan of the product.

[0003] Currently, defect detection in transparent materials typically employs optical imaging methods. However, existing technologies generally face a core challenge: when using a camera for inspection, because light passes through multiple layers of media with different refractive indices, the defect observed in the image is merely a two-dimensional projection. While single-camera or traditional binocular stereo vision systems can detect defects, they neglect the refraction of light as it exits the material surface, leading to significant errors in three-dimensional localization. This error prevents the detection system from accurately determining the specific layer in which the defect resides (e.g., whether it's in the protective glass layer, the functional adhesive layer, or the core substrate glass layer). If the depth (Z-axis) coordinates of the defect cannot be determined, it's impossible to assess whether the defect affects the core functional layers of the product, leading to quality control difficulties. This could result in products that are not unusable being mistakenly classified as defective, or products with fatal defects entering the market.

[0004] Therefore, the industry urgently needs a technical solution that can penetrate multiple layers of transparent materials, accurately measure the three-dimensional spatial coordinates of internal defects, and determine the physical layer in which they are located. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a method and system for high-precision three-dimensional localization of internal defects in multilayer transparent materials. This method combines stereoscopic vision triangulation with a refraction effect correction algorithm based on a physical model to solve the localization error problem caused by light refraction, thereby achieving accurate determination of the true physical location of the defect and identifying its specific material layer.

[0006] This invention proposes a stereoscopic vision detection method for internal defects in multilayer transparent materials, comprising the following steps:

[0007] Step S1: System calibration. Obtain the internal and external parameters of at least two cameras in the stereo vision system. The internal parameters include the intrinsic parameter matrix and distortion coefficients, and the external parameters include the rotation matrix and translation vector. By accurately obtaining the camera's internal and external parameters, an accurate camera imaging geometric model is established, reducing the defect positioning deviation caused by camera parameter errors and ensuring the accuracy of the conversion from three-dimensional space to two-dimensional image projection.

[0008] Step S2: Initial positioning. Match the same defect point in the images acquired by the at least two cameras, and calculate the apparent three-dimensional coordinates of the defect point using a triangulation algorithm. Utilize the parallax information from the multi-camera perspective to quickly obtain the approximate three-dimensional position of the defect, providing an initial guess close to the true value for subsequent refraction correction. This avoids the blindness of directly performing complex refraction calculations and improves the efficiency of the overall detection process.

[0009] Step S3: Refraction Correction. Using the apparent 3D coordinates as initial values, and based on the thickness and refractive index of each layer of the multilayer transparent material, forward ray tracing is performed. Starting from the currently guessed true defect location, an initial ray is emitted towards the optical center of the camera. The refraction direction of the ray after passing through the interface of each layer of the medium is calculated according to Snell's law. An optimization objective function is constructed, which is the reprojection error between the predicted pixel and the actual measured pixel. The true 3D coordinates are solved through an iterative optimization algorithm to minimize the reprojection error. The ray tracing is used to simulate the ray propagation path, and the refraction direction is accurately calculated using Snell's law. The difference between the prediction and the actual value is quantified using the optimization objective function, and the positioning error caused by ray refraction is eliminated through an iterative optimization algorithm, significantly improving the accuracy of the 3D defect positioning and making the positioning result closer to the true physical location.

[0010] Step S4: Layer Count Determination. The final calculated true 3D coordinates are compared with the physical boundaries of each layer of the material to determine the layer where the defect is located. Accurately determining the specific physical layer where the defect is located provides a direct and crucial basis for product quality assessment, avoiding misjudgments caused by the inability to determine the number of defective layers.

[0011] Furthermore, in the refraction correction step, the initial ray direction vector for forward ray tracing is: Where OL is the camera's optical center and Preal is the currently guessed true defect location, ensuring the accuracy of the starting point and direction of forward ray tracing and improving the accuracy of ray propagation path simulation.

[0012] Furthermore, in the refraction correction step, the direction vector after refraction is calculated according to the vector form of Snell's law. The vector form of Snell's law is: ;in Let η be the direction vector after refraction, and η be the ratio of the refractive indices of the two media, i.e., η = n1 / n2. The direction of the incident ray. This represents the interface normal vector. Employing the vector form of Snell's law allows for more accurate calculation of the refraction direction of light at the interface of different media. It considers the spatial vector characteristics of light propagation, further improving the accuracy of refraction direction calculation compared to the scalar form, thereby enhancing the precision of defect location.

[0013] Furthermore, the optimization objective function is: , where p L p R p represents the pixel coordinates of the actual measured defect point on the imaging planes of the left and right cameras. L ’ p R ’ The objective function represents the predicted pixel coordinates on the imaging planes of the left and right cameras after refraction, originating from the true 3D coordinates. This objective function quantitatively describes the difference between the predicted and actual measured pixels, providing a clear direction for iterative optimization. By minimizing this function value, the true 3D coordinates of the defect can be continuously approximated, ensuring the accuracy of the refraction correction results.

[0014] Furthermore, the iterative optimization algorithm employs a nonlinear optimization algorithm, which minimizes the objective function by iteratively updating the guessed values ​​of the true 3D coordinates. Nonlinear optimization algorithms can effectively handle complex nonlinear problems caused by light refraction. By continuously iterating and updating the guessed values ​​of the true three-dimensional coordinates, the objective function is minimized, thereby accurately solving for the true three-dimensional coordinates of the defect and further improving the positioning accuracy.

[0015] Furthermore, in the system calibration step, the camera intrinsic parameters are calibrated using the following formula: Where (u,v) are pixel coordinates, (X... C ,Y C Z C ) represents the 3D point coordinates in the camera coordinate system, s is the scale factor, and (f x ,f y (C) is the focal length. x C y () is the main point.

[0016] This formula accurately establishes the mapping relationship between three-dimensional points and image pixel coordinates in the camera coordinate system. By obtaining intrinsic parameters such as focal length and principal point through calibration, the accuracy of the transformation from three-dimensional points to two-dimensional pixel coordinates is ensured.

[0017] Furthermore, the triangulation algorithm is implemented by solving the following overdetermined system of equations: and , where s L and sR As the scaling factor, (u L ,v L ) and (u R ,v R K represents the pixel coordinates of the defect point in the left and right images, respectively. L and K R Let P be the intrinsic parameter matrix of the left and right cameras, R be the rotation matrix, and t be the translation vector. By solving the overdetermined equations and comprehensively utilizing the image information and camera parameters of the left and right cameras, the apparent 3D coordinates of the defect can be calculated more accurately, reducing the error caused by a single viewpoint.

[0018] This invention also provides a stereoscopic vision inspection system for internal defects in multilayer transparent materials, comprising: a light source module: located at the bottom layer, a programmable LCD or OLED screen, used to provide uniform backlighting upwards; a product support module for horizontally placing the multilayer transparent material to be tested; a stereoscopic vision module: at least two cameras, symmetrically arranged in a "V"-shaped convergence layout above the product, their optical axes intersecting at or near the plane of the product, forming a convergence angle of 30° to 60°, used to synchronously acquire product images; and a processing module connected to the stereoscopic vision module. The light source module provides uniform and programmable backlighting, ensuring clear defect imaging and improving defect recognition rate; the product support module stably places the material to be tested, preventing positional changes from affecting the detection results; the "V"-shaped convergence layout and suitable convergence angle of the stereoscopic vision module ensure sufficient parallax and measurement range, and synchronous acquisition ensures image time consistency; the processing module executes corresponding detection methods to achieve automated and high-precision defect detection. The collaborative work of each module improves the overall detection performance and reliability of the system.

[0019] In this invention, system calibration provides accurate camera parameters for initial positioning, the apparent coordinates of initial positioning serve as the initial values ​​for refraction correction, and the true coordinates obtained from refraction correction provide a basis for layer number determination, thus collaboratively realizing a complete detection process from coarse positioning to precise positioning and then to layer number determination.

[0020] In refraction correction, forward ray tracing, the application of Snell's law, the construction of the objective function, and the iterative optimization algorithm work together. Forward ray tracing provides the path for refraction calculation, Snell's law ensures the accuracy of the refraction direction calculation, the objective function provides the direction for iterative optimization, and the iterative optimization algorithm efficiently finds the optimal solution. Together, they eliminate refraction errors and achieve high-precision positioning.

[0021] The nonlinear screen appearance defect evaluation method of the present invention has the following technical effects:

[0022] High-precision positioning: By introducing a refraction correction model based on Snell's law, combined with forward ray tracing, optimization of objective function construction, and Levenberg-Marquardt iterative optimization algorithm, the measurement error caused by optical path refraction is greatly eliminated, achieving sub-millimeter or even micrometer-level three-dimensional positioning accuracy for internal defects, which is more accurate than traditional methods.

[0023] Precise layer determination: Based on high-precision depth coordinates, it can reliably determine the specific physical layer where the defect is located, providing a direct and crucial basis for product quality assessment, avoiding misjudgment and omission, and improving the accuracy of quality control.

[0024] Non-destructive testing: The entire testing process is a non-contact optical measurement that will not cause any physical damage to the product. It can be used for testing various precision and fragile multilayer transparent materials without affecting the subsequent use of the product.

[0025] Wide applicability: It is suitable for complex layered transparent materials containing multiple layers of glass and multiple layers of adhesive with different refractive indices. It can play a role in many fields such as display panels, special glass, and semiconductor packaging, and has a wide range of application prospects.

[0026] High degree of automation: The processing module executes a series of algorithms to achieve automated detection from image acquisition to result output, reducing manual intervention, improving detection efficiency, and meeting the batch detection needs in industrial production.

[0027] To more clearly illustrate the structural features and effects of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0028] Figure 1 This is a side view of a stereoscopic vision inspection system for internal defects in multilayer transparent materials according to the present invention.

[0029] Figure 2 This is a top view of a stereoscopic vision inspection system for internal defects in multilayer transparent materials according to the present invention. Detailed Implementation

[0030] The present invention will now be further described in conjunction with the accompanying drawings and relevant knowledge, and will be described clearly and completely. Obviously, the described applications are only some embodiments of the present invention, and not all embodiments.

[0031] Example 1: This invention provides a stereoscopic vision detection method for internal defects in multilayer transparent materials, comprising the following steps:

[0032] System calibration: Obtain the intrinsic and extrinsic parameters of at least two cameras in the stereo vision system. The intrinsic parameters include intrinsic parameter matrices and distortion coefficients, and the extrinsic parameters include rotation matrices and translation vectors.

[0033] Specifically, camera intrinsic parameter calibration involves obtaining the intrinsic parameter matrix K and distortion coefficients for each camera. The intrinsic parameter matrix describes the projection relationship from 3D points to a 2D image.

[0034] ;

[0035] Where (u,v) are pixel coordinates, (X) C ,Y C Z C ) represents the 3D point coordinates in the camera coordinate system, s is the scale factor, and (f x ,f y (C) is the focal length. x C y () is the main point.

[0036] Extrinsic parameter calibration: This involves determining the relative positional relationship between the two camera coordinate systems, i.e., the rotation matrix R and the translation vector t. The coordinates P of a 3D point in the left and right camera coordinate systems are also determined. L and P R Satisfying Relationship:

[0037] P R =R·P L +t;

[0038] Initial localization: Match the same defect point in the images acquired by the at least two cameras, and calculate the apparent three-dimensional coordinates of the defect point using a triangulation algorithm;

[0039] Specifically, this step is used to obtain an initial three-dimensional coordinate of the defect. Feature matching: In the images acquired by the two cameras, the same physical defect point is matched to obtain its pixel coordinates P in the left and right images. L =(u L ,v L ) and P R =(u R ,v R );

[0040] Furthermore, triangulation: Based on the calibrated parameters, an overdetermined system of equations is solved to find the spatial point P. When the world coordinate system is set at the left camera, the system of equations is as follows:

[0041] ;

[0042] ;

[0043] Solving this system of equations using the least squares method yields the apparent coordinates P, neglecting refraction. apparent .

[0044] Refraction Correction: Using the apparent 3D coordinates as initial values, and based on the thickness and refractive index of each layer of the multilayer transparent material, forward ray tracing is performed. Starting from the currently guessed true defect location, an initial ray is emitted towards the camera's optical center. The refraction direction of the ray after passing through the interface of each layer is calculated according to Snell's law. An optimization objective function is constructed, which is the reprojection error between the predicted pixel and the actual measured pixel. The true 3D coordinates are solved through an iterative optimization algorithm to minimize the reprojection error.

[0045] Specifically, this step is the core innovation of this invention, which uses iterative optimization to solve for the true physical location P of the defect. real Physical model: The thickness d of each layer of the product under test is known. i and refractive index n i Each interface layer is a parallel plane, and its normal vector is... .

[0046] More specifically: Iterative optimization process:

[0047] Initial conjecture: The obtained apparent coordinates P apparent As the real physical location P real The initial guess value.

[0048] Forward ray tracing: from the currently guessed true defect location P real Start from the camera's optical center (e.g., left camera O). L (Emit an initial ray with direction vector) .

[0049] Applying Snell's Law: The direction vector of a ray changes every time it passes through a medium interface. Calculate the refracted direction vector using the vector form of Snell's Law. If the direction of the incident ray is The interface normal vector is Let the refractive indices of the two media be n1 and n2 respectively (let η = n1 / n2), then:

[0050] This calculation starts from the layer where the defect is located and traces it outward layer by layer until the light enters the air.

[0051] Calculate the predicted pixel coordinates: The final ray of light, after refraction through all interfaces, intersects the camera's imaging plane, forming a predicted pixel coordinate. This process is performed on both the left and right cameras to obtain a pair of predicted coordinates (p...). L ’ ,p R’ ).

[0052] Constructing the optimization objective function: The goal of the optimization is to find a true coordinate P. real The objective function is to minimize the reprojection error between the predicted pixels and the actual measured pixels. real ) is defined as:

[0053] ;

[0054] Iterative optimization: Employ nonlinear optimization algorithms (such as Levenberg-Marquardt) to minimize the objective function E(P). real With the goal of continuously iterating and updating P real The guessed value. The final result is obtained when the error converges to its minimum value; the optimization process terminates at this point. This refers to the true and accurate three-dimensional coordinates of the defect inside the product after taking into account the refraction effect.

[0055] Layer number determination: The final calculated true 3D coordinates are compared with the physical boundaries of each layer of the material to determine the layer where the defect is located. Specifically, the defect layer number determination uses the final obtained true depth coordinates Z... real By comparing the defect with the known depth boundaries of each material layer, it can be determined which physical layer of the product the defect is located in.

[0056] By introducing a refraction correction model based on Snell's law and combining it with a nonlinear optimization algorithm, this invention greatly eliminates the measurement error caused by optical path refraction and achieves sub-millimeter or even micrometer-level three-dimensional positioning accuracy for internal defects.

[0057] Based on high-precision depth coordinates, this invention can reliably determine the specific physical layer where the defect is located, providing a direct and crucial basis for product quality assessment and avoiding misjudgment and omission.

[0058] The entire testing process of this invention is a non-contact optical measurement, which will not cause any physical damage to the product. It is not only applicable to double-glazed glass, but also to complex layered transparent materials containing multiple layers of glass and multiple layers of adhesive with different refractive indices.

[0059] Example 2, refer to Figure 1 , Figure 2As shown, the present invention also provides a stereoscopic vision inspection system for internal defects of multilayer transparent materials, comprising: a light source module: disposed at the bottom layer, which is a programmable LCD or OLED screen for providing uniform backlighting upwards; a product support module for horizontally placing the multilayer transparent material to be tested; a stereoscopic vision module: at least two cameras, symmetrically arranged in a "V"-shaped convergence layout above the product, with their optical axes intersecting at or near the plane where the product is located, forming a convergence angle of 30° to 60°, for synchronously acquiring product images; and a processing module: connected to the stereoscopic vision module.

[0060] In this invention, the detection system adopts a transmission-type stereotactic layout, with its core components arranged vertically, from top to bottom as follows:

[0061] Camera layer: Two or more high-resolution industrial cameras are arranged in a "V"-shaped convergent three-dimensional layout. The two cameras are distributed to the left and right of the product center as the axis of symmetry, and their optical axes are tilted inward and converge near the plane of the product being measured, forming a convergence angle of 30° to 60°. A fixed horizontal distance, i.e., baseline, is maintained between the optical centers of the two cameras.

[0062] Product layer: The multi-layer transparent material product to be tested is placed horizontally on a high-rigidity, optically transparent support platform.

[0063] Light source layer: An LCD or OLED display screen capable of displaying high-frequency stripe patterns, placed horizontally at the bottom of the system, serving as a uniform, programmable backlight.

[0064] Specific reference Figure 1 The side view of the detection system of this invention shows the vertical "V"-shaped convergence layout of the camera, product, and light source. Figure 2 The top view of the detection system of this invention shows the baseline distance between the two cameras and their overlapping stereo measurement area.

[0065] exist Figure 1 The image shows the core structure of the system from the side. At the bottom is a display screen 1, which serves as a backlight. The multi-layered glass product 2 to be tested is placed on the screen, and a defect point 3 is marked inside the product. At the top are a first camera 4 and a second camera 5, mounted in an inverted "V" shape, with a convergence angle 6 so that their fields of view are both focused on the product. Figure 1 The diagram schematically illustrates the path of light 7 emanating from the defect point to the two cameras.

[0066] Figure 2 The system's planar layout is shown from directly above. There is a horizontal baseline distance between the first camera 4 and the second camera 5. Each camera has its own field of view (FOV1, FOV2), and the overlapping area 8 of the two fields of view is the effective stereo measurement area, within which the defect must be located.

[0067] Example 3: A system for detecting internal defects in mobile phone cover glass, the hardware of which is based on... Figure 1 and Figure 2 The setup is as shown.

[0068] System components: Cameras: Two industrial cameras with a resolution of 5120x5120 pixels, namely the first camera 4 and the second camera 5.

[0069] Light source: A high-brightness OLED screen (display screen 1).

[0070] Product 2 to be tested: A multi-layered glass plate consisting of two 0.5mm thick glass layers bonded together with a 0.1mm thick OCA adhesive layer. The refractive index of each layer is known.

[0071] Implementation process:

[0072] 1. System calibration:

[0073] First, place a high-precision checkerboard calibration plate on the product support platform.

[0074] The first and second cameras were activated to capture images of the calibration board from 15 different poses.

[0075] Use calibration software (such as functions in the OpenCV library) to calculate the intrinsic parameter matrices K1, K2 and distortion coefficients of the two cameras respectively.

[0076] Using images captured in the same session, calculate the rotation matrix R and translation vector t from the first camera coordinate system to the second camera coordinate system. Save all calibration parameters.

[0077] 2. Defect image acquisition and initial localization:

[0078] Remove the calibration plate and place the multilayer glass plate to be tested on the support platform.

[0079] An OLED screen displays a pure white background, and two cameras simultaneously expose the image to capture images containing internal defects.

[0080] In the left and right images, the center of the defect is found using image processing algorithms (such as blotch detection), and its pixel coordinates P are obtained. L and P R .

[0081] Call the triangulation function, input the calibration parameters and pixel coordinate pairs (P L P R ), calculate the apparent three-dimensional coordinates P of the defect. apparent Assume the calculated result is (10.25, 20.41, -0.85) mm.

[0082] Refraction correction and precise positioning:

[0083] Start the refractive correction optimization program. Set P... apparent As the real coordinates P real The initial value.

[0084] The program enters an iterative loop, minimizing the reprojection error function using the Levenberg-Marquardt algorithm. .

[0085] In each iteration, the program adjusts the current P... real The guessed value is used to perform forward ray tracing using the vector form of Snell's law to calculate the predicted pixel P. L ’ and P R and update P real .

[0086] After the loop ends, the final optimized result is obtained. For example, the final coordinates are (10.23, 20.39, -0.58) mm.

[0087] Layer number determination:

[0088] Given that the Z-coordinate of the upper surface of the product is 0, the thickness of the first glass layer is 0.5mm, the thickness of the adhesive layer is 0.1mm, and the thickness of the second glass layer is 0.5mm.

[0089] The Z coordinates of each layer boundary are: 0mm (upper surface), -0.5mm (glass 1 - adhesive layer), -0.6mm (adhesive layer - glass 2), and -1.1mm (lower surface).

[0090] The final defect depth coordinate Z real =-0.58mm. Since -0.6 < -0.58 < -0.5, the system determines that the defect is located within the middle OCA adhesive layer. This result will be recorded and used for subsequent quality control decisions.

[0091] The technical principles of the present invention have been described above with reference to specific embodiments, which are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments; all technical solutions falling within the scope of the present invention's concept are within its protection scope. Those skilled in the art can conceive of other specific embodiments of the present invention without creative effort, and these embodiments will all fall within the protection scope of the present invention.

Claims

1. A stereoscopic vision detection method for internal defects in multilayer transparent materials, characterized in that, Includes the following steps: Step S1: Obtain the intrinsic and extrinsic parameters of at least two cameras in the stereo vision system. The intrinsic parameters include the intrinsic parameter matrix and distortion coefficients, and the extrinsic parameters include the rotation matrix and translation vector. Step S2: Match the same defect point in the images acquired by the at least two cameras, and calculate the apparent three-dimensional coordinates of the defect point using a triangulation algorithm; Step S3: Using the apparent 3D coordinates as initial values, and based on the thickness and refractive index of each layer of the multilayer transparent material, perform forward ray tracing. Starting from the currently guessed true defect location, emit an initial ray towards the camera's optical center. Calculate the refraction direction of the ray after passing through the interfaces of each medium layer according to Snell's law. Construct an optimization objective function, where the objective function is the reprojection error between the predicted pixel and the actual measured pixel. Solve for the true 3D coordinates through an iterative optimization algorithm to minimize the reprojection error. The initial ray direction vector for forward ray tracing is... Where OL is the optical center of the camera, and Preal is the currently guessed true defect location. The direction vector after refraction is calculated according to the vector form of Snell's law. The vector form of Snell's law is: ;in Let η be the direction vector after refraction, and η be the ratio of the refractive indices of the two media, i.e., η = n1 / n2, where n1 and n2 are the refractive indices of the two media. The direction of the incident ray. Let be the interface normal vector; the optimization objective function is: , where pL and pR are the actual measured pixel coordinates of the defect point on the imaging plane of the left and right cameras, and pL′ and pR′ are the predicted pixel coordinates on the imaging plane of the left and right cameras after refraction from the real three-dimensional coordinates; Step S4: Compare the final obtained true three-dimensional coordinates with the physical boundaries of each layer of the material to determine the layer where the defect is located.

2. The method according to claim 1, characterized in that, The iterative optimization algorithm employs a nonlinear optimization approach, iteratively updating the guessed values ​​of the true 3D coordinates to minimize the objective function. .

3. The method according to claim 1, characterized in that, In step S1, the camera's internal parameters are calibrated using the following formula: ; in, These are pixel coordinates. These are the coordinates of a three-dimensional point in the camera coordinate system. K is the scale factor, and K is the camera intrinsic parameter matrix. It's the focal length. This is the main point.

4. The method according to claim 1, characterized in that, The trigonometric measurement algorithm is implemented by solving the following overdetermined system of equations: and , where s L and s R As the scaling factor, (u L ,v L ) and (u R ,v R K represents the pixel coordinates of the defect point in the left and right images, respectively. L and K R Let P be the intrinsic parameter matrices of the left and right cameras, respectively, where P is the spatial point, R is the rotation matrix, and t is the translation vector.

5. A stereoscopic vision inspection system for internal defects in multilayer transparent materials, characterized in that, Includes: Light source module: located at the bottom layer, it is a programmable LCD or OLED screen used to provide uniform backlighting upwards; The product support module is used to horizontally place the multi-layer transparent material to be tested; the stereo vision module consists of at least two cameras symmetrically arranged in a "V"-shaped convergence layout above the product, with their optical axes intersecting at or near the plane where the product is located, forming a convergence angle of 30° to 60°, for synchronously acquiring product images; the processing module is connected to the stereo vision module and is used to execute the detection method described in any one of claims 1-4.

Images