A car paint simulation method and system supporting dynamic forward and reverse bidirectional rendering

Abstract

The application relates to the technical field of automobile painting, in particular to a car paint simulation method and system supporting dynamic forward and reverse bidirectional rendering, which comprises: forward rendering: inputting car paint formula detail information and analyzing paint layer rendering parameters and color master rendering parameters to output a rendering result in a car paint renderer; reverse rendering: inputting original measured data and performing reverse parameter analysis and parameter optimization to obtain car paint physical parameters; forward and reverse fusion: taking the car paint physical parameters as the analysis result of the newly input car paint formula detail information in the forward rendering, and taking the rendering result as a prior reference in the reverse rendering to constrain the range of the car paint physical parameters; performing forward and reverse consistency checking based on the rendering result and the original measured data to calculate a rendering deviation; and performing reverse propagation based on the rendering deviation to iteratively optimize the forward and reverse renderings until the rendering deviation is smaller than a threshold value. The application can improve the accuracy and effect of car paint rendering.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of automobile painting, in particular to a car paint simulation method and system supporting dynamic forward and reverse bidirectional rendering. BACKGROUND

[0002] With the surge of demand for personalized appearance in the automobile consumer market and the acceleration of the digital transformation of the automobile manufacturing industry, accurate simulation and efficient iteration of car paint appearance have become the core demand of the industry. However, the current car paint rendering technology in the industry still has significant limitations, which is difficult to match the actual application scenarios of the automobile industry. The main problems are as follows:

[0003] Static limitation of existing renderers: The mainstream car paint renderer takes "static effect reproduction" as the core, relies on preset material templates or existing car paint effect libraries, and only supports adjusting abstract material ball parameters (such as "highlight intensity" and "metallic degree" graphic parameters), which cannot be directly associated with the actual car paint formula details (such as clear paint thickness, aluminum powder particle size distribution, and color paint component ratio) in the automobile industry.

[0004] Disconnection with industrial processes: In the automobile industry, the core of car paint research and development and color design is the dynamic adjustment of formula parameters (such as increasing the aluminum powder proportion by 5% and adjusting the clear paint thickness), rather than modifying the material parameters at the graphic level. Due to the lack of mapping modeling of formula and optical properties, the existing renderers cannot automatically update the rendering effect when the formula changes, and manual re-debugging is required, which breaks the "formula adjustment-effect preview" closed loop and seriously affects the research and development efficiency.

[0005] Lack of dynamic response capability: The existing renderers cannot support interactive editing of formula details and dynamic linkage of rendering effects. The requirements such as "adjusting color master proportion to preview effect" in new car color development and "customizing particle distribution to view iridescent effect" in personalized color design cannot be met, and the industry still needs to rely on physical sample verification, causing cost waste and being contrary to the trend of digital research and development. SUMMARY

[0006] In order to improve the accuracy and effect of car paint rendering, the application provides a car paint simulation method and system supporting dynamic forward and reverse bidirectional rendering.

[0007] In the first aspect, the application provides a car paint simulation method supporting dynamic forward and reverse bidirectional rendering, which adopts the following technical scheme:

[0008] A car paint simulation method supporting dynamic forward and reverse bidirectional rendering, comprising the following steps:

[0009] Forward rendering: input car paint formula detail information, and parse paint layer rendering parameters and color master rendering parameters to output rendering results in a car paint renderer;

[0010] inverse rendering: input original measured data and perform inverse parameter analysis and parameter optimization to obtain the car paint physical parameters;

[0011] forward and inverse fusion: taking the car paint physical parameters as the analysis result of the new input car paint formula detail information in the forward rendering, and taking the rendering result as a prior reference to constrain the range of the car paint physical parameters in the inverse rendering;

[0012] Based on the rendering result and the original measured data, a forward and inverse consistency check is performed to calculate a rendering deviation;

[0013] Based on the rendering deviation, a backward propagation is performed to iteratively optimize the forward rendering and the inverse rendering until the rendering deviation is less than a threshold.

[0014] In some embodiments, the paint layer rendering parameters are analyzed, including the following steps:

[0015] The car paint formula detail information is decomposed into a physical layer structure including a primer layer, a color paint layer, a particle layer, and a clear coat layer;

[0016] The light transmission rule of each layer is modeled to assign independent per-layer rendering parameters, and the total reflectivity is calculated based on each per-layer rendering parameter, the per-layer rendering parameter including reflectivity, refractive index, thickness, absorption coefficient, particle density, particle size distribution, and orientation angle;

[0017] The paint layer rendering parameters are generated according to the per-layer rendering parameters and the total reflectivity.

[0018] In some embodiments, the color master rendering parameters are analyzed, including the following steps:

[0019] Based on the per-layer rendering parameters corresponding to the particle layer, the physical feature parameters of each type of particle color master in the particle layer are extracted;

[0020] According to the physical feature parameters and a preset Mie scattering model, the optical characteristics of the particle color master are calculated;

[0021] Based on a preset parameterized regression algorithm, a mapping relationship between the physical feature parameters and BSDF parameters is established;

[0022] Based on the optical characteristics and the mapping relationship, the color master rendering parameters are analyzed.

[0023] In some embodiments, the original measured data includes original multi-angle spectrum and original HDR image, and the original measured data is input and inverse parameter analysis and parameter optimization are performed to obtain the car paint physical parameters, including the following steps:

[0024] pre-process the original multi-angle spectrum and the original HDR image;

[0025] input the original measured data into a preset model based on a machine learning algorithm to back-propagate each layer material parameter and particle distribution feature corresponding to the original measured data;

[0026] minimize the difference between the initial output rendering result and the original measured data through a preset differentiable rendering engine, and adjust the each layer material parameter and the particle distribution feature based on the optimization result.

[0027] In some embodiments, the rendering result includes a rendering spectrum and a rendering HDR image, and forward-backward consistency verification is performed based on the rendering result and the original measured data to calculate a rendering deviation, including the following steps:

[0028] obtain a preset consistency verification rule, and based on the type of the consistency verification rule, compare and calculate the rendering spectrum and the original multi-angle spectrum, and compare and calculate the rendering HDR image and the original HDR, the type of the consistency verification rule at least including a waveband requirement, a color difference standard, and a texture detail comparison.

[0029] In some embodiments, based on the type of the consistency verification rule, compare and calculate the rendering spectrum and the original multi-angle spectrum, and compare and calculate the rendering HDR image and the original HDR, including the following steps:

[0030] based on the waveband requirement, calculate the mean square error between the rendering spectrum and the original multi-angle spectrum to generate a spectrum error result;

[0031] based on the color difference standard, calculate the color difference index between the rendering HDR image and the original HDR image to generate a color difference result;

[0032] based on the texture detail comparison, calculate the structural similarity index between the rendering HDR image and the original HDR image to generate a texture error result;

[0033] based on the spectrum error result, the color difference result, and the texture error result, generate the rendering deviation.

[0034] In some embodiments, based on the rendering deviation, perform back-propagation to iteratively optimize the forward rendering and the inverse rendering until the rendering deviation is less than a threshold value, including the following steps:

[0035] generate initial weight coefficients corresponding to each item in the consistency verification rule;

[0036] generate an optimization objective function based on the initial weight coefficients, the consistency check rule, and the car paint formula detail information;

[0037] determine whether iteration optimization is needed based on the size of the rendering deviation;

[0038] If needed, the rendering deviation is back-propagated to the optimization objective function to adjust and correct the initial weight coefficients in inverse rendering and the car paint formula detail information in forward rendering until the rendering deviation is less than a threshold value, at which point iteration is stopped and the final car paint formula detail information is output.

[0039] In some embodiments, the calculation formula of the optimization objective function is specifically:

[0040]

[0041] wherein θ represents a car paint formula parameter, is an initial weight coefficient corresponding to each item, represents rendering spectral data rendered based on the car paint formula parameter, represents original multi-angle spectral data, represents a color difference standard result, and SSIM represents a structural similarity index.

[0042] In some embodiments, inputting original measured data and performing inverse parameter analysis and parameter optimization to obtain car paint physical parameters further includes the following steps:

[0043] creating an anchor point library, the anchor point library storing optical characteristic data and car paint formula parameters corresponding to several typical car paints, the optical characteristic data including spectral curves, color information, and a structural similarity index;

[0044] extracting original spectral curves, original colors, and original structural similarity indexes based on the original measured data, and comparing them with the optical characteristic data to calculate characteristic distances with each anchor point;

[0045] determining whether there is a characteristic distance less than a preset value;

[0046] If so, selecting the anchor point with the smallest characteristic distance and extracting the car paint formula parameter corresponding thereto as the car paint physical parameter;

[0047] If not, obtaining the car paint physical parameter based on inverse parameter analysis and parameter optimization.

[0048] In some embodiments,

[0049] ​​​In a second aspect, the application provides a vehicle paint simulation system supporting dynamic forward and reverse bidirectional rendering, which adopts the following technical scheme:

[0050] A vehicle paint simulation system supporting dynamic forward and reverse bidirectional rendering is used to implement the method described above.

[0051] The technical scheme provided by the embodiment of the application has the following technical effects:

[0052] Through the bidirectional rendering architecture, the dynamic association between the vehicle paint formula and the optical effect is established in the forward direction, and the problem of how to feedback the formula from the known real vehicle paint effect is solved in the reverse direction, thereby providing support for quality detection and parameter optimization. Instead of focusing on the independent one-way rendering process, the closed loop of "generating effect through forward rendering → inversely analyzing and deducing parameters → fusing and correcting deviation through forward and reverse integration" is realized to achieve the dynamic linkage between the formula parameters and the visual effect. The output of each step provides a basis for the input of the next step, and through bidirectional interaction and iterative optimization, the unity of rendering accuracy and industrial practicality is ensured, thereby solving the core problem of "disconnection between formula and effect" in the traditional technology. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1 is a step schematic diagram of a vehicle paint simulation method supporting dynamic forward and reverse bidirectional rendering provided by the embodiment.

[0054] Figure 2 is a logic diagram of a vehicle paint simulation method supporting dynamic forward and reverse bidirectional rendering provided by the embodiment of the application. DETAILED DESCRIPTION

[0055] To make the objectives, technical schemes and advantages of the application clearer, the application is described and explained below in combination with the drawings and embodiments. However, it should be understood by those of ordinary skill in the art that the application can be implemented without these details. In some cases, to avoid unnecessary description, the aspects of the application become obscure. Well-known methods, processes, systems, components and / or circuits that have been described at a high level will not be described in detail. It is obvious for those of ordinary skill in the art that various changes can be made to the embodiments disclosed in the application, and the universal principles defined in the application can be applied to other embodiments and application scenarios without deviating from the principles and scope of the application. Therefore, the application is not limited to the shown embodiments, but conforms to the broadest range claimed in the application.

[0056] It should be noted that the description of the embodiments is intended to help understand the application, and is not intended to limit the application. Furthermore, the technical features involved in the various embodiments of the application described below can be combined with each other as long as they do not conflict with each other.

[0057] In the description of the present application, the meaning of one or more is one or more, the meaning of multiple is two or more, greater than, less than, more than, etc. are understood as not including the number, above, below, etc. are understood as including the number. If it is described as first, second, it is only used to distinguish technical features for the purpose, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features or implicitly indicating the sequence of indicated technical features.

[0058] In the description of the present application, the description of the terms "one embodiment", "some embodiments", "exemplary embodiment", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the description, the description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the described specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a combined manner.

[0059] As shown in Figure 1 and Figure 2 The embodiment of the present application discloses a car paint simulation method supporting dynamic forward and reverse bidirectional rendering, comprising the following steps:

[0060] Forward rendering:

[0061] S100, input car paint formula detail information, and parse paint layer rendering parameters and color master rendering parameters to output rendering results in a car paint renderer.

[0062] Forward rendering is a rendering process from formula to effect as a whole. The input is car paint formula detail data, and the optical rendering parameters of different paint layers and the color master rendering parameters of corresponding particle color master characteristics are decomposed through the formula. Finally, the output is the rendering of paint surface data and color master data through the car paint renderer to obtain the rendered spectral image and HDR image.

[0063] At the same time, the present application also supports dynamic rendering linkage, supports users to upload car paint formula detail information and further real-time edit car paint formula detail content (such as adjusting varnish thickness, aluminum powder proportion), wherein the action of real-time editing can be manually performed by the user, or can be intelligently performed based on the feedback information of the subsequent reverse propagation when linked with reverse rendering. The vehicle renderer automatically updates the rendered spectral and HDR images based on the change of the parameters, and outputs real-time and latest rendering results,

[0064] Inverse rendering:

[0065] S200, input the original measured data and perform inverse parameter analysis and parameter optimization to obtain the physical parameters of the car paint.

[0066] Inverse rendering is a rendering process from measured image data to parameters as a whole. The input is the original measured data composed of the spectral image actually measured by the user and the HDR image actually taken. The physical parameters of the car paint are output by a series of preprocessing, parameter analysis and further parameter optimization convergence on the original measured data. The physical parameters of the car paint directly correspond to the formula details information in the forward rendering.

[0067] Forward and inverse fusion:

[0068] S300, the physical parameters of the car paint are used as the analysis result of the new input car paint formula details information in the forward rendering, and the rendering result is used as the prior reference in the inverse rendering to constrain the range of the physical parameters of the car paint.

[0069] In this application, the forward rendering and the inverse rendering are not independent processes, but form a closed loop through data interaction, feedback verification and other actions to ensure that the effect of the forward rendering is credible and the parameters of the inverse analysis are available. Specifically,

[0070] The spectral / image features generated by the forward rendering are used as the "prior reference" for inverse analysis to constrain the parameter range. For example, when the measured data has an abnormality (such as a spectral jump caused by equipment failure), the abnormal value can be corrected by the forward result to narrow the parameter search range. The physical parameters (corresponding to the formula details) of the inverse analysis are fed back to the forward rendering model as "dynamic input" for the forward rendering to correct the model bias and reduce the deviation between the model assumption and the actual situation.

[0071] S400, perform forward-inverse consistency verification based on the rendering result and the original measured data to calculate the rendering deviation.

[0072] Further, a real-time feedback mechanism is designed to quantify the deviation and locate the problem.

[0073] According to the consistency verification of the rendering result (spectrum+HDR image) of the current forward rendering and the original measured data (spectrum+HDR image) actually detected, the main verification contents include the deviation verification of the spectrum, the color difference verification, and the texture detail difference verification of the particles, and the rendering deviation is calculated according to the verification result.

[0074] According to different rendering deviation results, the model can quickly quantify the result deviation of the formula-effect forward rendering and the image-parameter inverse rendering, and determine the specific problem causing the deviation difference based on the numerical relationship of the rendering deviation result.

[0075] S500, iteratively optimize the forward rendering and inverse rendering based on the rendering deviation until the rendering deviation is less than a threshold.

[0076] The rendering deviation is back-propagated to participate in the model action process of the forward rendering and the inverse rendering, and the parameter size corresponding to the forward rendering and the inverse rendering is adjusted and corrected based on the numerical relationship of the deviation. After each adjustment update, the forward rendering and the inverse rendering process are repeated and the rendering deviation is recalculated according to the new parameter information. In this way, the model parameters are iteratively optimized through continuous feedback and negative feedback until the final deviation is less than a threshold. Then the iteration is stopped and the final data is output.

[0077] Through the above method, through the bidirectional rendering architecture, the dynamic association between the car paint formula and the optical effect is established in the forward direction, and the problem of “how to feedback the formula from the known real car paint effect” is solved in the inverse direction, which provides support for quality detection and parameter optimization. At the same time, instead of focusing on independent one-way rendering process, the dynamic linkage of “formula parameters and visual effect” is realized through the closed loop of “forward rendering to generate effect → inverse analysis to feedback parameters → forward and inverse fusion to correct deviation”. The output of each step provides the basis for the input of the next step, and through bidirectional interaction and iterative optimization, the unity of rendering accuracy and industrial practicality is ensured, and the core problem of “formula and effect disconnection” in traditional technology is solved.

[0078] In some embodiments, the paint layer rendering parameters are analyzed, including the following steps:

[0079] S110, the car paint formula details information is decomposed into a physical layer structure including a primer layer, a color paint layer, a particle layer, and a clear paint layer.

[0080] Traditional car paint rendering usually regards car paint as a single “material ball”, but in the embodiments of the present application, in order to restore the physical structure characteristics of car paint as much as possible, layered modeling based on car paint formula details information is adopted to decompose the physical model of several car paint layers.

[0081] Primer layer: opaque, mainly to adhere to the car body and provide basic hiding power.

[0082] Color paint layer: translucent, containing pigment molecules (such as organic pigments), determining the basic hue of the car paint.

[0083] Particle layer: containing aluminum powder, mica sheet and other sheet / particle color masterbatch, which is the core source of iridescent effect (such as metallic luster, goniochromaticity).

[0084] Clear paint layer: transparent, covering the outermost layer, providing gloss and protection.

[0085] By hierarchical modeling, the transmission process of light in each layer of car paint can be decomposed into a complete path of "air-varnish interface reflection → varnish layer transmission → particle layer scattering → color paint layer absorption → primer layer reflection → reverse penetration of each layer", so as to facilitate subsequent simulation of real light effect to generate the corresponding optical characteristics of each layer of car paint.

[0086] S120, the light transmission rule of each layer is modeled to assign independent per-layer rendering parameters, and the total reflectivity is calculated based on the per-layer rendering parameters.

[0087] Based on the different regular characteristics of light in each layer, such as reflection, refraction, scattering, etc., the optical parameters required for rendering are added to each layer model and defined as per-layer rendering parameters. When the subsequent car paint renderer performs rendering based on the parameters, the per-layer rendering parameters provide the color basis and light effect basis of the rendered image. Specifically,

[0088] The optical parameters corresponding to the primer layer mainly include diffuse reflectivity (reflectivity is fixed, no angle dependence), refractive index, and thickness.

[0089] The optical parameters corresponding to the color paint layer include absorption coefficient (positively correlated with pigment concentration), refractive index, transmittance, and thickness.

[0090] The optical parameters corresponding to the particle layer include particle density, particle size distribution, orientation angle, refractive index, transmittance, absorption coefficient, and thickness.

[0091] The optical parameters corresponding to the varnish layer include refractive index, thickness, transmittance, and absorption coefficient.

[0092] The formula for generating the total reflectivity based on the above per-layer rendering parameters is as follows:

[0093] .

[0094] wherein,

[0095] .

[0096] characterized as total reflectivity; characterized as the equivalent reflectivity of the varnish layer and the underlying layer, which physically means that when light is incident from air to the varnish layer, it comprehensively reflects the total reflection effect of "direct reflection on the upper surface of the varnish layer" and "light transmitted through the varnish layer and reflected by the underlying layer structure and then penetrating out of the varnish layer again"; characterized as the equivalent reflectivity of the particle layer and the underlying layer, which physically means that when the light transmitted through the varnish layer reaches the particle layer, it comprehensively reflects the reflection effect of "direct reflection on the upper surface of the particle layer" and "light transmitted through the particle layer and reflected by the underlying layer and then penetrating out of the particle layer again", which is a basic parameter for calculating .

[0097] interface reflectance of air-varnish layer, interface reflectance of varnish layer-particle layer, interface reflectance of particle layer-color paint layer, transmittance of varnish layer, transmittance of particle layer, transmittance of color paint layer, reflectance of primer layer, refractive index of air (value 1.0), refractive index of varnish layer, refractive index of particle layer, refractive index of color paint layer, refractive index of primer layer, absorption coefficient of the ith layer, thickness of the ith layer.

[0098] Meanwhile, the specific calculation process of the above-mentioned partial parameters is as follows:

[0099] ;

[0100] ;

[0101] ;

[0102] ;

[0103] ;

[0104] ;

[0105] .

[0106] S130, generating paint layer rendering parameters according to each layer rendering parameter and total reflectance.

[0107] In some other embodiments, the color master rendering parameters are parsed, including the following steps:

[0108] S140, extracting physical characteristic parameters of each type of particle color master in the particle layer based on the corresponding each layer rendering parameter of the particle layer.

[0109] The optical effect (such as “goniochromatic”) of particle color master such as aluminum powder and mica sheet depends on its physical characteristics, but parameters such as “particle size 5-20 μm” and “aluminum powder content 10%” are commonly used in the industry to describe, which need to be converted into BSDF (Bidirectional Scattering Distribution Function) parameters that can be calculated by graphics algorithms.

[0110] Therefore, firstly, the particle layer containing the mica features is taken as an object, and the density (the number of particles in a unit volume, corresponding to the addition ratio of different particles in the formula), the particle size distribution (the distribution range and probability density of different particles), and the orientation (the inclination angle of the particle to the surface of the vehicle paint, affecting the direction of highlights) and other physical characteristic parameters of the particles such as aluminum powder and mica sheet are extracted according to the physical characteristic parameters of each layer corresponding to the layer rendering parameters.

[0111] In S150, the optical characteristics of the particle color master are calculated according to the physical characteristic parameters and a preset Mie scattering model.

[0112] Based on the Mie scattering theory (describing the scattering of spherical particles on light) and geometric optics (describing the specular reflection of sheet particles), the scattering and reflection characteristics of particles on light are calculated, including:

[0113] Scattering intensity (the scattering ability of particles on light of different wavelengths);

[0114] Angle dependence (the change law of scattered light with the incident angle and the observation angle, directly affecting the "angle-dependent color" effect);

[0115] Polarization characteristics (the polarization state of the scattered light of particles, affecting the "rainbow effect" of highlights).

[0116] Specifically, the core of the Mie scattering theory is to calculate the scattering efficiency factor through the particle size (d), the particle refractive index (n ), the refractive index of the surrounding medium (n , such as the refractive index of the paint layer ), and the incident light wavelength (λ ). The formula of the scattering efficiency factor is:

[0117] .

[0118] wherein, is the size parameter; , is the Mie scattering coefficient, which is determined by the refractive index difference between the particle and the medium and the size parameter; reflects the scattering ability of the particle on light, and is associated with the "scattering intensity parameter" in the subsequent BSDF parameter mapping.

[0119] In S160, a mapping relationship between the physical characteristic parameters and the BSDF parameters is established based on a preset parameterized regression algorithm.

[0120] The mapping between the physical characteristics of the particles and the BSDF parameters is established through the regression algorithm. Specifically,

[0121] The particle size distribution of aluminum powder is regressed to correspond to the "highlight bandwidth" of BSDF (the smaller the particle size, the more concentrated the highlight, and the stronger the metallic effect);

[0122] The thickness distribution of mica sheet is regressed to correspond to the "interference color shift" of BSDF (the thickness increases, and the reflected light wavelength shifts to long wave, such as from blue purple to golden yellow);

[0123] The particle density distribution is regressed to correspond to the "scattering intensity parameter" of BSDF (the higher the density, the stronger the reflected light, and the brighter the color).

[0124] These mapping relationships directly affect the scattering and reflection of light by the particle layer, and then determine the special optical effect finally presented by the car paint.

[0125] Taking the regression of the particle size distribution of aluminum powder to correspond to the "highlight bandwidth" of BSDF as an example, the mapping formula is:

[0126]

[0127] wherein, is the highlight angle range parameter in the BSDF model, characterizes the particle size distribution of the aluminum powder, is the particle size value, characterizes the probability corresponding to the particle size, is the average particle size, characterizes the standard deviation of the particle size distribution, , , is the regression coefficient (obtained by training experimental data), characterizes the regression error (minimized by training a large number of samples).

[0128] For new colorants (such as nanoparticles), the colorant parameterization calibrates the mapping model through a small amount of spectral data of physical samples, ensuring the accuracy of parameter conversion. This process is also carried out in the particle layer, because when the new colorant is added to the car paint, it mainly exists in the particle layer, and the optical effect of the new colorant is simulated and presented by adjusting the related parameters of the particle layer

[0129] S170, based on the optical characteristics and the mapping relationship, analyzing the colorant rendering parameters.

[0130] In the present application, the forward and inverse rendering share the same multi-layer BSDF model and particle Mie scattering model, ensuring physical consistency.

[0131] In some other embodiments, the original measured data includes original multi-angle spectrum and original HDR image, the original measured data is inputted, and inverse parameter analysis and parameter optimization are performed to obtain the physical parameters of the car paint, including the following steps:

[0132] S210, pre-process the original multi-angle spectrum and original HDR image.

[0133] Firstly, the pre-processing of the original multi-angle spectrum and the original HDR image includes multi-angle alignment, noise elimination and data normalization.

[0134] The multi-angle alignment is specifically to unify the angle coordinate system of the spectrum data of multiple angles (such as the reflectance spectrum under the incident light of 0 degrees, 45 degrees and 60 degrees) with the normal of the car paint surface as the reference to map to the same angle space, so as to avoid the analysis error caused by the deviation of the measurement angle.

[0135] The noise elimination is characterized by filtering the spectrum data by using Gaussian filtering, and filtering the HDR image by using bilateral filtering (while retaining the edge, removing the spot noise).

[0136] The data normalization is characterized by uniformly mapping the spectrum intensity (0, -1) and the HDR image pixel value (0-255) to the interval [0, 1] to provide consistent scale for subsequent fusion.

[0137] The above pre-processing process can greatly improve the analysis accuracy and provide high-quality input for the subsequent “reverse parameter analysis”.

[0138] S220, based on the machine learning algorithm, input the original measured data into the preset model to inversely deduce the material parameter of each layer and the particle distribution characteristics corresponding to the original measured data.

[0139] After pre-processing, the specific physical characteristic parameters of each layer of the car paint are inversely deduced from the measured data. Specifically, first, the optical characteristic parameters (such as total reflectance) analyzed in the forward rendering process are used as physical constraints to narrow the search range of the parameters.

[0140] The neural network model is trained, the input is the measured spectrum and image features, and the output is the parameters of each layer. The image information in the measured data is converted into specific physical characteristic parameters including the material parameters of each layer and the particle distribution characteristics by combining the physical model and the neural network results.

[0141] The machine learning algorithm plays an important role in the reverse rendering. Through the neural network model, the features of the measured data and the material parameters of each layer and the particle distribution characteristics are mapped.

[0142] For example, a trained neural network can infer the proportion of pigments in the color paint layer according to the change in the reflection intensity of a specific wavelength in the spectral data; according to the distribution and intensity of highlights in the HDR image, the particle size distribution, density, etc. of the aluminum powder or mica sheet in the particle layer are deduced. This mapping relationship is based on a large amount of experimental data and machine learning training, which can convert complex measurement data into specific formula parameters. In the training process, a large amount of paint measurement data with known formula details is input to let the neural network learn the internal relationship between the measurement data features and the formula parameters, so that in the actual inverse rendering, the corresponding formula parameters can be accurately deduced from the measurement data.

[0143] S230, minimizing the difference between the initial output rendering result and the original measured data by the preset differentiable rendering engine, and adjusting the material parameters and particle distribution characteristics of each layer based on the optimization result.

[0144] By minimizing the difference between the rendering result and the measured data through the differentiable rendering engine, the parameters are refined to ensure that the deduced formula details can be directly used for forward rendering.

[0145] Specifically, the differentiable rendering principle is to convert the rendering process into a derivable function, calculate the derivative (gradient) of the "difference between the rendering result and the measured data" with respect to the parameters, minimize the above derivable function, and adjust the physical parameters according to the gradient direction until the difference between the output rendering result and the original measured data is less than a threshold.

[0146] Among them, the explanation and description of the derivable function are specifically disclosed in the following.

[0147] In other embodiments, the rendering result includes a rendered spectrum and a rendered HDR image, and the forward-inverse consistency check is performed based on the rendering result and the original measured data to calculate the rendering deviation, including the following steps:

[0148] S410, obtaining a preset consistency check rule, and comparing and calculating the rendered spectrum and the original multi-angle spectrum based on the type of the consistency check rule, and comparing and calculating the rendered HDR image and the original HDR.

[0149] The rendering result driven by the inverse parameters is compared with the original measured data through the forward-inverse consistency check to calculate the deviation between the two, wherein the consistency check rule at least includes wavelength requirement, color difference standard, and texture detail comparison.

[0150] S420, calculating the mean square error between the rendered spectrum and the original multi-angle spectrum based on the wavelength requirement to generate a spectral error result.

[0151] 400-700nm waveband by waveband calculation of the mean square error (MSE) of the rendered spectrum and the measured spectrum reflects the basic color deviation (such as insufficient red pigment will cause large error at 600nm waveband).

[0152] S430, calculate the color difference index between the rendered HDR image and the original HDR image based on the color difference standard to generate a color difference result.

[0153] Based on the color difference index of human eye visual characteristics (human eye cannot distinguish when ΔE<1.5), used to evaluate the overall color consistency.

[0154] S440, calculate the structural similarity index between the rendered HDR image and the original HDR image based on texture detail comparison to generate a texture error result.

[0155] Evaluate the texture details (such as the uniformity of particle distribution), SSIM<0.95 indicates that there is a deviation in the particle layer parameters (such as particle size distribution).

[0156] S450, generate rendering deviation based on the spectrum error result, color difference result and texture error result.

[0157] Based on the different values and different states of the rendering deviation, the source of the deviation can be analyzed, such as the spectrum error concentrated in the blue light waveband, indicating that the blue pigment parameters of the paint layer are abnormal.

[0158] In some other embodiments, based on the rendering deviation, the reverse propagation is performed to iteratively optimize the forward rendering and inverse rendering until the rendering deviation is less than a threshold, including the following steps:

[0159] S510, generate initial weight coefficients corresponding to each item in the consistency check rule.

[0160] S520, generate an optimization objective function based on the initial weight coefficients, the consistency check rule and the car paint formula detail information.

[0161] S530, determine whether iterative optimization is needed based on the size of the rendering deviation.

[0162] S540, if needed, the rendering deviation is propagated to the optimization objective function to adjust and correct the initial weight coefficients in the inverse rendering and the car paint formula detail information in the forward rendering until the rendering deviation is less than the threshold, and the final car paint formula detail information is output.

[0163] Based on the deviation signal reverse propagation, dynamically adjust the weight factors of inverse analysis (such as spectrum / texture weight) and the paint layer rendering parameters and color master rendering parameters related to the car paint formula detail information during forward rendering until the deviation is less than the threshold.

[0164] The specific formula for calculating the objective function is as follows:

[0165] ,

[0166] Here, θ represents the paint formulation parameters, which are variables to be optimized. These include the clear coat thickness, the particle size distribution of aluminum powder / mica flakes, the absorption coefficient of the paint layer, and the particle density. These parameters determine the optical and physical properties of the paint. By adjusting θ, the rendering effect can be made closer to reality.

[0167] , , These are the initial weight coefficients corresponding to each item. This is a weighting factor for spectral error, controlling the importance of the difference between the rendered spectrum and the measured spectrum in the overall optimization objective. A larger value indicates greater emphasis on spectral accuracy. As a weighting factor for color difference, it controls the difference between the rendered color and the measured color. The influence weight of ")" in the objective function highlights the importance attached to the consistency of color perception by the human eye; As a weighting factor for texture differences, it controls the weight of "difference between rendered texture and measured texture (1−SSIM)" and emphasizes the attention to the consistency of the paint surface texture (such as particle distribution and gloss texture).

[0168] Characterized by the rendered spectral data based on the paint formulation parameter θ, and the virtual spectral data calculated by the forward rendering model (combining theories of light transmission and Mie scattering), it describes the reflection / transmission of different wavelengths of light by the paint and is the "expected optical performance" in the virtual environment.

[0169] Characterized by raw multi-angle spectral data, which is real vehicle paint spectral data obtained through actual measurement (using equipment such as spectrometers), it is a "real feedback" of the optical properties of vehicle paint in the physical world, serving as a "target reference" for optimization.

[0170] The result is represented by the color difference standard calculated based on the CIELAB color space, while SSIM is represented by the structural similarity index.

[0171] Based on the formula explanation, the process of iteratively optimizing model parameters through backpropagation using bias data includes:

[0172] Weighting factor adjustment: If the proportion of spectral error is high, increase the weighting factor. (Increase the weight of spectral fitting); if the texture difference is large, increase the weight. ;

[0173] Optimization of light transport parameters: If the interlayer reflection error is large (e.g., the calculated interfacial reflectivity T1 of the varnish-particle layer deviates), the refractive index in the reflectivity formula is corrected 、 );

[0174] Convergence condition: Repeat "reverse parameters → forward rendering → deviation calculation → parameter adjustment", when the three types of deviation are less than the threshold value (spectral error <1.2%, ΔE <1.5, SSIM >0.95), stop iteration, and output the final parameters.

[0175] It should be noted that both step S230 and step S540 are based on the above optimization objective function for parameter optimization and iteration, but the difference lies in:

[0176] The core of step S230 is to fine-tune the "formula parameters" obtained by reverse parameter analysis, and its positioning is "local optimization after reverse analysis", which only adjusts the physical parameters (such as varnish layer thickness, aluminum powder particle size distribution, etc.) obtained by reverse deduction, that is, it optimizes θ in the optimization objective function, while the weight factor 、 、 is a fixed value, and the purpose is to make the results generated by forward rendering as close to the measured data as possible;

[0177] The deviation threshold is the criterion for determining whether the reverse parameters are "available". When the deviation is less than the threshold, it means that the formula parameters obtained by reverse deduction are accurate enough to be directly used for forward rendering to reproduce the measured effect, meeting the demand of "deducing formula from measurement" in industrial scenarios (such as quality inspection and formula replication).

[0178] The core of step S540 is the global optimization of "forward and reverse rendering closed loop", and its positioning is "overall convergence after forward and reverse interaction", which not only includes the adjustment of reverse parameters, but also includes the dynamic correction of light transport parameters (such as interlayer transmittance) in forward rendering and multi-modal weight factors (the 、 、 ) in forward rendering, with the purpose of ensuring the physical consistency and effect consistency of forward rendering and reverse analysis.

[0179] The deviation threshold is the criterion for determining whether the forward and reverse rendering is "consistent". When the deviation is less than the threshold, it means that the forward rendering (based on formula parameters) and the reverse analysis (based on measured data) form a stable closed loop—forward rendering can accurately map formula changes, and reverse analysis can reliably deduce formula parameters, and the physical logic and effect output of the two are consistent, meeting the core demand of "dynamic editing and real-time feedback".

[0180] In other embodiments, the original measured data includes original multi-angle spectrum and original HDR image, the input original measured data is subjected to inverse parameter analysis and parameter optimization to obtain the physical parameters of the car paint, and the method further comprises the following steps:

[0181] S240, an anchor point library is created, and the anchor point library stores optical characteristic data and paint formula parameters corresponding to several typical car paints, the optical characteristic data including spectral curves, color information, and structural similarity indexes.

[0182] A certain number of classic car paints, typical car paints, and popular car paints (for example, pure black paint, bright silver paint, matte white paint, and pearl red paint) are pre-stored to form a relationship between the "formula-optical characteristic" to form an anchor point library, which serves as a "basic template" for subsequent rapid matching.

[0183] The structure of the anchor point library is as follows:

[0184] .

[0185] Among them, The full-band spectrum (400-700 nm, λ is the wavelength) is represented as, The formula parameter set is represented as.

[0186] S250, original spectral curves, original colors, and original structural similarity indexes are extracted based on the original measured data, and are compared with the optical characteristic data to calculate the feature distance between each anchor point.

[0187] The differences between the measured data and each anchor point in the anchor point library are quantified, and the most similar anchor point is found.

[0188] The optical characteristics of the new measured image are extracted, and the total distance is defined as the weighted sum of the spectral distance, the color distance, and the texture distance, specifically,

[0189] .

[0190] Among them, The spectral curve distance measured by the cosine similarity is represented as, and the value closer to 1 represents a higher similarity; The color distance calculated by the CIELAB color difference formula is represented as, and the smaller the value, the more similar, The texture difference distance is represented as, and the smaller the value, the more similar.

[0191] S260, it is determined whether there is a feature distance less than a preset value.

[0192] S270, if there is, the anchor point with the smallest feature distance is selected, and the corresponding paint formula parameter is extracted as the physical parameter of the car paint.

[0193] S280, if not, based on the inverse parameter analysis and parameter optimization to obtain the car paint physical parameters.

[0194] When the distance between the measured data and each anchor point is calculated, all data is traversed to find the anchor point that makes the smallest feature distance, while noting that the smallest needs to be less than the preset value, if not, it means that the existing anchor point data closest to the feature distance is also not similar or similar to the measured data.

[0195] When the smallest feature distance is less than the preset value, it means that the measured data is extremely similar to the optical feature corresponding to the anchor point. At this time, in order to change the gradient descent from blind search to directional refinement, the formula parameter corresponding to the anchor point can be directly selected as the car paint physical parameter corresponding to the inverse rendering, and the gradient descent iteration optimization process is directly performed. In this way, for some classic or common car paint, the inverse rendering result can be quickly searched based on the above method without going through the conventional repeated refinement iteration, reducing the iteration steps.

[0196] The application also discloses a car paint simulation system supporting dynamic forward and inverse bidirectional rendering, which is used to realize the above method.

[0197] Further, the corresponding basic acquisition device of the system includes an XRite MA5C multi-angle spectrophotometer, an optical transmittance measuring instrument, a paint film instrument and the like.

[0198] The efficiency index is specifically:

[0199] The single-frame rendering time during forward rendering is less than or equal to 0.5s, and the full-parameter optimization convergence time during inverse rendering is less than or equal to 10 minutes.

[0200] The implementation principle is:

[0201] Through the bidirectional rendering architecture, the dynamic association between the car paint formula and the optical effect is established during forward rendering, and the problem of how to feedback the formula from the known real car paint effect is solved during inverse rendering, which provides support for quality detection and parameter optimization. At the same time, instead of focusing on independent one-way rendering process, the closed loop of "forward rendering to generate effect → inverse analysis to deduce parameters → forward and inverse fusion to correct deviation" is realized to achieve the dynamic linkage of "formula parameters and visual effect". The output of each step provides the basis for the input of the next step, and through bidirectional interaction and iteration optimization, the unity of rendering accuracy and industrial practicality is ensured, and the core problem of "formula and effect disconnection" in traditional technology is solved.

[0202] It should be understood that, while the steps in the flow diagrams of the drawings are shown in sequential order, such need not be the order in which they are performed. Unless otherwise explicitly stated, the steps can be performed in any order, and need not be performed in the order shown.

[0203] The above are only preferred embodiments of the present application, not intended to limit the protection scope of the present application, therefore: any equivalent changes made according to the structure, shape, principle of the present application should be covered within the protection scope of the present application.

Claims

1. A car paint simulation method supporting dynamic forward and backward rendering, characterized in that, The method comprises the following steps: Forward rendering: inputting vehicle paint formula details and analyzing paint layer rendering parameters and color master rendering parameters to output rendering results in a vehicle paint renderer; Reverse rendering: inputting original measured data and performing reverse parameter analysis and parameter optimization to obtain vehicle paint physical parameters; Forward and reverse fusion: taking the vehicle paint physical parameters as the analysis results of the new input vehicle paint formula details in the forward rendering, and taking the rendering results as a prior reference in the reverse rendering to constrain the range of the vehicle paint physical parameters; Based on the rendering results and the original measured data, a rendering deviation is calculated; Based on the rendering deviation, back propagation is performed to iteratively optimize the forward rendering and the reverse rendering until the rendering deviation is less than a threshold value, specifically, Generating initial weight coefficients corresponding to each item in a preset consistency checking rule; Based on the initial weight coefficients, the consistency checking rule, and the vehicle paint formula details, an optimization objective function is generated; Based on the size of the rendering deviation, it is determined whether iterative optimization is needed; If so, the rendering deviation is back propagated to the optimization objective function to adjust and correct the initial weight coefficients in the reverse rendering and the vehicle paint formula details in the forward rendering, until the rendering deviation is less than the threshold value, at which point the iteration is stopped and the final vehicle paint formula details are output.

2. The car paint simulation method of claim 1, wherein, Analyzing paint layer rendering parameters, comprising the following steps: The vehicle paint formula details are decomposed into a physical layer structure including a primer layer, a color paint layer, a particle layer, and a clear coat layer; Modeling the light transmission rule of each layer to assign independent rendering parameters for each layer, and calculating the total reflectance based on each rendering parameter, wherein the rendering parameters include reflectance, refractive index, thickness, absorption coefficient, particle density, particle size distribution, and orientation angle; Generating the paint layer rendering parameters based on the rendering parameters and the total reflectance.

3. The car paint simulation method of claim 2, wherein, Analyzing color master rendering parameters, comprising the following steps: Based on the rendering parameters corresponding to the particle layer, the physical characteristic parameters of each type of particle color master in the particle layer are extracted; Based on the physical characteristic parameters and a preset Mie scattering model, the optical characteristics of the particle color master are calculated; Based on a preset parameterized regression algorithm, a mapping relationship between the physical characteristic parameters and BSDF parameters is established; Based on the optical characteristics and the mapping relationship, the color master rendering parameters are analyzed.

4. The car paint simulation method of claim 1, wherein, The original measured data includes original multi-angle spectra and original HDR images, and inputting the original measured data and performing reverse parameter analysis and parameter optimization to obtain vehicle paint physical parameters comprises the following steps: Pretreating the original multi-angle spectra and the original HDR images; Based on a machine learning algorithm, the original measured data is input into a preset model to back-propagate the material parameters and particle distribution characteristics corresponding to the original measured data; Through a preset differentiable rendering engine, the difference between the initial output rendering results and the original measured data is minimized, and the material parameters and particle distribution characteristics are adjusted based on the optimization results.

5. The car paint simulation method of claim 4, wherein, The rendering result includes a rendering spectrum and a rendering HDR image, and forward-backward consistency verification is performed based on the rendering result and original measured data to calculate a rendering deviation, including the following steps: A preset consistency verification rule is obtained, and the rendering spectrum and the original multi-angle spectrum are compared and calculated based on the type of the consistency verification rule, and the rendering HDR image and the original HDR image are compared and calculated, and the type of the consistency verification rule at least includes a wave band requirement, a color difference standard, and texture detail comparison.

6. The car paint simulation method of claim 5, wherein, The rendering spectrum and the original multi-angle spectrum are compared and calculated based on the type of the consistency verification rule, and the rendering HDR image and the original HDR are compared and calculated, including the following steps: Based on the wave band requirement, mean square error between the rendering spectrum and the original multi-angle spectrum is calculated to generate a spectrum error result; Based on the color difference standard, a color difference index between the rendering HDR image and the original HDR image is calculated to generate a color difference result; Based on the texture detail comparison, a structural similarity index between the rendering HDR image and the original HDR image is calculated to generate a texture error result; Based on the spectrum error result, the color difference result and the texture error result, the rendering deviation is generated.

7. The car paint simulation method of claim 6, wherein, The calculation formula of the optimization objective function is specifically: ， wherein θ represents a car paint formula parameter, , , are initial weight coefficients corresponding to each item, represents rendered spectral data rendered based on the car paint formula parameter, represents original multi-angle spectral data, represents a color difference standard result, and SSIM represents a structural similarity index.

8. The car paint simulation method of claim 4, wherein, The original measured data is input and reverse parameter analysis and parameter optimization are performed to obtain the physical parameters of the car paint, and the following steps are further included: An anchor point library is created, and the anchor point library stores optical characteristic data and car paint formula parameters corresponding to a plurality of typical car paints, and the optical characteristic data includes a spectrum curve, color information and a structural similarity index; Based on the original measured data, original spectrum curves, original colors and original structural similarity indexes are extracted, and compared with the optical characteristic data to calculate feature distances with each anchor point; It is judged whether there is a feature distance less than a preset value; If so, the anchor point with the smallest feature distance is selected, and the car paint formula parameter corresponding thereto is extracted as the physical parameter of the car paint; If not, the physical parameters of the car paint are obtained based on reverse parameter analysis and parameter optimization.

9. A car paint simulation system supporting dynamic forward and inverse bidirectional rendering, characterized in that, A method for implementing any one of claims 1-8.

Images