Physical model shooting illumination processing method and system based on intelligent light field

Abstract

The invention discloses an entity model shooting illumination processing method and system based on an intelligent light field, and the method comprises the steps: obtaining a light-emitting parameter used for controlling a light-emitting device, and the light-emitting parameter is determined according to the environment illumination in first multimedia information provided in advance; sending the light-emitting parameters of the light-emitting equipment to the light-emitting equipment; wherein the degree that the ambient light generated after the light emitting device emits light according to the light emitting parameters is close to the ambient light in the first multimedia information meets a preset condition; and after the light-emitting equipment emits light according to the light-emitting parameters, controlling shooting equipment to shoot. According to the method and the device, a light illuminator does not need to repeatedly input different light-emitting parameters for testing, so that the problem of low efficiency caused by continuous debugging mainly depending on the light illuminator when the light is adjusted according to the multimedia information in the prior art is solved, and the efficiency of light adjustment during shooting is improved.

Description

Technical Field

[0001] The present application relates to the field of optics, and more specifically, to a method and system for processing lighting for photographing a physical model based on an intelligent light field. Background Art

[0002] In today's film, television and animation industries, the use of physical models in pre-production and shooting is common and relatively mature, mainly in two ways: direct application and indirect application. Direct application is to directly shoot the physical model and shoot the characters in front of a green screen, and then synthesize the characters with the scene later; indirect application is to scan and digitize the physical model and use the digital model to complete subsequent work.

[0003] After the filming of a film, television or animation is completed, there may be a situation where additional shots are taken, for example, when a shot is missed or an error occurs during the filming process. In this case, additional shots are required.

[0004] Supplementary shooting requires that the scene be consistent with the original shooting scene, and also requires that the lighting conditions during shooting be basically consistent with the original lighting conditions. This can ensure better post-processing and prevent the audience from seeing any traces of supplementary shooting.

[0005] It is relatively easy to set up a shooting scene, but the lighting of the shooting scene requires very experienced lighting technicians to debug the lighting based on previously shot film and television animations. Lighting technicians usually repeatedly input lighting parameters to control the lighting, which makes lighting adjustment inefficient. Summary of the invention

[0006] The embodiments of the present application provide a method and system for lighting a physical model for shooting with an intelligent light field, which at least solves the problem of low efficiency in the related art caused by the lighting technicians having to perform continuous debugging when adjusting the lighting according to multimedia information, thereby improving the efficiency of lighting adjustment during shooting.

[0007] According to one aspect of the present application, a method for lighting processing for shooting a physical model based on an intelligent light field is provided, comprising: obtaining lighting parameters for controlling a lighting device, wherein the lighting device comprises a plurality of lighting entities at different positions, and each lighting entity can be independently controlled by the lighting parameters; the lighting parameters are determined according to ambient lighting in a first multimedia information provided in advance; the lighting parameters of the lighting device are sent to the lighting device; wherein the degree to which the ambient lighting generated by the lighting device after emitting light according to the lighting parameters is close to the ambient lighting in the first multimedia information satisfies a predetermined condition; after the lighting device emits light according to the lighting parameters, controlling a shooting device to shoot.

[0008] Furthermore, the luminous parameters are determined based on the ambient lighting in the first multimedia information provided in advance, including: determining the lighting information from the first multimedia information, wherein the lighting information is used as a basis for controlling the lighting of the light-emitting device; adjusting the luminous parameters of the light-emitting device according to the lighting information; after controlling the light-emitting device to emit light according to the lighting parameters, performing photographing processing to obtain second multimedia information; comparing the first multimedia information and the second multimedia information, and saving the luminous parameters of the light-emitting device if the difference between the first multimedia information and the second multimedia information meets predetermined conditions.

[0009] Furthermore, determining the lighting information from the first multimedia information includes: identifying an object model from the first multimedia information, and determining the lighting information based on the ambient lighting of the object model; or, identifying the overall ambient lighting of the first multimedia information from the entire first multimedia information, and determining the lighting information based on the overall ambient lighting.

[0010] Furthermore, identifying the object model from the first multimedia information includes: obtaining multiple groups of training data, wherein each group of training data includes input data and output data, the input data is media information, and the output data is the object model included in the media information; determining whether the number of the multiple groups of training data exceeds a threshold, wherein the threshold is pre-configured; if the threshold is exceeded, using the multiple groups of training data to train a neural network in a supervised machine learning training method; inputting the first media information into the neural network, and the neural network outputs the object model.

[0011] Furthermore, it also includes: after obtaining the neural network, continuing to obtain training data, and when the training data that continues to be obtained exceeds a second threshold, using the training data that continues to be obtained to perform incremental training on the neural network.

[0012] According to another aspect of the present application, a physical model shooting lighting processing system based on intelligent light field is also provided, including: a second acquisition module, used to acquire light-emitting parameters for controlling a light-emitting device, wherein the light-emitting device includes multiple light-emitting entities at different positions, and each light-emitting entity can be independently controlled by the light-emitting parameters; the light-emitting parameters are determined according to the ambient lighting in a first multimedia information provided in advance; a sending module, used to send the light-emitting parameters of the light-emitting device to the light-emitting device; wherein the degree to which the ambient lighting generated by the light-emitting device after emitting light according to the light-emitting parameters is close to the ambient light in the first multimedia information satisfies a predetermined condition; and a control module, used to control the shooting device to shoot after the light-emitting device emits light according to the light-emitting parameters.

[0013] Furthermore, the second acquisition module is used to: determine lighting information from the first multimedia information, wherein the lighting information is used as a basis for controlling the lighting of the light-emitting device; adjust the lighting parameters of the light-emitting device according to the lighting information; after controlling the lighting of the light-emitting device according to the lighting parameters, perform shooting processing to obtain second multimedia information; compare the first multimedia information and the second multimedia information, and save the lighting parameters of the light-emitting device when the difference between the first multimedia information and the second multimedia information meets predetermined conditions.

[0014] Furthermore, the second acquisition module is used to: identify an object model from the first multimedia information, and determine the lighting information based on the ambient lighting of the object model; or, identify the overall ambient lighting of the first multimedia information from the entire first multimedia information, and determine the lighting information based on the overall ambient lighting.

[0015] Furthermore, the second acquisition module is used to: obtain multiple groups of training data, wherein each group of training data includes input data and output data, the input data is media information, and the output data is an object model included in the media information; determine whether the number of the multiple groups of training data exceeds a threshold, wherein the threshold is pre-configured; if the threshold is exceeded, use the multiple groups of training data to train a neural network in a supervised machine learning training method; input the first media information into the neural network, and the neural network outputs the object model.

[0016] Furthermore, the second acquisition module is used to: after obtaining the neural network, continue to acquire training data, and when the training data that continues to be acquired exceeds a second threshold, use the training data that continues to be acquired to perform incremental training on the neural network.

[0017] In the embodiment of the present application, a method for obtaining light-emitting parameters for controlling a light-emitting device is adopted, wherein the light-emitting device includes a plurality of light-emitting entities at different positions, each of which can be independently controlled by the light-emitting parameters; the light-emitting parameters are determined according to the ambient lighting in the first multimedia information provided in advance; the light-emitting parameters of the light-emitting device are sent to the light-emitting device; wherein the degree of proximity between the ambient lighting generated by the light-emitting device after emitting light according to the light-emitting parameters and the ambient lighting in the first multimedia information satisfies a predetermined condition; after the light-emitting device emits light according to the light-emitting parameters, the shooting device is controlled to shoot. The present application eliminates the need for a lighting engineer to repeatedly input different light-emitting parameters for testing, thereby solving the problem of low efficiency caused by the lighting engineer's continuous debugging when adjusting the lighting according to multimedia information in the related art, and improving the efficiency of lighting adjustment during shooting. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent in conjunction with the accompanying drawings and with reference to the following specific embodiments. Throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale. The drawings that are part of this application are used to provide a further understanding of the application. The illustrative embodiments of the application and their descriptions are used to explain the application and do not constitute an improper limitation on the application. In the drawings:

[0019] Figure 1 A schematic diagram of a shooting illumination matching method based on an intelligent light field provided in an embodiment of the present application;

[0020] Figure 2 It is a flow chart of a lighting processing method for photographing a physical model based on an intelligent light field according to an embodiment of the present application. DETAILED DESCRIPTION

[0021] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0022] First, the technical terms involved in the following embodiments are explained.

[0023] Lux is the unit of luminance. Definition: When the luminous flux of an object evenly illuminated by light is 1 lumen on an area of ​​1 square meter, its illuminance is 1 lux. Application scenarios: The illuminance suitable for reading and sewing is about 60~100 lux. Alias: lux, legal symbol is lx. Lux, as a unit of illuminance, is widely used in fields such as photometry and ecology, and is one of the indispensable physical quantities in scientific research and daily life.

[0024] High Dynamic Range Imaging (HDRI or HDR) is a set of technologies used in computer graphics and film photography to achieve a larger exposure dynamic range (i.e., a larger difference between light and dark) than ordinary digital imaging technology. The purpose of high dynamic range imaging is to correctly represent the brightness of a large range in the real world, from direct sunlight to the darkest shadow.

[0025] The following embodiments also involve LEDs, which are short for light-emitting diodes and are commonly used light-emitting devices that emit light by releasing energy through the recombination of electrons and holes. They are widely used in the field of lighting. Light-emitting diodes can efficiently convert electrical energy into light energy and have a wide range of uses in modern society, such as lighting, flat panel displays, medical devices, etc. It should be noted that in the following embodiments, LEDs are used as an example for explanation, and other light-emitting devices or equipment can also achieve the same effect or function, which will not be described one by one here.

[0026] Reference Figure 1 Embodiments provide a shooting illumination matching method based on intelligent light field, Figure 1 A flowchart of a method for shooting illumination matching based on intelligent light field is provided in an embodiment of the present disclosure, such as Figure 1 As shown, the method includes the following steps:

[0027] In the first step, first obtain the first multimedia information. The first multimedia information includes at least one of the following: images, virtual 3D scenes, and videos. In this step, the first multimedia is multimedia that has been shot, which includes multimedia information. For example, images, videos, etc. are obtained from the original images obtained by shooting. Here, the information obtained from the first multimedia includes lighting information that needs to be restored. In one example, the first multimedia information is a video that mainly shows a scene in which the candlelight in the dark illuminates the protagonist. At this time, it is necessary to restore the lighting information of the candlelight scene to reshoot some scenes in the scene.

[0028] In the second step, the illumination information is determined from the first multimedia information, wherein the illumination information is used as a basis for controlling the illumination of the light-emitting device. It should be noted that the first multimedia information may have multiple situations. For example, the first multimedia information is a picture, and the picture has a high-rise building. In this case, a high-rise building model needs to be built, and then the illumination information corresponding to the high-rise building model is determined according to the light and shadow conditions of the high-rise building itself in the picture. In another example, the first multimedia information is a video of farmers working in the fields at dusk. At this time, after constructing the physical scenery such as the fields, the overall light and shadow conditions at dusk in the video are achieved by controlling the lights. The illumination information determined at this time is the overall illumination information of the scene. In this step, in order to realize the two situations in the above examples, determining the illumination information from the first multimedia information includes: identifying the object model from the first multimedia information, and determining the illumination information according to the ambient illumination of the object model; or identifying the overall ambient illumination of the first multimedia information from the entire first multimedia information; and determining the illumination information according to the overall ambient illumination. The object model here refers to an actual object existing in the first multimedia information, and the actual object may be a fixed object such as a building, or a moving object. Since a model is used to simulate the actual object existing in the first multimedia information during reshooting, the actual object existing in the first multimedia information is also called an object model (also referred to as a target object or target object).

[0029] As an optional implementation, the step of determining the lighting information from the first multimedia information can be combined with the two examples above, and the step includes: identifying one or more main objects from the first multimedia information, the process of re-shooting the main object is implemented using a model, that is, the main object is the object model (also referred to as the target object or target object), determining the lighting information of each main object respectively, determining one or more lighting parameters corresponding to each main object for the lighting information of each main object, deintersecting the lighting parameters corresponding to all the main objects to obtain a first lighting parameter, and the first lighting parameter is one or more; identifying the overall ambient lighting of the first multimedia information from the entire first multimedia information; determining the lighting information based on the overall ambient lighting, and determining the second lighting parameter based on the lighting information, and the second lighting parameter is one or more; taking the intersection of the first lighting parameter and the second lighting parameter to obtain a third lighting parameter, and the third lighting parameter is used as the basis for adjusting the lighting parameters of the light-emitting device in the third step.

[0030] In the third step, the lighting parameters of the lighting device are adjusted according to the lighting information (if the lighting parameters are determined, the lighting parameters are adjusted according to the lighting parameters); the lighting parameters here are required to send the lamp to emit light in the future. Therefore, when the lighting information is determined, the ambient lighting is read from the first multimedia information to generate a high dynamic range image; the pixel coordinates in the high dynamic range image are mapped to the spherical light position coordinates, and the color coordinates of the pixels or pixel blocks in the high dynamic range image are converted into the color and brightness of the light.

[0031] Obtain a first normal vector of the object model in the first multimedia information, and obtain first illumination information of highlights in the first multimedia information; perform preset arrangement calculations on the contrast in the first multimedia information to generate illumination information, and generate luminescence parameters based on the first illumination information and the illumination information; thus, the ambient illumination of the object model in the intelligent light field can be determined based on the first normal vector and the luminescence parameters.

[0032] In an optional implementation, there are generally multiple lamps, which can be arranged according to a dome, or can also be arranged on the spherical surface, so that a variety of lighting information can be achieved.

[0033] For example, the following lighting system including multiple lights can be used: the lighting system is called a panoramic shooting light box, including a box body, a fill light, a stage and a driving mechanism, the box body is provided with a storage cavity, one of the side walls of the box body is provided with an opening connected to the storage cavity, the opening is used for taking and placing objects and shooting by an external camera, the fill light is provided on multiple inner walls of the 360-degree panoramic shooting light box and is arranged toward the storage cavity, the stage is provided in the storage cavity and is rotatably connected to the box body, and the output shaft of the driving mechanism is connected to the stage and drives the stage to rotate. The box body has a front wall, a rear wall, a top wall, a bottom wall, a left wall and a right wall, the opening is provided on the front wall, and the inner sides of the rear wall, the top wall, the bottom wall, the left wall and the right wall are provided with fill lights.

[0034] The 360-degree panoramic shooting light box also includes a diffusion plate, which is arranged on the inner wall of the light box and corresponds to the fill light. The diffusion plate is located on the side of the fill light facing the accommodating cavity and covers the fill light. The 360-degree panoramic shooting light box also includes an upper light board, which is arranged at the opening of the box body and connected to the front wall and the top wall. There is an angle between the upper light board and the top wall and faces the stage. The 360-degree panoramic shooting light box can also include a left light board and a right light board. The bottom of the box body is provided with a left sliding mechanism and a right sliding mechanism. The left sliding mechanism is connected to the left light board, and the right sliding mechanism is connected to the right light board. The left light board can slide relative to the bottom wall through the left sliding mechanism, and the right light board can slide relative to the bottom wall through the right sliding mechanism.

[0035] In this step, the light field can be reconstructed in the following manner: obtain the three-dimensional coordinates of the object point on the target object in the camera coordinate system; according to the structural model of the focusing light field camera, perform ray tracing on each object point, reconstruct the original light field data image, and calculate the distance from the focused image point of the current object point in the main lens image to the main surface of the main lens; calculate the distance from the focused image point of the current object point in the main lens image to the microlens array surface according to the distance; calculate the image coordinate system coordinates of the imaging point of the current object point under each microlens in the microlens array surface according to the distance, the coordinates of the focused image point of the main lens and the center coordinates of the microlens; based on the image coordinate system coordinates of the imaging point of the current object point under each microlens in the microlens array surface, determine whether the current object point can be imaged under each microlens; and count the number of all microlenses that image the current object point. , and calculate the pixel value of the current object point under each microlens that images it; calculate the diffraction-limited spot size; compare the diffraction-limited spot size with the pixel size of the focusing light field camera; if the diffraction-limited spot size is larger than the pixel size of the focusing light field camera, the imaging point of the current object point on the CCD imaging area corresponds to more than one CCD pixel, and the image coordinate system coordinates of the imaging point of the current object point under each microlens that images it are used as the center of each CCD pixel of the current object point in the CCD imaging area, and generate the pixel block corresponding to the current object point; otherwise, the imaging point of the current object point on the CCD imaging area only corresponds to one CCD pixel on the CCD imaging area, and single pixel rendering is sufficient; perform the above operations on all object points to reconstruct the original light field data image.

[0036] In the fourth step, after the light-emitting device is controlled to emit light according to the light-emitting parameters, shooting processing is performed to obtain the second multimedia information.

[0037] In the fifth step, the first multimedia information and the second multimedia information are compared, and when the difference between the first multimedia information and the second multimedia information meets the predetermined conditions, the luminous parameters of the luminous device are saved. In this step, as an optional implementation, when the difference between the first multimedia information and the second multimedia information does not meet the predetermined conditions, the brightness, contrast and / or hue of the first multimedia information and the second multimedia information are compared, and the brightness, contrast and / or hue of the second multimedia information are adjusted. After the adjustment, the first multimedia information and the second multimedia information are compared again, and when the difference between the first multimedia information and the second multimedia information meets the predetermined conditions, the luminous parameters of the luminous device are saved.

[0038] The above steps solve the problem of artificial processing of restoring the lighting conditions of the shooting scene in the related art, thereby improving the efficiency and accuracy of restoring the lighting conditions.

[0039] In the above steps, the first multimedia information is associated with the lighting parameters of the lighting device and then saved, and the associated lighting parameters are identified as the lighting parameters of the first multimedia information. If there are multiple second multimedia information, the multiple second multimedia information are compared with the first multimedia information respectively, and the second multimedia information closest to the ambient lighting in the first multimedia information is saved, and the lighting parameters corresponding to the closest second multimedia information are saved.

[0040] As another optional implementation, the object model can also be generated as a software model, in which the above-mentioned illumination information is also stored, and the object model can be used by other users. The software model stores illumination information and luminous parameters, and the luminous parameters can also be used as luminous parameters of a virtual lamp in a software system, so as to simulate and obtain the second multimedia information in the software.

[0041] The above method steps are explained below with reference to an example.

[0042] The main purpose of this example is to solve the real-time lighting problem during the preview and shooting of physical models. It mainly serves the film, television, animation, game and other industries, and is used in the fields of film, television, animation, game pre-preview, film, television, animation shooting and production.

[0043] This example involves the service objects of intelligent light field technology, the status of physical model preview and shooting, and the storage and digital transformation of physical models. These are explained below.

[0044] Intelligent light field technology constructs a camera array to take multi-angle, high-precision photos, which not only restores the three-dimensional structure of the photographed person, but also obtains the reflection information of the face, so that the light effect of the face model can be reconstructed under different ambient lights. In essence, whether it is for research and development purposes or service objects, intelligent light field technology is mainly centered around people. Through a controllable light field photography scanning system, people are scanned in all directions to build 3D virtual characters (digital people). At the same time, since the scanning is performed in a space with a 360° variable and controllable light source, the scanned digital people take into account both accuracy and restoration while introducing the dimension of a real light source. Compared with traditional AI face-changing, the digital people scanned by intelligent light field technology have higher precision facial skin texture and texture, and the reflection of light on the face can match the changes in various ambient light.

[0045] In the film, television and animation industries, the use of physical models in pre-production and shooting is common and relatively mature, mainly in two ways: direct application and indirect application. Direct application is to directly shoot the physical model and shoot the characters in front of the green screen, and then synthesize the characters with the scene in the later stage; indirect application is to scan the physical model into digital form and use the digital model to complete the subsequent work. However, the current pre-production and shooting methods still have shortcomings, among which the image light source problem is quite critical. At present, the pre-production and shooting of physical miniature models, whether directly or indirectly, have disadvantages in the light source problem. Direct application is mainly reflected in the problem of unifying the light source of the characters and the scene, and indirect application is mainly reflected in the fact that the light source information can only be simulated by the computer in the later stage.

[0046] Because physical models occupy real physical space, preservation and destruction have long been a problem. Many used physical models often fall into the awkward dilemma of preservation and destruction. Specifically, if they are preserved, the physical models occupy physical space and need to be cleaned regularly due to dust. If they are destroyed, there is a fear that they may be used again in the future. Digital preservation and archiving of physical models is a solution.

[0047] In this example, the technical advantages of intelligent light field technology for scanning people are inherited. The dimension of light field is introduced in the early stage of physical model scanning. The light and shadow effects are more realistic than the digital models produced purely in post-production, and can more realistically simulate the light and shadow atmosphere, the sun and the moon, the morning and the evening in reality. In this example, the physical model is digitized, and a digital asset library of the model is established based on it. First, the digital asset library is more convenient for the preservation and application of physical models. Secondly, the digital model of the digital asset library retains the texture of the physical model while also having complete light field information, which provides more possibilities for later applications.

[0048] The light field acquisition of a physical model can be divided into four steps:

[0049] Step 11: Test the light field acquisition system and debug it to meet the light field environment required for scanning and acquisition of physical models.

[0050] Step 12: Place the physical model in a light field acquisition system for scanning and acquisition, and use a 360° light source system to simulate the state of the physical model under different light field conditions.

[0051] In step 13, the physical model with light field information scanned by the light field acquisition system is rendered, and the digital model is saved by a computer.

[0052] Step 14: The physical model has completed the digital transformation, the digital model in the computer is archived, and a digital model digital resource library is established.

[0053] The collected models in the digital asset library are further used in the pre-production and shooting processes.

[0054] In the above steps, 1) simulate any ambient lighting through the intelligent light field system: first read the ambient lighting from the image or virtual 3D scene to generate an HDRI image; then map the pixel coordinates to the spherical light position coordinates, and convert the color coordinates of the pixel or pixel block into the color and brightness of the light, thereby realizing ambient lighting simulation; 2) scan and reconstruct the object through the camera array on the spherical surface: first, 90 (18 cameras are installed on the horizontal circle with a 20° center angle interval and 5 columns in the vertical direction, for a total of 90 cameras. This method is a better distribution scheme obtained after multiple test results) 4K resolution cameras are arranged on the sphere to shoot the center area of ​​the sphere; then collect images of the object under test under different lighting; and then perform 3D reconstruction to form an asset library of the same object under different lighting.

[0055] The following is an optional performance indicator: 1) Controllable illumination range at the center of the sphere: 0-200,000 lux; 2) Controllable color temperature range: 3000K-8000K; 3) Controllable LED density on the spherical surface: 733 LEDs / square meter;

[0056] 4) Reconstruction accuracy: 3D point cloud 2mm, texture 0.1mm; 5) Lighting richness of a single asset: contains at least 20 types of scene lighting.

[0057] First of all, the physical miniature model scanned by the light field acquisition system retains the texture of the original model while also introducing the control of real-time light source effects, which solves many pain points in the film and television industry. For example, in the past, many movies chose to directly build a miniature model of the same scale and use it to simulate real scenes for shooting. The main reason is that the digital model that relies solely on post-production is not realistic enough and is difficult to put on the big screen. However, the disadvantage of this method is that it may produce inconsistent light and shadow effects between characters and scenes. Since the scales of characters and miniature models are different and they are shot separately, the lighting during shooting is often based on the experience of the staff, which is more artificial and empirical, and does not have rigorous technical support. In this example, the physical model is originally collected in an all-round way in a 360° light field. When shooting a person, you only need to use a light meter to record the real-time light source conditions on the scene, and import the recorded light source information into the digital model, so that the digital model generates the same light environment as when the person is shot, and the light uniformity of the person and the model is strictly controlled through technical intervention. This example provides new possibilities for the pre-production stage: based on the physical miniature model, it is digitally converted through the light field acquisition system to serve the pre-production stage. The real-time variable light effect is more intuitive and controllable than the previous method. Technically, the scanning method of the light field acquisition system (compared with the traditional scanning acquisition method) greatly expands the practicality and functionality. Although the scanned digital model cannot completely replace the physical model, the introduction of the early light field dimension makes its application scenarios richer. This proposal provides new ideas for the digital preservation of physical models and lays an important foundation for the establishment of a digital asset library for models.

[0058] In this example, the following technical points can be used: 1. High illumination - wide dynamic range: The instantaneous maximum illumination of the hardware system exceeds 200,000 Lux (the outdoor sunlight illumination at noon is 60,000 to 100,000 lux). It is the brightest light field system in the world and can achieve all-day and all-weather scene simulation. 2. High density - fine sampling: 73,332 independently controlled LEDs (757*97) can reproduce more details. Nearly 100 4K cameras can collect details of the object without blind spots. 3. Physical lighting - more realistic: The dimension of the light field is introduced in the early stage of scanning the physical model, and the light and shadow effects are more restored than the previous purely post-production digital models.

[0059] Through this example, the light field acquisition system is more convenient for the digital transformation of physical models, and can more realistically and objectively capture the real appearance of physical models, which is more conducive to the establishment of a digital model digital asset library. The intervention of light field factors in the scanning process makes the digital models scanned by the light field acquisition system more functional and comprehensive. Especially in the film, television, animation, and game industries, the digital models scanned by the light field acquisition system provide a more realistic, immersive light and shadow atmosphere, which not only improves efficiency, but also achieves light and shadow effects that were difficult to achieve purely through post-production in the past.

[0060] After saving the above-mentioned luminescence parameters, object model, lighting information and other data, they can be called up and used from the software system at any time when needed.

[0061] In this embodiment, a method for processing lighting for shooting a physical model based on an intelligent light field is also provided. Figure 2 is a flow chart of a method for photographing a physical model based on an intelligent light field according to an embodiment of the present application, such as Figure 2 As shown, the steps included in the method are described below.

[0062] Step S202, obtaining a lighting parameter for controlling a lighting device, wherein the lighting device includes a plurality of lighting entities at different positions, each lighting entity being independently controllable by the lighting parameter; the lighting parameter is determined according to the ambient light in the first multimedia information provided in advance;

[0063] Step S204, sending the lighting parameters of the lighting device to the lighting device; wherein the degree to which the ambient lighting generated by the lighting device after emitting light according to the lighting parameters is close to the ambient lighting in the first multimedia information satisfies a predetermined condition;

[0064] Step S206, after the light-emitting device emits light according to the light-emitting parameters, controlling the shooting device to shoot.

[0065] Through the above steps, it is no longer necessary for the lighting engineer to repeatedly input different lighting parameters for testing, which solves the problem of low efficiency caused by the lighting engineer's continuous debugging when adjusting the lighting according to multimedia information in the related art, thereby improving the efficiency of lighting adjustment during shooting.

[0066] There are many ways to determine lighting information. For example, the lighting parameters are determined based on the ambient lighting in the first multimedia information provided in advance, including: determining the lighting information from the first multimedia information, wherein the lighting information is used as a basis for controlling the lighting of the light-emitting device; adjusting the lighting parameters of the light-emitting device according to the lighting information; after controlling the light-emitting device to emit light according to the lighting parameters, performing shooting processing to obtain second multimedia information; comparing the first multimedia information and the second multimedia information, and saving the lighting parameters of the light-emitting device if the difference between the first multimedia information and the second multimedia information meets predetermined conditions.

[0067] Optionally, determining the lighting information from the first multimedia information includes: identifying an object model from the first multimedia information, and determining the lighting information based on the ambient lighting of the object model; or, identifying the overall ambient lighting of the first multimedia information from the entire first multimedia information, and determining the lighting information based on the overall ambient lighting.

[0068] There are many ways to identify an object model. For example, identifying an object model from the first multimedia information includes: obtaining multiple groups of training data, wherein each group of training data includes input data and output data, the input data is media information, and the output data is the object model included in the media information; determining whether the number of the multiple groups of training data exceeds a threshold, wherein the threshold is pre-configured; if the threshold is exceeded, using the multiple groups of training data to train a neural network in a supervised machine learning training method; inputting the first media information into the neural network, and the neural network outputs the object model.

[0069] Optionally, after obtaining the neural network, continue to acquire training data, and when the training data that continues to be acquired exceeds a second threshold, use the training data that continues to be acquired to perform incremental training on the neural network.

[0070] There are many ways to train a neural network (or neural network model), which are illustrated below:

[0071] Example 1

[0072] Step S1, the cloud server uses the first training data set to train the neural network model;

[0073] Step S2, the cloud server determines the accuracy of the test of the neural network model: if the accuracy is greater than a first preset threshold, go to step S3; if the accuracy is not greater than the first preset threshold, return to step S1 to continue training the neural network model; step S3, the cloud server sends the neural network model to the user end; step S4, the user end uses the second training data set to continue training the neural network model until the accuracy of the test of the neural network model is greater than a second preset threshold; step S5, the user end saves the neural network model as an available model for subsequent use; the second preset threshold is greater than the first preset threshold. In the accuracy of the test of the neural network model, the test refers to comparing the output data of the neural network model with the correct data, and the correct data here refers to the corresponding standard output data in the training data set used to train the neural network model.

[0074] In the subsequent use of the model, the user terminal continuously detects the number of inference errors of the available model, and re-executes steps S4 to S5 when the number of errors exceeds a third preset threshold. Inference here refers to outputting prediction results, and whether the inference is wrong is judged through user feedback. If the user feedback is wrong, it is recorded as an error.

[0075] In this example, by repeatedly training the neural network model with a large amount of training data, the final inference result of the neural network model meets the result required by the neural network model input data. The cloud server is remotely connected to multiple user terminals respectively; the cloud server collects the data uploaded by each user terminal as the first training data set. The data set of the cloud server refers to the collection of data sets collected by the cloud server from multiple user terminals, for example, the collection of data sets A, B, C, and D. The user terminal forms a second training data set by data collection. Data collection can include data collected by different sensors, or data directly input by the user. The training system of the neural network model includes a cloud server and a user terminal, and the cloud server is remotely connected to the user terminal. For the training process of a user terminal neural network model, the model is first trained on the cloud server, and the data set used is derived from the data set collected by the cloud server from each user terminal. When the accuracy of the neural network model exceeds the first preset threshold a%, the parameters of the neural network model are passed to the user terminal, and the user terminal continues to train using the data set collected by itself until the accuracy of the neural network model is higher than the second preset threshold b%, and the user starts to use the neural network model for reasoning. During the subsequent use of the neural network model, once the number of inference errors exceeds the third preset threshold (i.e., the inference failure threshold), it means that the generalization ability (ability to infer correctly) of the neural network model has decreased, and retraining is required until the training accuracy exceeds the user-side training accuracy threshold.

[0076] Example 2

[0077] This example may include the following steps:

[0078] Get a newly added sample set. The newly added sample set is a sample set used to train a neural network model that is not being trained for the first time. The newly added sample set may be different depending on the task to be performed by the neural network model that needs to be incrementally trained. For example, if the neural network model is a target detection model used for face detection, the newly added sample set may be a face image sample set, that is, a sample set consisting of a large number of face images and annotation information corresponding to each face image. For another example, if the neural network model is a target classification model used for image classification, the newly added sample set may be a sample set consisting of a large number of images of different preset categories and category information corresponding to each image.

[0079] During implementation, the user end can obtain a large number of training samples as new samples to form a new sample set according to the tasks to be performed by the neural network model that needs to be incrementally trained.

[0080] The newly added sample set is input into the target neural network model to obtain a feature data set output by at least one candidate segmentation layer in the target neural network model, wherein the target neural network model is a neural network model that has been trained at least once, and the candidate segmentation layer is a network layer in the target neural network model.

[0081] In implementation, when the user terminal establishes the initial neural network model corresponding to the target neural network model, according to the network structure of the initial neural network model, manually designates at least one network layer as a candidate segmentation layer. The network layer can be a convolutional layer, a pooling layer, etc. in the neural network. When the user terminal specifies the candidate segmentation layer, it must first exclude the input layer and the output layer of the initial neural network model. There is a kind of network layer in the neural network model. If it is removed from the neural network model, the neural network model will be divided into two independent parts. This kind of network layer can be selected first. It should be noted that this network layer with the above characteristics can be called a network connection layer. The selection of candidate segmentation layers is explained below by way of example.

[0082] In a neural network model in an example, there are 7 network layers, among which network layer 1 is the input layer and network layer 7 is the output layer. When selecting candidate segmentation layers, these two network layers can be excluded first. Among the remaining 5 network layers, if network layer 4 is removed, the neural network model will be divided into two, that is, network layer 4 is preferentially selected as the candidate segmentation layer. In another neural network layer, there are 8 network layers, among which network layer 1 is the input layer and network layer 8 is the output layer. Similarly, when selecting candidate segmentation layers, these two network layers can be excluded first. Among the remaining 6 network layers, there are no network layers of the above type, then these network layers can be used as candidate segmentation layers, or a preset number of network layers can be randomly selected from these network layers as candidate segmentation layers, or a preset number of network layers that are closest to the input layer can be selected from these network layers as candidate segmentation layers.

[0083] After the neural network model is established as described above, the process of selecting the candidate segmentation layer can be completed manually by a technician, or the electronic device can execute a pre-stored program to automatically complete the selection of the candidate segmentation layer. The selection method can be as follows: according to the position of each network layer in the target neural network model, a network connection layer is determined in the network layer in the target neural network model as a candidate segmentation layer; if the above network connection layer does not exist, the remaining network layers except the input layer and the output layer are all used as candidate segmentation layers, or, among the remaining network layers except the input layer and the output layer, a preset number of network layers closest to the input layer are selected as candidate segmentation layers.

[0084] After acquiring the newly added sample set, each newly added sample in the newly added sample set is input into the target neural network layer, and the corresponding feature data can be output from the above selected candidate segmentation layer. For each candidate segmentation layer, a corresponding feature data set can be obtained.

[0085] Step 103: determine the clustering degree of the feature data set output by each candidate segmentation layer, and determine a target segmentation layer corresponding to the clustering degree that meets a preset condition in the candidate segmentation layers.

[0086] Among them, the degree of clustering can be used to characterize the clustering ability of the segmentation layer and its previous network layers for data. In implementation, if the degree of clustering of the feature data set output by the candidate segmentation layer meets the preset conditions, the candidate segmentation layer can be used as the target segmentation layer. The following is an explanation of the method for determining the degree of clustering of the feature data set output by the candidate segmentation layer. For each candidate segmentation layer, the category to which each feature data in the feature data set output by the candidate segmentation layer belongs is determined according to the category to which each newly added sample in the newly added sample set belongs. For example, for a target classification model used to classify tiger images, lion images, and cat images, the newly added sample set includes multiple tiger images, multiple lion images, and multiple cat images, that is, the categories of the newly added samples in the newly added sample set are tiger, lion, and cat. Then, after the image (tiger image) with the category of tiger is input into the target neural network model, the category corresponding to the feature data output by the candidate segmentation layer is tiger.

[0087] In this way, the feature data of each category output by each candidate segmentation layer can form a feature data subset of this category. For each category of feature data subset, calculate the feature mean corresponding to each feature data in the feature data subset. Among them, the feature data can be a feature matrix, and the feature mean corresponding to each feature data in the feature data subset is calculated, that is, the mean of each feature matrix in the feature data subset is calculated, and the mean is also a feature matrix. Then, calculate the cluster distance between each feature data in the feature data subset of each category and the feature mean corresponding to the feature data subset. Among them, the cluster distance can be Euclidean distance, cosine distance, etc. Then, add the cluster distance calculated for each feature data and divide it by the number of feature data of this category, and you can get the first cluster distance mean corresponding to the feature data subset of this category.

[0088] Then, for each candidate segmentation layer, the first clustering distance means corresponding to the feature data subsets of each category are added and divided by the number of categories to obtain the second clustering distance mean corresponding to the feature data set output by the candidate segmentation layer. The second clustering distance mean corresponding to the feature data set output by each candidate segmentation layer is the clustering degree of the feature data set output by the candidate segmentation layer.

[0089] Then, among the second cluster distance means corresponding to the feature data sets output by all candidate segmentation layers, a target cluster distance mean that meets the preset conditions is determined. The candidate segmentation layer corresponding to the target cluster distance mean is determined as the target segmentation layer. There are many methods for determining the target segmentation layer among the candidate segmentation layers, and several of them are listed below for illustration.

[0090] Method 1: A third cluster distance mean value less than the first preset threshold value may be determined, and the minimum value among the third cluster distances may be determined as the target cluster distance mean value. The candidate segmentation layer corresponding to the target cluster distance mean value may be determined as the target segmentation layer.

[0091] Method 2: Determine the fourth cluster distance mean less than the second preset threshold value among the second cluster distance means corresponding to the feature data set output by all candidate segmentation layers. If there is a fifth cluster distance mean less than the third preset threshold value among the fourth cluster distance means, then among the candidate segmentation layers corresponding to the fifth cluster distance mean, the candidate segmentation layer farthest from the input layer in the target neural network model is determined as the target segmentation layer; if there is no fifth cluster distance mean less than the third preset threshold value, then the minimum value among the fourth cluster distance means is determined as the target cluster distance mean. The candidate segmentation layer corresponding to the target cluster distance mean is determined as the target segmentation layer.

[0092] Step 104: training the network layer after the target segmentation layer based on the newly added sample set.

[0093] In implementation, when training the target neural network model, the parameters corresponding to the target segmentation layer and the previous network layers determined above can be fixed, that is, in this training, they are not adjusted, but only the parameters of the network layers after the target segmentation layer are adjusted. It should be noted here that, depending on the scenario in which the target neural network model is used, the sample set used when training the target neural network model may also be different. Or different sample sets are used for different user-side permissions, so as to achieve the purpose of privacy protection. The following describes the training of the target neural network model in several possible scenarios.

[0094] Scenario 1: Scenario with high privacy protection requirements. The target neural network model is a neural network model trained using the image monitoring data of the first privacy scene as training data. Here, the first privacy scene can be a place with high privacy protection requirements, such as a school, hospital, etc. Then, after using the image monitoring data of the first privacy scene to train the neural network model to obtain the target neural network model, if the target neural network model is to be applied to the second privacy scene, a newly added sample set can be obtained, that is, the image monitoring data of the second privacy scene. Then, the target neural network model is trained using the image monitoring data of the second privacy scene, instead of training it together with the historical sample set (image monitoring data of the first privacy scene). Here, the second privacy scene can also be a school, hospital, etc.

[0095] For the case where the target neural network model is not trained in conjunction with the use of historical sample sets, there may also be the following scenario. The image recognition network downloaded by the user end on the AI ​​(Artificial Intelligence) open platform is the target neural network, and the user end needs to apply the image recognition network to the second privacy scenario. In this case, only the image monitoring data (newly added sample set) of the second privacy scenario stored locally can be used to train the image recognition network.

[0096] Scenario 2: Scenario with low privacy protection requirements or public scene. The target neural network model is a neural network model trained with the image monitoring data of the first non-privacy scene as training data. Here, the first non-privacy scene can be a place with low privacy protection requirements, such as a chain restaurant kitchen, public transportation, etc. Then, after the target neural network model is obtained by training the neural network model with the image monitoring data of the first non-privacy scene, the image monitoring data of the first non-privacy scene can be stored as a historical sample set for subsequent training. If the target neural network model is to be applied to the second non-privacy scene, a newly added sample set, i.e., the image monitoring data of the second non-privacy scene, can be obtained. At the same time, the stored historical sample set, i.e., the image monitoring data of the first non-privacy scene, can also be obtained. Then, the target neural network model is trained together with the image monitoring data of the second privacy scene and the image monitoring data of the first non-privacy scene. Here, the second privacy scene can also be a school, hospital, etc.

[0097] In addition, when the target neural network model is trained together with the newly added sample set and the historical sample set, the network layer after the target segmentation layer can also be trained by feature playback. That is, the newly added sample set is input into the target neural network model, and the feature data set outputted at the target segmentation layer is first obtained, and then the feature data set outputted by the stored historical samples at the target segmentation layer is obtained. Use these two feature data sets as sample inputs to train the network layer after the target segmentation layer. It should be noted here that each time a sample is used to train the neural network model, the sample and the feature data outputted by each network layer can be stored and deleted when the preset storage time is reached, or when the storage space reaches the preset threshold, the feature data with a longer storage time can be deleted.

[0098] After acquiring the newly added sample set, a feature data set output by at least one candidate segmentation layer in the target neural network model is obtained. Then, the target segmentation layer can be determined in the candidate segmentation layer based on the output feature data set. Subsequently, when the target neural network model is trained using the newly added samples, the network layer after the target segmentation layer can be directly trained without the need to train the target segmentation layer and the network layer before it. It can be seen that the network layers that need to be trained may not be all the network layers, which improves the training efficiency of the neural network model.

[0099] Through the above steps, it is no longer necessary for the lighting engineer to repeatedly input different lighting parameters for testing, which solves the problem of low efficiency caused by the lighting engineer's continuous debugging when adjusting the lighting according to multimedia information in the related art, thereby improving the efficiency of lighting adjustment during shooting.

[0100] In this embodiment, an electronic device is provided, including a memory and a processor. The memory stores a computer program, and the processor is configured to run the computer program to execute the method in the above embodiment.

[0101] The above program can be run in the processor, or it can also be stored in the memory (or computer-readable medium), which includes permanent and non-permanent, removable and non-removable media. Information storage can be achieved by any method or technology. Information can be computer-readable instructions, data structures, modules of programs or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.

[0102] These computer programs can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps of the functions specified in one or more blocks can be implemented by different modules corresponding to different steps.

[0103] Such a device or system is provided in this embodiment. The system or device is used to implement the functions of the method in the above embodiment, and each module in the system or device corresponds to each step in the method, which has been described in the method and will not be repeated here.

[0104] In this embodiment, a shooting illumination matching system based on an intelligent light field is provided, characterized in that it includes: an acquisition module, which acquires first multimedia information, wherein the first multimedia information includes at least one of the following: an image, a virtual 3D scene, and a video; a determination module, which determines illumination information from the first multimedia information, wherein the illumination information is used as a basis for controlling the illumination of a light-emitting device; an adjustment module, which is used to adjust the illumination parameters of the light-emitting device according to the illumination information; a shooting module, which is used to control the illumination of the light-emitting device according to the illumination parameters, and then perform shooting processing to obtain second multimedia information; and a saving module, which is used to compare the first multimedia information with the second multimedia information, and save the illumination parameters of the light-emitting device when the difference between the first multimedia information and the second multimedia information meets a predetermined condition.

[0105] Optionally, the determination module is used to: identify an object model from the first multimedia information; and determine the lighting information according to ambient lighting of the object model.

[0106] Optionally, the determination module is used to: identify the overall ambient lighting of the first multimedia information from the entire first multimedia information; and determine the lighting information according to the overall ambient lighting.

[0107] Optionally, the determination module is used to: read ambient lighting from the first multimedia information to generate a high dynamic range image; map pixel coordinates in the high dynamic range image to spherical light position coordinates, and convert color coordinates of pixels or pixel blocks in the high dynamic range image into light color and brightness.

[0108] Optionally, the shooting module is used to: when there are multiple groups of light-emitting parameters, use each group of light-emitting parameters to control the light-emitting device and shoot to obtain second multimedia information corresponding to the group of light-emitting parameters.

[0109] In this embodiment, a physical model shooting lighting processing system based on intelligent light field is also provided, including: a second acquisition module, used to obtain light-emitting parameters for controlling a light-emitting device, wherein the light-emitting device includes a plurality of light-emitting entities at different positions, and each light-emitting entity can be independently controlled by the light-emitting parameters; the light-emitting parameters are determined according to the ambient lighting in the first multimedia information provided in advance; a sending module, used to send the light-emitting parameters of the light-emitting device to the light-emitting device; wherein the degree to which the ambient lighting generated by the light-emitting device after emitting light according to the light-emitting parameters is close to the ambient lighting in the first multimedia information satisfies a predetermined condition; and a control module, used to control the shooting device to shoot after the light-emitting device emits light according to the light-emitting parameters.

[0110] Optionally, the second acquisition module is used to: determine lighting information from the first multimedia information, wherein the lighting information is used as a basis for controlling the lighting of a light-emitting device; adjust the lighting parameters of the light-emitting device according to the lighting information; after controlling the lighting of the light-emitting device according to the lighting parameters, perform photographing processing to obtain second multimedia information; compare the first multimedia information and the second multimedia information, and save the lighting parameters of the light-emitting device if the difference between the first multimedia information and the second multimedia information meets predetermined conditions.

[0111] Optionally, the second acquisition module is used to: identify an object model from the first multimedia information, and determine the lighting information based on the ambient lighting of the object model; or, identify the overall ambient lighting of the first multimedia information from the entire first multimedia information, and determine the lighting information based on the overall ambient lighting.

[0112] Optionally, the second acquisition module is used to: obtain multiple groups of training data, wherein each group of training data includes input data and output data, the input data is media information, and the output data is an object model included in the media information; determine whether the number of the multiple groups of training data exceeds a threshold, wherein the threshold is pre-configured; if the threshold is exceeded, use the multiple groups of training data to train a neural network in a supervised machine learning training method; input the first media information into the neural network, and the neural network outputs the object model.

[0113] Optionally, the second acquisition module is used to: after obtaining the neural network, continue to acquire training data, and when the training data that continues to be acquired exceeds a second threshold, use the training data that continues to be acquired to perform incremental training on the neural network.

[0114] The above-mentioned embodiments solve the problem of artificial processing of restoring the lighting conditions of the shooting scene in the related art, thereby improving the efficiency and accuracy of restoring the lighting conditions.

[0115] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A method for photographing and processing lighting for a physical model based on an intelligent light field, characterized in that: include: Acquire a lighting parameter for controlling a lighting device, wherein the lighting device includes a plurality of lighting entities at different positions, each lighting entity being independently controllable by the lighting parameter; the lighting parameter is determined according to ambient light in the first multimedia information provided in advance; Sending the lighting parameters of the lighting device to the lighting device; wherein the degree to which the ambient lighting generated by the lighting device after emitting light according to the lighting parameters is close to the ambient lighting in the first multimedia information satisfies a predetermined condition; After the light-emitting device emits light according to the light-emitting parameters, the photographing device is controlled to photograph.

2. The method according to claim 1, characterized in that The luminous parameters are determined according to the ambient lighting in the first multimedia information provided in advance and include: Determining illumination information from the first multimedia information, wherein the illumination information is used as a basis for controlling the illumination of the light emitting device; adjusting the light-emitting parameters of the light-emitting device according to the light information; After controlling the light-emitting device to emit light according to the light-emitting parameters, photographing and processing are performed to obtain second multimedia information; the first multimedia information and the second multimedia information are compared, and when the difference between the first multimedia information and the second multimedia information meets predetermined conditions, the light-emitting parameters of the light-emitting device are saved.

3. The method according to claim 2, characterized in that Determining illumination information from the first multimedia information includes: An object model is identified from the first multimedia information, and the illumination information is determined according to the ambient illumination of the object model; or, the overall ambient illumination of the first multimedia information is identified from the entire first multimedia information, and the illumination information is determined according to the overall ambient illumination.

4. The method according to claim 3, characterized in that Identifying an object model from the first multimedia information includes: Acquire multiple sets of training data, wherein each set of training data includes input data and output data, the input data is media information, and the output data is an object model included in the media information; Determining whether the number of the plurality of sets of training data exceeds a threshold, wherein the threshold is pre-configured; When the threshold is exceeded, a neural network is obtained by training the plurality of sets of training data in a supervised machine learning training mode; The first media information is input into the neural network, and the neural network outputs the object model.

5. The method according to claim 4, characterized in that Also includes: After the neural network is obtained, training data is continuously acquired, and when the training data continuously acquired exceeds a second threshold, incremental training is performed on the neural network using the training data continuously acquired.

6. A lighting processing system for photographing a physical model based on an intelligent light field, characterized in that: include: a second acquisition module, configured to acquire a light-emitting parameter for controlling a light-emitting device, wherein the light-emitting device includes a plurality of light-emitting entities at different positions, each of which can be independently controlled by the light-emitting parameter; and the light-emitting parameter is determined according to the ambient light in the first multimedia information provided in advance; A sending module, used to send the lighting parameters of the lighting device to the lighting device; wherein the degree to which the ambient lighting generated by the lighting device after emitting light according to the lighting parameters is close to the ambient lighting in the first multimedia information satisfies a predetermined condition; The control module is used to control the shooting device to shoot after the light-emitting device emits light according to the light-emitting parameters.

7. The system according to claim 6, characterized in that The second acquisition module is used for: Determining illumination information from the first multimedia information, wherein the illumination information is used as a basis for controlling the illumination of the light emitting device; adjusting the light-emitting parameters of the light-emitting device according to the light information; After controlling the light-emitting device to emit light according to the light-emitting parameters, photographing and processing are performed to obtain second multimedia information; the first multimedia information and the second multimedia information are compared, and when the difference between the first multimedia information and the second multimedia information meets predetermined conditions, the light-emitting parameters of the light-emitting device are saved.

8. The system according to claim 7, characterized in that The second acquisition module is used for: An object model is identified from the first multimedia information, and the illumination information is determined according to the ambient illumination of the object model; or, the overall ambient illumination of the first multimedia information is identified from the entire first multimedia information, and the illumination information is determined according to the overall ambient illumination.

9. The system according to claim 8, characterized in that The second acquisition module is used for: Acquire multiple sets of training data, wherein each set of training data includes input data and output data, the input data is media information, and the output data is an object model included in the media information; Determining whether the number of the plurality of sets of training data exceeds a threshold, wherein the threshold is pre-configured; When the threshold is exceeded, a neural network is obtained by training the plurality of sets of training data in a supervised machine learning training mode; The first media information is input into the neural network, and the neural network outputs the object model.

10. The system according to claim 9, characterized in that The second acquisition module is used for: After the neural network is obtained, training data is continuously acquired, and when the training data continuously acquired exceeds a second threshold, incremental training is performed on the neural network using the training data continuously acquired.

Images