Self-adaptive illumination method and system for 3D and AI visual sensing visible light movement

Abstract

The invention relates to the technical field of image illumination correction, and discloses a self-adaptive illumination method and system for a 3D and AI visual sensing visible light movement, and the method comprises the steps: cutting an image collection visual angle of the visible light movement, obtaining an image distribution grid system, collecting real-time image information, and substituting the real-time image information into the image distribution grid system, thereby achieving the self-adaptive illumination of the 3D and AI visual sensing visible light movement. The method comprises the steps of obtaining image distribution characteristics, analyzing the image distribution characteristics by using a pre-trained 3D structure recognition algorithm to form a 3D structure grid system, analyzing and executing an illumination correction scheme through a 3D structure correction algorithm to realize an adaptive illumination function, and remarkably improving the sense of reality and detail presentation of the image through 3D structure recognition and illumination correction. The adaptive illumination adjustment is realized, the image effect stability in different environments is ensured, the accurate positioning of the visual angle and the spatial information is enhanced through the gridding expression, and the problems that the adaptive illumination performance of the visible light movement in the prior art is lack of three-dimensional analysis and the illumination correction effect is poor are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of image illumination correction, and in particular to an adaptive illumination method and system for a 3D and AI visual sensing visible light core. Background Art

[0002] Visible light movement is one of the core components of modern optoelectronic technology and is widely used in smart cameras, surveillance equipment, drone vision, augmented reality (AR) and virtual reality (VR). Its function is to collect image information based on visible light and realize the understanding and presentation of the scene through digital processing.

[0003] In the prior art, automatic optical correction technology can be used to automatically correct the illumination of images captured by visible light movements. However, traditional image acquisition methods usually lack the precise distribution and analysis of the three-dimensional structure of the acquisition perspective, resulting in image information being limited to two-dimensional representation and unable to fully express spatial levels and structures, resulting in poor illumination correction effects. Summary of the invention

[0004] The purpose of the present invention is to provide an adaptive lighting method and system for a 3D and AI visual sensing visible light movement, aiming to solve the problem that the adaptive lighting performance of the visible light movement in the prior art lacks three-dimensional analysis and has poor lighting correction effect.

[0005] The present invention is implemented as follows. In a first aspect, the present invention provides an adaptive illumination method for a 3D and AI visual sensing visible light core, comprising: The image acquisition viewing angle of the visible light core is cut into viewing angle regions to obtain an image distribution grid system; wherein the image distribution grid system is used to express the spatial representative meaning of each specific position of the image acquisition viewing angle; Based on the visible light core, image information is collected for the image collection viewing angle to obtain real-time image information corresponding to the image collection viewing angle; Substituting the real-time image information into the image distribution grid system, and allowing the image distribution grid system to perform feature distribution analysis on the real-time image information, so as to obtain image distribution features of the real-time image information corresponding to the image distribution grid system; Performing recognition analysis of 3D structural relationships on the image distribution features according to a pre-trained 3D structural recognition algorithm, and adjusting the grid distribution of the image distribution grid system according to the results of the recognition analysis to obtain the 3D structural grid system of the real-time image information; According to the pre-trained 3D structure correction algorithm, the 3D structure grid system is subjected to node identification and illumination correction scheme analysis of key correction nodes, so as to obtain a plurality of key correction nodes of the 3D structure grid system and illumination correction schemes corresponding to each of the key correction nodes; Image illumination correction processing is performed on each of the key correction nodes according to the illumination correction scheme of each of the key correction nodes to realize the adaptive illumination function of the visible light core.

[0006] In a second aspect, the present invention provides an adaptive illumination system for a 3D and AI visual sensing visible light engine, which is used to implement an adaptive illumination method for a 3D and AI visual sensing visible light engine as described in any one of the first aspects, including: A viewing angle division module is used to cut the image acquisition viewing angle of the visible light core into viewing angle regions to obtain an image distribution grid system; wherein the image distribution grid system is used to express the spatial representative meaning of each specific position of the image acquisition viewing angle; An image acquisition module, used to acquire image information of the image acquisition viewing angle based on the visible light core, so as to obtain real-time image information corresponding to the image acquisition viewing angle; A feature decomposition module, used for substituting the real-time image information into the image distribution grid system, and allowing the image distribution grid system to perform feature distribution analysis on the real-time image information, so as to obtain image distribution features of the real-time image information corresponding to the image distribution grid system; A 3D recognition module, used to perform recognition and analysis of 3D structural relationships on the image distribution features according to a pre-trained 3D structure recognition algorithm, and to adjust the grid distribution of the image distribution grid system according to the result of the recognition and analysis, so as to obtain the 3D structural grid system of the real-time image information; A 3D correction module, used to perform node recognition and illumination correction scheme analysis on the key correction nodes of the 3D structure grid system according to a pre-trained 3D structure correction algorithm, so as to obtain a number of key correction nodes of the 3D structure grid system and illumination correction schemes corresponding to each of the key correction nodes; The correction execution module is used to perform image illumination correction processing on each of the key correction nodes according to the illumination correction scheme of each of the key correction nodes, so as to realize the adaptive illumination function of the visible light core.

[0007] The present invention provides an adaptive illumination method for a 3D and AI visual sensing visible light core, which has the following beneficial effects: The present invention obtains an image distribution grid system by cutting the image acquisition viewing angle of the visible light movement, collects real-time image information and substitutes it into the image distribution grid system to obtain image distribution characteristics, uses a pre-trained 3D structure recognition algorithm to analyze the image distribution characteristics to form a 3D structure grid system, analyzes and executes the lighting correction plan through a 3D structure correction algorithm to achieve an adaptive lighting function, and significantly improves the realism and detail presentation of the image through 3D structure recognition and lighting correction, achieves adaptive lighting adjustment, ensures stable image effects in different environments, and enhances the precise positioning of viewing angles and spatial information through grid expression, providing a more accurate data basis for subsequent image analysis, and solving the problems of lack of three-dimensional analysis and poor lighting correction effect in the adaptive lighting performance of the visible light movement in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 It is a schematic diagram of the steps of an adaptive illumination method for a 3D and AI visual sensing visible light core provided by an embodiment of the present invention; Figure 2 It is a structural schematic diagram of an adaptive lighting system for a 3D and AI vision sensing visible light movement provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0009] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0010] The implementation of the present invention is described in detail below in conjunction with specific embodiments.

[0011] Reference Figure 1 , Figure 2 As shown, a preferred embodiment of the present invention is provided.

[0012] In a first aspect, the present invention provides an adaptive illumination method for a 3D and AI visual sensing visible light core, comprising: S1: cutting the image acquisition angle of view of the visible light core into angle regions to obtain an image distribution grid system; wherein the image distribution grid system is used to express the spatial representative meaning of each specific position of the image acquisition angle of view; S2: collecting image information of the image collection viewing angle based on the visible light core to obtain real-time image information corresponding to the image collection viewing angle; S3: Substituting the real-time image information into the image distribution grid system, and allowing the image distribution grid system to perform feature distribution analysis on the real-time image information, so as to obtain image distribution features of the real-time image information corresponding to the image distribution grid system; S4: performing recognition analysis of 3D structural relationships on the image distribution features according to a pre-trained 3D structure recognition algorithm, and adjusting the grid distribution of the image distribution grid system according to the recognition analysis result to obtain the 3D structural grid system of the real-time image information; S5: performing node identification and illumination correction scheme analysis on the key correction nodes of the 3D structure grid system according to the pre-trained 3D structure correction algorithm, so as to obtain a plurality of key correction nodes of the 3D structure grid system and illumination correction schemes corresponding to each of the key correction nodes; S6: performing image illumination correction processing on each of the key correction nodes according to the illumination correction scheme of each of the key correction nodes, so as to realize the adaptive illumination function of the visible light core.

[0013] Specifically, in step S1 of the embodiment provided by the present invention, the image acquisition viewing angle of the visible light movement is determined, which is usually related to the field of view (FOV) of the camera and parameters such as the size and resolution of the image sensor. The viewing angle defines the spatial area that the camera can "see".

[0014] More specifically, after determining the image acquisition viewing angle, the next step is to divide the viewing angle into multiple small areas to form a viewing angle "grid". This step is equivalent to uniformly or non-uniformly cutting the viewing space, dividing the larger field of view area into multiple small units, each unit represents the spatial distribution of the image within the area.

[0015] More specifically, a grid system is constructed based on the cut viewing area. Each small unit of the grid system represents the spatial representation of a specific position in the image acquisition viewing angle. The structure of this grid system can be two-dimensional, three-dimensional, or a more complex distribution model based on the spatial coordinate system. The grid can adopt a regular grid (such as uniform division) or an irregular grid (such as dynamic adjustment according to factors such as image content and scene complexity).

[0016] More specifically, in the grid system, each grid unit corresponds to a specific spatial position in the viewing angle. Through the design of the grid system, each pixel, area or scene element in the image acquisition viewing angle can be mapped into the space, thereby obtaining the specific position and spatial representative meaning of the image acquisition viewing angle in the three-dimensional space. For example, a grid unit may correspond to an object or area in the image, while other units correspond to different objects or background areas.

[0017] It can be understood that through the perspective area cutting and the construction of the grid system, each position in the image acquisition perspective can be accurately mapped to a specific position in space, which enables each acquisition perspective to be accurately represented in space, facilitating further image processing and spatial analysis. This spatial mapping provides a basis for subsequent image analysis, three-dimensional modeling, target detection and other tasks. The establishment of the grid system not only helps to spatially represent the position of the image acquisition perspective, but also provides a framework for feature extraction from different areas of the image. Each grid unit can be used as a basic unit for image processing, which facilitates the analysis and processing of image features at different positions. The perspective area cutting and grid system design provide a flexible way to handle scenes of different complexities. For more complex scenes, the grid can be more detailed to better capture the details of the image. For simpler scenes, the grid can be appropriately simplified to improve computational efficiency.

[0018] It should be noted that the application scenarios of the visible light movement include fixed setting of viewing angle and instant setting of viewing angle. The fixed setting of viewing angle is to fix the visible light movement toward a certain viewing angle for long-term image acquisition, and the instant setting of viewing angle is to start the visible light movement and direct it toward a new viewing angle for real-time image acquisition. It is not difficult to see that no matter which application scenario, the visible light movement can automatically perform image acquisition viewing angles and subsequent analysis.

[0019] Specifically, in step S2 of the embodiment provided by the present invention, ensure that the image acquisition device is correctly configured so that an image that meets the required viewing angle can be captured. The settings of the visible light movement generally include: the position, orientation, focal length, exposure setting, resolution, etc. of the camera. After determining the viewing angle and configuring the image acquisition device, start collecting real-time image information. Real-time image acquisition is generally performed through camera sensors (such as CCD or CMOS sensors). These sensors convert the collected visible light signals into digital images. Through the image acquisition device, image data is obtained from the set viewing angle. These data will include the light intensity and color information of each pixel.

[0020] More specifically, preliminary image preprocessing may be performed on the collected real-time image information, such as denoising, illumination correction, white balance adjustment, etc., to ensure that the data quality meets the subsequent processing requirements.

[0021] It can be understood that by collecting images through the visible light movement, image data of the required perspective can be obtained in real time and continuously, which is suitable for applications such as dynamic scene monitoring, visual tracking, augmented reality, etc. Image acquisition is carried out in real time, which means that the system can reflect the latest image information of the current scene or environment.

[0022] Specifically, in step S3 of the embodiment provided by the present invention, the real-time image information (including image attributes such as illumination, color, texture, etc.) collected by the visible light movement is substituted into the previously defined image distribution grid system. At this time, each grid unit will be mapped to the corresponding area in the image, representing the image information at a specific location.

[0023] More specifically, the image distribution grid system disassembles and analyzes the image information through a series of image processing algorithms, which extract spatial features from the image, such as color distribution, brightness gradient, edge morphology, texture features, etc. These features can describe the spatial layout and structural characteristics of the image. The feature information of the image will be disassembled according to the division of the grid, and the image data in each grid unit will be extracted and processed to obtain the image features of the location. For example, a grid unit may extract the feature value of the area by analyzing the brightness and texture of its area.

[0024] More specifically, by decomposing the features of real-time image information, the image distribution grid system will output a new feature map, which is a summary of the spatial distribution of real-time image information in the grid system. This result shows the structure and distribution characteristics of the image at different spatial positions, including regional brightness, color changes, object edges, texture patterns, etc. Each grid unit now has a corresponding feature vector. The spatial distribution of these feature vectors can reflect the spatial structure and feature changes of the image in different regions.

[0025] More specifically, the final generated image distribution features can be stored in the system as the basis for subsequent image analysis, processing, or other tasks. At this time, the grid system provides the necessary spatial feature data for subsequent 3D structure analysis, lighting adjustment and other tasks.

[0026] It is understandable that by substituting image information into a grid system and performing feature decomposition, the structure and distribution of the image in space can be more clearly represented. This method can help the system understand the differences between different areas in the image, especially in complex or dynamic environments. Through grid processing, image features can be extracted at different resolutions and scales, allowing the system to capture spatial information at different levels and with different accuracies. This hierarchical feature extraction improves the flexibility and accuracy of image analysis, and the obtained image distribution features can provide rich data support for subsequent image analysis.

[0027] Specifically, in step S4 of the embodiment provided by the present invention, the image distribution features are analyzed using a pre-trained 3D structure recognition algorithm (such as a deep learning network, a convolutional neural network (CNN), an image registration algorithm, etc.). These algorithms can identify the three-dimensional spatial relationship between different areas in the image, such as depth information, object shape, spatial layout, etc. The 3D recognition algorithm will infer the depth information and spatial structure of the scene corresponding to the image based on information such as lighting, shadows, and texture changes in the image. This step will generate three-dimensional model data about the object's position, size, direction, and relative position to other objects.

[0028] More specifically, the layout of the image distribution grid system is adjusted according to the results of the 3D structure recognition analysis. This means that after obtaining the spatial features of the image, the original two-dimensional grid (or preliminary grid distribution) will be optimized to adapt to the three-dimensional structure of the image content. During the adjustment process, the size, shape, distribution density, etc. of the grid units may change to more accurately represent the three-dimensional space in the image. For example, the grid in the area far away from the camera can be adjusted to a larger size, while the grid in the close area can be refined to capture more subtle spatial features. For parts of the scene with complex structures or curved surface features, the grid can be non-uniformly encrypted to accurately capture details.

[0029] More specifically, through the adjusted grid layout, an image distribution grid system that matches the real three-dimensional structure is finally generated. This system can represent the accurate distribution of image information in three-dimensional space and reflect the depth and stereoscopic structure of the image. At this time, each grid unit not only represents the two-dimensional attributes of the image (such as color, texture, etc.), but is also closely related to its position in three-dimensional space. The adjusted 3D structure grid system will be the final output result, which can be further used in application scenarios such as image reconstruction, augmented reality, virtual reality, or as input for subsequent image analysis, object recognition and other tasks.

[0030] It can be understood that through 3D structure recognition and grid adjustment, the three-dimensional spatial structure of the scene corresponding to the image can be restored more accurately, providing a higher-precision description of spatial information. Through grid adjustment, the spatial layout of image data can be more effectively optimized, so that the distribution of image information is more in line with the actual three-dimensional scene, which helps to improve the accuracy of image analysis and processing. By associating image features with three-dimensional structures, the system can better understand and process depth information in the image, identify the spatial position, posture and form of objects, etc. This is particularly important for the processing of complex scenes. The generated 3D structure grid system provides more real and accurate three-dimensional spatial information for subsequent image reconstruction, object recognition, path planning, virtual reality and other applications, and supports more complex and sophisticated processing processes.

[0031] Specifically, in step S5 of the embodiment provided by the present invention, the previously generated 3D structure grid system is taken as input, including its spatial position, depth information, texture features, etc., and a pre-trained 3D structure correction algorithm, which may include a deep learning network, an image enhancement algorithm, etc., is used to analyze the 3D structure grid system. The algorithm can automatically identify key nodes in the 3D structure grid, which are usually areas where the spatial structure changes significantly and the lighting effects require special processing.

[0032] More specifically, by analyzing the spatial position, depth, color, lighting and other characteristics of the grid nodes, the key nodes that need to be corrected are identified. Based on the output results of the algorithm, several key correction nodes are determined. These nodes may be the edges of objects, surface protrusions, recessed areas, or parts with significant lighting changes.

[0033] More specifically, establish the lighting model of the current scene, including the light source position, light intensity, reflection characteristics, etc. By analyzing the lighting conditions of the 3D structure grid system, determine the lighting conditions of each key node, and design corresponding lighting correction solutions for each key correction node, which involves: Lighting balance: adjust the lighting uniformity to avoid areas with too bright or too dark lighting, shadow processing: optimize shadow effects, eliminate unnatural shadows or enhance shadow details in specific areas, reflection correction: handle light reflection and refraction, especially for the lighting effects of highlight areas and transparent objects.

[0034] More specifically, the illumination correction scheme obtained from the analysis is applied to the 3D structure grid system to generate 3D image data after correction. Through visualization and verification, it is checked whether the correction effect meets expectations. If necessary, further adjustments and optimizations are made.

[0035] It is understandable that by performing lighting correction on key nodes, the overall quality of the image can be significantly improved, making the lighting more natural and uniform and enhancing the visual effect. Lighting correction can enable the 3D model to maintain high-quality visual effects under different lighting conditions, enhance the realism and detail performance of the model, and design specific correction schemes for different key nodes can more effectively handle complex lighting scenes and avoid the computational overhead and poor results caused by global lighting adjustments.

[0036] Specifically, in step S6 of the embodiment provided by the present invention, image illumination correction processing is performed on each key correction node according to the illumination correction scheme of each key correction node to realize the adaptive illumination function of the visible light core.

[0037] The present invention provides an adaptive illumination method for a 3D and AI visual sensing visible light core, which has the following beneficial effects: The present invention obtains an image distribution grid system by cutting the image acquisition viewing angle of the visible light movement, collects real-time image information and substitutes it into the image distribution grid system to obtain image distribution characteristics, uses a pre-trained 3D structure recognition algorithm to analyze the image distribution characteristics to form a 3D structure grid system, analyzes and executes the lighting correction plan through a 3D structure correction algorithm to achieve an adaptive lighting function, and significantly improves the realism and detail presentation of the image through 3D structure recognition and lighting correction, achieves adaptive lighting adjustment, ensures stable image effects in different environments, and enhances the precise positioning of viewing angles and spatial information through grid expression, providing a more accurate data basis for subsequent image analysis, and solving the problems of lack of three-dimensional analysis and poor lighting correction effect in the adaptive lighting performance of the visible light movement in the prior art.

[0038] Preferably, the step of cutting the image acquisition viewing angle of the visible light core into viewing area sections to obtain an image distribution grid system comprises: S11: Acquire an image acquisition viewing angle of a visible light movement, acquire a plurality of viewing angle images based on the image acquisition viewing angle, and extract common image information from the plurality of viewing angle images to generate viewing angle background information of the image acquisition viewing angle; wherein the image acquisition viewing angle is used to describe the spatial range that can be included when the visible light movement acquires an image, and the viewing angle background information is used to describe the image background when the visible light movement acquires an image; S12: performing spatial quadrant cutting analysis on the perspective background information to obtain a spatial quadrant framework of the perspective background information, and performing spatial depth analysis on the perspective background information according to the spatial quadrant framework to obtain spatial distribution characteristics of the perspective background information; wherein the spatial quadrant framework is used to perform basic division of spatial regions on the perspective background information, and the spatial distribution characteristics are used to express spatial representative meanings of specific locations of the perspective background information; S13: Refining the frame grid and annotating the grid features of the spatial quadrant frame of the image acquisition viewing angle according to the spatial distribution characteristics, thereby obtaining an image distribution grid system composed of a plurality of spatial grids.

[0039] Specifically, the image acquisition viewing angle of the visible light movement is obtained. The image acquisition viewing angle describes the spatial range of images that the visible light movement can capture, including the depth, width, height, etc. of the visible area. The current viewing angle information is obtained from the movement, which is usually based on the field of view (FOV) of the sensor, camera or optical system and the actual position and posture of the visible light movement.

[0040] More specifically, based on the image acquisition perspective, multiple perspective images at different angles are collected. Each image represents image data obtained from different viewpoints at different angles and distances. Common image information is extracted from these perspective images, and repeated background features, object features, lighting changes, etc. in the images are identified and extracted as perspective background information. These common information are helpful for subsequent spatial analysis and grid construction.

[0041] More specifically, the extracted perspective background information is subjected to spatial quadrant cutting analysis. The entire perspective background space is divided into multiple regions or quadrants (e.g., four quadrants, eight quadrants, or finer divisions). This spatial division helps to understand the layout of the image in space, and the dimension and method of cutting can be determined based on factors such as the size of the image, depth information, and object distribution.

[0042] More specifically, by performing a depth analysis on each quadrant in the spatial quadrant framework, the spatial depth characteristics of each quadrant are obtained, which can be accomplished through stereo vision technology, LIDAR, or a depth estimation algorithm based on camera calibration to determine the spatial depth distribution of each area and evaluate the position and distance of objects in the perspective background, thereby further optimizing the spatial structure of the image distribution grid.

[0043] More specifically, based on the depth analysis results, the spatial distribution characteristics of each area of ​​the perspective background information are expressed. These characteristics represent the specific spatial meaning of each position in the image space, such as distance, depth level, object distribution, etc. Spatial feature values ​​are annotated for each spatial area (or quadrant), such as depth value, object density, lighting changes, etc., which helps the subsequent grid division.

[0044] More specifically, based on the spatial distribution characteristics, the original spatial quadrant framework is refined, and each quadrant is further divided into grids to ensure higher accuracy and more spatial details in different spatial regions. The grids are refined in key areas based on information such as depth and feature density to ensure that the grids are finer in important areas. Larger grids can be used in areas with less changes, and feature annotations are added to each refined grid, such as depth value, light intensity, object position, etc. In this way, each grid unit will contain rich spatial information, which will help subsequent image processing, rendering or analysis.

[0045] More specifically, through the processing of the above steps, an image distribution grid system composed of several refined spatial grids is finally obtained. This system can accurately describe the distribution and detailed features of the image in space, and provide accurate data support for subsequent image processing, reconstruction, analysis and other tasks.

[0046] It can be understood that through spatial cutting and depth analysis of the image acquisition perspective, the spatial information of each area in the image can be analyzed more accurately, reducing the error caused by perspective deviation. By combining spatial quadrant and depth information, different objects and backgrounds in the perspective can be effectively processed, and the accuracy of spatial reconstruction can be improved. The refined and annotated grid system can more accurately describe the spatial structure of the image, especially in scenes with complex backgrounds or object distributions. The refined grid can capture more detailed information, and the optimization of the grid enables higher quality data support to be provided in subsequent image processing (such as rendering, analysis, modeling, etc.).

[0047] Preferably, the step of substituting the real-time image information into the image distribution grid system, and allowing the image distribution grid system to perform feature distribution analysis on the real-time image information to obtain image distribution features of the real-time image information corresponding to the image distribution grid system includes: S31: mapping the real-time image information to correspond to the image distribution grid system, so that each specific location of the real-time image information is allocated to each spatial grid of the image distribution grid system, thereby realizing the substitution of the real-time image information to correspond to the image distribution grid system; S32: analyzing the spatial meaning of the real-time image information at each specific location according to the image distribution grid system to obtain the spatial meaning characteristics of each specific location of the real-time image information; S33: Integrate the spatial meaning features of each specific position of the real-time image information to obtain image distribution features of the real-time image information corresponding to the image distribution grid system.

[0048] Specifically, the current real-time image information is collected to obtain data containing features such as spatial position, light intensity, texture, depth, etc. This information needs to be preprocessed (such as denoising and normalization) to adapt to the input requirements of the image distribution grid system.

[0049] More specifically, coordinate mapping is performed on each specific position of the real-time image information, and it is matched one-to-one with the grid cells in the image distribution grid system. Based on geometric transformation or projection transformation, the coordinate information in the real-time image is matched to the spatial grid of the image distribution grid, and the pixel value, depth value, texture feature, etc. of the real-time image information are allocated to the corresponding spatial grid position, so that each grid can reflect the image information of the area. If some real-time image areas span multiple grids, they are allocated to the corresponding grid cells according to proportion or weight.

[0050] More specifically, feature analysis is performed on real-time image information, including: depth analysis: extracting the depth changes of real-time images in space, marking foreground and background information, texture analysis: analyzing the texture details of the image (such as edges, smooth areas) to enhance spatial understanding, lighting analysis: evaluating lighting distribution, including brightness, shadows and other features, spatial correlation analysis: combining the existing features of the image distribution grid system to perform a detailed analysis of the spatial meaning of the real-time image and clarify the physical or semantic meaning of the information contained in each grid unit.

[0051] More specifically, the spatial features of each grid unit are integrated into the overall features. For example, based on the geometric distribution of the grid, the overall depth feature distribution is calculated, the texture features are aggregated, and the texture distribution map of the image is generated. Combined with the spatial framework of the image distribution grid system, the feature distribution data of the entire real-time image information is generated.

[0052] More specifically, the integrated image distribution features are used as output for further image processing tasks (such as rendering, recognition, analysis, etc.). The output may include gridded spatial features, feature distribution maps, depth distribution maps, etc.

[0053] It can be understood that by mapping real-time image information to the image distribution grid system, dynamic data can be efficiently combined with the static grid framework, so as to quickly complete the integration and feature analysis of real-time information. Combined with the spatial framework of the image distribution grid system, the spatial position and semantic meaning of real-time image information can be accurately expressed, and the interpretability of image features can be enhanced. Through grid refinement and feature allocation, not only can image features be analyzed on a global scale, but also in-depth local areas can be used to refine the details. The output image distribution features can be directly used for deep learning model input, image rendering optimization, object recognition and other tasks, thereby improving the processing effect of these tasks.

[0054] Preferably, the training step of the 3D structure recognition algorithm includes: S401: Construct a convolutional neural network model consisting of an input layer, a convolutional layer, a pooling layer, several fully connected layers and an output layer; S402: preparing training data for model training of the convolutional neural network model; wherein the training data includes various features to be analyzed and target analysis features corresponding to the features to be analyzed, the features to be analyzed are image distribution features based on an image distribution grid system, and the target analysis features are 3D structural relationships corresponding to the image distribution features; S403: Substituting the training data into the input layer; S404: the input layer receives the training data and transmits the training data to the convolution layer, the convolution layer is used to collect 3D relationships in image distribution features of the training data to obtain 3D relationship mapping features of the training data; S405: the pooling layer is used to perform a pooling operation on the 3D relationship mapping features of the training data extracted by the convolution layer to reduce the size of the 3D relationship mapping features of the training data and retain key features of the 3D relationship mapping features of the training data; S406: Each of the fully connected layers is used to perform continuous vector flattening processing on the 3D relationship mapping features of the training data processed by the pooling layer, so as to flatten the 3D relationship mapping features of the training data into one-dimensional vector features; wherein the one-dimensional vector features are used to perform basic graphical expression on the 3D relationship mapping features of the training data; S407: the output layer receives and outputs the one-dimensional vector feature of the 3D relationship mapping feature, and uses the one-dimensional vector feature of the 3D relationship mapping feature as a 3D structure recognition mapping relationship; S408: Generate a mapping algorithm unit based on the 3D structure recognition mapping relationship, and perform algorithm object and algorithm target deduction processing on the mapping algorithm unit according to the image distribution grid system to obtain an algorithm object unit and an algorithm target unit of the mapping algorithm unit; wherein the algorithm object unit is used to substitute the image distribution features corresponding to the image distribution grid system corresponding to the real-time image information, and the algorithm target unit is used to output the 3D structure relationship corresponding to the image distribution features; S409: Combining the mapping algorithm unit, the algorithm object unit and the algorithm target unit to obtain a 3D structure recognition algorithm.

[0055] Specifically, a convolutional neural network model consisting of an input layer, a convolutional layer, a pooling layer, several fully connected layers and an output layer is constructed.

[0056] More specifically, the input layer is used to receive the data to be trained. The input data is the image distribution features based on the image distribution grid system. The convolution layer is responsible for extracting local features from the input image data, especially the features of the 3D structural relationship in the image. The convolution layer will perform convolution operations through convolution filters (or kernels) to capture the local spatial relationship of the image. The pooling layer is used to reduce the size of the feature map after the convolution operation (dimensionality reduction) while retaining the most important spatial feature information. When processing 3D structures, the pooling layer can help extract key depth and geometric information and remove unimportant details. The fully connected layer receives the features passed from the pooling layer and continuously flattens these features. This layer compresses high-dimensional 3D spatial features into one-dimensional vectors for further analysis and mapping. The output layer will receive the flattened vectors processed by the fully connected layer and output the final 3D structure recognition mapping relationship. This mapping relationship represents the relationship between the spatial structure and features of the input data.

[0057] More specifically, training data is prepared in advance, and the training data includes features to be analyzed and target analysis features. The features to be analyzed are image distribution features based on the image distribution grid system, and these features usually include the spatial distribution, depth information, texture, etc. of the image. The target analysis features are 3D structural relationships corresponding to the input image distribution features, and these target features may include the shape, depth, spatial layout, etc. of the object, and are used to train the model to learn how to convert the input 2D image information into corresponding 3D structural information.

[0058] More specifically, the preprocessed training data (including the features to be analyzed and the target analysis features) is input into the input layer of the network. After passing through the input layer, the input data is passed to the convolution layer. The convolution layer extracts the spatial features in the image through multiple convolution filters (convolution kernels), including local 3D structural relationships. The core of the convolution operation is to slide the convolution kernel on the input image and calculate the response of the image features, thereby extracting the local features of the image, such as edges, textures, lighting, etc.

[0059] More specifically, the pooling layer receives the feature map output by the convolutional layer and performs a pooling operation (such as maximum pooling or average pooling) on ​​it. The pooling operation can reduce the spatial size of the feature map while retaining the most important feature information. Especially in 3D structure recognition, pooling helps to improve the network's robustness to depth information. Through the pooling operation, the network compresses the 3D spatial information in the image, reduces the amount of computation, and retains key information about spatial relationships.

[0060] More specifically, the output of the pooling layer is flattened and converted into a one-dimensional vector. This step compresses the 3D relationship mapping features into a single vector representation that can be further analyzed. This one-dimensional vector will serve as the basic graphical expression, representing the 3D spatial relationship of the input image. In this way, the network converts 3D image information into a one-dimensional vector feature, making subsequent analysis easier.

[0061] More specifically, in the end, the flattened feature vector will be passed to the output layer, which converts the feature vector into a 3D structure recognition mapping relationship. This mapping relationship is the spatial features of the 3D structure predicted by the model, such as the object's position, size, shape, posture, etc.

[0062] It can be understood that through the convolution layer and the pooling layer, CNN can effectively extract key information such as depth information, texture information, and the spatial layout of objects from the input data, so that the network can understand the 3D structure of the image. While retaining important information, the pooling layer can effectively reduce the size of the features, reduce the computational complexity, and improve the robustness of the model to the input data. Through flattening, the 3D structural information can be converted into a one-dimensional vector form, making subsequent analysis and recognition easier and helping to reduce computational costs. Through the output layer, the model can output accurate 3D structure recognition results, provide a 3D structure mapping relationship corresponding to each input data, and achieve efficient spatial understanding and recognition tasks.

[0063] Preferably, the steps of performing recognition analysis of 3D structural relationships on the image distribution features according to a pre-trained 3D structure recognition algorithm, and adjusting the grid distribution of the image distribution grid system according to the result of the recognition analysis to obtain the 3D structural grid system of the real-time image information include: S41: performing recognition analysis of the 3D structural relationship of the image distribution features according to a pre-trained 3D structural recognition algorithm to obtain the 3D structural relationship of the image distribution features; S42: performing 3D structural feature annotation on each spatial grid of the image distribution grid system according to the 3D structural relationship to obtain a grid spatial feature set of the image distribution grid system; S43: According to the grid space feature set, the image distribution grid system is subjected to a fitting change in the spatial grid corresponding to the 3D structure, so that the spatial grid of the image distribution grid system presents a grid distribution form that fits the 3D structure, thereby obtaining the 3D structural grid system of the real-time image information.

[0064] Specifically, first, the real-time image information is processed by a pre-trained 3D structure recognition algorithm. The image distribution features usually include spatial position information, texture, depth and other features related to the 3D structure. These input image distribution features are analyzed by a trained convolutional neural network (CNN) or other deep learning algorithms. The model will identify the 3D structural relationship in the image based on the spatial laws learned during training, such as the shape, size, depth, relative position, perspective transformation, etc. of the object. Through recognition and analysis, the algorithm will generate a 3D structural relationship diagram, which contains specific information about the spatial layout and structural features reflected by the input image, such as three-dimensional shape, depth distribution, etc.

[0065] More specifically, the image distribution grid system is connected with the identified 3D structural relationship. The image distribution grid system is a system that divides the space into multiple grid units, where each grid unit represents a local area of ​​the image. This process usually requires setting the grid size and position according to the spatial distribution characteristics of the image. In each grid unit, the 3D spatial features of each grid unit are annotated according to the results of 3D structure recognition. These annotations include depth information, texture information, spatial relationships of objects, etc. The specific annotation process involves: depth annotation: the depth value of the grid unit, shape annotation: the shape information of the object in the grid, such as planes, curves, edges, etc., spatial relationship annotation: the relative position and posture relationship between objects in the grid. These annotation results will form a "grid spatial feature set", which describes the spatial attributes of each grid unit and how these attributes are distributed in the overall 3D structure.

[0066] More specifically, according to the annotated 3D structural features, the positions and shapes of each grid unit in the image distribution grid system are adjusted to make them correspond more accurately to the changes in the 3D structure. The core purpose of this adjustment is to make the distribution of the grid more in line with the actual 3D spatial layout and ensure that the grid system can truly reflect the three-dimensional structure of the image. By adjusting the size, shape and position of the grid, each grid unit is made more accurate in the corresponding spatial area. For example, in certain areas of the image (such as the edges of objects or complex structures), the grid may need to be refined, while in other relatively flat areas, the size of the grid unit can be expanded.

[0067] More specifically, for dynamic scenes or complex 3D structures, the grid distribution needs to be adjusted in real time. Through the feedback mechanism, the grid structure is dynamically optimized according to the changes in real-time image data to maintain its high consistency with the image.

[0068] More specifically, the adjusted grid system eventually forms an accurate grid distribution that conforms to the 3D structure. The grid system can reflect the spatial characteristics of image information in real time and display 3D information such as image depth, texture, and object layout. The 3D structure grid system is output as the final result and can usually be used in subsequent image analysis, object recognition, augmented reality (AR), virtual reality (VR) and other application scenarios.

[0069] It is understandable that by using the pre-trained 3D structure recognition algorithm, the 3D spatial structure relationship can be accurately extracted from the input image distribution features to ensure that the three-dimensional sense and spatial layout of the image information are truly restored. The image distribution grid system is dynamically adjusted according to the 3D structure relationship, so that the grid distribution can adapt to different image content and 3D structure, avoiding the error caused by the fixed grid. The grid adjustment can adapt to complex structures and dynamic scenes, thereby enhancing the flexibility and robustness of the model. The adjustment of the grid system can not only better represent the 3D structure of the image, but also help improve the accuracy of subsequent image processing tasks (such as object detection, spatial reconstruction, perspective conversion, etc.). The adjusted grid can more accurately correspond to the key features and details in the 3D structure. Through precise grid distribution and structural fit, the interference of irrelevant areas and redundant calculations is reduced, making subsequent processing more efficient. For example, in the object recognition task, the fine grid distribution can improve the recognition accuracy, reduce misidentification and calculation. Through this grid adjustment method, a deeper level of spatial understanding can be achieved, helping the system not only understand the shape of objects in the image, but also recognize the spatial relationship between objects. This is of great significance for the understanding of complex scenes and the construction of virtual worlds.

[0070] Preferably, the training steps of the 3D structure correction algorithm include: Construct a 3D structure correction algorithm consisting of an algorithm execution unit, an algorithm input unit and an algorithm output unit; wherein the algorithm execution unit includes a scene characteristic analysis unit and a node correction mode analysis unit, the algorithm input unit is used to receive a 3D structure grid system, the scene characteristic analysis unit is used to analyze the scene characteristics of the 3D structure grid system to obtain the scene characteristics fed back by the 3D structure grid system, the node correction mode analysis unit is used to retrieve the corresponding node extraction and illumination correction analysis mode according to the scene characteristics obtained by the scene characteristic analysis unit, and perform node identification and illumination correction scheme analysis of key correction nodes on the 3D structure grid system according to the node extraction and illumination correction mode, and the algorithm output unit is used to output a number of key correction nodes of the 3D structure grid system obtained by the node correction mode analysis unit and the illumination correction scheme corresponding to each of the key correction nodes; The training step of the scene characteristic analysis unit includes: S511: pre-collecting original data sets of several types of scene characteristics, and performing set division on the original data sets to obtain training sets, verification sets and test sets corresponding to the scene characteristics of each type; wherein the original data sets include several pairs of image data and annotation data, the image data is used to describe a 3D structure grid system, and the annotation data is used to describe the scene characteristics corresponding to the 3D structure grid system; S512: constructing a neural convolutional network model, and performing model training on the neural convolutional network model according to the training set to obtain model training parameters for feeding back the data correlation between the image data and the labeled data in the training set, and returning the model training parameters to the neural convolutional network model to perform model parameter configuration on the neural convolutional network model, thereby obtaining a preliminarily trained neural convolutional network model; S513: In the process of training the neural convolutional network model using the training set, using the verification set to verify the model parameters of the neural convolutional network model to supervise the effect of the model training of the training set; S514: after obtaining the preliminarily trained neural convolutional network model, testing the model parameters of the neural convolutional network model according to the test set, and performing model correction on the neural convolutional network model according to the test result; S515: extracting algorithms and encapsulating execution programs on the trained neural convolutional network model to obtain the scene characteristic analysis unit; The training step of the node correction mode analysis unit includes: S521: pre-collecting a number of training data sets, wherein the training data sets include correction objects and correction results, wherein the correction objects are 3D structural grid systems, and the correction results are key correction nodes and illumination correction schemes of the 3D structural grid systems; S522: analyzing the correlation between the correction object and the correction result according to a deep learning algorithm to obtain a correction execution process between the correction object and the correction result; S523: Perform feature induction on the correction execution process to obtain a corresponding node correction mode, and construct the node correction mode analysis unit based on the node correction mode.

[0071] Specifically, the algorithm execution unit includes a scene characteristic analysis unit and a node correction mode analysis unit. The scene characteristic analysis unit is used to analyze the scene characteristics in the 3D structure grid system and extract key information related to the scene (such as object type, spatial layout, lighting characteristics, etc.). The node correction mode analysis unit calls the appropriate node extraction and lighting correction mode according to the feedback information provided by the scene characteristic analysis unit, and corrects the key correction nodes in the 3D structure grid system, and processes the adjustment of node recognition position and lighting effect.

[0072] More specifically, the algorithm input unit receives data of the 3D structure grid system and inputs the data into the algorithm for further analysis, and the algorithm output unit outputs the corrected 3D structure grid system, including corrected key nodes and corresponding lighting correction solutions.

[0073] More specifically, we collect original data sets related to 3D scene characteristics. The data sets consist of paired image data and annotation data. The image data is used to describe the 3D structure grid system (such as depth map, texture map, etc.), while the annotation data is used to describe the scene characteristics corresponding to the grid system (such as object position, lighting direction, perspective change, etc.). The original data sets are divided into training sets, validation sets, and test sets. The training set is used for model training, the validation set is used for intermediate tuning, and the test set is used to evaluate the final effect of the algorithm.

[0074] More specifically, a neural convolutional network (CNN) model is designed and constructed. CNN can automatically extract image features and perform feature mapping, which plays a vital role in 3D scene feature analysis. This network will learn spatial relationships, object contours, lighting changes and other information from the image. The CNN model is trained using a training set. During the training process, the neural network adjusts the weights through the back-propagation algorithm to minimize the prediction error. During training, appropriate optimization algorithms (such as Adam, SGD, etc.) need to be used to improve the performance of the model. The model is evaluated on the validation set, and the model hyperparameters are adjusted according to the results of the validation set to ensure that the model can generalize to different scene data.

[0075] More specifically, according to the scene characteristics (such as object position, lighting characteristics, etc.) obtained from the scene characteristic analysis unit, the node correction mode analysis unit will call different correction modes. The correction modes mainly include: Node extraction mode: According to the key nodes in the 3D structure grid system, the position information and corresponding scene characteristics of the key nodes are extracted; Lighting correction mode: Considering the impact of lighting conditions on the image and 3D structure, correction adjustments such as lighting intensity and direction are made.

[0076] More specifically, during the correction process, we analyze which nodes need to be corrected in the current scene, and make corresponding adjustments based on the scene characteristics. For example, some nodes may produce shadows due to insufficient lighting, and need to be corrected for lighting. Finally, the algorithm outputs several key correction nodes and provides corresponding lighting correction solutions for each key node. The output result is a revised version of the 3D structure grid system, including: the corrected node position, the effect of lighting correction (such as brightness, reflectivity, etc.).

[0077] It is understandable that the algorithm can adaptively adjust the grid system and lighting conditions according to the characteristics of different scenes to ensure that the 3D structure is more realistic and natural. Through the training of neural networks, scene characteristics and node correction mode analysis, efficient correction processing can be achieved, and the geometric shape and lighting effects of key nodes in the 3D structure grid can be optimized to ensure the accuracy of image quality and spatial sense. Through the lighting correction mode, the algorithm can effectively solve the image distortion problem caused by insufficient or uneven lighting, making the corrected image more visually realistic.

[0078] Preferably, the step of performing image illumination correction processing on each of the key correction nodes according to the illumination correction scheme of each of the key correction nodes to realize the adaptive illumination function of the visible light core includes: S61: performing node annotation processing on the real-time image information according to each of the key correction nodes to obtain a plurality of node annotations on the real-time image information; wherein the node annotations are used to determine the key correction nodes on the real-time image information; S62: Correcting the illumination parameters of the positions corresponding to the node labels according to the illumination correction schemes of the key correction nodes, so as to perform image illumination correction processing on the real-time image information.

[0079] Specifically, the positions of key correction nodes are determined in the real-time image information. Based on the known key correction nodes, the corresponding positions or areas are extracted from the real-time image, and these positions are labeled with nodes. The labeling information can indicate the specific positions of the correction nodes in the image. These node labels will help locate and process subsequent lighting correction operations.

[0080] More specifically, according to the lighting correction scheme, the lighting parameters of the positions corresponding to the node annotations are corrected. According to the lighting correction scheme of each key correction node, the lighting information of the corresponding nodes in the image is analyzed, and the parameters such as lighting intensity, color temperature, light source position, and lighting direction are adjusted to ensure that the lighting effect in the image is more in line with the actual situation of the scene. These parameters are optimized through mathematical modeling or lighting models (such as the Phong model, Blinn-Phong model, etc.) to eliminate errors caused by uneven lighting or environmental changes.

[0081] More specifically, lighting correction is performed on real-time image information to enhance the visual effect of the image and make it more consistent with the actual lighting scene. The corrected lighting parameters are used to re-render the image, adjust the brightness, shadows, highlights, etc., to achieve the effect of adaptive lighting. Combined with node labeling, it is ensured that lighting correction only affects the position of key nodes and avoids affecting other areas. The adjusted image will be more natural, highlighting the details of the corrected nodes and improving the overall quality and expressiveness of the image.

[0082] It is understandable that through the dynamic adjustment of the lighting correction scheme, the algorithm can automatically optimize the lighting effects of key nodes according to different lighting conditions in the real-time image. This can not only eliminate the impact caused by changes in ambient light, but also make the correction nodes more coordinated with the surrounding scenes. Through precise lighting correction processing, the realism of the image can be significantly improved. The corrected image can be more in line with the natural lighting environment of visual perception and improve the overall visual quality. This lighting adjustment based on node correction can make the key nodes in the 3D image more prominent, making the image richer and more realistic when presenting three-dimensional sense, depth and lighting levels, and enhancing the immersiveness of the 3D image. Lighting correction is not just processing the entire image, but through precise node annotation, personalized adjustments are made to the lighting requirements of each correction node, thereby improving the visual effects of key nodes. Through real-time calculation and adaptive adjustment, this correction method can adapt to a variety of complex lighting environments, ensure the accuracy and visual performance of the image in various environments, and has greater flexibility and robustness.

[0083] Reference Figure 2 As shown, in a second aspect, the present invention provides an adaptive illumination system for a 3D and AI visual sensing visible light engine, which is used to implement an adaptive illumination method for a 3D and AI visual sensing visible light engine as described in any one of the first aspects, including: A viewing angle division module is used to cut the image acquisition viewing angle of the visible light core into viewing angle regions to obtain an image distribution grid system; wherein the image distribution grid system is used to express the spatial representative meaning of each specific position of the image acquisition viewing angle; An image acquisition module, used to acquire image information of the image acquisition viewing angle based on the visible light core, so as to obtain real-time image information corresponding to the image acquisition viewing angle; A feature decomposition module, used for substituting the real-time image information into the image distribution grid system, and allowing the image distribution grid system to perform feature distribution analysis on the real-time image information, so as to obtain image distribution features of the real-time image information corresponding to the image distribution grid system; A 3D recognition module, used to perform recognition and analysis of 3D structural relationships on the image distribution features according to a pre-trained 3D structure recognition algorithm, and to adjust the grid distribution of the image distribution grid system according to the result of the recognition and analysis, so as to obtain the 3D structural grid system of the real-time image information; A 3D correction module, used to perform node recognition and illumination correction scheme analysis on the key correction nodes of the 3D structure grid system according to a pre-trained 3D structure correction algorithm, so as to obtain a number of key correction nodes of the 3D structure grid system and illumination correction schemes corresponding to each of the key correction nodes; The correction execution module is used to perform image illumination correction processing on each of the key correction nodes according to the illumination correction scheme of each of the key correction nodes, so as to realize the adaptive illumination function of the visible light core.

[0084] In this embodiment, for the specific implementation of each module in the above system embodiment, please refer to the above method embodiment, which will not be repeated here.

[0085] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An adaptive illumination method for 3D and AI visual sensing visible light core, characterized in that: include: The image acquisition viewing angle of the visible light core is cut into viewing angle regions to obtain an image distribution grid system; wherein the image distribution grid system is used to express the spatial representative meaning of each specific position of the image acquisition viewing angle; Based on the visible light core, image information is collected for the image collection viewing angle to obtain real-time image information corresponding to the image collection viewing angle; Substituting the real-time image information into the image distribution grid system, and allowing the image distribution grid system to perform feature distribution analysis on the real-time image information, so as to obtain image distribution features of the real-time image information corresponding to the image distribution grid system; Performing recognition analysis of 3D structural relationships on the image distribution features according to a pre-trained 3D structural recognition algorithm, and adjusting the grid distribution of the image distribution grid system according to the results of the recognition analysis to obtain the 3D structural grid system of the real-time image information; According to the pre-trained 3D structure correction algorithm, the 3D structure grid system is subjected to node identification and illumination correction scheme analysis of key correction nodes, so as to obtain a plurality of key correction nodes of the 3D structure grid system and illumination correction schemes corresponding to each of the key correction nodes; Image illumination correction processing is performed on each of the key correction nodes according to the illumination correction scheme of each of the key correction nodes to realize the adaptive illumination function of the visible light core.

2. The adaptive illumination method of the 3D and AI visual sensing visible light core according to claim 1, characterized in that: The steps of cutting the image acquisition viewing angle of the visible light core into viewing area to obtain the image distribution grid system include: Acquire an image acquisition viewing angle of a visible light movement, acquire a plurality of viewing angle images based on the image acquisition viewing angle, and extract common image information from the plurality of viewing angle images to generate viewing angle background information of the image acquisition viewing angle; wherein the image acquisition viewing angle is used to describe the spatial range that can be included when the visible light movement acquires an image, and the viewing angle background information is used to describe the image background when the visible light movement acquires an image; Performing spatial quadrant cutting analysis on the perspective background information to obtain a spatial quadrant framework of the perspective background information, and performing spatial depth analysis on the perspective background information according to the spatial quadrant framework to obtain spatial distribution characteristics of the perspective background information; wherein the spatial quadrant framework is used to perform basic division of spatial regions on the perspective background information, and the spatial distribution characteristics are used to express spatial representative meanings of specific locations of the perspective background information; The spatial quadrant framework of the image acquisition viewing angle is subjected to frame grid refinement and grid feature annotation according to the spatial distribution characteristics, thereby obtaining an image distribution grid system composed of a plurality of spatial grids.

3. The adaptive illumination method of the 3D and AI visual sensing visible light core according to claim 1, characterized in that: Substituting the real-time image information into the image distribution grid system, and allowing the image distribution grid system to perform feature distribution analysis on the real-time image information to obtain image distribution features of the real-time image information corresponding to the image distribution grid system, comprises: Mapping the real-time image information to the image distribution grid system so that each specific location of the real-time image information is allocated to each spatial grid of the image distribution grid system, thereby implementing the substitution of the real-time image information to the image distribution grid system; Analyzing the spatial meaning of the real-time image information at each specific location assigned according to the image distribution grid system to obtain the spatial meaning characteristics of each specific location of the real-time image information; An integration operation is performed on the spatial meaning features of each specific position of the real-time image information to obtain image distribution features of the real-time image information corresponding to the image distribution grid system.

4. The adaptive illumination method of the 3D and AI visual sensing visible light core according to claim 1, characterized in that: The training steps of the 3D structure recognition algorithm include: Construct a convolutional neural network model consisting of an input layer, a convolutional layer, a pooling layer, several fully connected layers, and an output layer; Preparing training data for model training of the convolutional neural network model; wherein the training data includes various features to be analyzed and target analysis features corresponding to the features to be analyzed, the features to be analyzed are image distribution features based on an image distribution grid system, and the target analysis features are 3D structural relationships corresponding to the image distribution features; Substituting the training data into the input layer; The input layer receives the training data and transmits the training data to the convolution layer, and the convolution layer is used to collect 3D relationships in image distribution features of the training data to obtain 3D relationship mapping features of the training data; The pooling layer is used to perform a pooling operation on the 3D relationship mapping features of the training data extracted by the convolution layer to reduce the size of the 3D relationship mapping features of the training data and retain key features of the 3D relationship mapping features of the training data; Each of the fully connected layers is used to perform continuous vector flattening processing on the 3D relationship mapping features of the training data processed by the pooling layer, so as to flatten the 3D relationship mapping features of the training data into one-dimensional vector features; wherein the one-dimensional vector features are used to perform basic graphical expression on the 3D relationship mapping features of the training data; The output layer receives and outputs the one-dimensional vector feature of the 3D relationship mapping feature, and uses the one-dimensional vector feature of the 3D relationship mapping feature as a 3D structure recognition mapping relationship; A mapping algorithm unit is generated based on the 3D structure recognition mapping relationship, and an algorithm object and an algorithm target are deduced for the mapping algorithm unit according to the image distribution grid system to obtain an algorithm object unit and an algorithm target unit of the mapping algorithm unit; wherein the algorithm object unit is used to substitute the image distribution features corresponding to the image distribution grid system corresponding to the real-time image information, and the algorithm target unit is used to output the 3D structure relationship corresponding to the image distribution features; The mapping algorithm unit, the algorithm object unit and the algorithm target unit are combined to obtain a 3D structure recognition algorithm.

5. The adaptive illumination method of the 3D and AI visual sensing visible light core according to claim 1, characterized in that: The steps of performing recognition analysis of 3D structural relationships on the image distribution features according to a pre-trained 3D structure recognition algorithm, and adjusting the grid distribution of the image distribution grid system according to the recognition analysis results to obtain the 3D structural grid system of the real-time image information include: Performing recognition and analysis of the 3D structural relationship of the image distribution features according to a pre-trained 3D structural recognition algorithm to obtain the 3D structural relationship of the image distribution features; Annotating 3D structural features of spatial grids at various locations of the image distribution grid system according to the 3D structural relationship to obtain a grid spatial feature set of the image distribution grid system; According to the grid space feature set, the image distribution grid system is subjected to fitting changes in the spatial grid corresponding to the 3D structure, so that the spatial grid of the image distribution grid system presents a grid distribution form that fits the 3D structure, thereby obtaining the 3D structural grid system of the real-time image information.

6. The adaptive illumination method of the 3D and AI visual sensing visible light core as claimed in claim 1, characterized in that: The training steps of the 3D structure correction algorithm include: Construct a 3D structure correction algorithm consisting of an algorithm execution unit, an algorithm input unit and an algorithm output unit; wherein the algorithm execution unit includes a scene characteristic analysis unit and a node correction mode analysis unit, the algorithm input unit is used to receive a 3D structure grid system, the scene characteristic analysis unit is used to analyze the scene characteristics of the 3D structure grid system to obtain the scene characteristics fed back by the 3D structure grid system, the node correction mode analysis unit is used to retrieve the corresponding node extraction and illumination correction analysis mode according to the scene characteristics obtained by the scene characteristic analysis unit, and perform node identification and illumination correction scheme analysis of key correction nodes on the 3D structure grid system according to the node extraction and illumination correction mode, and the algorithm output unit is used to output a number of key correction nodes of the 3D structure grid system obtained by the node correction mode analysis unit and the illumination correction scheme corresponding to each of the key correction nodes; The training step of the scene characteristic analysis unit includes: Collecting several types of original data sets of scene characteristics in advance, and dividing the original data sets into sets to obtain training sets, verification sets and test sets corresponding to various types of scene characteristics; wherein the original data sets include several pairs of image data and annotation data, the image data is used to describe the 3D structure grid system, and the annotation data is used to describe the scene characteristics corresponding to the 3D structure grid system; Constructing a neural convolutional network model, and performing model training on the neural convolutional network model according to the training set to obtain model training parameters for feeding back the data correlation between the image data and the labeled data in the training set, and returning the model training parameters to the neural convolutional network model to perform model parameter configuration on the neural convolutional network model, thereby obtaining a preliminarily trained neural convolutional network model; In the process of training the neural convolutional network model using the training set, the model parameters of the neural convolutional network model are verified using the verification set to supervise the effect of the model training of the training set; After obtaining the preliminarily trained neural convolutional network model, testing the model parameters of the neural convolutional network model according to the test set, and modifying the neural convolutional network model according to the test results; Extracting algorithms and encapsulating execution programs on the trained neural convolutional network model to obtain the scene characteristic analysis unit; The training step of the node correction mode analysis unit includes: Collecting several training data sets in advance, wherein the training data sets include correction objects and correction results, wherein the correction objects are 3D structural grid systems, and the correction results are key correction nodes and illumination correction schemes of the 3D structural grid systems; Analyzing the correlation between the correction object and the correction result according to a deep learning algorithm to obtain a correction execution process between the correction object and the correction result; The correction execution process is characterized and summarized to obtain a corresponding node correction mode, and the node correction mode analysis unit is constructed based on the node correction mode.

7. The adaptive illumination method of the 3D and AI visual sensing visible light core as claimed in claim 1, characterized in that: The steps of performing image illumination correction processing on each of the key correction nodes according to the illumination correction scheme of each of the key correction nodes to realize the adaptive illumination function of the visible light core include: Performing node annotation processing on the real-time image information according to each of the key correction nodes to obtain a plurality of node annotations on the real-time image information; wherein the node annotations are used to determine the key correction nodes on the real-time image information; According to the illumination correction scheme of each of the key correction nodes, the illumination parameter correction processing is performed on the position corresponding to each of the node annotations, so as to perform image illumination correction processing on the real-time image information.

8. An adaptive lighting system for 3D and AI visual sensing visible light movement, characterized in that: The method for implementing an adaptive illumination method of a 3D and AI visual sensing visible light core according to any one of claims 1 to 7 comprises: A viewing angle division module is used to cut the image acquisition viewing angle of the visible light core into viewing angle regions to obtain an image distribution grid system; wherein the image distribution grid system is used to express the spatial representative meaning of each specific position of the image acquisition viewing angle; An image acquisition module, used to acquire image information of the image acquisition viewing angle based on the visible light core, so as to obtain real-time image information corresponding to the image acquisition viewing angle; A feature decomposition module, used for substituting the real-time image information into the image distribution grid system, and allowing the image distribution grid system to perform feature distribution analysis on the real-time image information, so as to obtain image distribution features of the real-time image information corresponding to the image distribution grid system; A 3D recognition module, used to perform recognition and analysis of 3D structural relationships on the image distribution features according to a pre-trained 3D structure recognition algorithm, and to adjust the grid distribution of the image distribution grid system according to the result of the recognition and analysis, so as to obtain the 3D structural grid system of the real-time image information; A 3D correction module, used to perform node recognition and illumination correction scheme analysis on the key correction nodes of the 3D structure grid system according to a pre-trained 3D structure correction algorithm, so as to obtain a number of key correction nodes of the 3D structure grid system and illumination correction schemes corresponding to each of the key correction nodes; The correction execution module is used to perform image illumination correction processing on each of the key correction nodes according to the illumination correction scheme of each of the key correction nodes, so as to realize the adaptive illumination function of the visible light core.

Images