Multi-angle shooting implementation method adopting 3D and AI visual sensing visible light movement

Abstract

The invention relates to the technical field of image processing, and discloses a multi-angle shooting implementation method adopting a 3D and AI visual sensing visible light movement, which comprises the following steps of: acquiring an initial image of a shot object by using the visible light movement, and analyzing the initial image through an image shooting visual angle identification model to obtain an initial visual angle; analyzing the difference between the target vision angle and the initial vision angle based on the initial vision angle to obtain vision angle difference feature distribution, generating a basic simulation image according to the difference feature distribution, carrying out cross 3D feature recognition and angle conversion to generate an expanded simulation image, and obtaining a supplementary simulation image by using a multi-angle supplementary model, according to the method, through the 3D and AI visual technology, the multi-angle and high-quality shot images are automatically generated, the comprehensiveness and accuracy of image acquisition are effectively improved, and the problem that in the prior art, an ideal visual angle image cannot be obtained when the image is limited by a scene is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a method for implementing multi-angle shooting using a 3D and AI visual sensing visible light movement. Background Art

[0002] With the rapid development of computer vision technology and artificial intelligence, image processing and 3D reconstruction technology have been widely used in many fields. Traditional shooting methods rely on image acquisition from a single angle, which is usually limited by the shooting angle of view, resulting in incomplete shooting effects or missing details. In some cases, due to the scene limitations of the scene where the subject is located, the photographer cannot obtain the best image visual angle, and lacks image acquisition opportunities from more angles such as side view and overhead view, resulting in the photographer being unable to obtain the ideal perspective image. Summary of the invention

[0003] The purpose of the present invention is to provide a method for realizing multi-angle shooting using a 3D and AI visual sensing visible light movement, aiming to solve the problem in the prior art that an ideal viewing angle image cannot be obtained when limited by scene restrictions.

[0004] The present invention is implemented in this way. In a first aspect, the present invention provides a method for implementing multi-angle shooting using a 3D and AI visual sensing visible light core, comprising: Capturing an image of the photographed object through a visible light engine to obtain an initial visual image of the photographed object; Performing visual angle analysis processing on the initial visual image according to a pre-trained image shooting angle recognition model to obtain an initial visual angle of the initial visual image; Acquire a plurality of target visual angles for the photographed object, and perform visual angle difference analysis processing on each of the target visual angles based on the initial visual angle to obtain a visual angle difference feature distribution; wherein the visual angle difference feature distribution includes a plurality of visual angle difference features, and the visual angle difference feature is used to describe the visual angle difference between the initial visual angle and the target visual angle; Performing corresponding visual angle conversion processing on the initial visual image according to the visual angle difference characteristic distribution, so as to obtain basic simulation images corresponding to each of the target visual angles; Based on the basic simulated images of each target visual angle, cross-cutting 3D feature recognition and visual angle conversion are performed to obtain extended simulated images corresponding to each target visual angle; Performing image supplementation at multiple visual angles on the basic simulated image according to a pre-trained multi-angle supplementation model of the photographed object, so as to obtain supplemented simulated images corresponding to each of the target visual angles; The extended simulation image and the supplementary simulation image of the same target visual angle are fused to obtain an angle shot image of the target visual angle, and the angle shot images of the target visual angles together constitute a multi-angle shooting set of the shooting object; wherein the multi-angle shooting set is used to perform multi-angle image visual depiction of the shooting object.

[0005] The present invention provides a method for implementing multi-angle shooting using a 3D and AI visual sensing visible light core, which has the following beneficial effects: The present invention uses a visible light movement to collect an initial image of a photographed object, analyzes the initial image through an image shooting perspective recognition model to obtain an initial visual angle, analyzes the difference between the target visual angle and the initial visual angle based on the initial visual angle, obtains a visual angle difference feature distribution, generates a basic simulation image according to the difference feature distribution, performs cross-3D feature recognition and angle conversion to generate an extended simulation image, uses a multi-angle supplement model to obtain a supplementary simulation image, fuses the extended and supplementary simulation images, and generates a multi-angle shooting image set. The method automatically generates multi-angle, high-quality shooting images through 3D and AI vision technology, effectively improves the comprehensiveness and accuracy of image acquisition, and solves the problem in the prior art that an ideal perspective image cannot be obtained due to scene restrictions. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Figure 1 It is a schematic diagram of the steps of a method for implementing multi-angle shooting using a 3D and AI visual sensing visible light movement provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0007] 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.

[0008] The implementation of the present invention is described in detail below in conjunction with specific embodiments.

[0009] Reference Figure 1 As shown, a preferred embodiment of the present invention is provided.

[0010] The present invention provides a method for realizing multi-angle shooting using a 3D and AI visual sensing visible light core, comprising: S1: Capturing an image of a photographed object through a visible light engine to obtain an initial visual image of the photographed object; S2: performing visual angle analysis processing on the initial visual image according to a pre-trained image shooting angle recognition model to obtain an initial visual angle of the initial visual image; S3: Acquire a plurality of target visual angles for the photographed object, and perform visual angle difference analysis processing on each of the target visual angles based on the initial visual angle to obtain a visual angle difference feature distribution; wherein the visual angle difference feature distribution includes a plurality of visual angle difference features, and the visual angle difference feature is used to describe the visual angle difference between the initial visual angle and the target visual angle; S4: performing corresponding visual angle conversion processing on the initial visual image according to the visual angle difference characteristic distribution, so as to obtain basic simulation images corresponding to each of the target visual angles; S5: performing cross-3D feature recognition and visual angle conversion based on the basic simulated images of each target visual angle to obtain an extended simulated image corresponding to each target visual angle; S6: supplementing the basic simulated image at multiple visual angles according to a pre-trained multi-angle supplementation model of the photographed object, so as to obtain supplemented simulated images corresponding to each of the target visual angles; S7: Perform image fusion on the extended simulation image and the supplementary simulation image of the same target visual angle to obtain an angle shot image of the target visual angle, wherein the angle shot images of the target visual angles together constitute a multi-angle shooting set of the shooting object; wherein the multi-angle shooting set is used to perform multi-angle image visual depiction of the shooting object.

[0011] Specifically, in step S1 of the embodiment provided by the present invention, a visible light engine is used to capture images of the subject to obtain an initial visual image, which is the starting point of the entire image processing flow and provides a basic view of the subject at a specific visual angle, ensuring that the captured image provides a clear view of the object and serves as the basis for subsequent visual angle analysis and conversion.

[0012] It can be understood that the technical solution provided by the present invention is based on the initial visual image of the photographed object, and deduces more visual angles to obtain theoretical images of the photographed object at other visual angles. In this deduction process, 3D structure analysis technology and AI model technology will be used.

[0013] Specifically, in step S2 of the embodiment provided by the present invention, the initial visual image is analyzed using a pre-trained image shooting perspective recognition model to identify the visual angle of the object captured in the image, i.e., the "initial visual angle". By automatically identifying the perspective of the object captured, basic data is provided for subsequent visual angle difference analysis.

[0014] More specifically, the initial visual angle is the visual angle used when capturing an image of a subject. The image content displayed by the subject at the initial visual angle is the initial visual image. At other visual angles, the subject will display other forms of image content, and this other form of image content has a high correlation with the content in the initial visual image. This correlation can be obtained by analyzing the visual angle difference between the initial visual angle and the target visual angle.

[0015] Specifically, in step S3 of the embodiment provided by the present invention, several target visual angles for the photographed object are obtained, and visual angle difference analysis processing is performed on each target visual angle based on the initial visual angle to obtain a visual angle difference feature distribution; wherein the visual angle difference feature distribution includes several visual angle difference features, and the visual angle difference feature is used to describe the visual angle difference between the initial visual angle and the target visual angle.

[0016] It can be seen that the same photographed object will show different image contents at different visual angles, and the correlation between these image contents shown at different visual angles is related to the differences between the various visual angles.

[0017] Specifically, the photographed object has its 3D structure, and the 3D structure will show different image contents in different directions in space. By analyzing the image content and 3D structure at a specified visual angle, it can be concluded what kind of image content the 3D structure of the photographed object will show in other directions.

[0018] More specifically, the target visual angle is the visual angle corresponding to the image content that needs to be deduced. The initial visual image can be the front, side or other sides of the subject, and the target visual angle is the target angle deduced based on the initial visual image. The initial visual angle can also be to shoot the subject from the front but with a slightly tilted angle, and set the front of the subject as the target visual angle.

[0019] Specifically, in step S4 of the embodiment provided by the present invention, the initial visual image is subjected to visual angle conversion according to the aforementioned visual angle difference characteristics to generate basic simulation images of multiple target visual angles. The basic simulation images provide a preliminary perspective transformation effect for subsequent further image conversions, so that each target visual angle image has a rough preliminary version.

[0020] It can be understood that the basic simulated image is obtained by analyzing the 3D structure of the photographed object and judging the image content that will be directly displayed by the photographed object at the target visual angle under the effect of its 3D structure based on the analysis result.

[0021] It is not difficult to see that since the basic simulation image is obtained by directly analyzing the initial visual image and the initial visual angle, the basic simulation image has the highest credibility. However, a complete composition cannot be formed by relying solely on the basic simulation image. At this time, the basic simulation image needs to be expanded and supplemented to form a complete image content at the target visual angle.

[0022] Specifically, in step S5 of the embodiment provided by the present invention, cross-3D feature recognition and visual angle conversion are performed based on the basic simulation images of each target visual angle to obtain extended simulation images corresponding to each target visual angle. This step can be understood as taking each basic simulation image as the initial visual image of a new round, repeating the deduction of the basic simulation image performed in the previous round, and summarizing and analyzing the content of the deduced basic simulation image to obtain the extended simulation image of the target visual angle.

[0023] More specifically, part of the image content that cannot be directly obtained through direct conversion of the initial visual image at the initial visual angle can be acquired through secondary deduction of the deduced basic simulation image. This is because the basic simulation image contains information content that is deduced from the initial visual image but not directly included. Information that was previously unavailable can be acquired through secondary deduction of the new information content.

[0024] Specifically, in step S6 of the embodiment provided by the present invention, a pre-trained multi-angle supplementation model of the photographed object is used to perform image supplementation at multiple visual angles on the basic simulated image. The supplementation model will automatically fill in the missing parts or details in the image according to the existing image and target angle information, and generate more image details that meet the target visual angle. The supplemented image includes the supplementation of the shape and structure of the object, and also involves enhancement of details in texture, lighting, shadow, etc.

[0025] It can be understood that the multi-angle supplement model of the photographed object is a deep learning model, and its training method is to collect multi-angle images of the training object in advance, and substitute the collected images into the deep learning model for model training to obtain the regular expression form of the training object under various perspectives. The regular expression form is matched and supplemented with the basic simulation image obtained in the previous step to obtain a supplemented simulation image, wherein the training object is of the same type as the photographed object to be supplemented, which can be an object, scene or portrait, etc.

[0026] Specifically, in step S7 of the embodiment provided by the present invention, the extended simulated image and the supplementary simulated image under the same target visual angle are fused. Image fusion generally refers to merging useful information in two images together, eliminating duplicate information or detail conflicts, and ensuring the integrity of the final image. The fusion method can adopt image fusion technology and optimize the fusion result according to the different characteristics of the images.

[0027] It can be understood that the fused image can comprehensively expand the advantages of the simulated image and the supplementary simulated image, so as to generate a more delicate, seamless and detail-rich target visual angle image, and the angle shooting images of each target visual angle together constitute a multi-angle shooting set of the shooting object, and the multi-angle shooting set is used for multi-angle image visual depiction of the shooting object.

[0028] The present invention provides a method for implementing multi-angle shooting using a 3D and AI visual sensing visible light core, which has the following beneficial effects: The present invention uses a visible light movement to collect an initial image of a photographed object, analyzes the initial image through an image shooting perspective recognition model to obtain an initial visual angle, analyzes the difference between the target visual angle and the initial visual angle based on the initial visual angle, obtains a visual angle difference feature distribution, generates a basic simulation image according to the difference feature distribution, performs cross-3D feature recognition and angle conversion to generate an extended simulation image, uses a multi-angle supplement model to obtain a supplementary simulation image, fuses the extended and supplementary simulation images, and generates a multi-angle shooting image set. The method automatically generates multi-angle, high-quality shooting images through 3D and AI vision technology, effectively improves the comprehensiveness and accuracy of image acquisition, and solves the problem in the prior art that an ideal perspective image cannot be obtained due to scene restrictions.

[0029] Preferably, the pre-training step of the image shooting perspective recognition model includes: S21: constructing a convolutional neural network model and collecting training data for model training of the convolutional neural network model; wherein the training data includes an object visual image of a photographed object of a specified type and a visual angle corresponding to the object visual image; S22: Divide the training data to obtain a training set, a validation set, and a test set; S23: performing model training on the convolutional neural network model according to the training set to obtain mapping association features between the object visual image and the visual angle in the training data, and performing feedback adjustment of model parameters of the convolutional neural network model according to the mapping association features; S24: performing a verification process on the model parameters of the convolutional neural network model after feedback adjustment of the model parameters according to the verification set, so as to evaluate the model performance of the convolutional neural network model according to the result of the verification process, and performing hyperparameter tuning and training progress supervision on the convolutional neural network model according to the evaluation result; S25: Performing a model performance evaluation on the convolutional neural network model after model training with the training set and the validation set according to the test set to obtain a final performance of the convolutional neural network model; when the final performance meets a preset standard, the pre-training of the convolutional neural network model is terminated.

[0030] Specifically, a convolutional neural network (CNN) model is designed and built. The model should be suitable for image processing tasks and be able to effectively learn the mapping relationship between images and visual angles. Usually, the CNN model includes convolutional layers, pooling layers, fully connected layers, etc., which can capture the key features in the image through multi-level feature extraction and learning. The CNN model can automatically learn the spatial structure and texture information in the image, and is particularly suitable for processing complex image data and high-dimensional features, so that the model can extract useful features from visual images for angle recognition.

[0031] More specifically, training data is collected for training convolutional neural networks. The training data should contain visual images of specified types of photographed objects and the visual angle corresponding to each visual image. These data come from real photography, simulation generation or existing image datasets. Each image should have a clear label, that is, the corresponding perspective (for example: front, side, top view, top view, etc.). By collecting images of different perspectives and different types of objects, the diversity and representativeness of the training data can be ensured, thereby improving the model's adaptability to various angles and objects. Training data containing multiple angles and multiple object types helps to train the robustness of the model, enabling it to cope with a variety of actual situations.

[0032] More specifically, the collected training data is divided into training set, validation set and test set. Generally speaking, the training set is used for model training, the validation set is used to adjust model hyperparameters and evaluate performance during training, and the test set is used to finally evaluate the generalization ability of the model. The division ratio is usually: 70%~80% for the training set, 10%~15% for the validation set, and 10%~15% for the test set. The division of the data set ensures the scientificity and reliability of the training process and avoids overfitting of the model on the training data. The introduction of the validation set and the test set can effectively evaluate the performance of the model on unseen data and help monitor and optimize the generalization ability of the model.

[0033] More specifically, the convolutional neural network model is trained using a training set. The goal of the training is to adjust the model parameters (such as weights and biases) through the back-propagation algorithm so that the model can learn the mapping relationship between the image and the corresponding visual angle. During the training process, the convolutional neural network continuously optimizes its internal parameters through multiple iterations to minimize the error on the training data (for example, using the cross entropy loss function, mean square error, etc.). Through a large number of training samples, the model can learn the features of different perspectives in the image and establish a mapping relationship between the visual image and the perspective. The model optimizes parameters through back-propagation and gradient descent algorithms to gradually improve the recognition accuracy of visual angles. The training can gradually extract high-level abstract features, so that the convolutional neural network can effectively recognize the perspective information in the image.

[0034] More specifically, after each round of training, the loss function is calculated using the data from the training set, and the model parameters are adjusted through the back-propagation algorithm. These adjustments reflect the mapping relationship established by the model between images and visual angles. The parameters in the network are continuously adjusted based on feedback to optimize the accuracy of the mapping. Through parameter adjustment, the model can more accurately identify and predict the visual angle of the image. The continuously optimized parameters can increase the convergence speed of the model, reduce overfitting, and thus improve the generalization ability.

[0035] More specifically, during the training process, the validation set is used to regularly evaluate the performance of the model. The evaluation results of the validation set can be used to determine whether the model is overfitting, whether it has converged, and whether it has achieved the desired effect. Based on the evaluation results, the hyperparameters (such as learning rate, batch size, network depth, etc.) are tuned to adjust the progress and strategy of model training. Regular evaluation of the validation set can monitor model performance in real time, avoid overfitting, and improve the stability of the model. Hyperparameter tuning can significantly improve the efficiency and accuracy of model training, ensuring that the model can achieve optimal performance in different training scenarios.

[0036] More specifically, the test set is used to conduct a final performance evaluation of the trained model. The performance of the model on unseen data is evaluated by calculating evaluation indicators such as accuracy, precision, recall, and F1 score. If the model performance meets the preset standard (for example, the accuracy rate reaches more than 90%), the model training is considered complete. The evaluation of the test set can truly reflect the performance of the model in actual applications and avoid overfitting of the model to the validation set or training set. The evaluation results help determine whether the model has reached the usable standard in actual applications.

[0037] More specifically, when the final performance of the model on the test set meets the preset standards, the model training process is ended, and the final training model and its parameters are saved for subsequent use, ensuring that the trained model has high accuracy and stability in practical applications, and can be quickly and efficiently applied to image perspective recognition tasks. The model has been fully trained, verified and tested, has good generalization capabilities, and can adapt to a variety of different shooting objects and perspectives.

[0038] It is understandable that convolutional neural networks can automatically extract important image features from training data, identify visual angles, and thus establish an efficient mapping relationship between images and perspectives. Convolutional neural networks that have been fully trained and tuned can accurately predict the visual angles of images, and are particularly suitable for application scenarios with multi-perspective images. By dividing the training set, validation set, and test set, it ensures that the model can perform well in different data sets and practical applications, and has strong robustness. Through hyperparameter tuning, progress supervision, and feedback adjustment, it can efficiently train high-performance models and improve the application value of the models.

[0039] Preferably, the step of performing visual angle difference analysis processing on each of the target visual angles based on the initial visual angle to obtain a visual angle difference feature distribution includes: S31: performing coordinate positioning of the photographed object according to the initial visual angle to obtain object positioning coordinates of the photographed object and initial angle positioning coordinates of the initial visual angle; S32: positioning and extending the space where the photographed object is located based on the object positioning coordinates and the initial angle positioning coordinates to obtain a spatial positioning system of the photographed object; S33: Substituting each of the target visual angles into the spatial positioning system respectively, so as to perform spatial positioning processing on each of the target visual angles according to the spatial positioning system, and obtaining the target angle positioning coordinates of each of the target visual angles in the spatial positioning system; S34: performing visual angle difference analysis on the initial angle positioning coordinates according to each of the target angle positioning coordinates to obtain visual angle difference features between each of the target visual angles and the initial visual angle, and combining each of the visual angle difference features to obtain a visual angle difference feature distribution.

[0040] Specifically, the coordinates of the photographed object are located according to the initial visual angle. Specifically, given the initial visual angle (such as the angle of the observer in the image), the position of the photographed object in a certain reference coordinate system is determined, and the object positioning coordinates of the object (usually the coordinates in three-dimensional space, and the initial angle positioning coordinates, that is, the representation of the initial visual angle in the spatial coordinate system) are calculated through the visual angle and the spatial position of the photographed object.

[0041] It can be understood that this process can convert the visual information (i.e., perspective) in an image or scene into actionable coordinate information, laying the foundation for subsequent spatial analysis and angle difference processing. Coordinate positioning establishes a connection between the visual angle and the position and direction in the actual space, ensuring that the analysis is based on accurate spatial information.

[0042] More specifically, the object positioning coordinates and initial angle positioning coordinates of the photographed object are used to determine the space where the photographed object is located. This process extends the spatial positioning through mathematical models (such as three-dimensional coordinate transformation, translation and rotation, etc.) to construct a spatial positioning system. The spatial positioning system is a coordinate system that can be used to represent the position and orientation of the photographed object in three-dimensional space. The system usually takes into account the transformation relationship of the object in three-dimensional space, such as the relative position change between the object and the observer.

[0043] It can be understood that the spatial positioning system provides a unified reference framework for the subsequent visual angle difference analysis. By establishing an accurate coordinate system in space, the spatial positioning of different target visual angles can be consistent. This step ensures that the subsequent analysis of different visual angles is based on the same coordinate system, effectively eliminating the impact of spatial position differences on angle difference analysis.

[0044] More specifically, each target visual angle is substituted into the constructed spatial positioning system. For each target visual angle, spatial positioning processing is performed in combination with the spatial position and direction corresponding to the target visual angle. In this step, spatial positioning processing needs to be performed on each target visual angle to obtain the spatial position change of each target angle relative to the initial angle. Finally, the target angle positioning coordinates of each target visual angle in the spatial positioning system are output, that is, the three-dimensional coordinate value corresponding to the target angle.

[0045] It can be understood that through this process, the spatial relationship of different target visual angles can be quantified into three-dimensional coordinates for easy comparison and analysis. This step provides accurate data support for visual angle difference analysis, allowing the spatial information of different target angles to be processed under the same framework.

[0046] More specifically, the target angle positioning coordinates of each target visual angle are compared and analyzed with the initial angle positioning coordinates of the initial visual angle. This analysis step quantifies the changes between different visual angles by calculating the angle difference and the spatial distance difference. Specifically, the following methods can be used for analysis: Angle difference calculation: calculate the Euclidean distance or angle difference between the target angle and the initial angle (for example, using cosine similarity, angle calculation, etc.); Spatial difference calculation: consider the spatial distance difference between the target angle positioning coordinates and the initial angle positioning coordinates, and further analyze the spatial displacement caused by the change in visual angle.

[0047] It is understandable that through difference analysis, the differences between different visual angles can be quantified, providing data support for subsequent feature extraction, classification or clustering tasks. The extraction of visual angle difference features can reveal the complex relationship between the photographed object and the viewing angle, and thus provide strong support for tasks such as angle classification, object recognition, and scene understanding.

[0048] More specifically, the difference features of the visual angles of each target are combined to form an overall visual angle difference feature distribution. This feature distribution can reflect the regularity and spatial distribution pattern of visual angle differences, and provide a basis for subsequent angle recognition, target detection and other tasks. Through the combination of visual angle difference features, a global understanding of visual angle differences can be formed, and the feature change patterns of the objects photographed at different visual angles can be further revealed. This feature distribution provides effective input for subsequent angle recognition and scene understanding, and helps to improve the accuracy and stability of the model.

[0049] Preferably, the step of performing corresponding visual angle conversion processing on the initial visual image according to the visual angle difference characteristic distribution to obtain a basic simulated image corresponding to each of the target visual angles includes: S41: performing image analysis of 3D visual features on the initial visual image to obtain a 3D visual feature map of the initial visual image; wherein the 3D visual feature map includes 3D visual features corresponding to specific positions of the photographed object in the initial visual image, and the 3D visual features are used to describe 3D structural information fed back by the specified specific position of the photographed object in the initial visual image; S42: performing a conversion analysis of an image content expression mode on the 3D visual feature map according to the visual angle difference feature, so as to obtain a set of 3D structure expression vectors of the photographed object corresponding to the visual angle difference feature; wherein the set of 3D structure expression vectors includes a plurality of 3D structure expression vectors, and the 3D structure expression vectors are used to describe the correlation between the image content displayed by the photographed object in the initial visual image and the image content that can be displayed by the photographed object at the target visual angle; S43: performing image expression content conversion on the initial visual image according to the 3D structure expression vector to obtain a basic simulated image of the target visual angle.

[0050] Specifically, the three-dimensional visual features are extracted from the initial visual image, and the three-dimensional structural information contained therein is analyzed. The core goal of this step is to extract the spatial structure and depth information of the photographed object from the initial visual image, and generate a 3D visual feature map through a certain algorithm. The 3D visual feature map includes 3D visual features related to the spatial relationship of the photographed object at different positions in the image. These features are used to describe the structural information of the object in the initial visual image, such as depth, relative position, shape, texture, etc. Each feature corresponds to the photographed object at a specific position in the image. This step can extract key three-dimensional spatial information from the two-dimensional image, so that the subsequent visual angle conversion is not just a simple image transformation, but an accurate simulation based on the three-dimensional structure. The 3D visual feature map provides reliable three-dimensional spatial data support for image generation at the target visual angle, which can ensure that the converted image is consistent with the target visual angle in three-dimensional structure, laying the foundation for subsequent angle conversion processing, and ensuring that reasonable spatial structure and visual content are displayed at different angles.

[0051] More specifically, after obtaining the 3D visual feature map of the initial visual image, the visual angle difference feature is used to guide the expression mode conversion of the image content. Specifically, the visual angle difference feature represents the difference between the initial visual angle and the target visual angle. These differences will be used to adjust the 3D visual features of the initial image, so as to obtain an expression method that meets the requirements of the target visual angle. In this stage, for each target visual angle, the 3D feature map is spatially transformed and re-expressed through the difference feature to generate a set of new 3D structure expression vectors. The set consists of several 3D structure expression vectors, each of which describes the association between different parts of the photographed object and its spatial structure features at the target visual angle. This process can achieve fine adjustment of the three-dimensional visual feature map according to the visual angle difference feature, so that the initial image can show a more realistic spatial expression at the target angle. Through this conversion, the structural changes of the photographed object at the target angle can be captured more accurately, ensuring the spatial authenticity and visual consistency of the target image at the target angle, providing a very detailed and high-dimensional structured representation for the subsequent image content conversion, so that the generation of the target visual angle image is no longer a simple geometric transformation, but is based on complex three-dimensional spatial information adjustment.

[0052] More specifically, the generated 3D structure expression vector set is used to convert the image content of the initial visual image. The core of this conversion process is to map each 3D structure expression vector in the image to the target visual angle, so that the content of the initial image can show a reasonable visual effect at the new angle. Through this stage of conversion, a basic simulation image corresponding to each target visual angle is finally generated. These simulated images are generated based on the spatial characteristics and visual expression requirements at the target angle, and can accurately simulate the visual transition from the initial perspective to the target perspective. This step ensures that the initial visual image can be converted into a reasonable visual expression based on the three-dimensional structure and spatial relationship at the target angle, and generates an accurate target visual angle image. Through precise structural adjustment and image content conversion, the perspective distortion, spatial inconsistency and other problems that may exist in traditional methods can be avoided. The basic simulation image generated by this method is not just a conversion of visual effects, but a deep optimization based on three-dimensional structure and angle differences, which ensures the naturalness and realism of the image.

[0053] Preferably, the step of performing conversion analysis of the image content expression mode on the 3D visual feature map according to the visual angle difference feature to obtain a set of 3D structure expression vectors of the photographed object corresponding to the visual angle difference feature includes: S421: analyzing the 3D structural relationship of the 3D visual features at each specific position in the 3D visual feature map according to the visual angle difference feature, so as to obtain a visual masking relationship between the 3D visual features at each specific position in the 3D visual feature map and the visual angle difference feature; S422: dividing the 3D visual feature map into visual areas according to the visual masking relationship to obtain a displayable visual area and a non-displayable visual area of ​​the 3D visual feature map corresponding to the visual angle difference feature; S423: analyzing the visual angle conversion relationship of the presentable visual area according to the visual angle difference feature to obtain a 3D structure expression vector of the presentable visual area corresponding to the visual angle difference feature; S424: performing combination processing on each of the 3D structure expression vectors to obtain a set of 3D structure expression vectors of the photographed object corresponding to the visual angle difference feature.

[0054] Specifically, based on the visual angle difference characteristics, the spatial relationship between the 3D visual features at each specific position in the 3D visual feature map is analyzed. The focus of this analysis is to determine the three-dimensional structural information such as the relative position, depth, and occlusion between each feature and the target visual angle. The 3D structural relationship analysis involves identifying and establishing the spatial structural relationship between the visual features at different positions and the target angle by calculating the relative changes of each part of the photographed object at different visual angles. At this stage, the spatial change law at the target angle can be extracted from the three-dimensional features of the initial visual image, providing the depth and spatial structure that can be correctly expressed at the target angle, which helps to identify the occlusion phenomenon in the initial image and provide the necessary data support for the subsequent occlusion relationship analysis, ensuring that the three-dimensional structural adjustment of the target visual angle can be more accurate in space and avoid unnatural image changes or errors.

[0055] More specifically, after analyzing the structural relationship of 3D visual features, the occlusion relationship of these features is analyzed based on the visual angle difference characteristics. Specifically, this analysis is used to identify which parts of the visual features will be obscured and which parts will be exposed at the target visual angle. Visual occlusion relationship analysis usually involves judging that under different viewing angles, some visual areas may be obscured by other parts (such as the foreground or background of the object), or some areas may no longer be visible due to different viewing angles. Through in-depth analysis of visual occlusion relationships, visible and invisible visual areas at the target angle can be effectively distinguished. This is crucial for the accurate conversion of 3D visual features, because occlusion information can affect the presentation of the final image content, and helps to optimize the processing of occluded areas at different angles during the image content conversion process, so that the target image is more visually realistic and in line with physical laws, and can avoid erroneous "perspective distortion" or inappropriate object occlusion in the perspective conversion.

[0056] More specifically, according to the visual occlusion relationship in the previous step, the 3D visual feature map is divided into a displayable visual area and a non-displayable visual area. Specifically, the displayable visual area refers to those parts that will not be obstructed and can be observed at the target visual angle, while the non-displayable visual area refers to those parts that are obstructed or invisible at the target angle. This division helps to accurately determine which areas need to update or reconstruct the visual content at the target visual angle. Through precise visual area division, the amount of calculation of irrelevant areas can be effectively reduced, the calculation efficiency can be improved, and only the displayable area can be converted, thereby accelerating the processing process. The division of displayable and non-displayable areas can accurately control which parts need to be dynamically adjusted at the target visual angle, avoid unnecessary calculations and image reconstruction, and improve the quality and expressiveness of the final image.

[0057] More specifically, according to the visual angle difference characteristics, the displayable visual area is analyzed to determine the structural transformation and perspective conversion relationship of the area at the target visual angle. This process includes determining the 3D structural changes, perspective transformation and other possible visual effects of the area. The goal of this analysis is to obtain the accurate expression of these displayable areas at the target visual angle and generate corresponding 3D structural expression vectors so that the visual content at the target angle can be accurately generated in the subsequent image conversion process. Through this analysis, it can be ensured that the displayable area at the target visual angle can accurately express the spatial structure in the initial image, avoiding perspective distortion or deformation of the object shape. The generated 3D structural expression vector can effectively combine the three-dimensional features at the target angle with the content in the initial image, providing a higher-precision image conversion effect, making the conversion of images at different angles more realistic, and enhancing the image's sense of depth and space.

[0058] More specifically, the previously generated 3D structure expression vectors are combined to form a complete 3D structure expression vector set. This vector set includes the spatial information and visual content expression of each part of the photographed object at the target visual angle. The combination of these structure vectors forms a comprehensive expression of the object at the target visual angle, which can serve as the basis for subsequent image conversion and reconstruction. Through this step, the final generated 3D structure expression vector set can fully and accurately express the three-dimensional structure and visual content of the photographed object at the target angle. These structure vectors provide an efficient spatial data representation, which can be used for subsequent image content conversion to ensure the accuracy and authenticity of the generated image at the target visual angle. The generation of the vector set makes the conversion of the target visual angle not only rely on local features, but also comprehensively considers the global three-dimensional spatial relationship, thereby obtaining a more refined and comprehensive visual expression.

[0059] Preferably, the step of performing cross-3D feature recognition and visual angle conversion based on the basic simulated images of each target visual angle to obtain an extended simulated image corresponding to each target visual angle includes: S51: marking the basic simulation image corresponding to the designated target visual angle as an extended object, and marking the remaining basic simulation images corresponding to the target visual angles as analysis objects respectively; S52: performing visual angle difference analysis on the extended object and each of the analyzed objects respectively, so as to obtain visual angle difference features between the extended object and each of the analyzed objects; S53: performing image analysis of 3D visual features on each of the analysis objects respectively to obtain a 3D visual feature map of each of the analysis objects; S54: performing image expression mode conversion analysis on the 3D visual feature maps of each of the analysis objects according to the visual angle difference features between each of the analysis objects and the extended objects, so as to obtain a set of 3D structure expression vectors corresponding to the visual angle difference features of each of the analysis objects; S55: performing image expression content conversion on each of the analysis objects according to the 3D structure expression vector set of each of the analysis objects, so as to obtain a conversion simulation image of the extended object corresponding to each of the analysis objects; S56: Performing comprehensive verification processing on the converted simulation image of the extended object corresponding to the analysis object to determine the extended simulation image corresponding to the extended object.

[0060] Specifically, a basic simulation image corresponding to a target visual angle is selected and marked as an extended object. This image is used as a reference for conversion at other visual angles. The remaining basic simulation images corresponding to each target visual angle are marked as analysis objects. Through this division, the relationship between different images is determined, which provides a clear structure for subsequent difference analysis and conversion. By clearly dividing the extended objects and analysis objects, the image conversion process can be precisely controlled to ensure the accurate conversion and verification of each object at the target angle. This marking method is helpful to compare the feature differences between different angles in the subsequent steps, so as to better map and convert the 3D structure.

[0061] More specifically, a visual angle difference analysis is performed on the expanded object and each analyzed object. The goal of this analysis is to identify the visual angle difference characteristics between each analyzed object and the expanded object. Specifically, it is necessary to study the differences in visual angles between the two images, such as perspective changes, light and shadow changes, object deformation, etc. This process usually derives the specific characteristics of visual angle changes by comparing the perspective differences, depth information and relative positions of each point in the image. The visual angle difference analysis can accurately reveal the changing patterns of images at different visual angles, and provides the necessary feature information for subsequent 3D structure transformations. Through this analysis, the system can identify the differences in each visual angle, so that the conversion from one angle to another can accurately capture these changes, avoiding image generation results that do not conform to physical laws.

[0062] More specifically, an image analysis of the 3D visual features is performed on each analyzed object to generate a corresponding 3D visual feature map. This map is based on the three-dimensional structural features of each image, and mainly involves the spatial position of each point, depth information, object shape, etc. The analysis of the 3D feature map can capture the spatial relationship in the image and provide information about the depth, contour, surface features, etc. of the object. The generated 3D visual feature map can comprehensively reflect the spatial structure of the target image, and help to accurately restore the performance of the object at different visual angles. By constructing a 3D visual feature map for each analyzed object, detailed spatial data support is provided for subsequent image content conversion and expression mode conversion.

[0063] More specifically, according to the visual angle difference characteristics between each analysis object and the expanded object, a conversion analysis of the image expression mode is performed on the 3D visual feature map of each analysis object. The main purpose of this conversion analysis is to adjust the spatial structure and expression form of the image by identifying the angle difference, so that each analysis object can be correctly expressed at the target visual angle. The conversion analysis mainly involves spatial transformation, perspective transformation, etc. of each point in each 3D feature map to meet the needs of the target visual angle. Through this conversion analysis, accurate conversion from one visual angle to another can be achieved, so that the depth, perspective and object shape in the image at the target angle are more natural and in line with physical laws. A systematic method is provided to ensure that the image content at different angles can retain its three-dimensional structural characteristics after conversion without deformation or unreality.

[0064] More specifically, based on the set of 3D structural expression vectors of each analysis object, the image expression content is converted for each analysis object respectively. This process involves applying the 3D structural expression vector under the target visual angle to the actual image content to generate a conversion simulation image corresponding to each analysis object. The conversion simulation image of each analysis object reflects the image content under the target visual angle through the conversion of the expression mode and the combination of 3D structural features. Through the conversion of the image expression content, the three-dimensional structure of each analysis object under the target visual angle can be accurately mapped to the image, thereby generating an image that meets the visual angle requirements. This process makes the target image conversion not just a simple processing of the image surface content, but a reconstruction of the entire 3D spatial structure, ensuring that the image conversion result is more realistic and in line with the actual visual experience.

[0065] More specifically, a comprehensive verification process is performed on the converted simulated images corresponding to the extended object and each analyzed object. The goal of this verification process is to detect whether the conversion effect between the extended object and each analyzed object is natural and accurate at the target visual angle, and whether the spatial relationship and visual coherence of the objects are retained. The verification process may include comparing visual consistency, depth accuracy, morphological consistency, etc. to ensure the authenticity of the target image. Through comprehensive verification, the quality control of the finally generated extended simulated image can be performed to ensure that it meets the actual visual needs. This verification step ensures the high precision and visual quality of the final result, avoids errors or distortion problems that may occur during the conversion process, and ensures that the performance of the extended simulated image at the target visual angle meets expectations.

[0066] More specifically, according to the result of the verification process, the extended simulated image corresponding to the extended object is determined. This image is the final result after conversion and verification, and can accurately reflect the content at the target visual angle. After the image generation is completed, it can be further displayed or used for practical applications. Through this series of processing, the final extended simulated image has high accuracy and authenticity, and can show the image effect at the target visual angle.

[0067] Preferably, the pre-training of the multi-angle supplementary model of the photographed object includes: S61: constructing a convolutional neural network model, and collecting training data for model training of the convolutional neural network model; wherein the training data includes an angular visual image of a specified visual angle of the photographed object; S62: Divide the training data to obtain a training set, a validation set, and a test set; S63: performing model training on the convolutional neural network model according to the training set to obtain mapping association features between image regions at various locations in the angle visual image of the photographed object in the training data, and performing feedback adjustment of model parameters of the convolutional neural network model according to the mapping association features; S64: performing a verification process on the model parameters of the convolutional neural network model after feedback adjustment of the model parameters according to the verification set, so as to evaluate the model performance of the convolutional neural network model according to the result of the verification process, and performing hyperparameter tuning and training progress supervision on the convolutional neural network model according to the evaluation result; S65: Performing a model performance evaluation on the convolutional neural network model after model training with the training set and the validation set according to the test set to obtain a final performance of the convolutional neural network model; when the final performance meets a preset standard, ending the pre-training of the convolutional neural network model.

[0068] Specifically, a convolutional neural network (CNN) model is constructed, which is designed to process multi-angle visual images, identify features of different areas in the image, and perform feature mapping according to angle differences. Training data is collected, especially angular visual images of different specified visual angles of the photographed object. These image data should cover the visual performance of the photographed object from multiple different angles so that the training model can identify the spatial relationship and visual differences between multiple angles. By constructing a convolutional neural network model, the powerful image feature extraction capability of CNN can be used to identify local and global features of the image, especially spatial relationships, depth information and changes in object morphology. The training data of multi-angle visual images can help the network learn the transformation rules from one angle to another and form a more comprehensive visual feature representation.

[0069] More specifically, the collected training data is divided into training set, validation set and test set. The division ratio of each data set can be adjusted according to actual needs, but usually the training set accounts for the majority, the validation set is used for tuning, and the test set is used for the final model evaluation. By reasonably dividing the data set, the independence of training, validation and testing can be ensured to avoid overfitting or underfitting of the model. The training set is used for model learning, the validation set is used to evaluate and adjust the hyperparameters of the model, and the test set is used as the final verification of the model's generalization ability to ensure that the model has good performance.

[0070] More specifically, the convolutional neural network model is trained using a training set. The training process is mainly to allow the model to learn the mapping association features between the angle visual images of the object by continuously optimizing parameters. These mapping features reflect the visual transformation rules from one angle to another, such as changes in the shape of the object, changes in illumination, changes in viewing angle, etc. During the training process, the network will adjust the convolution kernel, weights and other parameters according to the training error through a feedback mechanism to better fit the data features. Through training, the convolutional neural network can automatically extract high-order features from the image, and gradually learn how to associate visual images from different angles, so that the network can capture the full picture of the object from a set of visual angle images. Feedback adjustment enables the network to efficiently optimize parameters, gradually reduce errors, and ensure that the model can extract meaningful features from complex image data.

[0071] More specifically, after the model is trained with the training set, the validation set is used to validate the feedback-adjusted convolutional neural network model. The validation set data will not participate in the training process, but can be used to evaluate the performance of the model on unseen data and help determine whether the model can be generalized. According to the results of the validation process, the model's hyperparameters are tuned (such as learning rate, number of convolutional layers, filter size, etc.), and the model's training progress is monitored. Hyperparameter tuning can be performed through methods such as grid search or Bayesian optimization. The use of the validation set ensures the generalization ability of the model, not just limited to the features in the training set. Through the validation process, the performance of the current model on different data sets can be evaluated, and the model can be optimized in a targeted manner. Hyperparameter tuning optimizes the overall performance of the model, making the network more stable during the training process, avoiding the problem of training too fast or too slow, thereby improving training efficiency and final model performance.

[0072] More specifically, after the model is trained and verified and tuned, the test set is used to evaluate the performance of the final convolutional neural network model. The data in the test set is used to evaluate the performance of the model on unknown data and measure its accuracy, robustness and generalization ability. The evaluation criteria include accuracy, recall rate, F1 score, etc., depending on the specific task requirements. If the performance of the model on the test set meets the preset standard, it means that the model has good generalizability and can be used for subsequent applications. The performance evaluation of the test set can effectively test the final effect of the model and ensure that it can work normally in the real environment. The evaluation results of the test set help developers clarify the actual performance of the model, whether it has achieved the predetermined accuracy target, and whether there are areas that need further improvement.

[0073] More specifically, when the model reaches the preset performance standard on the test set, the pre-training of the model is ended. At this point, the convolutional neural network model is ready for practical applications and can be further debugged, deployed or integrated into the multi-angle supplementation model. After comprehensive training and verification, the model can accurately identify the visual features of the photographed object at different angles, ensuring the accuracy and robustness of the multi-angle supplementation model. This process ensures the efficiency and stability of the convolutional neural network model, and can provide strong support for subsequent image processing tasks.

[0074] Preferably, the step of fusing the extended simulated image and the supplemented simulated image of the same target visual angle to obtain the angle-shot image of the target visual angle comprises: S71: performing source tracing processing on the image regions of the extended simulated image and the supplemented simulated image respectively, and performing confidence assessment on the image regions of the extended simulated image and the supplemented simulated image respectively according to the source tracing results, so as to obtain image confidence distributions of the extended simulated image and the supplemented simulated image; S72: performing a weighted comprehensive analysis of the image regions of the extended simulation image and the supplemented simulation image according to the image confidence distribution of the extended simulation image and the supplemented simulation image to obtain an angle-captured image of the target visual angle.

[0075] Specifically, the extended simulated image and the supplemented simulated image are respectively subjected to the traceability processing of the image regions. The purpose of the traceability processing is to track and calibrate the source information of each region in the image, to ensure that the spatial relationship of each part of the image can be accurately processed during the fusion process. The traceability processing is used to determine the source of the image information, whether it comes from the image extension processing or the image supplementation processing, to accurately understand the geometric structure of the image and its spatial layout, and to ensure that the spatial relationship between the image region and the object is clear, which is crucial for subsequent fusion, weighted analysis and regional evaluation.

[0076] More specifically, based on the tracing results, a confidence assessment is performed on each area in the extended simulated image and the supplementary simulated image. This step mainly analyzes the characteristics of the image area to evaluate the reliability or confidence of the area in the overall image. The confidence assessment may be performed in a variety of ways, such as clarity assessment based on regional contrast, texture features, and edge information, or depth and spatial consistency assessment based on local areas. The confidence assessment can provide a reliability weight for each image area for the subsequent image fusion process, ensuring that more attention is paid to areas with high confidence and strong stability during fusion. In this way, low-quality or uncertain areas can be effectively avoided from being added to the final image, thereby improving the quality and accuracy of the final image.

[0077] More specifically, based on the confidence evaluation results, the image confidence distribution of the extended simulated image and the supplemented simulated image is calculated. This distribution reflects the degree of trust or confidence of each image region, usually expressed in the form of probability or weight. The distribution is a pixel-by-pixel weight map that marks the confidence of each region in the image. These weights reflect the relative importance of the image region in generating the final target visual angle image. The calculation of the image confidence distribution provides a basis for weighted fusion, ensuring that in the image synthesis process, regions with higher confidence contribute more to the final image. This distribution can automatically adapt to different image qualities and regional characteristics, avoid manual intervention, and improve the automation and intelligence level of image fusion.

[0078] More specifically, a weighted comprehensive analysis is performed on the image areas based on the image confidence distribution of the extended simulated image and the supplementary simulated image. In this process, the confidence of each image area is used as a weight to perform weighted fusion on the areas of the two images. The weighted comprehensive analysis can adopt different weighted fusion strategies, such as the weighted average method, the smooth fusion method based on Gaussian blur, or the image fusion technology in the deep learning network. In the weighted process, the areas with high confidence will occupy a larger proportion in the final synthesized image, thereby ensuring that the image fusion result is more accurate.

[0079] More specifically, through weighted comprehensive analysis, images from different sources can be reasonably fused according to their credibility, thereby improving the detail richness and realism of the target visual angle image. This weighted fusion method can retain useful detail information in the image while reducing the interference of uncertain or blurred areas on the final result, thereby enhancing the quality of the synthesized image.

[0080] More specifically, based on the weighted fusion results of the above steps, an angled image of the target visual angle is generated. This image should contain complete information of the target object at the target visual angle, with high detail expression and realism. The generated angled image not only maintains the structure, texture and other details of the target object, but also ensures the seamless fusion of information from different perspectives, reducing the transition boundaries and stitching traces between different image sources.

[0081] 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 principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for implementing multi-angle shooting using a 3D and AI visual sensing visible light movement, characterized in that: include: Capturing an image of the photographed object through a visible light engine to obtain an initial visual image of the photographed object; Performing visual angle analysis processing on the initial visual image according to a pre-trained image shooting angle recognition model to obtain an initial visual angle of the initial visual image; Acquire a plurality of target visual angles for the photographed object, and perform visual angle difference analysis processing on each of the target visual angles based on the initial visual angle to obtain a visual angle difference feature distribution; wherein the visual angle difference feature distribution includes a plurality of visual angle difference features, and the visual angle difference feature is used to describe the visual angle difference between the initial visual angle and the target visual angle; Performing corresponding visual angle conversion processing on the initial visual image according to the visual angle difference characteristic distribution, so as to obtain basic simulation images corresponding to each of the target visual angles; Based on the basic simulated images of each target visual angle, cross-cutting 3D feature recognition and visual angle conversion are performed to obtain extended simulated images corresponding to each target visual angle; Performing image supplementation at multiple visual angles on the basic simulated image according to a pre-trained multi-angle supplementation model of the photographed object, so as to obtain supplemented simulated images corresponding to each of the target visual angles; The extended simulation image and the supplementary simulation image of the same target visual angle are fused to obtain an angle shot image of the target visual angle, and the angle shot images of the target visual angles together constitute a multi-angle shooting set of the shooting object; wherein the multi-angle shooting set is used to perform multi-angle image visual depiction of the shooting object.

2. The method for implementing multi-angle shooting using 3D and AI visual sensing visible light core as claimed in claim 1, characterized in that: The pre-training step of the image shooting perspective recognition model includes: Constructing a convolutional neural network model and collecting training data for model training of the convolutional neural network model; wherein the training data includes an object visual image of a photographed object of a specified type and a visual angle corresponding to the object visual image; Dividing the training data into a training set, a validation set, and a test set; Performing model training on the convolutional neural network model according to the training set to obtain mapping association features between the visual image of the object and the visual angle in the training data, and performing feedback adjustment of model parameters of the convolutional neural network model according to the mapping association features; Performing model parameter verification processing on the convolutional neural network model that has been adjusted by feedback of the model parameters according to the verification set, evaluating the model performance of the convolutional neural network model according to the result of the verification processing, and performing hyperparameter tuning and training progress supervision on the convolutional neural network model according to the evaluation result; The model performance of the convolutional neural network model after model training with the training set and the validation set is evaluated according to the test set to obtain the final performance of the convolutional neural network model. When the final performance meets the preset standard, the pre-training of the convolutional neural network model is terminated.

3. The method for implementing multi-angle shooting using 3D and AI visual sensing visible light core as claimed in claim 1, characterized in that: The step of performing visual angle difference analysis processing on each of the target visual angles based on the initial visual angle to obtain a visual angle difference feature distribution comprises: Positioning the photographed object according to the initial visual angle to obtain object positioning coordinates of the photographed object and initial angle positioning coordinates of the initial visual angle; Positioning and extending the space where the photographed object is located based on the object positioning coordinates and the initial angle positioning coordinates to obtain a spatial positioning system of the photographed object; Substituting each of the target visual angles into the spatial positioning system respectively, so as to perform spatial positioning processing on each of the target visual angles according to the spatial positioning system, and obtaining the target angle positioning coordinates of each of the target visual angles in the spatial positioning system; Visual angle difference analysis is performed on the initial angle positioning coordinates according to each of the target angle positioning coordinates to obtain visual angle difference features between each of the target visual angles and the initial visual angle, and each of the visual angle difference features is combined to obtain a visual angle difference feature distribution.

4. The method for implementing multi-angle shooting using 3D and AI visual sensing visible light core as claimed in claim 1, characterized in that: The steps of performing corresponding visual angle conversion processing on the initial visual image according to the visual angle difference characteristic distribution to obtain basic simulated images corresponding to each of the target visual angles include: Performing image analysis of 3D visual features on the initial visual image to obtain a 3D visual feature map of the initial visual image; wherein the 3D visual feature map includes 3D visual features corresponding to specific locations of the photographed object in the initial visual image, and the 3D visual features are used to describe 3D structural information fed back by the photographed object at a specified specific location in the initial visual image; Performing a conversion analysis of the image content expression mode on the 3D visual feature map according to the visual angle difference feature to obtain a set of 3D structure expression vectors of the photographed object corresponding to the visual angle difference feature; wherein the set of 3D structure expression vectors includes a plurality of 3D structure expression vectors, and the 3D structure expression vectors are used to describe the correlation between the image content displayed by the photographed object in the initial visual image and the image content that can be displayed by the photographed object at the target visual angle; The image expression content of the initial visual image is converted according to the 3D structure expression vector to obtain a basic simulated image at the target visual angle.

5. The method for implementing multi-angle shooting using 3D and AI visual sensing visible light core as claimed in claim 4, characterized in that: The step of performing conversion analysis of the image content expression mode on the 3D visual feature map according to the visual angle difference feature to obtain a set of 3D structure expression vectors of the photographed object corresponding to the visual angle difference feature comprises: Analyzing the 3D structural relationship of the 3D visual features at each specific position in the 3D visual feature map according to the visual angle difference feature, so as to obtain a visual masking relationship between the 3D visual features at each specific position in the 3D visual feature map and the visual angle difference feature; Dividing the 3D visual feature map into visual areas according to the visual masking relationship to obtain a displayable visual area and a non-displayable visual area of ​​the 3D visual feature map corresponding to the visual angle difference feature; Analyzing the visual angle conversion relationship of the presentable visual area according to the visual angle difference feature to obtain a 3D structure expression vector of the presentable visual area corresponding to the visual angle difference feature; The 3D structure expression vectors are combined to obtain a set of 3D structure expression vectors corresponding to the visual angle difference features of the photographed object.

6. The method for implementing multi-angle shooting using 3D and AI visual sensing visible light core as claimed in claim 1, characterized in that: The steps of performing cross-3D feature recognition and visual angle conversion based on the basic simulated images of each target visual angle to obtain the extended simulated images corresponding to each target visual angle include: Marking the basic simulation image corresponding to the specified target visual angle as an extended object, and marking the remaining basic simulation images corresponding to each of the target visual angles as analysis objects; Performing visual angle difference analysis on the extended object and each of the analyzed objects respectively to obtain visual angle difference features between the extended object and each of the analyzed objects; Performing image analysis of 3D visual features on each of the analysis objects to obtain a 3D visual feature map of each of the analysis objects; According to the visual angle difference characteristics between each of the analysis objects and the extended objects, a conversion analysis of the image expression mode is performed on the 3D visual feature map of each of the analysis objects to obtain a set of 3D structure expression vectors corresponding to the visual angle difference characteristics of each of the analysis objects; According to the 3D structure expression vector set of each of the analysis objects, image expression content conversion is performed on each of the analysis objects to obtain a conversion simulation image of the extended object corresponding to each of the analysis objects; A comprehensive verification process is performed on the converted simulation image of the extended object corresponding to the analysis object to determine the extended simulation image corresponding to the extended object.

7. The method for implementing multi-angle shooting using 3D and AI visual sensing visible light core as claimed in claim 1, characterized in that: The pre-training of the multi-angle supplement model of the photographed object includes: Constructing a convolutional neural network model and collecting training data for model training of the convolutional neural network model; wherein the training data includes an angular visual image of a specified visual angle of a photographed object; Dividing the training data into a training set, a validation set, and a test set; Performing model training on the convolutional neural network model according to the training set to obtain mapping association features between image regions at various locations in the angle visual image of the photographed object in the training data, and performing feedback adjustment of model parameters of the convolutional neural network model according to the mapping association features; Performing model parameter verification processing on the convolutional neural network model that has been adjusted by feedback of the model parameters according to the verification set, evaluating the model performance of the convolutional neural network model according to the result of the verification processing, and performing hyperparameter tuning and training progress supervision on the convolutional neural network model according to the evaluation result; The model performance of the convolutional neural network model after model training with the training set and the validation set is evaluated according to the test set to obtain the final performance of the convolutional neural network model. When the final performance meets the preset standard, the pre-training of the convolutional neural network model is terminated.

8. The method for implementing multi-angle shooting using 3D and AI visual sensing visible light core as claimed in claim 1, characterized in that: The step of fusing the extended simulated image and the supplemented simulated image of the same target visual angle to obtain the angle-shot image of the target visual angle comprises: Performing source tracing processing on the image regions of the extended simulated image and the supplemented simulated image respectively, and performing confidence assessment on the image regions of the extended simulated image and the supplemented simulated image respectively according to the source tracing results, so as to obtain image confidence distributions of the extended simulated image and the supplemented simulated image; According to the image confidence distribution of the extended simulation image and the supplemented simulation image, a weighted comprehensive analysis of the image area is performed on the extended simulation image and the supplemented simulation image to obtain an angle-shot image of the target visual angle.

Images