Image data acquisition and processing method based on computer vision

Abstract

The invention provides an image data acquisition processing method and system based on computer vision, and relates to the technical field of computer vision, and the method comprises the steps: obtaining original acquisition images of a target recognition object at different angles; performing image de-noising on each original acquisition image to obtain a de-noised image; performing image enhancement on the de-noised image to obtain an enhanced image; performing feature extraction on the enhanced image to obtain image features corresponding to different angles; performing feature fusion on the image features to obtain fused features; and inputting the fusion feature into a trained multi-view target recognition model to obtain a processing result. According to the invention, the target identification capability and the image data acquisition and processing efficiency can be improved.

Description

Technical Field

[0001] This invention relates to the field of computer vision technology, and in particular to an image data acquisition and processing method based on computer vision. Background Technology

[0002] With the rapid development of information technology, image data is playing an increasingly important role in various fields (such as healthcare, security, transportation, and industry). Image data is not only used for recording and analysis, but also plays a key role in automated decision-making, intelligent monitoring, and human-computer interaction.

[0003] Traditional image acquisition methods typically rely on manual operation, using a single or limited number of cameras to capture images. This method is not only inefficient but also susceptible to human error, leading to unstable data quality.

[0004] With the advancement of computer vision technology, especially the rapid development of deep learning and image processing algorithms, automated processing of image data has become possible. Computer vision technology can extract useful information from images, enabling functions such as object detection, image classification, and image segmentation.

[0005] A patent application with application number 202410714760.2 discloses a data acquisition and reading method based on computer vision images. The method includes the following steps: acquiring target recognition task information and controlling a camera to perform image acquisition at a primary shooting angle to obtain image information; recognizing the image information using a preliminary image recognition model to obtain preliminary image recognition information; comparing the preliminary image recognition information with the target description information in the target recognition task information using a similarity formula to obtain a similarity coefficient; determining whether the similarity coefficient is greater than or equal to a preset similarity threshold; and when the similarity coefficient is determined to be greater than or equal to the preset similarity threshold, then recognizing the image information using a precise image recognition model to obtain precise image recognition information. This patent automatically acquires data through computer vision image recognition, improving the efficiency and accuracy of data acquisition. However, this method relies on a primary shooting angle for image acquisition and performs poorly under different lighting conditions, background complexity, and viewing angle changes. This can lead to a decrease in the accuracy of image recognition. Summary of the Invention

[0006] In order to overcome the shortcomings of the prior art, the purpose of this invention is to provide an image data acquisition and processing method and system based on computer vision, which can improve the target recognition capability and the efficiency of image data acquisition and processing.

[0007] To achieve the above objectives, the present invention provides the following solution:

[0008] A computer vision-based image data acquisition and processing method includes:

[0009] Acquire raw images of the target object from different angles;

[0010] Denoise the original acquired images to obtain denoised images;

[0011] The denoised image is then enhanced to obtain an enhanced image;

[0012] Feature extraction is performed on the enhanced image to obtain image features corresponding to different angles;

[0013] The image features are fused to obtain fused features;

[0014] The fused features are input into the trained multi-view target recognition model to obtain the processing result.

[0015] Preferably, acquiring original images of the target object at different angles and heights includes:

[0016] A three-dimensional visualization model of the target object is obtained, and the surface information of the three-dimensional visualization model is obtained; the surface information includes vertices, edges, and faces;

[0017] Determine the optimal shooting point based on the surface information;

[0018] Using an image acquisition device, the original image of the target object is acquired at the optimal shooting point.

[0019] Preferably, determining the optimal shooting point based on the surface information includes:

[0020] For each face of the 3D visualization model, calculate the surface normal N of the face. k The surface normal N k The calculation formula for N is: k = (P2-P1)×(P3-P1); P1, P2, and P3 are the three vertices on the k-th face;

[0021] The surface normal N k The normal vector is standardized; the standardization formula is:

[0022] A set of candidate shooting positions V is generated based on the center and size of the 3D visualization model; the candidate shooting positions V are placed on a sphere outside the 3D visualization model or inside a sphere with the center of the 3D visualization model as the origin.

[0023] For each candidate shooting position V, calculate the view vector L from the viewpoint to the center of the model; where L = CV; C is the center of the 3D visualization model;

[0024] For each candidate shooting position V, calculate the angle between it and the normal of the 3D visualization model to evaluate the surface quality of the viewpoint coverage; the scoring formula for evaluating the surface quality of the viewpoint coverage is: s = ∑ k cos(α k ); s is the score of the shooting position, k is the number of all surfaces of the model traversed, and · represents the dot product.

[0025] Based on the rating of the shooting location, the locations with the highest scores are selected as the best shooting points.

[0026] Preferably, the image denoising is performed based on a Gaussian filtering method.

[0027] Preferably, the denoised image is enhanced to obtain an enhanced image, including:

[0028] A moving image enhancement window is constructed based on the scale of the denoised image;

[0029] According to formula F ij =|x ij -y ij | Obtain the pixel difference value for each pixel in each image enhancement window; where, F ij x represents the pixel difference at position (i, j) in the denoised image. ij y represents the pixel value at position (i, j) in the original acquired image. ij This represents the pixel value at position (i, j) in the denoised image;

[0030] According to the formula The pixel difference is corrected for brightness uniformity to obtain the correction difference; where D ij D is the correction difference for each pixel. dark Let K(i) be the dark signal value of the pixel, and K(i) be the correction coefficient.

[0031] Pixels whose correction difference is greater than the feature threshold are used as feature points and their number is recorded.

[0032] The proportion of feature points to all pixels under the image enhancement window is used as the percentage value;

[0033] When the proportion value within the moving image enhancement window is greater than a preset threshold, median denoising is performed on the denoised image within the corresponding moving image enhancement window to obtain median denoised data.

[0034] Slide the moving image enhancement window until the entire median denoised data has been traversed to obtain the enhanced image.

[0035] Preferably, feature extraction is performed on the enhanced image to obtain image features corresponding to different angles, including:

[0036] The enhanced image is then subjected to image scaling and image normalization to obtain a normalized image;

[0037] The normalized image is used to extract features using a pre-defined VGG16 network to obtain various image features.

[0038] Preferably, the fusion feature is obtained through a weighted average fusion method; the formula for the weighted average fusion is:

[0039]

[0040] Among them, f i For the image features at the i-th angle, w i f is the weight of the i-th image feature, where n is the number of image features. fused For the fusion feature, C(θ i ) represents the coverage of the i-th angle, and j is the index of all angles in the feature fusion, with the value of j ranging from 1 to n.

[0041] Preferably, the formula for calculating the coverage is:

[0042]

[0043] Among them, A i Let A be the visible surface area of ​​the 3D visualization model corresponding to the target object as observed from the i-th viewpoint. total The total surface area of ​​the three-dimensional visualization model corresponding to the target object is denoted as .

[0044] A computer vision-based image data acquisition and processing system includes:

[0045] The image acquisition module is used to acquire raw images of the target object from different angles;

[0046] An image denoising module is used to denoise each of the original acquired images to obtain a denoised image;

[0047] The image enhancement module is used to enhance the denoised image to obtain an enhanced image;

[0048] The feature extraction module is used to extract features from the enhanced image to obtain image features corresponding to different angles;

[0049] The feature fusion module is used to perform feature fusion on the image features to obtain fused features;

[0050] The recognition module is used to input the fused features into the trained multi-view target recognition model to obtain the processing result.

[0051] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0052] This invention provides a computer vision-based image data acquisition and processing method and system. The method includes: acquiring original images of a target object from different angles; denoising each of the original images to obtain a denoised image; enhancing the denoised image to obtain an enhanced image; extracting features from the enhanced image to obtain image features corresponding to different angles; fusing the image features to obtain fused features; and inputting the fused features into a trained multi-view target recognition model to obtain a processing result. This invention can improve target recognition capabilities and the efficiency of image data acquisition and processing. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 A flowchart of the method provided in an embodiment of the present invention;

[0055] Figure 2 This is a schematic diagram of the system structure provided in an embodiment of the present invention. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] The purpose of this invention is to provide a computer vision-based image data acquisition and processing method and system, which has significant beneficial effects in improving target recognition capabilities, robustness, and applicability, and provides a good foundation for the development of related technologies.

[0058] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0059] Figure 1 The method flowchart provided in the embodiments of the present invention is as follows: Figure 1 As shown, the present invention provides an image data acquisition and processing method based on computer vision, comprising:

[0060] Step 100: Acquire raw images of the target object from different angles;

[0061] Step 200: Denoise each of the original acquired images to obtain a denoised image;

[0062] Step 300: Perform image enhancement on the denoised image to obtain an enhanced image;

[0063] Step 400: Extract features from the enhanced image to obtain image features corresponding to different angles;

[0064] Step 500: Perform feature fusion on the image features to obtain fused features;

[0065] Step 600: Input the fused features into the trained multi-view target recognition model to obtain the processing result.

[0066] Specifically, this implementation can design the shooting angle and height based on the geometric features and size of the target object. For example, for three-dimensional objects, multiple angles (such as front, side, and top) and different heights (such as bottom, middle, and top) can be selected for shooting to ensure that complete information of the target is captured. Furthermore, this embodiment also acquires a three-dimensional model of the target object and automatically calculates the optimal shooting angle and height using a model generation algorithm. The optimal shooting position is determined using the surface normals and viewpoint position of the three-dimensional model.

[0067] Preferably, acquiring original images of the target object at different angles and heights includes:

[0068] A three-dimensional visualization model of the target object is obtained, and the surface information of the three-dimensional visualization model is obtained; the surface information includes vertices, edges, and faces;

[0069] Determine the optimal shooting point based on the surface information;

[0070] Using an image acquisition device, the original image of the target object is acquired at the optimal shooting point.

[0071] Preferably, determining the optimal shooting point based on the surface information includes:

[0072] For each face of the 3D visualization model, calculate the surface normal N of the face. k The surface normal N k The calculation formula for N is: k = (P2-P1)×(P3-P1); P1, P2, and P3 are the three vertices on the k-th face;

[0073] The surface normal N k The normal vector is standardized; the standardization formula is:

[0074] A set of candidate shooting positions V is generated based on the center and size of the 3D visualization model; the candidate shooting positions V are placed on a sphere outside the 3D visualization model or inside a sphere with the center of the 3D visualization model as the origin.

[0075] For each candidate shooting position V, calculate the view vector L from the viewpoint to the center of the model; where L = CV; C is the center of the 3D visualization model;

[0076] For each candidate shooting position V, calculate the angle between it and the normal of the 3D visualization model to evaluate the surface quality of the viewpoint coverage; the scoring formula for evaluating the surface quality of the viewpoint coverage is: s = ∑ k cos(α k ); s is the score of the shooting position, k is the number of all surfaces of the model traversed, and · represents the dot product.

[0077] Based on the rating of the shooting location, the locations with the highest scores are selected as the best shooting points.

[0078] Specifically, this embodiment can automatically calculate the optimal shooting angle and position for capturing the details of the object based on the 3D model. This embodiment not only improves the quality of image acquisition but also adapts to different target shapes and complexities, making it a flexible and efficient data acquisition strategy.

[0079] Furthermore, in this embodiment, the position with the highest score is selected as the optimal shooting point based on the calculated score. This position ensures that the shooting angle is nearly perpendicular to the normal direction, allowing surface details to be fully captured.

[0080] Preferably, the image denoising is performed based on a Gaussian filtering method.

[0081] Specifically, this embodiment can also select other filtering methods, such as median filtering, a nonlinear filtering technique particularly suitable for removing salt-and-pepper noise. This method effectively preserves edge information and reduces noise impact by replacing the value of each pixel with the median value of its neighboring pixels. Median filtering performs well when processing images containing sharp noise. Bilateral filtering can also be selected, a denoising method that considers both spatial distance and pixel value similarity. It uses a weighted average of neighboring pixels, with the weights determined by both spatial distance and pixel value differences. This method can preserve edge details while denoising, making it suitable for applications that require maintaining image structure. This embodiment can also select wavelet transform, a frequency domain denoising technique that separates components of different frequencies by converting the image to the wavelet domain. Then, high-frequency components can be thresholded to remove noise, and finally, the processed components are inversely transformed back to the spatial domain. Wavelet denoising is highly effective when processing images with multi-scale features. In addition, this embodiment can also select nonlocal means (NLM), a denoising method that denoises by considering the similarity of all pixels in the image. This method calculates a weighted average for each pixel, with weights based on the similarity to other pixels similar to that pixel. This approach effectively preserves image details while removing noise and is suitable for various types of noise.

[0082] Preferably, the denoised image is enhanced to obtain an enhanced image, including:

[0083] A moving image enhancement window is constructed based on the scale of the denoised image;

[0084] According to formula F ij =|x ij -y ij | Obtain the pixel difference value for each pixel in each image enhancement window; where, F ij x represents the pixel difference at position (i, j) in the denoised image. ij y represents the pixel value at position (i, j) in the original acquired image. ij This represents the pixel value at position (i, j) in the denoised image;

[0085] According to the formula The pixel difference is corrected for brightness uniformity to obtain the correction difference; where D ij D is the correction difference for each pixel. dark Let K(i) be the dark signal value of the pixel, and K(i) be the correction coefficient.

[0086] Pixels whose correction difference is greater than the feature threshold are used as feature points and their number is recorded.

[0087] The proportion of feature points to all pixels under the image enhancement window is used as the percentage value;

[0088] When the proportion value within the moving image enhancement window is greater than a preset threshold, median denoising is performed on the denoised image within the corresponding moving image enhancement window to obtain median denoised data.

[0089] Slide the moving image enhancement window until the entire median denoised data has been traversed to obtain the enhanced image.

[0090] Preferably, feature extraction is performed on the enhanced image to obtain image features corresponding to different angles, including:

[0091] The enhanced image is then subjected to image scaling and image normalization to obtain a normalized image;

[0092] The normalized image is used to extract features using a pre-defined VGG16 network to obtain various image features.

[0093] Specifically, this embodiment uses image processing libraries (such as OpenCV, PIL, etc.) to scale the enhanced image to a uniform size (e.g., 224x224 pixels) to adapt to the input requirements of the preset convolutional neural network. Furthermore, this embodiment normalizes the pixel values ​​of the scaled image, scaling them to the range of [0,1] or [-1,1].

[0094] Furthermore, in addition to the VGG16 network, this embodiment can also select other pre-trained convolutional neural networks (such as ResNet50, InceptionV3, etc.) and load the model. This embodiment uses a deep learning framework (such as TensorFlow or PyTorch) for implementation.

[0095] Preferably, the fusion feature is obtained through a weighted average fusion method; the formula for the weighted average fusion is:

[0096]

[0097] Among them, f i For the image features at the i-th angle, w i f is the weight of the i-th image feature, where n is the number of image features. fused For the fusion feature, C(θ i ) represents the coverage of the i-th angle, and j is the index of all angles in the feature fusion, with the value of j ranging from 1 to n.

[0098] Preferably, the formula for calculating the coverage is:

[0099]

[0100] Among them, A i Let A be the visible surface area of ​​the 3D visualization model corresponding to the target object as observed from the i-th viewpoint. total The total surface area of ​​the three-dimensional visualization model corresponding to the target object is denoted as .

[0101] Optionally, this embodiment considers that images from different angles cover different parts of the target object, and assigns weights based on the viewpoint coverage. This embodiment uses a 3D model for calculation, evaluating coverage by calculating the surface area observed from the i-th viewpoint. The method for calculating the visible surface area in this embodiment is using ray casting technology. The basic idea is to emit a ray from the viewpoint towards the target object and calculate the intersection point with the target object.

[0102] Furthermore, the calculation steps for the visible surface area in this embodiment are as follows:

[0103] 1. Define viewpoint: Set the observation viewpoint position P i For example, P i (x i ,y i ,z i ).

[0104] 2. Generate a ray from viewpoint P. i Rays are emitted in all directions; the directions of the rays can be randomly generated or uniformly distributed, such as using a spherical coordinate system.

[0105] 3. Calculate the intersection points. Determine the intersection point between each ray and the target object, and calculate the number of intersection points N. For each ray, if it intersects the object, record the intersection point.

[0106] 4. Calculate the visible surface area, assuming that the intersection of each ray with the surface forms a small area element. The visible surface area can be calculated by multiplying the average of the small areas by the number of intersections.

[0107] visible surface area A i The calculation formula is: A i =k·N;

[0108] Where k is the small area corresponding to each intersection point (depending on the angle and distribution between rays), and the unit is usually a square unit (such as a square centimeter).

[0109] Furthermore, the multi-view target recognition model in this embodiment employs a deep learning architecture to effectively process input features from different perspectives and perform target recognition. Specifically, the model's structure is based on a convolutional neural network (CNN), combined with a multi-view feature fusion mechanism, and its implementation process is as follows:

[0110] First, the model in this embodiment receives fused features from different angles, obtained through multi-angle image features extracted in previous steps. The fused features are then fed into multiple convolutional layers of the model, which are responsible for extracting higher-level feature information. Each convolutional layer is typically followed by a pooling layer to reduce the dimensionality of the feature map while preserving important spatial features. Through layer-by-layer extraction, the model can capture various features of the target, such as shape, texture, and structure, from different viewpoints.

[0111] The model in this embodiment also uses a feature fusion layer to merge features from multiple perspectives. This fusion can be achieved through simple concatenation, weighted averaging, or more complex feature selection mechanisms to ensure that the model can fully utilize multi-perspective information.

[0112] In the final stage of the model, the convolutional and fused features are fed into a fully connected layer for further nonlinear transformation, ultimately outputting a probability distribution representing the likelihood of each category. The model is trained using the cross-entropy loss function to optimize the prediction results. To improve the accuracy and robustness of recognition, the model can also incorporate data augmentation techniques and regularization methods (such as dropout or L2 regularization) to prevent overfitting.

[0113] During the training phase, a labeled multi-view image dataset is used to train the model, and the model parameters are continuously updated through the backpropagation algorithm. After training, the model can accurately identify targets in new input data, regardless of whether the data comes from different perspectives of the same target or changes in environmental conditions. This deep learning multi-view target recognition model enables efficient and accurate target recognition processing.

[0114] Corresponding to the above methods, such as Figure 2 As shown, this embodiment also provides an image data acquisition and processing system based on computer vision, including:

[0115] The image acquisition module is used to acquire raw images of the target object from different angles;

[0116] An image denoising module is used to denoise each of the original acquired images to obtain a denoised image;

[0117] The image enhancement module is used to enhance the denoised image to obtain an enhanced image;

[0118] The feature extraction module is used to extract features from the enhanced image to obtain image features corresponding to different angles;

[0119] The feature fusion module is used to perform feature fusion on the image features to obtain fused features;

[0120] The recognition module is used to input the fused features into the trained multi-view target recognition model to obtain the processing result.

[0121] The beneficial effects of this invention are as follows:

[0122] (1) This invention effectively reduces noise interference in the original acquired image by denoising the image, thereby improving the image's clarity and usability. After denoising, the image enhancement step further improves the image's contrast and brightness, making the features more obvious and facilitating subsequent feature extraction.

[0123] (2) The images processed by this invention after denoising and enhancement provide richer feature information, thereby making the feature extraction process more efficient and accurate. With higher quality input, the model can capture more details and reflect the multi-dimensional features of the target, such as shape, texture and color.

[0124] (3) This invention can effectively improve the accuracy of target recognition by fusing image features from different angles. The fused features not only take into account the impact of viewpoint changes on target features, but also overcome the information gaps that may exist under a single viewpoint, so as to obtain a more comprehensive target description.

[0125] (4) This invention can cope with changes in different environmental conditions, such as changes in lighting and viewing angle. The acquisition and processing of multi-view images greatly improves the robustness of the system and can effectively reduce the false recognition rate. The model can maintain good performance when facing complex scenarios in real-world applications.

[0126] (5) This invention can adapt to a variety of different application scenarios and target objects, and has strong scalability. With the development of technology, more image processing technologies can be introduced or the model architecture can be adjusted to continuously improve recognition performance and adaptability.

[0127] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0128] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A computer vision-based image data acquisition and processing method, characterized in that: include: Acquire original images of the target recognition object at different angles; Performing image denoising on each of the original collected images to obtain a denoised image; Performing image enhancement on the denoised image to obtain an enhanced image; Extracting features from the enhanced image to obtain image features corresponding to different angles; Performing feature fusion on the image features to obtain fused features; The fused features are input into a trained multi-view target recognition model to obtain a processing result.

2. The computer vision-based image data acquisition and processing method according to claim 1, characterized in that: Get the original images of the target object at different angles and heights, including: Acquire a three-dimensional visualization model of the target recognition object, and acquire surface information of the three-dimensional visualization model; the surface information includes vertices, edges and faces; determining an optimal shooting point according to the surface information; An image acquisition device is used to acquire an original acquisition image of the target recognition object at the optimal shooting point.

3. The computer vision-based image data acquisition and processing method according to claim 2, characterized in that: Determining the best shooting point according to the surface information includes: For each face of the 3D visualization model, calculate the surface normal N of the face k ; The surface normal N k The calculation formula is: N k =(P2-P1)×(P3-P1); P1, P2 and P3 are the three vertices on the kth face respectively; The surface normal N k The normal vector of is normalized; the normalization formula is: Generate a set of candidate shooting positions V according to the center and size of the three-dimensional visualization model; the candidate shooting positions V are placed on a spherical surface outside the three-dimensional visualization model or in a sphere with the center of the three-dimensional visualization model as the origin; For each candidate shooting position V, a viewing angle vector L from the viewpoint to the center of the model is calculated; wherein L=CV; C is the center of the three-dimensional visualization model; For each candidate shooting position V, the angle with the normal of the 3D visualization model is calculated to evaluate the surface quality covered by the viewpoint; the scoring formula for evaluating the surface quality covered by the viewpoint is: s = ∑ k cos(α k ), s is the score of the shooting position, k is the number of all surfaces traversed in the model, · represents the dot product, Based on the scores of the shooting locations, multiple locations with the highest scores are selected as optimal shooting points.

4. The computer vision-based image data acquisition and processing method according to claim 1, characterized in that: The image denoising is performed based on a Gaussian filtering method.

5. The computer vision-based image data acquisition and processing method according to claim 1, characterized in that: Performing image enhancement on the denoised image to obtain an enhanced image includes: Constructing a moving image enhancement window according to the scale of the denoised image; According to formula F ij =|x ij -y ij |Get the pixel difference of each pixel under each image enhancement window; where F ij represents the pixel difference of the denoised image at position (i, j), x ij Represents the pixel value of the original captured image at position (i, j), y ij Represents the pixel value of the denoised image at position (i, j); According to the formula Perform brightness uniformity correction on the pixel difference to obtain a corrected difference; wherein, D ij is the correction difference of the pixel point, D dark is the dark signal value of the pixel point, K(i) is the correction coefficient; Pixel points whose correction difference is greater than the feature threshold are taken as feature points and the number is recorded; The ratio of the feature points to all the pixels in the image enhancement window is taken as a ratio value; When the proportion value in the moving image enhancement window is greater than a preset threshold, median denoising is performed on the denoised image in the corresponding moving image enhancement window to obtain median denoising data; The moving image enhancement window is slid until the entire median denoising data is traversed to obtain the enhanced image.

6. The computer vision-based image data acquisition and processing method according to claim 1, characterized in that: Feature extraction is performed on the enhanced image to obtain image features corresponding to different angles, including: Performing image scaling and image normalization processing on the enhanced image to obtain a normalized image; The preset VGG16 network is used to perform feature extraction on the normalized image to obtain various image features.

7. The computer vision-based image data acquisition and processing method according to claim 1, characterized in that: The fusion feature is obtained by weighted average fusion; the formula of weighted average fusion is: Among them, f i is the image feature of the i-th angle, w i is the weight of the i-th image feature, n is the number of image features, f fused is the fusion feature, C(θ i ) is the coverage of the i-th angle, j is the index of all angles in feature fusion, and the value range of j is from 1 to n.

8. The computer vision-based image data acquisition and processing method according to claim 7, characterized in that: The calculation formula of the coverage is: Among them, A i is the visible surface area of ​​the three-dimensional visualization model corresponding to the target recognition object observed from the i-th viewing angle, A total The total surface area of ​​the three-dimensional visualization model corresponding to the target identification object.

9. An image data acquisition and processing system based on computer vision, characterized in that: include: An image acquisition module is used to acquire original images of the target recognition object at different angles; An image denoising module, used for performing image denoising on each of the original collected images to obtain a denoised image; An image enhancement module, used for performing image enhancement on the denoised image to obtain an enhanced image; A feature extraction module, used to extract features from the enhanced image to obtain image features corresponding to different angles; A feature fusion module, used for fusing the image features to obtain fused features; The recognition module is used to input the fusion features into a trained multi-view target recognition model to obtain a processing result.

Images