A data processing method for AR intelligent camera device

Abstract

The present invention discloses a data processing method for an AR smart camera device, the method specifically comprising: obtaining the original image data captured in real time by the AR smart camera device, processing the original image data using an adaptive illumination compensation algorithm to obtain first image data; performing denoising on the first image data to obtain second image data; determining whether any athlete in the second image data is blocked, and if blocked, completing the data according to the athlete's historical trajectory to form third image data; marking the illegal behavior in the third image data to form fourth image data; placing the fourth image data into an AR information display template for fusion processing to generate AR display image data. The present invention can effectively improve the viewing and interactivity of live broadcasts of sports events, provide viewers with high-quality, personalized AR enhanced experience, and greatly improve user satisfaction and participation.

Description

Technical Field

[0001] The present invention relates to the field of AR smart camera technology, and in particular to a data processing method for an AR smart camera device. Background Art

[0002] With the rapid development of science and technology, augmented reality (AR) technology is increasingly being used in live broadcasts of sports events, bringing viewers an unprecedented immersive viewing experience. AR smart cameras analyze camera images in real time and overlay virtual information, such as athlete statistics and path trajectories, greatly enriching the audience's viewing perspective. However, the complexity and variability of sports event scenes pose many challenges to the data processing of AR smart cameras.

[0003] First, the environment of sports events is usually noisy, and the lighting conditions vary greatly, such as changeable weather and flickering lights on site. These factors may lead to unstable camera image quality. Traditional image processing algorithms often find it difficult to achieve ideal results when faced with such changing lighting conditions, which increases the difficulty of image preprocessing. Therefore, how to effectively deal with lighting changes and improve image quality has become a major problem in the application of AR smart camera devices in sports events.

[0004] Secondly, the rapid movement and frequent overlap of athletes in sports events place extremely high demands on target recognition and tracking algorithms. Traditional target recognition algorithms are prone to target loss and confusion when dealing with such complex scenarios, which greatly reduces the accuracy and real-time nature of AR information. Especially in scenarios with dense athletes and fast-changing movements, how to accurately distinguish and track different athletes has become the key to improving the AR viewing experience.

[0005] Furthermore, occlusions often occur in sports events, such as occlusions by other athletes, referees, or equipment on the sidelines, which brings great uncertainty to target tracking. Traditional tracking algorithms often have difficulty in continuously and effectively tracking targets when faced with occlusions, resulting in a disconnect between AR information and the real picture, affecting the audience's viewing experience. Therefore, how to effectively deal with occlusions and maintain the continuity and accuracy of target tracking has become an important issue that AR smart camera devices need to solve.

[0006] In addition, the rules and characteristics of different sports vary significantly. For example, football, basketball, track and field, etc., differ in the number of athletes, field size, scoring rules, etc. These differences lead to the need to adjust the content and presentation of AR information accordingly to meet the needs of different sports. However, traditional AR fusion algorithms often lack flexibility and are difficult to adjust according to the characteristics of different sports, resulting in a stereotyped AR effect and an inability to provide a personalized viewing experience. Summary of the invention

[0007] The purpose of the present invention is to provide a data processing method for an AR smart camera device, which can effectively improve the viewing experience and interactivity of live sports events, provide viewers with high-quality, personalized AR enhanced experience, greatly improve user satisfaction and participation, and solve at least one of the above-mentioned prior art problems.

[0008] In a first aspect, the present invention provides a data processing method for an AR smart camera device, the method specifically comprising:

[0009] Acquire raw image data captured in real time by the AR smart camera device, and process the raw image data using an adaptive illumination compensation algorithm to obtain first image data;

[0010] Using a pre-built image denoising model, denoising the first image data to obtain second image data;

[0011] Using a convolutional neural network-based target detection algorithm to extract athlete features from the second image data to obtain athlete position information;

[0012] According to the athlete's position information, a Kalman filter algorithm is used to predict the athlete's movement trajectory to obtain the athlete's movement trend information;

[0013] Based on the sports trend information of the athletes, determining whether any athlete in the second image data is blocked, and if blocked, completing the third image data according to the historical trajectory of the athlete;

[0014] Based on a preset sports rules library, marking the illegal behaviors in the third image data to form fourth image data;

[0015] According to the user preference information, a corresponding AR information display template is determined from a preset AR information display template library, and the fourth image data is placed into the AR information display template for fusion processing to generate AR display image data.

[0016] In a second aspect, the present invention provides a data processing system for an AR smart camera device, the system specifically comprising:

[0017] A first data processing module is used to obtain raw image data captured by the AR smart camera device in real time, and process the raw image data using an adaptive illumination compensation algorithm to obtain first image data;

[0018] A second data processing module, configured to use a pre-built image denoising model to perform denoising on the first image data to obtain second image data;

[0019] A third data processing module is used to extract athlete features from the second image data using a target detection algorithm based on a convolutional neural network to obtain athlete position information;

[0020] A fourth data processing module is used to predict the athlete's movement trajectory using a Kalman filter algorithm based on the athlete's position information to obtain the athlete's movement trend information;

[0021] a fifth data processing module, for determining whether any athlete in the second image data is blocked based on the athlete movement trend information, and if blocked, completing the block according to the athlete's historical trajectory to form third image data;

[0022] a sixth data processing module, configured to mark the illegal behaviors in the third image data based on a preset sports rule library to form fourth image data;

[0023] The seventh data processing module is used to determine the corresponding AR information display template from the preset AR information display template library according to the user preference information, put the fourth image data into the AR information display template for fusion processing, and generate AR display image data.

[0024] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor and a computer program stored in the memory, and when the computer program is executed on the processor, it implements a data processing method for an AR smart camera device as described in any one of the above methods.

[0025] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the data processing method for an AR smart camera device as described in any one of the above methods is implemented.

[0026] Compared with the prior art, the present invention has at least one of the following technical effects:

[0027] 1. The present invention can effectively improve the viewing and interactivity of live sports events, provide viewers with high-quality, personalized AR enhanced experience, and greatly improve user satisfaction and participation.

[0028] 2. The present invention adopts an image segmentation algorithm based on region growing, which can process different lighting areas in the image more finely and avoid the excessive or insufficient problems that may be caused by global illumination compensation; the illumination compensation model established by the machine learning algorithm can automatically adjust the compensation parameters according to the illumination characteristics of the image, so as to obtain clear and stable images under various lighting conditions.

[0029] 3. The present invention uses a convolutional neural network model for image denoising training, which can learn the distribution and characteristics of image noise, thereby more effectively removing noise interference in the image.

[0030] 4. The present invention uses the pre-trained YOLOv3 model for fast and accurate target detection, which can obtain the athlete's position information in real time. By combining Canny edge detection and region growing algorithm, it can further accurately extract the athlete's contour and features, thereby improving the accuracy of target recognition.

[0031] 5. The present invention uses the Kalman filter algorithm to predict the athlete's movement trajectory, which can smoothly track the athlete's movement and maintain a good tracking effect even when the athlete moves quickly or is blocked, providing continuous and stable athlete position information for subsequent AR information display.

[0032] 6. The present invention can automatically identify and mark athletes' violations by matching image features with a sports rule library, which not only improves the fairness of the game but also provides viewers with more information about game rules and athlete behavior.

[0033] 7. The present invention selects the most suitable AR information display template according to the user's personalized preference information, and can provide the user with a viewing experience that better meets his or her needs and preferences.

[0034] 8. The present invention allocates the attention weight of the display area through the deep convolution generative adversarial network algorithm, which can further optimize the display effect of AR information and make it more prominent and vivid. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0036] Figure 1 It is a flowchart of a data processing method for an AR smart camera device provided by an embodiment of the present invention;

[0037] Figure 2 It is a structural schematic diagram of a data processing system for an AR smart camera device provided by one embodiment of the present invention;

[0038] Figure 3 It is a structural schematic diagram of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0039] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0040] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0041] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0042] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0043] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0044] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0045] In the embodiment of the present application, the execution subject of the process includes a terminal device, which includes but is not limited to: a server, a computer, a smart phone, a tablet computer, and other devices capable of executing the method disclosed in the present application. Figure 1 A flow chart of a data processing method for an AR smart camera device disclosed in an embodiment of the present invention is shown, and the details are as follows:

[0046] S101, obtaining original image data captured in real time by an AR smart camera device, and processing the original image data using an adaptive illumination compensation algorithm to obtain first image data.

[0047] In this embodiment, the AR smart camera device is fixed at the scene of the sports event to ensure that its field of view covers the activity area of ​​all athletes. The camera device is turned on to acquire and transmit the original image data in real time. The adaptive illumination compensation algorithm is based on image segmentation and machine learning technology. First, the original image data is divided into multiple regions, and each region is processed separately to deal with the problem of uneven illumination. Then, the illumination features of each region, such as brightness, contrast, etc., are extracted to form an illumination feature vector. Next, a machine learning algorithm (such as support vector machine, random forest, etc.) is used to train the illumination compensation model, and the model input is the illumination feature vector and the output is the illumination compensation parameter. Using the trained illumination compensation model, the illumination compensation parameters of each region, such as gain, offset, etc., are automatically adjusted according to the illumination feature vector. Finally, all processed regional images are spliced ​​together to form the first image data.

[0048] In this embodiment, the adaptive illumination compensation algorithm can accurately process images under different illumination conditions, effectively improve the problem of uneven illumination in the image, and make the image clearer and brighter. Since the algorithm can automatically adjust the compensation parameters according to the illumination characteristics of the image, the stability and consistency of the image can be maintained even when the illumination conditions change rapidly. The image data processed by the adaptive illumination compensation algorithm has more uniform and stable illumination conditions, which is conducive to the efficient execution of subsequent processing steps such as target detection, tracking, and AR information display.

[0049] S102: Using a pre-built image denoising model, perform denoising processing on the first image data to obtain second image data.

[0050] In this embodiment, an image denoising model based on deep learning can be selected, such as a denoising model based on a convolutional neural network (CNN), a denoising model based on a generative adversarial network (GAN), or a denoising model based on SwinUNet. The first image data processed by the adaptive illumination compensation algorithm is input into the pre-built image denoising model. After receiving the input data, the model automatically identifies and removes the noise in the image. The denoising process may involve multiple steps, such as feature extraction, noise estimation, noise removal, and image reconstruction. After denoising, the model outputs the second image data, that is, the denoised image. By comparing the image data before and after denoising, it is verified whether the denoising effect meets expectations. The evaluation can be carried out by combining subjective evaluation (such as visual observation) and objective evaluation (such as peak signal-to-noise ratio, structural similarity index, etc.). If the denoising effect is not ideal, the pre-built image denoising model can be optimized. The optimization method includes adjusting model parameters, increasing training data, improving model structure, etc.

[0051] In this embodiment, by using a pre-built image denoising model to perform denoising on the first image data, noise and spots in the image can be effectively removed, making the image clearer and smoother. This helps to improve the accuracy and efficiency of subsequent image analysis, target detection and recognition tasks. Denoising can not only remove noise, but also enhance the image details to a certain extent, which makes the edges, textures and other features in the image more prominent, which is conducive to the precise execution of subsequent processing steps.

[0052] S103: extract athlete features from the second image data using a target detection algorithm based on a convolutional neural network to obtain athlete position information.

[0053] In this embodiment, image data of various scenes and sports types including athletes are collected to ensure the diversity and representativeness of the data. The image data is manually annotated to mark the position (such as a bounding box) and category (such as different athletes) of the athletes. A classic convolutional neural network model is selected, such as Faster R-CNN, YOLO series or SSD series, which perform well in target detection tasks. A model pre-trained on a large data set (such as ImageNet, COCO, etc.) is used as a starting point to improve the generalization ability and detection accuracy of the model. The image data is enhanced, such as rotation, scaling, flipping, etc., to increase the diversity of the data and improve the robustness of the model. The enhanced image data and the annotated data are input into the convolutional neural network model for end-to-end training. The parameters of the model are optimized by the back propagation algorithm so that the model can accurately detect the position and category of the athlete. The second image data is feature extracted using the trained convolutional neural network model. The model will automatically learn the unique features of the athlete, such as shape, texture, color, etc. The model will output the athlete's position information, usually in the form of a bounding box. By parsing the coordinates of the bounding box, we can get the exact position of the player in the image.

[0054] In this embodiment, the target detection algorithm based on the convolutional neural network can accurately detect the position and category of the athlete, and can maintain a high detection accuracy even in complex backgrounds and multiple sports types.

[0055] S104, predicting the athlete's movement trajectory using a Kalman filter algorithm based on the athlete's position information to obtain the athlete's movement trend information.

[0056] In this embodiment, the position information of the athlete in the continuous frame image is obtained from the target detection algorithm based on the convolutional neural network, including the coordinates of the athlete's bounding box (usually the coordinates of the center point or the coordinates of the upper left corner and the lower right corner). The position (such as x coordinate, y coordinate) and speed (such as vx, vy) of the athlete are used as state variables, that is, the state vector X. According to the motion law of the athlete, the state transfer equation is defined to describe the change of the state variable from time k to time k+1. Usually, this can be a simple linear motion model, such as uniform linear motion or uniformly accelerated linear motion. The athlete position information output by the target detection algorithm is used as the observation quantity Z. The observation matrix H is defined to map the state vector X to the observation quantity Z. The system error matrix Q and the observation error matrix R are set according to experience or experimental data, which respectively describe the uncertainty of the state transfer equation and the observation quantity. The initial state vector X0 and the initial covariance matrix P0 are set. These initial values ​​can be set according to the actual situation, and will be updated by the Kalman filter algorithm later. According to the state transfer equation and the current state vector Xk-1, the state vector Xk_pred and the covariance matrix Pk_pred of the next moment are predicted. According to the observation quantity Zk, the observation matrix H, the system error matrix Q and the observation error matrix R, the Kalman gain Kk is calculated, and then the state vector Xk and the covariance matrix Pk are updated. By continuously applying the Kalman filter algorithm, the predicted position of the athlete at the future moment can be obtained, thereby drawing the athlete's movement trajectory. Based on the predicted position information, the athlete's movement trend can be analyzed, such as speed change, direction change, etc.

[0057] In this embodiment, the Kalman filter algorithm can effectively smooth the athlete's motion trajectory, reduce the impact of noise and errors, and improve the accuracy and continuity of the trajectory. The Kalman filter algorithm is real-time and can update and predict the state immediately after each frame of image is acquired, which is suitable for real-time tracking and prediction tasks.

[0058] S105, based on the athlete movement trend information, determine whether any athlete in the second image data is blocked, and if blocked, complete it according to the historical trajectory of the athlete to form third image data.

[0059] In the present embodiment, in the second image data, it is checked whether the image of each athlete is complete. If the image of a certain athlete is partially or completely missing, it is judged that the athlete may be blocked by other athletes or stadium facilities. Combined with the motion trend information, the occurrence of blocking can be further confirmed. For example, if the motion trend of an athlete shows that he should appear in a certain position, but he cannot be found in the image of the position, then it is likely that he is blocked. Once it is determined that a certain athlete is blocked, his historical trajectory information is used to complete the current image data. The historical trajectory information can include data such as the position, speed, direction, etc. of the athlete in the past several image frames. Through an interpolation algorithm or a machine learning model, the possible position of the athlete at the current blocked moment can be predicted according to these historical data, and his image can be drawn accordingly in the third image data. The completed athlete image is merged with the original image data or other unblocked athlete images to form a complete third image data. This set of data not only contains the panoramic information of the stadium, but also ensures that the image of each athlete is complete and accurate.

[0060] In this embodiment, the blocked athlete images are supplemented by historical trajectories, thereby ensuring the integrity of the third image data, which enables analysts to more accurately understand the running situation and tactical execution effect of each athlete.

[0061] S106: Based on a preset sports rule library, marking the illegal behaviors in the third image data to form fourth image data.

[0062] In this embodiment, computer vision technology, such as a deep learning model or an image processing algorithm, is used to perform a detailed analysis of the third image data. Key information such as the position, action, and relationship between the athlete and the ball are identified. The information obtained from the image analysis is matched with a preset sports rule library. According to the judgment conditions in the rule library, it is determined whether there is a violation in the image. If a violation is detected, a specific annotation method (such as a border, an arrow, a text, etc.) is used to mark the violation and the violating athlete on the image. The annotation information includes information such as the type of violation, the time of occurrence, and the athletes involved, and then the annotated image data is saved as the fourth image data.

[0063] In this embodiment, by combining computer vision technology and a pre-set sports rule library, violations in the game can be more accurately identified and marked, which reduces the subjectivity and errors of manual detection and improves the objectivity and accuracy of detection.

[0064] S107, determining a corresponding AR information display template from a preset AR information display template library according to the user preference information, putting the fourth image data into the AR information display template for fusion processing, and generating AR display image data.

[0065] In this embodiment, when the AR live broadcast application is first started or set by the user, the user's preference information is collected, such as the type of AR information (such as player data, tactical analysis, violation prompts, etc.), AR information display style (such as concise, detailed, dynamic, etc.), color preference, etc. According to common user needs and design aesthetics, a library containing multiple AR information display templates is established. Each template contains different information display layouts, color schemes, animation effects, etc. to meet the preferences of different users. According to the collected user preference information, the most suitable template is selected from the AR information display template library, and the matching process can be based on the weight of the user preference and the similarity of the template features. The fourth image data (i.e., the game screen containing the violation mark) is used as the background image. According to the selected AR information display template, the relevant AR information (such as player name, number, violation prompt, etc.) is superimposed on the background image. Image processing techniques such as transparency adjustment, shadow effect, animation transition, etc. are used to make the AR information and the background image naturally merge. The fused image data is saved as AR display image data, and the AR display image data contains both the real-time screen of the game and the AR information preferred by the user.

[0066] In this embodiment, through the well-designed AR information display template, key information can be presented to users in an intuitive and clear manner, which helps users to obtain and understand important information in the game more quickly and improves the readability and practicality of the information. The natural integration of AR information with the game screen can bring users a more immersive viewing experience, as if users are at the game scene, and can participate in the game more deeply, which enhances the attractiveness and viewing value of the game.

[0067] In some embodiments, in the above step S101, the adopting the adaptive illumination compensation algorithm to process the original image data to obtain the first image data specifically includes:

[0068] Using an image segmentation algorithm based on region growing, the original image data is segmented into multiple regions to form multiple region images;

[0069] Extract the illumination features of each region image respectively to form an illumination feature vector;

[0070] Collecting first image samples under different lighting conditions, and marking corresponding lighting compensation parameters according to various lighting features of the first image samples to form a lighting compensation training data set;

[0071] Taking the illumination compensation training data set as input, using a machine learning algorithm for modeling training to establish an illumination compensation model;

[0072] Inputting the illumination characteristic vector of each regional image into the illumination compensation model to obtain illumination compensation parameters of each regional image;

[0073] According to the illumination compensation parameters, illumination compensation processing is performed on each regional image respectively, and all regional images after illumination compensation processing are spliced ​​to form first image data.

[0074] In this embodiment, an image is segmented into multiple regions by an image segmentation algorithm based on region growing to obtain segmented region images. The region growing algorithm starts from a selected seed pixel and gradually merges adjacent pixels into the same region according to the similarity criterion between pixels until the merging condition is no longer met. For each segmented region image, the illumination features of the region are extracted, including features such as brightness mean and contrast, to obtain the illumination feature vector of the region. By collecting image samples under different illumination conditions in advance, the illumination features are extracted, and the illumination compensation model is trained and established corresponding to the manually annotated illumination compensation parameters. The illumination feature vector of each region is input into the illumination compensation model, and the illumination compensation parameters of the region are obtained by model calculation. According to the illumination compensation parameters of each region, the region image is subjected to illumination compensation processing to obtain the region image after illumination compensation. The illumination compensation processing adjusts the gray value of the pixel so that the image presents a consistent visual effect under the illumination conditions of different regions. All the region images after illumination compensation are spliced ​​to obtain a complete illumination compensation image as the first image data.

[0075] For example, taking the image of an indoor sports event scene as an example, the average brightness, contrast and color temperature of each area can be extracted. The vector composed of these features can fully describe the lighting conditions of the area and provide a basis for subsequent lighting compensation. For example, a darker corner area may have a lower average brightness value and a higher contrast value. The construction of the lighting compensation training data set is crucial to the performance of the model. Samples can be collected by taking photos of the same scene at different times and in different weather conditions. For example, for a football sports event scene, it can be taken under different lighting conditions such as sunny days, cloudy days, and dusk, and the corresponding lighting compensation parameters are marked. These parameters include exposure compensation value, white balance adjustment, etc. Machine learning algorithms play an important role in the establishment of lighting compensation models. Algorithms such as support vector machines (SVM) or deep neural networks can be selected for training. The input is the lighting feature vector and the output is the corresponding lighting compensation parameter. In this way, the model can learn the best compensation strategy under different lighting conditions. Light compensation processing is a key step in improving image quality. For each segmented region, adjustments are made based on the compensation parameters output by the model. For example, for a window area in an indoor photo, the brightness may need to be reduced to avoid overexposure, while the brightness of the shadow area needs to be increased to show more details. This regional adjustment can more accurately improve the overall lighting effect of the image. The final image stitching step reassembles all the compensated regions into a complete image. This process requires attention to the smooth transition of region boundaries to avoid obvious stitching traces. For example, techniques such as gradient blending or Poisson fusion can be used to achieve a natural transition effect. This region-based lighting compensation method has significant advantages over global adjustment. It can personalize the lighting characteristics of different regions in the image, thereby maximizing the visual quality of each region while maintaining the overall harmony of the image. This is especially important for image processing of complex scenes, such as high dynamic range (HDR) scenes or mixed indoor and outdoor scenes with complex lighting conditions. In this way, the local overexposure or underexposure problems that are difficult to handle with traditional global adjustments can be effectively solved, and a more natural and detailed image can be obtained.

[0076] In some embodiments, in the above step S102, the use of a pre-built image denoising model to perform denoising processing on the first image data to obtain second image data may also include:

[0077] Collecting second image samples containing different noise interference data, marking the noise type and noise distribution of the second image samples, and forming an image denoising training data set;

[0078] The image denoising training data set is used as input, and a convolutional neural network model is used for modeling training to form an image denoising model.

[0079] In this embodiment, image samples containing different noise interferences are obtained, and for each image sample, the noise type and noise distribution are determined and annotated to obtain a set of image samples with noise annotation information. The set of image samples with noise annotation information is preprocessed, the image size and format are unified, and divided into a training set and a verification set, and a training data set and a verification data set for image denoising are constructed. According to the goal of image denoising, the architecture of the convolutional neural network model is designed, and parameters such as the number of convolutional layers, the size of the convolutional kernel, and the setting of the pooling layer are determined to build an image denoising convolutional neural network model. The constructed training data set is input into the image denoising convolutional neural network model, and the model parameters are continuously adjusted through the forward propagation and back propagation processes to perform modeling training to obtain a preliminary image denoising model. The performance of the preliminary image denoising model obtained by training is evaluated using the verification data set, and the peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) indicators before and after denoising are calculated to judge the denoising effect of the model. If the denoising performance of the model does not meet expectations, the hyperparameters of the convolutional neural network model, such as learning rate, batch size, etc., are adjusted according to the performance evaluation results, and the training is repeated until an image denoising model that meets the denoising requirements is obtained. The trained image denoising model is used to denoise the image to be processed that contains noise interference. Through the forward propagation process of the convolutional neural network, image features are automatically extracted, noise interference is eliminated, and a clear denoised image is obtained.

[0080] For example, we first need to collect image samples containing various noise interferences, which may include common types such as Gaussian noise, salt and pepper noise, Poisson noise, and special noise caused by equipment defects or environmental factors. For example, night photos taken under low light conditions may contain obvious granular noise, and photos of fast-moving objects may appear motion blurred. It is crucial to accurately annotate the noise types and distributions of the collected image samples. This process requires not only professional knowledge, but also the help of automated tools. For example, for Gaussian noise, the noise intensity can be estimated by calculating the local variance of the image; for salt and pepper noise, the noise density can be determined by detecting the distribution of extreme pixel points. During the annotation process, the distribution of noise in different areas of the image must also be considered, because some noise may be more obvious in dark or bright areas. Organize the annotated image samples into a training data set, which should cover all possible noise situations. For example, it can include images with different lighting conditions, different shooting equipment, and different scene types to ensure the generalization ability of the model. The diversity of the data set directly affects the actual application effect of the model, so it may be necessary to expand the samples through data enhancement techniques, such as rotation, scaling, or adding artificial noise. Using convolutional neural network (CNN) to train image denoising models is one of the most effective methods at present. The multi-layer structure of CNN can automatically learn the hierarchical features of the image, so as to effectively distinguish noise from useful information. In model design, it is possible to consider using residual connections to retain the original image information, or using attention mechanisms to focus on areas with severe noise. For example, a typical denoising network may contain multiple convolutional layers, pooling layers, and upsampling layers. The input is a noisy image, and the output is a clear image after denoising. During the training process, it is crucial to choose a suitable loss function. Commonly used loss functions include mean square error (MSE) and structural similarity index (SSIM). The former focuses on pixel-level differences, while the latter focuses more on the image quality perceived by the human eye. By adjusting the weights of these loss functions, a balance can be achieved between denoising effect and image detail retention. In addition, the use of appropriate optimizers and learning rate scheduling strategies can also significantly improve model performance. The image denoising model finally trained can not only effectively remove various types of noise, but also maintain image details and texture. This deep learning-based method has obvious advantages over traditional filtering technology. It can better adapt to complex noise distribution and improve clarity while maintaining the overall structure of the image. For example, for a landscape photo containing subtle textures, traditional methods may over-smooth the image and cause loss of details, while deep learning models can better distinguish between noise and meaningful texture information, thereby retaining the natural texture of the image while denoising.

[0081] In some embodiments, in the above step S103, the extracting athlete features from the second image data using a target detection algorithm based on a convolutional neural network to obtain the athlete position information specifically includes:

[0082] Using a pre-trained YOLOv3 convolutional neural network model, perform model inference on the second image data to obtain a candidate region bounding box of the athlete target in the second image data and a corresponding confidence level;

[0083] The candidate region bounding box with a confidence greater than a preset confidence threshold is taken as a candidate target region, and the target contour features of the candidate target region are extracted using the Canny edge detection algorithm to obtain a binary contour image;

[0084] Using the geometric center of the contour image as a seed point, segmenting the candidate target region using a region growing algorithm to obtain segmented connected regions, and using the connected regions as the precise region of the athlete target;

[0085] Extracting HSV color histogram features and LBP texture features from the precise area to construct a visual feature vector;

[0086] A support vector machine classifier is used to classify and identify the visual feature vector, obtain position information of different athlete targets, and form athlete position information.

[0087] In this embodiment, the second image data is obtained and input into the pre-trained YOLOv3 convolutional neural network model for model inference processing. The candidate region bounding box and the corresponding confidence of the athlete target in the second image data are obtained by model inference. It is determined whether the confidence of the candidate region bounding box is greater than the preset confidence threshold. If it is, the candidate region bounding box is determined as the candidate target region. For the determined candidate target region, the Canny edge detection algorithm is used to perform edge detection processing on it, and the target contour features of the candidate target region are extracted. According to the extracted target contour features, a binary contour image of the candidate target region is generated to obtain a binary representation of the target contour. The geometric center of the contour image is determined, and the geometric center is used as the seed point of the region growing algorithm. With the geometric center as the seed point, the candidate target region is segmented by the region growing algorithm to obtain the segmented connected region. It is determined whether the segmented connected region meets the characteristics of the athlete target. If it meets, the connected region is determined as the precise region of the athlete target; if it does not meet, the contour image of the candidate target region is returned to be re-acquired. The HSV color histogram features and LBP texture features are extracted from the precise region of the athlete target to construct a visual feature vector. According to the pre-established support vector machine classifier model, the visual feature vector is classified and identified to obtain the identity information of the athlete target. According to the classification and identification results, the location information of different athlete targets is obtained, and the location information is associated with the identity information of the athlete target to form the athlete location information. The athlete location information is output for subsequent applications such as athlete tracking and behavior analysis.

[0088] For example, in a video frame of a football game, YOLOv3 can quickly detect the positions of all players, referees, and footballs on the field. The bounding box output by the model usually contains the approximate area of ​​the target, but it may not be accurate enough. To improve the positioning accuracy, a confidence threshold can be set to filter the candidate area. For example, the threshold is set to 0.7, and only the detection results with a confidence level greater than 70% are retained, effectively reducing the false detection rate. The Canny edge detection algorithm is applied to the filtered candidate area to extract the contour features of the athlete. The Canny algorithm can effectively capture the edge of the target and form a binary contour image by calculating the image gradient, non-maximum suppression, and double threshold method. After obtaining the contour image, the region growing algorithm is applied with its geometric center as the seed point for fine segmentation. Region growth starts from the seed point and gradually incorporates similar pixels into the region until it can no longer expand. This method can more accurately outline the outline of the athlete and is particularly suitable for target segmentation in complex backgrounds. For example, on a crowded field, it can accurately separate the outline of a single player, even if there is partial occlusion between the players. Extracting features from the precise region obtained by segmentation is the key to identifying different athletes. The HSV color histogram can effectively characterize the color distribution of the athlete's clothing, while the LBP texture feature can capture the texture information of the clothing. The combination of these two features forms a powerful visual feature vector. For example, in a basketball game, the color feature can distinguish players from different teams, while the texture feature helps to identify different players in the same team. Finally, the feature vector is classified using a support vector machine (SVM) classifier. SVM distinguishes different categories by finding the optimal hyperplane and has good generalization ability. In practical applications, the SVM model can be pre-trained to identify all participants in a specific game. For example, in a tennis doubles match, SVM can accurately distinguish four players even if they are wearing similar clothing. In this way, the precise position information of each athlete in the video frame can be obtained, laying the foundation for subsequent advanced applications such as motion trajectory analysis and tactical research.

[0089] In some embodiments, in the above step S104, the step of using a Kalman filter algorithm to predict the athlete's motion trajectory based on the athlete's position information to obtain the athlete's motion trend information specifically includes:

[0090] According to the player position information of each second image data, a video stream of continuous time frames is formed;

[0091] In the video stream, the predicted value of the state quantity of the previous time frame is used as the prior estimation of the current time frame, and combined with the observed value of the current time frame, the Kalman algorithm is used for prediction analysis to obtain the position coordinates of the players in each time frame;

[0092] The athlete position information of each time frame is spliced ​​together to form the athlete movement trend information.

[0093] In this embodiment, image data of each time frame in the video stream is obtained, and the athlete position information is extracted for each time frame image to obtain the athlete position coordinates of the time frame; the athlete position coordinates of the previous time frame are used as the prior estimate of the current time frame, and the athlete position observation value of the current time frame is obtained at the same time; based on the prior estimate and the observation value, the Kalman filter algorithm is used to predict the athlete position state quantity of the current time frame to obtain the athlete position coordinates of the time frame; it is determined whether there is a next time frame, and if so, the athlete position coordinates of the current time frame are used as the prior estimate of the next frame, and the processing is returned to continue; the athlete position coordinates of all time frames are spliced ​​in chronological order to form a position coordinate sequence reflecting the athlete's motion trajectory; the position coordinate sequence is smoothed to remove the trajectory jitter caused by the prediction error to obtain a smooth athlete motion trajectory; according to the smoothed athlete motion trajectory, the athlete's motion trend information, such as motion direction, speed change, etc., is analyzed and the result is output.

[0094] Exemplarily, first, the player position data at each time point is arranged in chronological order to form a continuous image sequence. For example, in a football game, there may be 25 frames of images per second, each frame containing the position coordinates of 22 players. These coordinate data are converted into visual images and then combined into a video stream in chronological order. The Kalman filter algorithm plays an important role in the prediction of player positions. The algorithm continuously optimizes the prediction results by combining prior estimates and current observations. In practical applications, suppose that a player's position in the 10th frame is (100, 200) and the speed is (5, 3). The Kalman filter predicts the position of the 11th frame based on this information. When the actual observation value of the 11th frame is obtained, the algorithm weighs the predicted value and the observed value to obtain a more accurate estimate. This method is particularly suitable for dealing with noise interference, such as position errors caused by camera shaking or occlusion. The formation of motion trend information is the ultimate goal of the analysis. By splicing the position data of each time frame, the movement trajectory of the athlete can be drawn. For example, in a basketball game, by connecting all the position points of the defender within 30 seconds, his running route can be clearly seen. This trajectory not only shows the position change, but also reflects information such as speed and acceleration. For example, dense trajectory segments indicate that the athlete is slow, while sparse segments mean fast movement. The advantage of this analysis method lies in its comprehensiveness and accuracy. By analyzing continuous frames, instantaneous changes can be captured. For example, in a tennis match, key actions such as sudden stops and turns of players can be accurately located, which is crucial for tactical analysis. In addition, long-term trajectory analysis can reveal the player's range of activities and habitual movement patterns, providing a basis for coaches to formulate tactics. In general, this athlete position analysis method based on continuous video streams and Kalman filtering provides in-depth technical support for sports events. It can not only track the athlete's position in real time, but also predict future movements, providing strong data support for tactical analysis, training optimization and game strategy formulation.

[0095] In some embodiments, in the above step S105, the marking of the illegal behavior in the third image data based on the preset sports event rule library to form the fourth image data specifically includes:

[0096] Acquire sports image features of the third image data, match the sports image features with a preset sports event rule library, and determine the event type and event rule data corresponding to the third image data;

[0097] Based on the competition rule data, analyzing the third image data to determine whether any athlete in the third image data has violated the rules;

[0098] If an athlete has violated the rules, the display content indicating the violation will be marked to form fourth image data.

[0099] In this embodiment, the third image data is obtained, and sports image features therein are extracted, including venue identification, athlete position, score and other information. The extracted sports image features are matched with the preset sports project rule library, and the specific event type corresponding to the third image data is determined by feature similarity calculation. According to the determined event type, the corresponding event rule data is obtained, including foul action definition, violation penalty standards and other contents. The target detection algorithm is used to detect athletes on the third image data to obtain the position coordinates and action posture of each athlete. The detected athlete action posture is compared with the foul action definition in the event rule data. If the similarity between the athlete action and the foul action exceeds the preset threshold, it is determined that the athlete has violated the rules. For athletes determined to have violated the rules, a violation mark box is added to the corresponding display position of the third image data, and a specific text description is determined according to the violation penalty standards. The image data with the violation mark added is output as the fourth image data for subsequent event live broadcast or playback analysis. Through the machine learning model, violation cases are continuously accumulated, the rule library and penalty standards are optimized, and the recognition accuracy of violations is improved.

[0100] For example, taking a football match as an example, the system can identify features such as the lines of the field, the positions of the players, and the trajectory of the ball. Through deep learning algorithms, such as convolutional neural networks, these features can be extracted from the image. For example, for an image of a football field, the system can identify the white lines of the field, the green grass, and the color and position distribution of the players. Matching the extracted features with the preset rule library is an important part of determining the type and rules of the event. The rule library contains feature descriptions and rule sets of various sports. For example, if a rectangular field, two goals and 22 players are detected, the system will match it as a football match. At the same time, the rule library also contains definitions of specific rules such as offside and handball. Based on the matched rule data, the system performs violation analysis. In a football match, the system may pay attention to offside, fouls, etc. For example, by tracking the position of the players, the system can determine whether the offensive player is in an offside position. If at the moment of passing the ball, the offensive player is in front of the second-to-last defensive player and closer to the opponent's goal line, the system will mark this potential offside situation. For detected violations, the system will mark them on the image. This includes drawing eye-catching boxes around illegal players, adding text descriptions, or using arrows to indicate illegal actions. For example, in the case of offside, the system may draw a red box around the offside player and mark it with the words "suspected offside". This automated analysis and annotation system has multiple meanings in sports events. First, it can assist referees in making more accurate judgments and reduce human errors. Secondly, it provides richer visual information for live broadcasts and replays of events, enhancing the audience experience. Furthermore, coaches and players can use this data for tactical analysis and technical improvements. In general, this sports event rule determination system based on image analysis has brought technological innovation to modern sports events by combining computer vision and sports expertise. It not only improves the fairness and viewing experience of the game, but also opens up new avenues for sports data analysis.

[0101] In some embodiments, in the above step S107, determining the corresponding AR information display template from a preset AR information display template library according to the user preference information, placing the fourth image data into the AR information display template for fusion processing, and generating AR display image data specifically includes:

[0102] Obtaining personalized preference information of the user, preprocessing the personalized preference information to form a user preference feature vector, wherein the user preference feature vector includes user age, user gender, hobbies and historical browsing content style;

[0103] Matching the user preference feature vector with a preset AR information display template library to determine the AR information display template with the highest matching degree with the user preference;

[0104] Filling the fourth image data into the AR information display template to form initial AR display image data;

[0105] A deep convolutional generative adversarial network algorithm is used to allocate attention weights to different display areas on the initial AR display image data, determine the display effect degree of different display areas on the initial AR display image data, and form target AR display image data.

[0106] In this embodiment, the personalized preference information of the user in the application is obtained, including basic attributes such as age and gender filled in by the user, as well as behavioral data such as interest tags and historical browsing content styles. The obtained personalized preference information of the user is preprocessed, noise data is removed, key features are extracted, and a user preference feature vector is constructed. According to business needs, a variety of AR information display templates are preset, and each template has different visual styles and interaction methods for different groups of people. The similarity between the user preference feature vector and the features of each AR display template is calculated to obtain a matching score. The matching scores are sorted to determine the AR display template with the highest matching degree as the personalized AR information display solution for the user. In the AR scene, the AR display template with the highest matching degree is called, and the corresponding AR content and interactive components are dynamically generated to realize the personalized AR information display of the user's preference. The user's interactive behavior and content browsing data in the AR scene are continuously tracked, and the user preference feature vector is updated to realize real-time dynamic optimization of AR information display.

[0107] Fill the fourth image data into the AR information display template to generate initial AR display image data. Use a deep convolutional neural network to extract features from the initial AR display image data to obtain image feature data. Input the image feature data into the generative adversarial network, and optimize the display effect of the initial AR display image data through adversarial learning between the generator and the discriminator. During the training process of the generative adversarial network, an attention mechanism is introduced to allocate attention weights to different display areas of the initial AR display image data. According to the attention weight allocation result, the degree of display effect of different display areas on the initial AR display image data is determined to obtain the optimized AR display image data. The optimized AR display image data is used as the target AR display image data for subsequent AR information display.

[0108] Exemplarily, the system can collect user data through user registration information, questionnaires or behavior analysis. For example, a 30-year-old male user may fill in his favorite sports and technology news when registering, and the system may also record that he often browses outdoor sports equipment and the latest smartphone-related content. The preprocessing stage converts the raw data into a structured user preference feature vector. This vector may include dimensions such as age (30), gender (male), hobbies (sports, technology), and historical browsing content style (outdoor equipment, smartphone). In this way, the system can quantify complex user information into a computable data form. The AR information display template library is a series of templates pre-designed by the system for different types of users and content. For example, for young male users who like sports, there may be a dynamic template with a sports scene as the background and an information layout that highlights important data. The system selects the best matching template by calculating the similarity between the user preference feature vector and each template. In this example, the system may select a template that combines sports elements and a sense of technology. The fourth image data (i.e., the AR information to be displayed) is filled into the selected template to form an AR display image. Through this series of steps, the system finally generates an AR display image that not only meets the user's personal preferences but also highlights key information.

[0109] Reference Figure 2 An embodiment of the present invention provides a data processing system 2 for an AR smart camera device, and the system 2 specifically includes:

[0110] The first data processing module 201 is used to obtain the original image data captured by the AR smart camera device in real time, and process the original image data using an adaptive illumination compensation algorithm to obtain first image data;

[0111] A second data processing module 202 is used to use a pre-built image denoising model to perform denoising on the first image data to obtain second image data;

[0112] A third data processing module 203 is used to extract athlete features from the second image data using a target detection algorithm based on a convolutional neural network to obtain athlete position information;

[0113] The fourth data processing module 204 is used to predict the athlete's motion trajectory using a Kalman filter algorithm according to the athlete's position information to obtain the athlete's motion trend information;

[0114] A fifth data processing module 205 is used to determine whether any athlete in the second image data is blocked based on the athlete movement trend information, and if blocked, to complete the block according to the athlete's historical trajectory to form third image data;

[0115] A sixth data processing module 206, configured to mark the illegal behaviors in the third image data based on a preset sports rule library to form fourth image data;

[0116] The seventh data processing module 207 is used to determine the corresponding AR information display template from the preset AR information display template library according to the user preference information, put the fourth image data into the AR information display template for fusion processing, and generate AR display image data.

[0117] It is understandable that if Figure 1 The contents of the data processing method embodiment for an AR smart camera device shown in the figure are all applicable to the data processing system embodiment for an AR smart camera device. The functions specifically implemented by the data processing system embodiment for an AR smart camera device are similar to those of the embodiment shown in the figure. Figure 1 The data processing method embodiment for the AR smart camera device shown in the figure is the same as that in the embodiment of the present invention, and the beneficial effects achieved are the same as those in the embodiment of the present invention. Figure 1 The beneficial effects achieved by the data processing method embodiment for the AR smart camera device shown are also the same.

[0118] It should be noted that the information interaction, execution process and other contents between the above-mentioned systems are based on the same concept as the embodiment of the method of the present invention. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0119] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0120] Reference Figure 3The embodiment of the present invention further provides a computer device 3, comprising: a memory 302 and a processor 301 and a computer program 303 stored in the memory 302. When the computer program 303 is executed on the processor 301, a data processing method for an AR smart camera device as described in any one of the above methods is implemented.

[0121] The computer device 3 may be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server. The computer device 3 may include, but is not limited to, a processor 301 and a memory 302. Those skilled in the art will appreciate that Figure 3 It is only an example of computer device 3 and does not constitute a limitation on computer device 3. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components, for example, it may also include input and output devices, network access devices, etc.

[0122] The processor 301 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0123] In some embodiments, the memory 302 may be an internal storage unit of the computer device 3, such as a hard disk or memory of the computer device 3. In other embodiments, the memory 302 may also be an external storage device of the computer device 3, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the computer device 3. Further, the memory 302 may also include both an internal storage unit and an external storage device of the computer device 3. The memory 302 is used to store an operating system, an application program, a boot loader, data, and other programs, such as the program code of the computer program. The memory 302 may also be used to temporarily store data that has been output or is to be output.

[0124] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the data processing method for an AR smart camera device as described in any one of the above methods is implemented.

[0125] In this embodiment, if the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium can at least include: any entity or device that can carry the computer program code to the camera / terminal device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. For example, USB flash drive, mobile hard disk, disk or optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.

[0126] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0127] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0128] In the embodiments disclosed in the present application, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are only schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0129] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

Claims

1. A data processing method for an AR smart camera device, characterized in that: The method specifically comprises: Acquire raw image data captured in real time by the AR smart camera device, and process the raw image data using an adaptive illumination compensation algorithm to obtain first image data; Using a pre-built image denoising model, denoising the first image data to obtain second image data; Using a convolutional neural network-based target detection algorithm to extract athlete features from the second image data to obtain athlete position information; According to the athlete's position information, a Kalman filter algorithm is used to predict the athlete's movement trajectory to obtain the athlete's movement trend information; Based on the sports trend information of the athletes, determining whether any athlete in the second image data is blocked, and if blocked, completing the third image data according to the historical trajectory of the athlete; Based on a preset sports rules library, marking the illegal behaviors in the third image data to form fourth image data; Determine a corresponding AR information display template from a preset AR information display template library according to the user preference information, put the fourth image data into the AR information display template for fusion processing, and generate AR display image data; The step of extracting athlete features from the second image data using a target detection algorithm based on a convolutional neural network to obtain athlete position information specifically includes: Using a pre-trained YOLOv3 convolutional neural network model, perform model inference on the second image data to obtain a candidate region bounding box of the athlete target in the second image data and a corresponding confidence level; The candidate region bounding box with a confidence greater than a preset confidence threshold is taken as a candidate target region, and the target contour features of the candidate target region are extracted using the Canny edge detection algorithm to obtain a binary contour image; Using the geometric center of the contour image as a seed point, segmenting the candidate target region using a region growing algorithm to obtain segmented connected regions, and using the connected regions as the precise region of the athlete target; Extracting HSV color histogram features and LBP texture features from the precise area to construct a visual feature vector; A support vector machine classifier is used to classify and identify the visual feature vector, obtain position information of different athlete targets, and form athlete position information.

2. The method according to claim 1, characterized in that The adopting the adaptive illumination compensation algorithm to process the original image data to obtain the first image data specifically includes: Using an image segmentation algorithm based on region growing, the original image data is segmented into multiple regions to form multiple region images; Extract the illumination features of each region image respectively to form an illumination feature vector; Collecting first image samples under different lighting conditions, and marking corresponding lighting compensation parameters according to various lighting features of the first image samples to form a lighting compensation training data set; Taking the illumination compensation training data set as input, using a machine learning algorithm for modeling training to establish an illumination compensation model; Inputting the illumination characteristic vector of each regional image into the illumination compensation model to obtain illumination compensation parameters of each regional image; According to the illumination compensation parameters, illumination compensation processing is performed on each regional image respectively, and all regional images after illumination compensation processing are spliced ​​to form first image data.

3. The method according to claim 1, characterized in that The method further comprises: performing denoising processing on the first image data by using a pre-built image denoising model to obtain second image data; Collecting second image samples containing different noise interference data, marking the noise type and noise distribution of the second image samples, and forming an image denoising training data set; The image denoising training data set is used as input, and a convolutional neural network model is used for modeling training to form an image denoising model.

4. The method according to claim 1, characterized in that The method of using a Kalman filter algorithm to predict the athlete's motion trajectory based on the athlete's position information to obtain the athlete's motion trend information specifically includes: According to the player position information of each second image data, a video stream of continuous time frames is formed; In the video stream, the predicted value of the state quantity of the previous time frame is used as the prior estimation of the current time frame, and combined with the observed value of the current time frame, the Kalman algorithm is used for prediction analysis to obtain the position coordinates of the players in each time frame; The athlete position information of each time frame is spliced ​​together to form the athlete movement trend information.

5. The method according to claim 1, characterized in that The marking of the illegal behavior in the third image data based on the preset sports rules library to form the fourth image data specifically includes: Acquire sports image features of the third image data, match the sports image features with a preset sports event rule library, and determine the event type and event rule data corresponding to the third image data; Based on the competition rule data, analyzing the third image data to determine whether any athlete in the third image data has violated the rules; If an athlete has violated the rules, the display content indicating the violation will be marked to form fourth image data.

6. The method according to claim 1, characterized in that The determining a corresponding AR information display template from a preset AR information display template library according to the user preference information, placing the fourth image data into the AR information display template for fusion processing, and generating AR display image data specifically includes: Obtaining personalized preference information of the user, preprocessing the personalized preference information to form a user preference feature vector, wherein the user preference feature vector includes user age, user gender, hobbies and historical browsing content style; Matching the user preference feature vector with a preset AR information display template library to determine the AR information display template with the highest matching degree with the user preference; Filling the fourth image data into the AR information display template to form initial AR display image data; A deep convolutional generative adversarial network algorithm is used to allocate attention weights to different display areas on the initial AR display image data, determine the display effect degree of different display areas on the initial AR display image data, and form target AR display image data.

7. A data processing system for an AR smart camera device, characterized in that: The system specifically comprises: A first data processing module is used to obtain raw image data captured by the AR smart camera device in real time, and process the raw image data using an adaptive illumination compensation algorithm to obtain first image data; A second data processing module, configured to use a pre-built image denoising model to perform denoising on the first image data to obtain second image data; A third data processing module is used to extract athlete features from the second image data using a target detection algorithm based on a convolutional neural network to obtain athlete position information; A fourth data processing module is used to predict the athlete's movement trajectory using a Kalman filter algorithm based on the athlete's position information to obtain the athlete's movement trend information; a fifth data processing module, for determining whether any athlete in the second image data is blocked based on the athlete movement trend information, and if blocked, completing the block according to the athlete's historical trajectory to form third image data; a sixth data processing module, configured to mark the illegal behaviors in the third image data based on a preset sports rule library to form fourth image data; a seventh data processing module, configured to determine a corresponding AR information display template from a preset AR information display template library according to the user preference information, put the fourth image data into the AR information display template for fusion processing, and generate AR display image data; The step of extracting athlete features from the second image data using a target detection algorithm based on a convolutional neural network to obtain athlete position information specifically includes: Using a pre-trained YOLOv3 convolutional neural network model, perform model inference on the second image data to obtain a candidate region bounding box of the athlete target in the second image data and a corresponding confidence level; The candidate region bounding box with a confidence greater than a preset confidence threshold is taken as a candidate target region, and the target contour features of the candidate target region are extracted using the Canny edge detection algorithm to obtain a binary contour image; Using the geometric center of the contour image as a seed point, segmenting the candidate target region using a region growing algorithm to obtain segmented connected regions, and using the connected regions as the precise region of the athlete target; Extracting HSV color histogram features and LBP texture features from the precise area to construct a visual feature vector; A support vector machine classifier is used to classify and identify the visual feature vector, obtain position information of different athlete targets, and form athlete position information.

8. A computer device, characterized in that: include: A memory and a processor and a computer program stored in the memory, when the computer program is executed on the processor, implements the data processing method for an AR smart camera device as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the data processing method for an AR smart camera device as described in any one of claims 1 to 6 is implemented.

Images