Anti-shooting method and device for display screen

Abstract

The invention relates to a display screen anti-shooting method and device, equipment and a medium. The method comprises the following steps: acquiring an environment image in front of a display screen; inputting the environment image into a YOLO V11 model subjected to migration and incremental learning training to obtain a detection result corresponding to the environment image; the detection result comprises at least one of whether a human face and a human body are detected in the environment image or not and whether shooting equipment is detected or not and a retention time length; determining a display strategy corresponding to the display screen according to the detection result; the display strategy is used for indicating the display screen to display corresponding content. By adopting the method, the detection precision can be improved, and real-time and accurate anti-shooting control is realized.

Description

Technical Field

[0001] The present invention belongs to the field of computer vision, and in particular relates to a method and a device for preventing a display screen from being photographed. Background Art

[0002] With the development of information security and privacy protection needs, preventing screen content from being illegally photographed has become an important research topic. Traditional anti-shooting methods usually rely on physical means, such as anti-peeping films, but there are problems such as limited viewing angles and poor visual effects.

[0003] Existing anti-shooting technologies mainly rely on image acquisition devices to collect video streams in real time and identify and detect the photography behavior. Existing detection of handheld stealing devices is mainly based on target detection technology, that is, identifying faces and mobile phones and other shooting devices in the image. However, it is affected by complex environments, as well as interference from factors such as illumination, occlusion, and the coexistence of similar objects during imaging. In particular, when detecting the shooting behavior of shooting devices, commonly used detection models (such as YOLOv5, YOLOv8, etc.) have limited performance in complex scenes. Summary of the invention

[0004] Based on this, it is necessary to provide a display screen anti-shooting method and device that can improve detection accuracy in response to the above technical problems.

[0005] In a first aspect, the present application provides a method for preventing a display screen from being photographed, comprising:

[0006] Acquire the environment image in front of the display screen;

[0007] Input the environment image into the trained improved YOLO V11 model to obtain a detection result corresponding to the environment image; the detection result includes at least one of whether a face is detected in the environment image, whether a human body is detected, and whether a shooting device is detected;

[0008] A display strategy corresponding to the display screen is determined according to the detection result; the display strategy is used to instruct the display screen to display corresponding content.

[0009] In one embodiment, determining a display strategy corresponding to the display screen according to the detection result includes:

[0010] When the detection result is that there is a shooting device in the environment image, the display strategy is to display a blocking interface on the display screen; the blocking interface is used to block the normal interface of the display screen, and / or the blocking interface is used to display a warning prompt;

[0011] When the detection result is that there are multiple human bodies or faces in the environment image, the display strategy is to display an occlusion interface on the display screen; the occlusion interface is used to block the normal interface of the display screen, and / or the occlusion interface is used to display a warning prompt;

[0012] When the detection result is that there is a face or a human body in the environment image and there is no shooting device, the display strategy is to display a normal interface on the display screen;

[0013] When the detection result is that there is no face in the environment image, the display strategy is to turn off the display screen.

[0014] In one of the embodiments, the detection result also includes whether the face in the environment image is the face corresponding to the preset user;

[0015] When the detection result is that the face in the environment image is not the face corresponding to the preset user, the display strategy is to display a warning prompt on the display screen and send a corresponding prompt message to the preset terminal.

[0016] In one embodiment, the improved YOLO V11 model is obtained by the following method:

[0017] Acquire multiple training images, where the training images include scenes in front of the display screen under different lighting conditions, different backgrounds, and different shooting angles;

[0018] Performing labeling processing on each training image to obtain a labeling file corresponding to each training image; the labeling file is labeled with at least one of a shooting device and a face in the training image;

[0019] The training images and the corresponding annotation files are used as training data sets. Based on the multi-task learning framework, the YOLO V11 model is trained using the training data sets to obtain a trained improved YOLO V11 model.

[0020] In one embodiment, the annotation file is compatible with the COCO format.

[0021] In one embodiment, acquiring an environment image in front of a display screen includes:

[0022] Collecting a real-time video stream in front of a display screen, the real-time video stream includes image frames arranged continuously in time sequence;

[0023] Based on GPU acceleration, the image frames in the real-time video stream are preprocessed in real time to obtain the environment image.

[0024] In one embodiment, preprocessing image frames in a real-time video stream in real-time based on GPU acceleration to obtain an environment image includes:

[0025] Use GPU as the default computing device to obtain real-time video stream and read image frames in the real-time video stream frame by frame;

[0026] Convert each image frame into an initial image in a preset format, where the preset format is the input format of the improved YOLO V11 model;

[0027] For each initial image, the image processing function provided by the deep learning framework is used to normalize the size of the initial image to obtain the environment image.

[0028] In a second aspect, the present application also provides a display screen anti-shooting device, comprising:

[0029] An acquisition module, used to acquire an environment image in front of the display screen;

[0030] A detection module is used to input the environment image into the trained improved YOLO V11 model to obtain a detection result corresponding to the environment image; the detection result includes at least one of whether a face is detected in the environment image, whether a human body is detected, and whether a shooting device is detected;

[0031] The display module is used to determine a display strategy corresponding to the display screen according to the detection result; the display strategy is used to instruct the display screen to display corresponding content.

[0032] In a third aspect, the present application further provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any of the aforementioned methods when executing the computer program.

[0033] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of any of the aforementioned methods when executed by a processor.

[0034] The above-mentioned display screen anti-shooting method, device, computer equipment and storage medium obtain the environmental image in front of the display screen; input the environmental image into the trained improved YOLO V11 model to obtain the detection result corresponding to the environmental image; the detection result includes whether a face is detected in the environmental image and whether at least one of the shooting equipment is detected; determine the display strategy corresponding to the display screen according to the detection result; the display strategy is used to instruct the display screen to display the corresponding content, which can improve the detection accuracy of the environmental image in front of the display screen, thereby realizing precise control of the display interface of the display screen, and realizing real-time and precise anti-shooting control. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are 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 1A schematic diagram of a flow chart of a method for preventing a display screen from being photographed in one embodiment;

[0037] Figure 2 is a schematic diagram of an occlusion interface when a detection result is that a mobile phone exists in an environment image in an embodiment;

[0038] Figure 3 FIG. 1 is a structural block diagram of a device for preventing a display screen from being photographed in one embodiment. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0040] In real life, there is a need to protect the display screen from being filmed in many application scenarios, such as:

[0041] Financial institutions: Prevent sensitive customer information from being photographed on ATM and business hall screens. Company meetings: Protect screen content during meetings to avoid leaking confidential information. Medical facilities: Prevent unauthorized filming when using screens to display patient information in hospitals. Cinemas: During movie viewing, if a camera is detected pointing at the screen for a long time, it may be considered an illegal act of recording a movie. In order to protect the copyright of movies and the interests of producers, cinemas need to prevent viewers from using cameras or other devices to secretly photograph the content of the movie.

[0042] The screen anti-shooting method in traditional technology has the following main defects:

[0043] 1. Limitations of physical anti-peeping technology

[0044] Traditional privacy film:

[0045] Limited protection range: The privacy film can only block peeping from both sides of the screen, but cannot effectively prevent frontal shooting, and it is difficult to cope with the various angles of modern smartphones and camera equipment.

[0046] Visual experience is affected: The privacy film will reduce screen brightness and clarity, affecting the visual experience of normal users, especially in scenes that require high-resolution display.

[0047] Easy to damage: The privacy film is easily damaged or worn and needs to be replaced regularly, which increases the user's usage and maintenance costs.

[0048] 2. Limitations of screen content blur protection

[0049] False triggering and recognition errors: Face detection accuracy is limited, and it may misjudge normal user operations as peeping behavior and trigger a blur effect, affecting normal user use.

[0050] Insufficient real-time performance: The fuzzification process usually relies on software processing. If the processing speed is not fast enough, fast and real-time protection cannot be achieved. Especially under high-resolution or complex display content, processing delays affect the user experience.

[0051] Dependence on ambient light and angle: In dark light or at an angle, the accuracy of face recognition drops significantly, resulting in misjudgment or failure to trigger the protection effect, affecting the actual anti-shooting effect.

[0052] 3. Limitations of Anti-shooting Methods Based on Simple Object Detection

[0053] Limited applicability: These methods can only identify specific types of devices (such as smartphones with obvious lenses) and cannot effectively deal with new types of devices (such as hidden cameras).

[0054] High false detection rate: In complex backgrounds, features such as reflections can easily be misidentified as camera devices, resulting in false alarms and affecting normal user use.

[0055] Detection effect depends on hardware: This type of method usually relies on a high-resolution camera to capture reflective features, resulting in high system costs and is not suitable for widespread application on ordinary screens.

[0056] 4. Performance bottleneck of real-time detection

[0057] High computing resource requirements: Many existing protection technologies rely on traditional image detection algorithms, which have poor real-time processing and require high computing resources. Traditional anti-shooting technologies fail to fully utilize GPU parallel computing resources to improve detection speed and real-time performance, resulting in the following defects:

[0058] Large delay: When processing a large number of video frames, it is difficult for existing solutions to ensure real-time detection, and delays often occur, affecting detection accuracy.

[0059] Not suitable for complex scenarios: In multi-person environments or complex backgrounds, the detection speed is further reduced, which cannot meet the needs of high-frequency and low-latency detection, and is prone to detection misses.

[0060] Strong hardware dependence: Traditional detection technology requires high CPU and memory resources. In some scenarios, it is difficult to run on ordinary hardware devices and it is difficult to meet general application needs.

[0061] 5. Lack of multi-scenario adaptability due to intelligent judgment

[0062] Inability to intelligently judge multiple situations: Many protection systems can only work effectively in specific situations and lack intelligent judgment mechanisms under multiple conditions. Most existing technologies only perform single-dimensional detection of peepers or filming behaviors, and lack comprehensive judgment of faces and filming equipment. This brings the following problems:

[0063] Unable to effectively distinguish between normal and abnormal behaviors: Existing technologies make it difficult to intelligently determine whether there is a user operating normally in front of the screen, or whether the device is in an unmanned scene, resulting in false triggering of protection, or the screen continues to display when no one is around, increasing the risk of leakage.

[0064] Lack of algorithms that can adapt to complex scenarios: In complex situations such as users frequently switching scenes and multi-person environments, the existing system finds it difficult to intelligently judge the real protection needs, which often leads to false alarms or missed alarms, limiting the flexibility and scope of application of protection.

[0065] 6. Anti-shooting warning feedback is not timely

[0066] Feedback delay of existing methods: Many systems have large delays in alarm or screen protection mechanisms, making it difficult for users to respond in time, which may result in information being captured. This situation is particularly disadvantageous in high-risk environments (such as finance, medical care, etc.), where there is a lack of timely protection response.

[0067] Lack of clear prompts to users: When existing technologies detect potential voyeuristic photography, they often only use a single method (such as blurring or turning off the screen) for protection. There is a lack of clear prompts to users, which can easily lead to misunderstanding or neglect.

[0068] Based on this, an embodiment of the present application provides a method for preventing a display screen from being photographed, so as to at least partially solve the above-mentioned problem.

[0069] In one embodiment, Figure 1 As shown, a method for preventing a display screen from being photographed is provided. This embodiment takes the method applied to a terminal as an example for illustration. It can be understood that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server.

[0070] In this embodiment, the method includes the following steps:

[0071] Step 102: Acquire the environment image in front of the display screen.

[0072] Specifically, the terminal can obtain the environmental image in front of the display screen through the image acquisition device. Among them, the display screen can be a physical display screen such as an LED screen installed on various terminal devices, or a virtual screen projected by a projection device. For example, terminals such as laptops and ATM machines can obtain environmental images through their own cameras; display screens such as projection screens used in scenes such as conference rooms can obtain environmental images through cameras set above the display screens, or multiple cameras can be set in various directions of the conference room to obtain environmental images from multiple angles.

[0073] Step 104, input the environment image into the trained improved YOLO V11 model to obtain a detection result corresponding to the environment image; the detection result includes at least one of whether a face is detected in the environment image, whether a human body is detected, and whether a shooting device is detected.

[0074] Among them, the YOLO V11 (“You Once Look Once Version 11”) model is a target detection model, which mainly includes the following parts:

[0075] 1. C3K2 module: It is an enhancement of the CSP (Cross Stage Partial) block in the previous version. This module uses different kernel sizes (such as 3x3 or 5x5) and channel separation strategies to optimize the extraction of more complex features. Its outer layer is dominated by the C2F structure, and the inner layer is a C3 structure. The C3K2 structure is based on the C2F structure, and the feature extraction layer is replaced by the Bott l eneck layer with the C3K layer. The C3K layer is consistent with the C3 structure, but the Bott l eneck convolution kernel size is adjusted to transform it into two convolution layers with a kernel size of 3.

[0076] SPFF module: It is an optimized version of the SPP (Spatial Pyramid Poo ling) module. This module allows the model to perform better by capturing the properties of objects at different scales, improves the neck structure, and helps to better detect small objects.

[0077] C2PSA block: It provides more effective feature extraction by combining channel and spatial information, and works with the multi-head attention mechanism to achieve more accurate perception of objects. It is able to optimize the feature maps of the previous layer and enrich them with the attention mechanism to improve the performance of the model, making more accurate detection possible in complex scenes.

[0078] In actual use, the original YOLO V11 model can be used for multi-target training to obtain an improved YOLO V11 model that can simultaneously identify faces, human bodies, and shooting devices in the output image; the structure of the YOLO V11 model can also be adjusted before training, such as introducing an attention mechanism, improving the backbone network, and adding a detection head, etc., to adjust the structure of the YOLO V11 model to adapt to the actual detection needs, and then further train to obtain an improved YOLO V11 model. Exemplarily, on the basis of the original YOLO V11 model, a lightweight pre-trained model yo l o11 n.pt can be used for fine-tuning through the method of transfer learning. This not only inherits the high efficiency of YOLO V11, but also makes the model more suitable for special target detection in anti-shooting scenarios.

[0079] For example, the improved YOLO V11 model in actual application can not only detect faces, human bodies and shooting devices in environmental images at the same time, but also further identify who the user corresponding to the face is, and detect other devices that can take pictures, such as cameras and phone watches. It can also detect objects such as water cups and bottles and give prompts.

[0080] Step 106, determining a display strategy corresponding to the display screen according to the detection result; the display strategy is used to instruct the display screen to display corresponding content.

[0081] Specifically, the terminal can determine the display strategy corresponding to the display screen according to the detection result. Among them, the display strategy is used to instruct the display screen to display the corresponding content. Exemplarily, when the detection result indicates that there is a shooting device in the environmental image, that is, there may be user shooting behavior, the corresponding display strategy can be determined to be a black screen to prevent the content on the screen from being shot.

[0082] An embodiment of the present application provides a method for preventing a display screen from being photographed, which comprises acquiring an environment image in front of a display screen; inputting the environment image into a trained improved YOLO V11 model to obtain a detection result corresponding to the environment image; the detection result includes whether a face is detected in the environment image and whether at least one of a photographing device is detected; determining a display strategy corresponding to the display screen according to the detection result; the display strategy is used to instruct the display screen to display corresponding content, which can improve the detection accuracy of the environment image in front of the display screen, thereby achieving precise control of the display interface of the display screen, and achieving real-time and precise anti-shooting control.

[0083] In one embodiment, determining a display strategy corresponding to the display screen according to the detection result may include the following steps:

[0084] Step 202, when the detection result is that there is a shooting device in the environmental image, the display strategy is to display a blocking interface on the display screen; the blocking interface is used to block the normal interface of the display screen, and / or the blocking interface is used to display a warning prompt.

[0085] Specifically, in order to prevent the camera of a device such as a shooting device from aiming at the screen to shoot, the detection result may be that there is a shooting device in the environmental image, and the display strategy is to display a blocking interface on the display screen. The blocking interface is used to block the normal interface of the display screen, and / or the blocking interface is used to display a warning prompt. For example, a black interface can be used to block the normal interface of the display screen to prevent it from being photographed by the user.

[0086] In addition, only a single method (such as blurring or turning off the screen) is used for protection, lacking clear prompts to users, which is easy to cause misunderstanding or neglect. Therefore, the shielding interface of the present application can also be used to display warning prompts.

[0087] like Figure 2 As shown, in an exemplary application scenario, when the presence of a shooting device is detected in the environmental image, relevant warning prompts such as "No shooting" or "Please move your phone" can be displayed in the occlusion interface to clearly inform the user that the current interface change is due to the presence of a shooting device or other device detected in front of the screen, so as to prevent the user from mistakenly thinking that the terminal is damaged.

[0088] Step 204, when the detection result is that there are multiple human bodies or faces in the environmental image, the display strategy is to display an occlusion interface on the display screen; the occlusion interface is used to block the normal interface of the display screen, and / or the occlusion interface is used to display a warning prompt.

[0089] Specifically, when the detection result shows that there are multiple human bodies or faces in the environment image, it means that in addition to the preset user, there may be multiple users viewing the content on the screen together. However, in many scenarios, for example, the user's balance and other private information are displayed on the bank ATM machine, and the user's physical examination report or medical history and other private information are displayed on the hospital terminal.

[0090] In such cases, when multiple human bodies or faces are detected in the environmental image, the display strategy can be determined to display an occlusion interface on the display screen; the occlusion interface is used to block the normal interface of the display screen, and / or the occlusion interface is used to display a warning prompt.

[0091] Exemplarily, the warning prompt may be a text prompt such as "You do not have viewing authority for the current terminal / content, please move away from the screen", and the warning prompt may be used to cover the normal screen to remind the current user to move away and protect the content displayed on the screen from being leaked.

[0092] Step 206, when the detection result is that there is a face in the environment image and there is no shooting device, the display strategy is to display a normal interface on the display screen.

[0093] Specifically, when the terminal detects that there is a face in the environmental image and no shooting device, it means that the user is using the terminal normally. Therefore, the display strategy can be determined to display a normal interface on the display screen, and the screen displays content normally.

[0094] Step 208: When the detection result is that there is no human face in the environment image, the display strategy is to turn off the display screen.

[0095] Specifically, when the detection result is that there is no human face in the environmental image, which means that no one is currently using the display screen, the display policy can be determined to turn off the display screen, thereby reducing the risk of leakage.

[0096] In one embodiment, the detection result also includes whether the face in the environment image is the face corresponding to the preset user. Determining the display strategy corresponding to the display screen according to the detection result may also include the following steps:

[0097] Step 210, when the detection result is that the face in the environment image is not the face corresponding to the preset user, the display strategy is to display a warning prompt on the display screen and send a corresponding prompt message to the preset terminal.

[0098] Specifically, if the aforementioned improved YOLO V11 model is trained to recognize whether the face is a normal user or a user with permission to use, then when the detection result is that the face in the environmental image is not the face corresponding to the preset user, the display strategy is to display a warning prompt on the display screen and send the corresponding prompt information to the preset terminal. Exemplarily, the warning prompt can be a text prompt such as "You do not have the viewing permission of the current terminal / content, please stay away from the screen", and the normal screen is blocked with a warning prompt to remind the current user to stay away and protect the content displayed on the screen from being leaked. Optionally, when a face corresponding to a non-preset user is detected, it means that there may be a situation of deliberately peeping at confidential content, so a corresponding prompt information can also be sent to the preset terminal to prompt the corresponding user to come to the display screen to check the situation. The facial image corresponding to the currently detected non-preset user can also be sent to the preset terminal to facilitate the subsequent investigation of relevant infringing personnel.

[0099] Furthermore, when the same camera is detected in multiple consecutive environmental images, that is, when the camera is detected to be pointed at the display screen for a long time in the continuous video stream, it indicates that there may be illegal recording. In this case, the corresponding prompt information is also sent to the preset terminal to prompt the corresponding user to come to the vicinity of the display screen to check the situation. If a facial image is detected at the same time, the facial image corresponding to the currently detected non-preset user can also be sent to the preset terminal to facilitate the subsequent investigation of relevant infringing persons.

[0100] A display screen protection method for preventing shooting is provided in an embodiment of the present application. By setting the judgment logic and screen display strategy in combination with the actual application scenario, it can flexibly respond to different scenarios. When the presence of a shooting device is detected, a blocking interface is displayed, when it is detected that no one is using the screen, the screen is turned off, or when a non-preset face is detected, a blocking interface is displayed and a warning prompt is issued. This ensures that the screen content is not shot and is not spied on by irrelevant personnel, thereby further enhancing information security.

[0101] In one embodiment, the improved YOLO V11 model is obtained by the following method:

[0102] Step 302: Acquire multiple training images, where the training images include scenes in front of the display screen with different lighting conditions, different backgrounds, and different shooting angles.

[0103] Specifically, for specific scenarios of preventing screen shooting, training images covering a variety of lighting conditions, background complexity, and shooting angles can be collected and constructed to form a training data set. Optionally, the training images can pay special attention to the diverse features of faces, bodies, and shooting devices.

[0104] Optionally, the training images can be expanded to include detection of various categories such as mobile phones, cameras, and handheld devices, with approximately 2,000 additional training images per category. This greatly enriches the model's detection capabilities, enabling it to accurately identify a variety of possible shooting devices in complex environments.

[0105] Step 304 , annotate each training image to obtain an annotation file corresponding to each training image; the annotation file is annotated with at least one of the shooting device and the face in the training image.

[0106] To ensure detection accuracy, the targets in the data set (such as shooting equipment, faces, etc.) are accurately labeled.

[0107] In one embodiment, the annotation format is compatible with the COCO format and meets the input requirements of YOLO V11.

[0108] Step 306 , using the training images and the corresponding annotation files as training data sets, based on a multi-task learning framework, using the training data sets to train the YOLO V11 model, and obtaining a trained improved YOLO V11 model.

[0109] Specifically, by optimizing the model using a multi-task learning framework, the YOLO V11 model can simultaneously detect faces, bodies, and camera equipment, thereby improving detection efficiency. For example, during the training process, online data enhancement technology is used to perform enhancement operations such as rotation, scaling, and brightness adjustment on the image to improve the generalization ability of the model and adapt it to the anti-shooting needs in different environments.

[0110] A display screen protection method provided in an embodiment of the present application overcomes the problem of low detection accuracy of traditional models in complex environments by adding training data and optimizing models for shooting devices.

[0111] In one embodiment, obtaining the environment image in front of the display screen may include the following steps:

[0112] Collecting a real-time video stream in front of a display screen, the real-time video stream includes image frames arranged continuously in time sequence;

[0113] Based on GPU acceleration, the image frames in the real-time video stream are preprocessed in real time to obtain the environment image.

[0114] Specifically, in order to meet the real-time requirements, the present invention uses GPU parallel computing to accelerate the reasoning process of the YOLO V11 model. The parallel processing unit of the GPU can process multiple image frames of the input video stream at the same time and transmit the video frame stream to the YOLO V11 model in real time. The acceleration process is as follows:

[0115] Video stream preprocessing: The GPU is responsible for real-time preprocessing of the captured video stream, resizing the video frames according to the input requirements of the YOLO V11 model (for example, 640×640) and normalizing them.

[0116] Model inference acceleration: The preprocessed video frames are sent to the YOLO V11 model in batches, and the parallel processing capability of the GPU is used to efficiently complete the model inference process and realize real-time detection of multiple targets.

[0117] Detection result processing: When processing the inference results, the GPU first analyzes the face and body detection results. If normal conditions are met, the screen continues to display; if a camera device is detected, a screen prompt is triggered and a warning log is recorded.

[0118] In one embodiment, preprocessing the image frames in the real-time video stream in real-time based on GPU acceleration to obtain the environment image may include the following steps:

[0119] Step 402, using the GPU as a default computing device, obtaining a real-time video stream, and reading image frames in the real-time video stream frame by frame;

[0120] Step 404, converting each image frame into an initial image in a preset format, where the preset format is an input format of the improved YOLO V11 model;

[0121] Step 406: For each initial image, use the image processing function provided by the deep learning framework to normalize the size of the initial image to obtain an environment image.

[0122] Specifically, for each frame of the image read, it is converted into a format suitable for processing by the deep learning framework, usually by converting the image data into a tensor. Then, the image can be resized using the image processing functions provided by the deep learning framework or custom image processing operations. If other preprocessing operations are required, such as normalizing pixel values, cropping, etc., they can be performed at this step.

[0123] By using GPU to accelerate the processing of real-time video streams and normalize the image size, the processing speed of the real-time video stream of the environmental image can be improved. Through the above GPU acceleration steps, the embodiment of the present application can achieve real-time detection under high frame rate and high resolution video streams. By utilizing the parallel computing capability of the GPU, the model achieves a processing speed of 20ms per frame in the inference stage, meeting the needs of real-time detection.

[0124] In order to further illustrate the scheme of the embodiment of the present application, a specific example is used below. In a specific embodiment, the display screen anti-shooting method in each of the above embodiments can be integrated into a software anti-shooting system. The software anti-shooting system runs directly on the original device (such as Wi ndows, domestic information creation system, MacOS), uses the device's own camera to perform real-time analysis of the environment in front of the screen, and dynamically controls the display content through the desktop application, slightly occupying the original system resources.

[0125] The specific process is as follows:

[0126] 1. Environmental Information Collection and Preprocessing

[0127] 1. Use the device's built-in camera to capture the environment image or video stream in front of the screen in real time.

[0128] 2. Preprocess the collected video stream, including:

[0129] The image resolution is adjusted to the standard size required by the algorithm input.

[0130] format to support the input format of deep learning models.

[0131] 2. Target Detection and Behavior Analysis

[0132] 1. Use the improved YOLO V11 model to detect and classify targets in environmental images:

[0133] Detection target: Identify targets such as faces, bodies, and camera equipment.

[0134] Behavior analysis: Based on the number, location, and dynamic behavior of detected targets, determine whether the following situations exist: 1) Single or multiple people are operating. 2) Whether there is a mobile phone or other camera. 3) The face of an unauthorized user.

[0135] 3. Display strategy control

[0136] 1. Execute dynamic display control strategy based on the detection results:

[0137] 1) Normal display: If no abnormal situation is detected (such as camera equipment, multiple people or unauthorized users), the screen continues to display content normally.

[0138] 2) Display warning: If the presence of a camera or multiple people is detected, a warning message will pop up on the screen to alert the user.

[0139] 3) Screen off: If illegal filming behavior is detected (such as pointing the phone at the screen for a long time), the desktop application is directly controlled to turn off the screen content to protect information security.

[0140] 4. User Tips and Feedback

[0141] 1. Display the detection status and prompt information in real time through the desktop application interface.

[0142] 2. Support sending notifications (such as warning prompts or detection logs) to administrators or preset terminals for easy tracking and management.

[0143] The above-mentioned display screen anti-shooting method and software anti-shooting system have the following characteristics and advantages:

[0144] 1. Real-time: With the help of GPU acceleration and efficient model inference, real-time target detection and analysis can be achieved locally on the device.

[0145] 2. Lightweight: The software occupies only a small amount of original system resources, has little impact on system performance, and is suitable for running on ordinary devices.

[0146] 3. Intelligent control: Dynamically adjust the display strategy according to the detection results to implement various protection measures such as watermark prompts, warning information pop-ups or screen turning off.

[0147] 4. High compatibility: supports multiple mainstream operating systems (Windows, domestic information technology systems, MacOS), and can run directly on existing devices without the need for additional hardware.

[0148] 5. Wide range of applicable scenarios: Applicable to scenarios such as finance, medical care, and conferences, suitable for needs that require convenient deployment and real-time protection.

[0149] In another embodiment, the above-mentioned display screen anti-shooting methods can be integrated into a hardware anti-shooting system. The hardware anti-shooting system can collect video data of the original system through the HDMI I N interface, and at the same time combine the built-in camera to perform real-time analysis of the environmental information in front of the screen to achieve dynamic protection of the screen content. The specific workflow is as follows:

[0150] 1. Video data acquisition and input

[0151] 1. The hardware device collects the screen video signal of the original system through the HDMI I N interface. 2. The video data is converted into a standard format suitable for deep learning model input through the acquisition card inside the device.

[0152] II. Environmental Information Collection and Analysis

[0153] 1. The built-in high-definition camera monitors the environment in front of the screen in real time and captures images or video streams in the environment.

[0154] 2. Preprocess the collected environmental information (such as size normalization and format conversion) and input it into the improved YOLO V11 model for detection.

[0155] 3. Target Detection and Behavior Analysis

[0156] 1. Use the deep learning algorithm consistent with the software version to identify and classify targets (such as faces, bodies, and shooting devices) in environmental images. 2. Analyze user behavior based on the detection results to determine whether there is shooting behavior, multiple targets, or unauthorized users.

[0157] 4. Dynamic watermark addition and display strategy control

[0158] Dynamically process the video signal collected by HDMI according to the detection results and implement the corresponding display strategy:

[0159] 1. Add watermark: superimpose dynamic watermark on the original video signal to prevent the screen content from being illegally filmed.

[0160] 2. Display warning: When a camera or unauthorized user is detected, a warning message will be displayed.

[0161] 3. Screen off: When obvious illegal shooting behavior is detected (such as the phone being left on for a long time), the screen content will be turned off to protect information security.

[0162] 5. Processing result output and integration

[0163] 1. The processed video signal is output to the display screen through the HDMIOUT interface to realize the control of the screen display content. 2. The hardware device runs on the Linux system and supports seamless integration with the original system without changing the existing operating system or application program, and has plug-and-play characteristics.

[0164] The above-mentioned hardware anti-shooting system has the following characteristics and advantages:

[0165] 1. Real-time performance: Based on the GPU-accelerated target detection algorithm, millisecond-level response time can be achieved. 2. High compatibility: Adapt to various display devices through the HDMI interface, completely independent of the original system, and does not affect system performance. 3. Intelligence: Combining environmental monitoring and video data processing, it can dynamically add watermarks, display warnings or turn off the screen to meet the needs of various scenarios. 4. Security: Through multi-level protection mechanisms (watermarks, warnings, screen off, etc.), it effectively prevents the screen content from being illegally photographed.

[0166] This hardware solution provides an efficient and intelligent solution for information security and privacy protection, and is suitable for scenarios that require a high level of security, such as conferences, financial institutions, and medical terminals.

[0167] The present application provides a display screen anti-shooting method, which realizes efficient coordination of algorithms and hardware through GPU acceleration, and improves the overall performance of the system. According to the detection results, the content of the display screen is intelligently adjusted, providing a balance between personalization and security.

[0168] In one embodiment, a display screen anti-shooting method provided by the present application can be applied to the following test environments and device performances:

[0169] 1. Hardware configuration

[0170] Pure software environment CPU: Intel Core i 7-12900K Memory: 64GB DDR4 GPU: NVID IAGeForce RTX 3060 laptop graphics card, 6GB GDDR6 video memory, support CUDA parallel computing.

[0171] Embedded devices: RK3588 chip, 8g memory + 64G, with high-performance AI computing capabilities, suitable for resource-constrained devices.

[0172] 2. Software environment

[0173] Operating system: Windows 11, Ubuntu 22.04

[0174] Deep learning framework: PyTorch 2.5.1+CUDA 11.8 support.

[0175] CUDA and cuDNN versions: CUDA 11.8, cuDNN 8.2.

[0176] 3. Importance

[0177] CPU usage: When using GPU computing, the CPU usage is almost negligible, which shows that the model mainly relies on GPU for computing, reducing the burden on the CPU.

[0178] CPU performance: When using the CPU for calculations, the CPU occupancy rate is 5-10%, and the processing time per frame is about 40ms, which further reflects the advantage of GPU acceleration.

[0179] Embedded device performance: On the RK3588 embedded device, the model can run at 30ms per frame, showing the good adaptability of the model in resource-constrained environments.

[0180] In one embodiment, a display screen anti-shooting method provided by the present application is applied to a CPU, a GPU and an embedded device respectively, and the performance comparison results are as follows:

[0181] 1. Significant advantages of GPU acceleration

[0182] High performance: On the RTX 3060 laptop graphics card, the model achieves a processing speed of 20ms per frame, which is approximately equal to 50 frames per second (FPS), meeting the real-time requirements.

[0183] Resource Utilization: The use of GPU significantly reduces CPU usage, allowing the system to handle other tasks simultaneously and improving overall efficiency.

[0184] 2. Feasibility of CPU calculation

[0185] Low CPU usage: Even when using only the CPU, the CPU usage is only 5-10%, and the processing time for each frame is about 40ms, which is approximately equal to 25 frames per second (FPS). It can still be applied in some scenarios where real-time requirements are not high.

[0186] 3. Applicability of embedded devices

[0187] RK3588 performance: On the RK3588 embedded device, the model processing time per frame is 30ms, which is approximately equal to 33 frames per second (FPS), which is a very excellent performance indicator for embedded applications.

[0188] In one embodiment, the present application provides a display screen anti-shooting method, which compares the performance differences with the traditional model in various implementation environments, and the results are as follows:

[0189] 1. Comparison under GPU environment

[0190] Performance of YOLOv5 and YOLOv8: On the same RTX 3060 laptop graphics card, the per-frame processing time of the YOLOv5 and YOLOv8 models is approximately 25ms and 28ms, respectively.

[0191] Improved the performance of the YOLO V11 model: Under the same hardware conditions, the processing time per frame is 20ms, which is 20% and 28.5% faster than YOLOv5 and YOLOv8 respectively.

[0192] 2. Comparison under CPU environment

[0193] Performance of YOLOv5 and YOLOv8: When using only CPU, the processing time per frame for YOLOv5 and YOLOv8 is approximately 60ms and 70ms respectively.

[0194] Improved performance of the YOLO V11 model: 40ms per frame processing time on CPU, 33% and 42.8% faster than YOLOv5 and YOLOv8 respectively.

[0195] 3. Comparison on embedded devices

[0196] Limitations of traditional models: YOLOv5 and YOLOv8 on the embedded device RK3588 have large models and high computational complexity, and the processing time per frame may exceed 50ms, which cannot meet the real-time requirements.

[0197] Improved the performance of the YOLO V11 model: A processing speed of 30ms per frame was achieved on RK3588, meeting the needs of real-time detection and demonstrating the high efficiency and lightweight of the model.

[0198] In one embodiment, a display screen anti-shooting method provided in the embodiment of the present application can also be accelerated based on NPU (neural processing unit). Among them, the advantages of NPU acceleration include:

[0199] 1. Efficient inference speed: With the acceleration of RK3588's NPU, the model can run at a speed of 30ms per frame, which is approximately equal to 33 frames per second (FPS), meeting the needs of real-time detection.

[0200] 2. Low power consumption and high performance: NPU is optimized for neural network computing and can provide high computing performance with low power consumption, which is suitable for embedded and mobile devices.

[0201] In addition, NPU is an independent chip, which also has the following significance:

[0202] 1. Independent and controllable: As an independently developed chip, RK3588 has independent intellectual property rights, which helps to avoid the risks of overseas chip supply chain.

[0203] 2. Optimized hardware adaptation: The model is optimized and adapted for the NPU of RK3588, so that the model can fully utilize the hardware acceleration capabilities and improve the overall performance.

[0204] The adaptation and optimization of the aforementioned improved YOLO V11 model on the autonomous NPU may include the following steps:

[0205] 1. Model quantization and conversion

[0206] Model quantization: To adapt to the computing characteristics of the NPU, the model may be INT8 quantized to reduce the model size and computational complexity.

[0207] Model conversion: Use specific tools (such as RKNN Tool kit) to convert the PyTorch model into a model format executable by NPU.

[0208] 2. Performance optimization

[0209] Operator optimization: The model is adjusted for the set of operators supported by the NPU to ensure that all operations in the model can be accelerated on the NPU.

[0210] Parallel computing: Use the parallel computing capabilities of NPU to improve the reasoning speed of the model.

[0211] The embodiments of the present application provide a display screen anti-shooting method and an improved YOLO V11 model, which can support multiple hardware platforms: the model can not only run on high-performance GPUs, but also run efficiently on CPUs and NPUs, demonstrating the cross-platform compatibility of the model. Adapt to a variety of computing architectures: The design and optimization of the model take into account different computing architectures such as GPU, CPU, and NPU, and has a wide range of applicability. Furthermore, it can meet different application scenarios, for resource-constrained devices: on resource-constrained embedded devices such as RK3588, the model can still achieve real-time detection, which is suitable for scenarios such as IoT devices and smart terminals.

[0212] In addition, the improved YOLO V11 model has good adaptability to NPU: after optimization, it can run smoothly on the NPU of RK3588, making full use of the hardware acceleration capabilities. It also has high performance: under NPU acceleration, the model achieves a processing speed of 30ms per frame, which is better than the performance of traditional models on similar devices.

[0213] In one of the embodiments, in order to prevent illegal recording in large-scale scenes such as movie theaters, a display screen anti-shooting method provided in the embodiment of the present application uses advanced computer vision technology for real-time detection to become an effective solution for detecting small details in a large field of view. The YOLOv11 model plays a key role in this solution with its excellent detection ability for small targets in a large field of view. The following details how to achieve real-time detection and recording of illegal video recording in the audience seats based on YOLOv11, combined with multiple imaging acquisitions by a rotating pan-tilt head, target dwell time detection, and GPU acceleration.

[0214] 1. System Overview

[0215] Objective: To detect and identify illegal recordings in large-scale scenarios (such as movie theaters) in real time, and take timely measures to prevent the spread of pirated content.

[0216] Core technologies:

[0217] YOLOv11 target detection model: It has the ability to efficiently and accurately detect small detail targets (such as mobile phones and cameras in the hands of viewers) in a large field of view.

[0218] The rotating pan-tilt head collects images multiple times: it expands the monitoring range, covers the entire cinema, and captures small targets in the audience seats.

[0219] Target dwell time detection: Identify continuous illegal recording behavior by detecting how long the target stays in the same location.

[0220] GPU acceleration: Improves the speed of image processing and model reasoning to meet real-time requirements.

[0221] 2. System Architecture

[0222] Hardware

[0223] High-resolution cameras: Equipped with low-light or infrared capabilities and mounted on a pan-tilt head, they can capture clear images of the auditorium in dimly lit theater environments, with enough detail to identify small devices.

[0224] Rotating PTZ: supports all-round rotation of the camera, covering the entire audience area and achieving wide-viewing-field monitoring.

[0225] GPU server: A high-performance computing device equipped with a GPU (such as the NVID IA RTX series) for real-time image processing and running the YOLOv11 model.

[0226] Software

[0227] Image acquisition and preprocessing module: controls the rotation of the camera and image acquisition, preprocesses the image, and enhances the clarity of small targets.

[0228] YOLOv11 object detection model: specially trained to excel in detecting small details in a large field of view.

[0229] Target tracking and residence time calculation module: Continuously track the detected target and calculate its residence time at the same location.

[0230] Detection and alarm module: Determine whether there is illegal recording behavior based on the detection results and residence time, and trigger corresponding processing measures.

[0231] 3. Detailed steps of the testing process

[0232] 1. System initialization

[0233] Equipment startup: Start the pan / tilt head and camera, set the rotation angle range and step angle to ensure that the entire auditorium is covered.

[0234] Model loading: Load the optimized YOLOv11 model and target tracking algorithm on the GPU server to prepare for processing image data.

[0235] 2. Image Acquisition

[0236] Rotational scanning: The rotating pan / tilt moves the camera according to a preset scanning strategy (such as rotating a certain angle each time) to achieve full coverage of the auditorium.

[0237] High-resolution imaging: At every angle, the camera captures high-resolution images, ensuring that small objects are clearly visible in a large field of view.

[0238] 3. Image Preprocessing

[0239] Small target enhancement: For targets with small details, image enhancement processing such as sharpening, contrast adjustment, and denoising is performed to improve the visibility of small targets.

[0240] Resizing: Resize the image to the input size required by the YOLOv11 model (e.g., 640×640 pixels) while maintaining the details of small objects.

[0241] Format conversion: Convert the image to the data format required by the model, preparing to input the model for detection.

[0242] 4. Object Detection

[0243] Model reasoning: The preprocessed image is input into the YOLOv11 model. Thanks to YOLOv11's enhanced ability to detect small targets, the model can accurately identify small recording devices in a large field of view.

[0244] Test content:

[0245] Small recording device identification: Detect devices such as mobile phones and cameras that may be used for illegal recording, even if they occupy a small pixel in a large scene.

[0246] Suspicious behavior detection: Identify actions such as raising arms or facing the screen for a long time in the audience. These subtle behavioral changes can be accurately captured even in a large field of view.

[0247] 5. Target tracking and dwell time calculation

[0248] Target association: Using the target tracking algorithm, each detected target is assigned a unique ID and the position and status of the target are tracked in consecutive frames.

[0249] Residence time calculation:

[0250] Position stability judgment: By analyzing the position changes of the target in consecutive frames, it is determined whether it remains stationary in the same position.

[0251] Time accumulation: Records the time the target stays at the same location. If the stay time exceeds the preset threshold (e.g. 10 seconds), it is considered to be suspected of illegal recording.

[0252] 6. Result analysis and judgment

[0253] Comprehensive assessment: Combined with information such as the target’s category, dwell time, and behavioral characteristics, a comprehensive judgement is made as to whether there is any illegal recording.

[0254] False alarm filtering: Set reasonable thresholds to avoid misjudging normal behaviors (such as briefly checking your phone) as illegal recordings.

[0255] 7. Alarm and processing

[0256] Triggering alarm:

[0257] Screen prompt: Display a warning message on the movie screen to remind viewers that illegal recording is prohibited.

[0258] Sound warning: A warning sound is played through the theater sound system to attract the audience's attention.

[0259] Notify staff: Send information such as the location and image of suspicious behavior to theater management for timely intervention.

[0260] Evidence records: Save relevant images and video clips as a basis for subsequent processing.

[0261] 8. Cycle detection

[0262] Continuous monitoring: The pan / tilt head continues to move according to the preset strategy and repeats the above steps to achieve continuous monitoring of the audience seats.

[0263] Dynamic adjustment: Based on the detection results, the system can dynamically adjust the scanning strategy, increase the scanning frequency of suspicious areas, and focus on monitoring.

[0264] 4. Key technical details

[0265] 1. Advantages of the YOLOv11 model in detecting small targets in a large field of view

[0266] Improved feature extraction: YOLOv11 introduces modules such as C3K2, SPFF, and C2PSA, which enhance the feature extraction capability of small objects and enable the model to accurately identify small details in a large field of view.

[0267] High-resolution support: When processing high-resolution images, the model can maintain sensitivity to small objects and avoid the loss of small object information due to downsampling.

[0268] Optimized detection head: For the detection of small targets, the structure and parameters of the detection head are adjusted to improve the detection accuracy of small-sized objects.

[0269] 2. Model training and optimization

[0270] Dedicated dataset: A large number of images containing small objects in a large field of view are collected and annotated, covering different lighting conditions, audience behaviors, and shooting angles.

[0271] Data enhancement: Use data enhancement techniques such as scaling, cropping, and rotation to improve the model's robustness in detecting small targets.

[0272] Transfer learning: Based on pre-trained lightweight models (such as yo l o11 n.pt), transfer learning is performed to accelerate the training process and improve model performance.

[0273] 3. Target tracking and dwell time detection

[0274] Efficient tracking algorithm: DeepSORT, a small target tracking algorithm suitable for real-time tracking, is used to ensure that small targets can still be accurately tracked in the presence of multiple targets and multiple interferences.

[0275] Dwell time threshold setting: According to the theater environment and audience behavior characteristics, the dwell time threshold is set reasonably to distinguish between normal use and illegal recording.

[0276] 4. GPU acceleration and real-time guarantee

[0277] Parallel computing: Make full use of the parallel computing capabilities of GPU to accelerate model reasoning and tracking algorithms to meet real-time requirements.

[0278] Resource optimization: Rationally allocate computing resources to avoid performance degradation caused by resource competition and ensure stable system operation.

[0279] 5. System advantages

[0280] Excellent small target detection capability: YOLOv11 significantly improves the detection accuracy of small details in a large field of view, and can accurately identify small recording devices and suspicious behaviors in the audience.

[0281] High real-time performance: Combined with GPU acceleration and optimization algorithms, it achieves millisecond-level detection and response, and promptly discovers and handles illegal recording behaviors.

[0282] Comprehensive coverage: The use of a rotating pan-tilt head enables all-round, multi-angle monitoring of the auditorium, avoiding blind spots.

[0283] Intelligence and automation: The system automatically completes detection, tracking, judgment and alarm, reducing human intervention and improving monitoring efficiency.

[0284] 6. Notes

[0285] Privacy protection: During the monitoring process, strictly abide by relevant laws and regulations, protect the privacy of viewers, and ensure the legality and compliance of the system.

[0286] Parameter tuning: According to different theater environments, model parameters and thresholds are adjusted regularly to adapt to environmental changes and reduce false alarm rates.

[0287] System maintenance: Regularly check and maintain hardware equipment, update software systems and models, and ensure that the system continues to operate efficiently.

[0288] 7. Extended Application

[0289] Concerts and theaters: Prevent live performances from being illegally recorded and distributed, and protect intellectual property rights.

[0290] Exhibitions and press conferences: protect the confidentiality of exhibits and new product information to prevent early disclosure.

[0291] Meeting places: prevent commercial secrets and sensitive information from being secretly photographed and ensure information security.

[0292] 8. Conclusion

[0293] By combining the YOLOv11 model's excellent detection capability for small details in a large field of view with technologies such as multiple acquisitions of images by a rotating pan-tilt head and target dwell time detection, the system can efficiently and accurately identify and record illegal recording behaviors in large scenes. This solution not only improves detection accuracy and real-time performance, but also effectively protects content copyright and information security, providing strong technical support for preventing the spread of piracy.

[0294] The overall process of the above display screen anti-shooting method is as follows:

[0295] 1. System startup and initialization

[0296] The camera and pan / tilt are started, and the GPU server loads the optimized YOLOv11 model.

[0297] 2. Pan-tilt scanning

[0298] The gimbal rotates according to a preset strategy, and the camera continuously captures high-resolution images at each position to ensure that small targets are clearly visible.

[0299] 3. Image preprocessing

[0300] Perform small object enhancement, resize, and format conversion on the image to prepare for input into the model.

[0301] 4. Object Detection

[0302] The YOLOv11 model accurately detects small recording devices and suspicious behaviors in a large field of view.

[0303] 5. Target tracking

[0304] Track detected targets, assign unique IDs, and record location changes.

[0305] 6. Calculation of residence time

[0306] Calculate the time the target stays in the same location to identify continuous illegal recording behavior.

[0307] 7. Illegal behavior judgment

[0308] Comprehensive information such as target category and dwell time is used to determine whether there is any illegal recording behavior.

[0309] 8. Alarm and processing

[0310] Trigger alarm measures, such as screen prompts, sound warnings, notify staff, and preserve evidence.

[0311] 9. Cycle monitoring

[0312] Continue to rotate and scan, update detection and tracking in real time, and dynamically adjust monitoring strategies.

[0313] Through the above process, the system fully leverages the advantages of YOLOv11 in detecting small targets with a large field of view, realizes efficient and accurate monitoring of illegal recording behaviors, and provides reliable technical guarantees for information security and copyright protection in large-scale scenarios.

[0314] Based on the same inventive concept, the embodiment of the present application also provides a display screen anti-shooting device for implementing the display screen anti-shooting method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more display screen anti-shooting device embodiments provided below can refer to the limitations of the display screen anti-shooting method above, and will not be repeated here.

[0315] In an exemplary embodiment, Figure 3 As shown, a display screen anti-shooting device is provided, comprising:

[0316] An acquisition module 301 is used to acquire an environment image in front of a display screen;

[0317] A detection module 302 is used to input the environment image into the trained improved YOLO V11 model to obtain a detection result corresponding to the environment image; the detection result includes at least one of whether a face is detected in the environment image, whether a human body is detected, and whether a shooting device is detected;

[0318] The display module 303 is used to determine a display strategy corresponding to the display screen according to the detection result; the display strategy is used to instruct the display screen to display corresponding content.

[0319] In one embodiment, the display module is further used for:

[0320] When the detection result is that there is a shooting device in the environment image, the display strategy is to display a blocking interface on the display screen; the blocking interface is used to block the normal interface of the display screen, and / or the blocking interface is used to display a warning prompt;

[0321] When the detection result is that there are multiple human bodies or faces in the environment image, the display strategy is to display an occlusion interface on the display screen; the occlusion interface is used to block the normal interface of the display screen, and / or the occlusion interface is used to display a warning prompt;

[0322] When the detection result is that there is a face in the environment image and there is no camera, the display strategy is to display a normal interface on the display screen;

[0323] When the detection result is that there is no face in the environment image, the display strategy is to turn off the display screen.

[0324] In one of the embodiments, the detection result also includes whether the face in the environment image is the face corresponding to the preset user;

[0325] The display module is also used to:

[0326] When the detection result is that the face in the environment image is not the face corresponding to the preset user, the display strategy is to display a warning prompt on the display screen and send a corresponding prompt message to the preset terminal.

[0327] In one embodiment, the improved YOLO V11 model is obtained by the following method:

[0328] Acquire multiple training images, where the training images include scenes in front of the display screen under different lighting conditions, different backgrounds, and different shooting angles;

[0329] Performing labeling processing on each training image to obtain a labeling file corresponding to each training image; the labeling file is labeled with at least one of a shooting device and a face in the training image;

[0330] The training images and the corresponding annotation files are used as training data sets. Based on the multi-task learning framework, the YOLO V11 model is trained using the training data sets to obtain a trained improved YOLO V11 model.

[0331] In one embodiment, the annotation file is compatible with the COCO format.

[0332] In one embodiment, the acquisition module is further used to:

[0333] Collecting a real-time video stream in front of a display screen, the real-time video stream includes image frames arranged continuously in time sequence;

[0334] Based on GPU acceleration, the image frames in the real-time video stream are preprocessed in real time to obtain the environment image.

[0335] In one embodiment, the acquisition module is further used to:

[0336] Use GPU as the default computing device to obtain real-time video stream and read image frames in the real-time video stream frame by frame;

[0337] Convert each image frame into an initial image in a preset format, where the preset format is the input format of the improved YOLO V11 model;

[0338] For each initial image, the image processing function provided by the deep learning framework is used to normalize the size of the initial image to obtain the environment image.

[0339] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of a power supply safety management method as described above when executing the computer program.

[0340] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0341] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can refer to the partial description of the method embodiments. The device embodiments described above are only schematic, wherein the components described as separate parts may or may not be physically separated, and the parts displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without paying any creative work.

[0342] The above-mentioned embodiments only express several implementation methods of the embodiments of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the patent of the embodiments of the present application. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the embodiments of the present application, and these all belong to the protection scope of the embodiments of the present application.

Claims

1. A method for preventing a display screen from being photographed, characterized in that: The method comprises: Acquire the environment image in front of the display screen; Input the environment image into the trained improved YOLO V11 model to obtain a detection result corresponding to the environment image; the detection result includes at least one of whether a face is detected in the environment image, whether a human body is detected, and whether a shooting device is detected; A display strategy corresponding to the display screen is determined according to the detection result; the display strategy is used to instruct the display screen to display corresponding content.

2. The method according to claim 1, characterized in that The determining, according to the detection result, a display strategy corresponding to the display screen includes: When the detection result is that there is a shooting device in the environmental image, the display strategy is to display a blocking interface on the display screen; the blocking interface is used to block a normal interface of the display screen, and / or the blocking interface is used to display a warning prompt; When the detection result is that there are multiple human bodies or faces in the environmental image, the display strategy is to display a blocking interface on the display screen; the blocking interface is used to block a normal interface of the display screen, and / or the blocking interface is used to display a warning prompt; When the detection result is that there is a face or a body in the environment image, and there is no shooting device, the display strategy is to display a normal interface on the display screen; When the detection result is that there is no human face in the environment image, the display strategy is to turn off the display screen.

3. The method according to claim 1, characterized in that The detection result also includes whether the face in the environment image is the face corresponding to the preset user; When the detection result is that the face in the environment image is not the face corresponding to the preset user, the display strategy is to display a warning prompt on the display screen and send a corresponding prompt message to a preset terminal.

4. The method according to claim 1, characterized in that The improved YOLO V11 model is obtained by the following method: Acquire multiple training images, wherein the training images include scenes in front of the display screen under different lighting conditions, different backgrounds, and different shooting angles; Performing annotation processing on each of the training images to obtain an annotation file corresponding to each of the training images; The annotation file is annotated with at least one of a shooting device and a face in the training image; The training images and the corresponding annotation files are used as training data sets, and based on a multi-task learning framework, the YOLO V11 model is trained using the training data sets to obtain the trained improved YOLO V11 model.

5. The method according to claim 4, characterized in that: The annotation file is compatible with the COCO format.

6. The method according to any one of claims 1 to 5, characterized in that: The step of acquiring the environment image in front of the display screen comprises: Collecting a real-time video stream in front of the display screen, wherein the real-time video stream includes image frames arranged continuously in time sequence; The image frames in the real-time video stream are preprocessed in real time based on GPU acceleration to obtain the environment image.

7. The method according to claim 6, characterized in that The preprocessing of the image frames in the real-time video stream based on GPU acceleration to obtain the environment image includes: Using the GPU as a default computing device, acquiring the real-time video stream, and reading image frames in the real-time video stream frame by frame; Convert each of the image frames into an initial image in a preset format, where the preset format is the input format of the improved YOLO V11 model; For each of the initial images, an image processing function provided by a deep learning framework is used to normalize the size of the initial image to obtain the environment image.

8. A display screen anti-shooting device, characterized in that: The device comprises: An acquisition module, used to acquire an environment image in front of the display screen; A detection module, configured to input the environment image into a trained improved YOLO V11 model to obtain a detection result corresponding to the environment image; the detection result includes at least one of whether a face is detected in the environment image, whether a human body is detected, and whether a shooting device is detected; The display module is used to determine a display strategy corresponding to the display screen according to the detection result; the display strategy is used to instruct the display screen to display corresponding content.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images