Display method of vehicle screen and vehicle

Abstract

The invention provides a display method of a vehicle screen and a vehicle, and relates to the technical field of image detection.The display method comprises the steps that in response to a video playing instruction in the vehicle, image frames in a video to be displayed are detected, an image detection result is obtained, and the image detection result is used for indicating whether the image frames comprise violation content or not; when the image detection result indicates that the image frame comprises the illegal content, shielding the illegal content in the image frame to obtain a target video to be displayed; and displaying the target video to be displayed. According to the display method, the influence of illegal contents on the vehicle screen on the physical and psychological health of passengers can be avoided, and the user experience is improved.

Description

Technical Field

[0001] This application relates to the field of image detection technology, and more specifically, to a method for displaying a vehicle screen and a vehicle within the field of image detection technology. Background Technology

[0002] With the widespread adoption of in-vehicle intelligent entertainment systems, the ability to project images from mobile devices to the vehicle's infotainment system has become a mainstream feature. For example, users can stream video from their phones or tablets to the vehicle's screen. However, if the video stream contains violent, nude, or other inappropriate content, displaying it on the vehicle screen could negatively impact the physical and mental health of passengers.

[0003] Therefore, how to prevent illegal content displayed on vehicle screens from affecting the physical and mental health of passengers is a technical problem that needs to be solved. Summary of the Invention

[0004] This application provides a display method for a vehicle screen and a vehicle. The display method can obscure illegal content in the image frames of the video to be displayed, thereby preventing the illegal content from having an adverse effect on the physical and mental health of passengers and improving the user experience.

[0005] Firstly, a display method for a vehicle screen is provided, the display method including: In response to a video playback command in the vehicle, image frames in the video to be displayed are detected to obtain image detection results, which are used to indicate whether the image frame contains illegal content; When the image detection result indicates that the image frame contains the illegal content, the illegal content in the image frame is occluded to obtain the target video to be displayed. The target video to be displayed is shown.

[0006] In the embodiments of this application, image frames of the video to be displayed are detected, and when the image detection result indicates that there is illegal content in the image frame, the illegal content is occluded; the occluded target video to be displayed is then displayed. Compared to related technologies that directly output the video stream to the vehicle's display screen in response to the vehicle's video playback command, potentially causing illegal content in the video stream frames to affect the physical and mental health of passengers, this application detects image frames in the video to be displayed before displaying it, obtaining image detection results to ensure that illegal content in the image frames can be detected in advance. If illegal content is found in the image frame, it is occluded, ensuring effective filtering of illegal information in the video stream and preventing illegal content from affecting the physical and mental health of passengers.

[0007] Furthermore, this application only precisely obscures the infringing content, rather than globally obscuring or prohibiting the output of the video to be displayed. This ensures that non-infringing content in the video frames can be displayed normally, avoiding the inability to display the video due to global intervention. Thus, it can effectively filter infringing content while taking into account the user's viewing experience.

[0008] In conjunction with the first aspect, in certain implementations of the first aspect, the detection of image frames in the video to be displayed to obtain image detection results includes: Determine the target risk parameters of the image frame; If the target risk parameter is greater than the first threshold, the image frame is divided into blocks to obtain at least two image blocks; The image detection result is obtained based on the image frame and the image block.

[0009] In the embodiments of this application, the target risk parameter is a risk value obtained by quickly screening the violation risk of image frames using a lightweight model. If the target risk parameter of an image frame is greater than a first threshold, it indicates that the image frame has a violation probability (i.e., the image frame is a suspicious frame). Therefore, further detection is performed based on the image frame and its image blocks to obtain the image detection result. This ensures that only image frames with a possibility of violation are processed in blocks and subjected to more refined detection to obtain the image detection result. Compared to performing block processing and detection frame by frame, this method can effectively reduce the computing power consumption of the vehicle's system and avoid problems such as screen stuttering and frame rate drop caused by inference latency, thereby balancing the detection efficiency and detection accuracy of image frames.

[0010] In one implementation, if the target risk parameter is less than a second threshold, the image detection result indicates that there is no illegal content in the image frame.

[0011] In the embodiments of this application, if the target risk parameter is less than the second threshold, it indicates that there is no probability of violation in the image frame. Therefore, the image detection result is directly determined to indicate that there is no illegal content in the image frame, avoiding misjudgment of safe image frames with no possibility of violation and performing unnecessary intervention operations.

[0012] In combination with the first aspect and the above implementation methods, in some implementation methods of the first aspect, determining the target risk parameter of the image frame includes: The resolution of the image frame is adjusted to a preset first resolution, which is smaller than the original resolution of the image frame; The first resolution image frame is input into the pre-trained target model to obtain the first risk parameter; The target risk parameter is obtained based on the first risk parameter.

[0013] In the embodiments of this application, the resolution of the image frame is adjusted to a preset first resolution, and the image frame at the first resolution is input into the pre-trained target model to obtain the first risk parameter. By reducing the resolution of the image frame from the original resolution to the first resolution, the computing power consumption of the vehicle system can be reduced, avoiding any impact on the normal operation of the vehicle system. In addition, reducing the resolution of the image frame allows the target model to focus on the global structure and core features of the image, suppressing the influence of high-frequency noise in high-resolution images on the detection process of the target model, improving the generalization ability of the target model, and reducing the risk of overfitting.

[0014] In combination with the first aspect and the above implementation methods, in some implementation methods of the first aspect, obtaining the target risk parameter based on the first risk parameter includes: If the first risk parameter is greater than the first threshold, the first risk parameter is determined as the target risk parameter; If the first risk parameter is less than or equal to the first threshold and the first risk parameter is greater than the second threshold, the resolution of the image frame is adjusted to a preset second resolution; the image frame with the second resolution is input into the target model to obtain the target risk parameter, wherein the second resolution is less than the original resolution of the image frame and greater than the first resolution.

[0015] In the embodiments of this application, if the first risk parameter is greater than the first threshold, it indicates that the target model can detect the probability of a violation in the image frame when detecting a lower resolution image frame. Therefore, the possibility of a violation in the image frame is determined, and the first risk parameter is determined as the target risk parameter to ensure that the target risk parameter can accurately represent the probability of a violation in the image frame. If the first risk parameter is less than or equal to the first threshold and greater than the second threshold, it indicates that the target model cannot accurately determine the probability of a violation in the image frame when detecting a lower resolution image frame. Therefore, the resolution of the image frame is increased from the first resolution to the second resolution. By reducing the compression ratio of the down-resolution, the detection accuracy is improved, and misjudgment of the probability of a violation in the image frame is avoided in the case of excessive down-resolution and feature loss.

[0016] In conjunction with the first aspect and the above implementation methods, in some implementation methods of the first aspect, determining the image detection result based on the image frame and the image block includes: Determine a first evaluation parameter corresponding to the image frame, the first evaluation parameter being used to indicate whether the image frame has a risk of violation; Determine a second evaluation parameter for the image block corresponding to the image frame, the second evaluation parameter being used to indicate whether the image block includes an image element of the target type; The image detection result is obtained based on the first evaluation parameter and the second evaluation parameter.

[0017] In the embodiments of this application, a first evaluation parameter for the image frame and a second evaluation parameter for the image block are determined, and the image detection result is determined by comprehensively considering the first evaluation parameter and the second evaluation parameter. This ensures that both the overall image frame and local regions of the image frame (i.e., image blocks of the image frame) are considered, thereby obtaining a more accurate target detection result and avoiding misjudgment of illegal content in the image frame.

[0018] In conjunction with the first aspect and the above-described implementations, in some implementations of the first aspect, the display method further includes: Identify the image semantics of the image frame; The process of obtaining the image detection result based on the first evaluation parameter and the second evaluation parameter includes: The image detection result is obtained based on the image semantics, the first evaluation parameter, and the second evaluation parameter.

[0019] In the embodiments of this application, using only the first evaluation parameter can easily lead to missed detection of local violations or misjudgment of globally similar scenes; using only the second evaluation parameter can easily lead to misjudgment due to isolated elements; and using only image semantics can easily lead to misjudgment due to abstract judgments lacking concrete evidence. Therefore, this application identifies the image semantics of image frames; and obtains image detection results based on the image semantics, the first evaluation parameter, and the second evaluation parameter. This ensures that three-dimensional fusion and mutual verification can be performed, thereby effectively avoiding missed and false positives of violations in image frames and ensuring that more accurate target detection results can be obtained.

[0020] In combination with the first aspect and the above implementation methods, in some implementation methods of the first aspect, obtaining the image detection result based on the image semantics, the first evaluation parameter, and the second evaluation parameter includes: If the image semantics indicate that the image frame is a violent image, and the first evaluation parameter indicates that the image frame has a violation risk, and the image block includes image elements of the target type, then the image detection result indicates that the image frame includes the violation content. When the first evaluation parameter indicates that the image frame does not have a violation risk, or the image block does not contain image elements of the target type, the image detection result indicates that the image frame does not contain the violation content.

[0021] In the embodiments of this application, if the image semantics indicate that the image frame is a violent image, and the first evaluation parameter indicates that the image frame has a global violation risk, and the second evaluation parameter detects an image element of the target type in the corresponding image block, the image detection result is determined to be that the image frame contains violation content. This ensures that both the global detection result of the image frame (first evaluation parameter) and the local detection result of the image block of the image frame (second evaluation parameter) are considered simultaneously. Bidirectional verification is performed based on the first and second evaluation parameters to avoid misjudgments of violation content caused by considering only a single dimension, thereby reducing the probability of false positives and false negatives in mixed scenes and improving the accuracy of the image detection results.

[0022] In combination with the first aspect and the above implementation methods, in some implementation methods of the first aspect, obtaining the image detection result based on the image semantics, the first evaluation parameter, and the second evaluation parameter includes: If the image semantics indicate that the image frame is a pornographic image, a weighting coefficient is determined based on the proportion of skin color pixels in the image frame and the human body posture in the image frame; The image detection result is determined based on the first evaluation parameter, the second evaluation parameter, and the weighting coefficient.

[0023] In the embodiments of this application, in scenarios such as sports and fitness, the amount of exposed skin is high but does not constitute illegal content that needs to be occluded. If the judgment is made solely based on the first and second evaluation parameters, it may lead to the normal scenes in sports and fitness scenarios being judged as containing illegal content, thereby triggering false occlusion and affecting the user's viewing experience. Therefore, when the image semantic indication is that the image frame is a pornographic image, a weight coefficient is determined based on the proportion of skin color pixels and the human posture. The image detection result is determined by combining the first evaluation parameter, the second evaluation parameter, and the weight coefficient. This ensures that the judgment of whether it is a normal sports and fitness scene can be made based on the user's posture and the proportion of skin color pixels, avoiding the judgment of normal scenes as containing illegal content, thereby ensuring the accuracy of the image detection result.

[0024] In conjunction with the first aspect and the above implementation methods, in some implementation methods of the first aspect, when the image detection result indicates that the image frame includes the illegal content, the illegal content in the image frame is occluded to obtain the target video to be displayed, including: If the image detection results of a consecutive preset number of image frames all indicate that the image frames contain the illegal content, the illegal content in the preset number of image frames is occluded to obtain the target video to be displayed.

[0025] In the embodiments of this application, since the presence of illegal content in the video to be displayed usually occurs in consecutive image frames, if only a single frame's image detection result indicates the presence of illegal content, it may be a misjudgment of the single frame's image detection result due to noise or compression artifacts. Therefore, this application ensures that the image detection results for a predetermined number of consecutive image frames all indicate that the image frames contain illegal content, thus occluding the illegal content in consecutive image frames. Compared to occlusion processing based on the image detection results of a single image frame, this method helps filter out transient compression noise and does not immediately trigger occlusion. Furthermore, for illegal content that persists across multiple frames, it can accumulate evidence from multiple frames, ensuring more accurate occlusion.

[0026] Secondly, a display device for a vehicle screen is provided, the display device comprising: The detection module is used to detect image frames in the video to be displayed in response to a video playback command in the vehicle, and obtain image detection results. The image detection results are used to indicate whether the image frame contains illegal content. The processing module is configured to, when the image detection result indicates that the image frame contains the illegal content, perform occlusion processing on the illegal content in the image frame to obtain a target video to be displayed; and display the target video to be displayed.

[0027] Thirdly, a vehicle is provided, the vehicle including a memory and a processor, the memory for storing executable program code, and the processor for calling and running the executable program code from the memory, causing the vehicle to perform the display method in the first aspect or any possible implementation of the first aspect.

[0028] Fourthly, a computer program product is provided, comprising: computer program code, which, when run on a computer, causes the computer to execute the display method described in the first aspect or any possible implementation thereof.

[0029] Fifthly, a computer-readable storage medium is provided that stores computer program code, which, when executed on a computer, causes the computer to perform the display method described in the first aspect or any possible implementation thereof. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of a scenario provided in an embodiment of this application; Figure 2 This is a schematic flowchart illustrating a display method for a vehicle screen provided in an embodiment of this application; Figure 3This is a schematic flowchart illustrating another vehicle screen display method provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of a vehicle screen display device provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of a vehicle provided in an embodiment of this application. Detailed Implementation

[0031] The technical solutions in this application will be clearly and thoroughly described below with reference to the accompanying drawings. In the description of the embodiments of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B. "And / or" in the text is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Furthermore, in the description of the embodiments of this application, "multiple" refers to two or more than two.

[0032] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature.

[0033] With the widespread adoption of in-vehicle intelligent entertainment systems, the ability to project images from mobile devices to the vehicle's infotainment system has become a mainstream feature. For example, users can stream video from their phones or tablets to the vehicle's screen. However, if the video stream contains violent, nude, or terrifying content, displaying it on the vehicle screen could negatively impact the physical and mental health of passengers.

[0034] Therefore, how to prevent illegal content displayed on vehicle screens from affecting the physical and mental health of passengers is a technical problem that needs to be solved.

[0035] In view of this, this application provides a display method for a vehicle screen and a vehicle. Through embodiments of this application, image frames of the video to be displayed are detected, and when the image detection result indicates the presence of illegal content in the image frame, the illegal content is occluded. This ensures that illegal content in the image frame can be detected before the video to be displayed. Occluding illegal content when it is detected ensures effective filtering of illegal information in the video stream, preventing illegal content from affecting the physical and mental health of passengers. Furthermore, this application only precisely occludes illegal content, rather than globally occluding or prohibiting the output of the video to be displayed, ensuring that non-illegal content in the video frame can be displayed normally, avoiding the inability to display the video due to global intervention. Thus, while effectively filtering illegal content, the viewing experience of users inside the vehicle is also considered.

[0036] The following is combined Figures 1 to 3 The display method for a vehicle screen provided in the embodiments of this application will be described in detail.

[0037] Figure 1 This is a schematic diagram of a scenario provided in an embodiment of this application.

[0038] like Figure 1 As shown, scenario 100 includes a vehicle screen 110. When the vehicle screen is playing a video, if inappropriate content (e.g., content containing violence, gore, pornography, nudity, horror, etc.) is detected in the video's image frames, the inappropriate content is obscured by obscuring block 111. Furthermore, during obscuring, a prompt message can be displayed on the vehicle screen, such as "Sensitive content has been obscured."

[0039] Optionally, when obscuring inappropriate content, one can use... Figure 1 The occlusion blocks (i.e., masking strips) shown obscure the infringing content. Alternatively, Gaussian blur or pixelation can be used to cover the infringing content, or a user-predefined image can be used for occlusion (e.g., cartoon images, landscape images, etc.). Gaussian blur refers to a weighted average blurring of pixels based on a Gaussian distribution. Pixelation involves dividing the image into a grid, covering it with blocks of a single color, and assigning the color of each pixel block the average value of its internal pixels.

[0040] Understandably, when inappropriate or illegal content exists in a video's image frame, this application only obscures the corresponding local area, while non-illegal content remains displayed normally. This effectively prevents illegal content from impacting users' mental and physical health, while ensuring the normal display of non-illegal content within the image frame, avoiding the impact on user experience caused by obscuring the entire screen or halting video playback.

[0041] It should be noted that, Figure 1 The display interface in this application is a schematic illustration of the display interface of a vehicle screen, used to describe the display interface when illegal content on the vehicle screen is obscured. This application does not limit the specific content of the display interface or the method of obscuring the illegal content.

[0042] Figure 2 This is a schematic flowchart illustrating a method for displaying a vehicle screen according to an embodiment of this application.

[0043] For example, Figure 2 The display method 200 shown can be executed by the vehicle's infotainment system, or by the vehicle itself, or by the vehicle's processor or chip.

[0044] like Figure 2 As shown, the vehicle screen display method 200 includes S210 to S230. The vehicle screen display method shown in S210 to S230 will be described in detail below.

[0045] S210, in response to a video playback command in the vehicle, detects image frames in the video to be displayed and obtains image detection results.

[0046] Among them, video playback instructions refer to control instructions to start or continue playing video content, which are used to trigger the decoding and display process of the video stream to be displayed; the instructions usually include information such as the source identifier of the playback content, the playback start time, and the playback window parameters.

[0047] For example, video playback commands can be triggered by a user or the vehicle's infotainment system, and different methods can be used to trigger them in different scenarios. For instance, in a vehicle-mounted screen mirroring scenario, when projecting a video from a terminal device (e.g., a mobile phone) to the vehicle's infotainment screen, the user can use the screen mirroring function on their terminal device to transmit the video stream to the infotainment system, thus triggering the video playback command. Alternatively, the user can select a video source from a connected terminal device on the infotainment system to trigger the video playback command. When playing local videos or online videos connected to the network, the user can select a video on the infotainment system and click play to trigger the video playback command, or the infotainment system can automatically trigger the video playback command at scheduled intervals.

[0048] For example, upon detecting a video playback command in the vehicle, in response to the command, the video source and playback parameters are parsed to obtain the video to be displayed. Image frames of the video to be displayed are then captured according to a preset capture period, and the captured image frames are detected to obtain image detection results. These image detection results indicate whether the image frames contain illegal content.

[0049] Among them, prohibited content includes, but is not limited to, bloody, violent, pornographic, nude, and terrifying content, or prohibited content can specifically refer to content that is unsafe / inappropriate to view in the workplace (Not Safe For Work, NSFW). In addition, users can also customize the prohibited content that needs to be obscured to suit the personalized needs of different users.

[0050] For example, when capturing image frames from a video at a preset period, the preset period can be a fixed value or a value dynamically determined based on the performance of the vehicle's infotainment system. For instance, if the preset period is a fixed 50ms, then the Android native screenshot interface of the infotainment system is used to capture image frames every 50ms. If the preset period is a dynamic value, the period is determined based on the current performance indicators of the infotainment system. If the system has sufficient computing power, a smaller period is set to improve detection accuracy and avoid missing image frames containing illegal content due to an excessively large capture period. If the system has low computing power, a larger period is set, and frequent image frame captures consume a lot of resources, causing problems such as system lag and frame rate drops.

[0051] For example, taking a car infotainment system projection scenario, when capturing image frames, the native Android screenshot interface (e.g., ImageReader interface or SurfaceControl.screenshot interface) can be used to directly copy an image from the projection function's buffer without modifying the rendering chain. Because the image frame capture is performed on a separate asynchronous thread—that is, video decoding and projection display are in one thread, while screenshotting, cropping, and scaling operations are in another—they are independent and do not interfere with each other, ensuring a stable chain and not affecting the existing projection effect.

[0052] The image frame detection process will be further explained below.

[0053] Specifically, the process of detecting image frames in the video to be displayed and obtaining image detection results includes: determining the target risk parameter of the image frame; if the target risk parameter is greater than a first threshold, dividing the image frame into blocks to obtain at least two image blocks; and obtaining image detection results based on the image frame and the image blocks.

[0054] For example, if the target risk parameter of an image frame is greater than a first threshold, it indicates that the image frame has a probability of violation (i.e., the image frame is a suspicious frame). Therefore, further detection is performed based on the image frame and its image blocks to obtain the image detection result. This ensures that only image frames with a possibility of violation are processed in blocks and subjected to more refined detection to obtain the image detection result. Compared to processing and detecting each frame in blocks, this method can effectively reduce the computing power consumption of the vehicle's system and avoid problems such as screen stuttering and frame rate drop caused by inference latency, thus balancing the detection efficiency and accuracy of image frames.

[0055] Optionally, if the target risk parameter is less than the second threshold, the image detection result indicates that there is no illegal content in the image frame. Since a target risk parameter less than the second threshold indicates that there is no probability of illegality in the image frame, directly determining that the image detection result indicates that there is no illegal content in the image frame avoids misjudging safe image frames with no possibility of illegality and performing unnecessary intervention operations.

[0056] For example, when dividing an image frame into blocks, the image frame can be evenly divided according to a fixed number of rows and columns. For instance, the image frame can be divided into 9 image blocks of the same size in a 3×3 grid (i.e., 9 image blocks in 3 rows and 3 columns); or it can be divided into 16 image blocks of the same size in a 4×4 grid.

[0057] For example, image frames can be segmented using a sliding window. A window of a fixed size is pre-set and slides across the image frame at a preset step size, generating one image block with each slide. By adjusting the window size and step size to adapt to image elements of different sizes, it ensures the detection of image elements with significant scale variations that pose a risk of violation (e.g., small knives or large bloody areas). Overlapping windows improve the detection accuracy of edge targets.

[0058] Alternatively, image frames can be segmented using an image pyramid and a fixed grid. An image pyramid is a set of images generated by scaling an image at different ratios. For example, an image frame can be reduced to a first image frame at a first ratio, and then enlarged to a second image frame at a second ratio. The first image frame, the second image frame, and the original-sized image frame can then be segmented. The first ratio can be 0.75 times, and the second ratio can be 1.5 times. The fixed grid segmentation can be done in a 3×3 or 4×4 pattern.

[0059] Understandably, by scaling image frames at different ratios to create a set of image blocks, multi-scale adaptation can be achieved, avoiding the problem of single-scale blocks failing to cover image elements of different sizes. For example, smaller image elements can be covered by scaling at 0.75x, and larger image elements by scaling at 1.5x. This cross-verification of multi-scale blocks avoids misjudgments caused by noise in single-scale blocks. Furthermore, by scaling image frames at a second ratio, the detailed features of intrusive elements (such as the texture of skin tones and the edges of weapons) can be enhanced, improving the detection rate of blurry or low-resolution target-type image elements.

[0060] The method for determining target risk parameters will be further explained below.

[0061] In one implementation, determining the target risk parameter of an image frame includes: adjusting the resolution of the image frame to a preset first resolution, the first resolution being smaller than the original resolution of the image frame; inputting the image frame at the first resolution into a pre-trained target model to obtain the first risk parameter; and obtaining the target risk parameter based on the first risk parameter.

[0062] The target model is a lightweight model, such as a lightweight neural network model (MobileNetV3-Small model). This lightweight neural network model is a lightweight classifier used for fast binary or multi-class classification, and outputs corresponding indicators after detection. The lightweight model is used to quickly screen the violation risk of image frames to obtain target risk parameters. The model output is a violation score or probability, used to indicate whether an image frame is suspicious.

[0063] It should be noted that due to the limited computing power of in-vehicle systems, they cannot support high-precision analysis of high-resolution images. Therefore, inputting raw resolution image frames into the target model for analysis may lead to a decrease in the frame rate, system lag, longer latency, and prolonged display of inappropriate content. Furthermore, since the core visual features of inappropriate content (e.g., human posture, weapon shape, skin color areas) are low- to mid-frequency features, these features can still be effectively extracted by the model at low resolution. However, high-frequency details in high-resolution images (e.g., texture, noise, compression artifacts) do not contribute to violation detection and instead increase the learning difficulty and inference noise of the model. For example, in a 1080P image, exposed areas can still be identified by skin color areas and human contours at a 224×224 resolution, while high-frequency information such as clothing texture and background details contribute little to violation detection.

[0064] Therefore, downscaling image frames from their original resolution to a lower resolution before inputting them into the target model directly reduces the amount of input data, thereby reducing computational consumption at the source. This ensures a significant reduction in computational overhead for model inference without substantial loss of detection accuracy, meeting the real-time requirements of in-vehicle systems. For example, scaling an image frame to a smaller resolution of 224×224 and inputting the scaled image frame into the target model allows the model to output a risk value for the image frame. This risk value represents the safety risk of the image frame; a higher risk value indicates a greater probability of violation.

[0065] In the embodiments of this application, the resolution of the image frame is adjusted to a preset first resolution, and the image frame at the first resolution is input into the pre-trained target model to obtain the first risk parameter. By reducing the resolution of the image frame from the original resolution to the first resolution, the computing power consumption of the vehicle system can be reduced, avoiding any impact on the normal operation of the vehicle system. In addition, reducing the resolution of the image frame allows the target model to focus on the global structure and core features of the image, suppressing the influence of high-frequency noise in high-resolution images on the detection process of the target model, improving the generalization ability of the target model, and reducing the risk of overfitting.

[0066] For example, if the first risk parameter is greater than the first threshold, the first risk parameter is determined as the target risk parameter; if the first risk parameter is less than or equal to the first threshold and the first risk parameter is greater than the second threshold, the resolution of the image frame is adjusted to a preset second resolution; the image frame with the second resolution is input into the target model to obtain the target risk parameter, wherein the second resolution is less than the original resolution of the image frame and greater than the first resolution.

[0067] Understandably, if the first risk parameter is greater than the first threshold, it means that the target model can detect the probability of a violation in a lower-resolution image frame. Therefore, the probability of a violation in the image frame is determined, and the first risk parameter is set as the target risk parameter to ensure that the target risk parameter can accurately represent the probability of a violation in the image frame. If the first risk parameter is less than or equal to the first threshold and greater than the second threshold, it means that the target model cannot accurately determine the probability of a violation in a lower-resolution image frame. Therefore, the resolution of the image frame is increased from the first resolution to the second resolution. By reducing the compression ratio of the down-resolution, the detection accuracy is improved, and misjudgment of the probability of a violation in the image frame is avoided in the case of excessive down-resolution and feature loss.

[0068] For example, the first threshold is 0.5, and the second threshold is 0.2. The original resolution image frame is down-resolutiond to obtain a 224×224 resolution image frame (i.e., the first resolution image frame). This 224×224 resolution image frame is input into the target model to obtain the first risk value. If the first risk value is 0.6, which is greater than the first threshold, it indicates that the image frame has a possibility of violation, and the first risk value is directly determined as the target risk value. When the first risk value is less than 0.2, it indicates that the image frame has no probability of violation. When the first risk value is 0.3, it is impossible to accurately determine whether the image frame has a possibility of violation; therefore, the original resolution image frame is down-resolutiond to 400×400 to obtain the second resolution image frame. This second resolution image frame is then input into the target model again to obtain the target risk parameter.

[0069] It should be noted that the above are illustrative examples of the first threshold, the first resolution, and the second resolution of the second threshold, used to describe the process of determining the target risk parameters. This application does not limit the specific values ​​of the first threshold, the first resolution, and the second resolution of the second threshold.

[0070] Optionally, after inputting the image frame at the second resolution into the target model, if the obtained target risk parameters are still between the first and second thresholds, it is determined that the image frame may have a violation. The image frame is then divided into blocks, and the image detection result is determined based on the image blocks and the image frame. Here, an image block is a specific region of interest (ROI) in the image frame, which is the key area for image analysis and processing.

[0071] In another implementation, determining the target risk parameters of an image frame includes: adjusting the resolution of the image frame to a preset first resolution, which is smaller than the original resolution of the image frame; and inputting the image frame at the first resolution into a pre-trained target model to obtain the target risk parameters.

[0072] It should be noted that the target risk parameter is the result of preliminary detection of image frames, which enables initial screening of image frames. Failure to perform preliminary screening based on the target risk parameter may result in high-precision detection of image frames that pose no security risk, leading to unnecessary computational consumption and causing vehicle infotainment system lag or malfunctions. Therefore, this application only performs further detection based on image frames and image blocks when the target risk parameter exceeds a first threshold, obtaining the target detection result. The process of obtaining the target detection result through further detection is explained below.

[0073] In one implementation, a first evaluation parameter corresponding to an image frame is determined, which is used to indicate whether the image frame has a violation risk; a second evaluation parameter corresponding to an image block is determined, which is used to indicate whether the image block includes an image element of the target type; and an image detection result is obtained based on the first evaluation parameter and the second evaluation parameter.

[0074] The target type of image elements refers to a predefined set of image elements that have illegal attributes and need to be detected and intervened. Target type image elements include violent image elements, pornographic image elements, and other types of image elements; violent image elements include weapons of attack such as knives, guns, and clubs, as well as bloody elements such as human blood, wounds, and dismembered limbs; pornographic image elements include exposed areas such as human torsos and private parts that violate public order and good morals; other types of image elements include horror and suspense image elements, image elements containing prohibited drugs, and image elements representing shady organizations.

[0075] For example, the first evaluation parameter is obtained by evaluating the entire image frame, and the second evaluation parameter is obtained by evaluating the image patches corresponding to the image frame. The first and second evaluation parameters are obtained using different detection methods. The first evaluation parameter refers to the result of evaluating the overall violation risk of the image frame using a target model, while the second evaluation parameter refers to the result of detecting image elements in each image patch corresponding to the image frame using a target detector. The target detector can be a lightweight single-detection model (You Only Look Once-Nano, YOLO-Nano). The YOLO Nano model, by optimizing the network structure and parameters, is used to identify specific targets in the image frame, such as knives, guns, clubs, and areas with significant bloodstains. It can accurately identify the borders, categories, and confidence levels of image elements, achieving high detection accuracy while reducing the model's computational complexity and memory usage.

[0076] It should be noted that determining the image detection result solely based on the first evaluation parameter (global violation risk) or solely based on the second evaluation parameter (local violation risk) of the image frame can easily lead to misjudgments. For example, if the target detection result is determined solely based on the first evaluation parameter, a misjudgment may occur if the image frame as a whole contains similar violation content due to lighting, color, or composition, but lacks specific violation elements (i.e., image elements of the target type). If the target detection result is determined solely based on the second evaluation parameter, a misjudgment may occur if there are violation elements in a local area of ​​the image (such as swords in a corner or a small area of ​​blood) but the image frame as a whole does not have a violation risk. For example, if there are swords or a small area of ​​blood, but the scene is a normal movie or medical scene, the normal scene may be misjudged as a violation scene, triggering false occlusion.

[0077] In the embodiments of this application, a first evaluation parameter for the image frame and a second evaluation parameter for the image block are determined, and the image detection result is determined by comprehensively considering the first evaluation parameter and the second evaluation parameter. This ensures that both the overall image frame and local regions of the image frame (i.e., image blocks of the image frame) are considered, thereby obtaining a more accurate target detection result and avoiding misjudgment of illegal content in the image frame.

[0078] In another implementation, the image semantics of the image frame are identified; based on the image semantics, the first evaluation parameter, and the second evaluation parameter, the image detection result is obtained.

[0079] For example, when identifying the semantics of an image frame, a lightweight deep learning model is used to extract and classify the global content features of the entire frame image, and finally output the semantic type corresponding to the image. The semantic types include violent semantics and pornographic semantics, etc.

[0080] For example, to pre-train a lightweight deep learning model (referred to as the lightweight model) on a training set to address the detection requirements of illegal semantics, screenshots are taken from the video stream output to the vehicle screen. The resulting image frames are scaled to the standard input size of the lightweight model, and the color gamut of the image frames is converted from RGB to the gamut used during model training (e.g., YCbCr or HSV) to reduce the impact of differences between screen lighting and the screen's color gamut. The pixel values ​​of the image frames are scaled to a specific range to match the requirements of the lightweight model (e.g., the specific range could be 0 to 1 or -1 to 1). Gaussian filtering is applied to noise such as compression artifacts and abrupt changes in lighting to improve the stability of feature extraction. After the lightweight model outputs confidence scores for various semantic types, the image semantics are determined according to pre-defined judgment rules. The determination rules include: if only a single semantic type has a confidence level greater than the confidence threshold (for example, the confidence threshold is 0.6), then the semantic type is determined to be image semantic; if multiple semantic types have a confidence level greater than the confidence threshold, then the semantic type with the highest confidence level is determined to be image semantic.

[0081] For example, when training the lightweight model, image frame samples commonly seen in in-vehicle scenarios are collected and labeled with semantic types, including illegal semantic types (violence, pornography, horror, etc.) and normal semantic types (fitness, swimwear, cartoons, daily life, etc.) to ensure that the samples cover the actual situation of different scenarios and different resolutions. The model is trained using these image frame samples, with the goal of having the model output confidence scores for various semantic types. The trained model is then lightweighted (where lightweighting refers to the process of converting the 32-bit floating-point numbers commonly used in deep learning models into 8-bit integers (INT8) to improve model running speed, reduce memory usage and power consumption). The lightweight model is then converted into a format that can be deployed in the vehicle's infotainment system and deployed in the vehicle's infotainment system.

[0082] It is understandable that using only the first evaluation parameter can easily lead to missed detections of local violations or misjudgments of globally similar scenes; using only the second evaluation parameter can easily lead to misjudgments due to isolated elements; and using only image semantics can easily lead to misjudgments due to abstract judgments lacking concrete evidence. Therefore, this application identifies the image semantics of image frames; and obtains image detection results based on the image semantics, the first evaluation parameter, and the second evaluation parameter. This ensures that three-dimensional fusion and mutual verification can be performed, thereby effectively avoiding missed and false positives of violations in image frames and ensuring relatively accurate target detection results.

[0083] It should be noted that common violations of semantics include two types: violent semantics (i.e., image semantics indicating that an image frame is a violent image) and pornographic semantics (i.e., image semantics indicating that an image frame is a pornographic image). When determining the target detection result, violent images and pornographic images can be determined using different methods. The two methods for determining the target detection result are further explained below.

[0084] In one implementation: if the image semantics indicate that the image frame is a violent image, and the first evaluation parameter indicates that the image frame has a violation risk and the image block includes image elements of the target type, then the image detection result indicates that the image frame includes violation content; if the first evaluation parameter indicates that the image frame does not have a violation risk, or the image block does not include image elements of the target type, then the image detection result indicates that the image frame does not include violation content.

[0085] Violent images refer to images that contain visual features that harm others or depict bloody and terrifying scenes, which are likely to have a negative impact on the user's physical and mental health.

[0086] For example, the visual characteristics of violent images include, but are not limited to, offensive weapons (e.g., knives, guns, sticks, controlled implements, etc.), physical conflict and injury characteristics (e.g., punching, kicking, dragging, fighting, weapon attacks, etc., or the presence of bloody features such as human injury, bleeding, and mutilated limbs), and horrific violence characteristics (e.g., horrific images with violence as their theme, bloody masks, and cartoon characters with violent themes).

[0087] For example, when the first evaluation parameter (the entire frame violation risk score) is greater than a first set threshold, it is determined that the image frame has a violation risk; when the second evaluation parameter (the detection confidence of the target type image element) is greater than a second set threshold, it is determined that the target type image element exists. When the first evaluation parameter is less than or equal to the first set threshold, it is determined that the image frame does not have a violation risk; when the second evaluation parameter is less than or equal to the second set threshold, it is determined that the target type image element does not exist.

[0088] For example, the first threshold is set to 0.6, the second threshold to 0.7, and the confidence threshold to 0.65. If a user plays an action movie clip while projecting the image onto a car screen, and a particular image frame shows two characters fighting with machetes in the street, with obvious physical conflict and knife displays, the lightweight model determines that the image frame is a violent image if the confidence level is greater than the confidence threshold. The entire frame classification model scores the image frame for violation risk, and the first evaluation parameter is 0.78, which is greater than the first threshold, indicating that the image frame has a violation risk. Image element detection is performed on the resulting image blocks obtained by segmenting the image frame. A knife element is detected in the block in row 2, column 3 with a confidence level of 0.85; a blood clot element is detected in the block in row 3, column 1 with a confidence level of 0.81. Since the confidence levels of both image elements are greater than the second threshold, the second evaluation parameter indicates that the image block contains a target type element (violent element), and the final image detection result indicates that the image frame contains illegal content.

[0089] For example, if the image frame is a kitchen cooking scene, with a character holding a knife to cut vegetables and no physical conflict, the first evaluation parameter is a violation risk score of 0.4, which is lower than the first set threshold of 0.6; and the confidence level of detecting the knife element is 0.85. The second evaluation parameter indicates that it contains elements of the target type. However, since the first evaluation parameter is at the first set threshold, even if the knife element is detected, it is determined to be a safe frame, thus avoiding misjudging normal life scenes as violations.

[0090] In the embodiments of this application, if the image semantics indicate that the image frame is a violent image, when the first evaluation parameter indicates global risk and the second evaluation parameter detects a target type violation element in the corresponding image block, the image detection result is determined to be that the image frame contains violation content, thereby reducing the probability of false alarms and false negatives in mixed scenes.

[0091] In another implementation: if the image semantics indicate that the image frame is a pornographic image, the weight coefficient is determined based on the proportion of skin color pixels in the image frame and the human body pose in the image frame; the image detection result is determined based on the first evaluation parameter, the second evaluation parameter and the weight coefficient.

[0092] Pornographic images refer to images that contain visual features such as excessive exposure of sensitive human body parts or physical postures that do not conform to public order and good morals, which are likely to have a negative impact on the physical and mental health of users.

[0093] For example, the visual features of pornographic images include, but are not limited to, the direct exposure of sensitive areas such as the human torso and private parts, or the indirect display of sensitive parts through clothing (e.g., overly revealing clothing); and the presence of body postures in the images that are intended to be sexually suggestive (e.g., intimate behavior that goes beyond the scope of social etiquette).

[0094] For example, when determining the weight coefficient, if the skin color pixel count is high (greater than the preset percentage) and the human body posture is an exposed posture such as lying down, the weight coefficient is determined to be a weight value greater than 1; if the skin color pixel count is high and the human body posture is a fitness or exercise posture, the weight coefficient is determined to be a weight value less than 1; if the skin color count is less than the preset percentage and the human body posture is a posture other than the above-mentioned exposed posture, fitness posture and exercise posture, the weight coefficient is determined to be 1.

[0095] For example, human body keypoints can be obtained through lightweight pose network models, and the human pose can be determined based on the positions of these keypoints. For instance, the MoveNet or BlazePose-Tiny models can be used to determine human pose; MoveNet and BlazePose-Tiny are lightweight artificial intelligence models used for analyzing human pose.

[0096] For example, when determining the image detection result based on the first evaluation parameter, the second evaluation parameter, and the weighting coefficient, the product of the first evaluation parameter and the weighting coefficient is determined as the weighted first evaluation parameter; the product of the second evaluation parameter and the weighting coefficient is determined as the weighted second evaluation parameter. If the weighted first evaluation parameter is greater than a first preset threshold and the weighted second evaluation parameter is greater than a second preset threshold, the image detection result indicates that there is illegal content in the image frame; conversely, if the weighted first evaluation parameter is less than or equal to the first preset threshold, or the weighted second evaluation parameter is less than or equal to the second preset threshold, the image detection result indicates that there is no illegal content in the image frame.

[0097] Optionally, when determining the image detection result based on the first evaluation parameter, the second evaluation parameter, and the weight coefficient, the product of the first evaluation parameter, the second evaluation parameter, and the weight coefficient can also be determined as the final risk value; if the final risk value is greater than the third set threshold, the image detection result indicates that the image frame contains illegal content; if the final risk value is less than the third set threshold, the image detection result indicates that the image frame does not contain illegal content.

[0098] It is understandable that in scenarios such as sports and fitness, athletes often have a high degree of skin exposure, but this does not constitute content that needs to be obscured. Judging solely based on the first and second evaluation parameters might lead to incorrectly identifying normal scenes in sports and fitness scenarios as containing inappropriate content, thus triggering false obscuration and affecting the user's viewing experience. Therefore, when an image frame is semantically identified as pornographic, a weighting coefficient is determined based on the proportion of skin color pixels and human posture. This coefficient is then combined with the first and second evaluation parameters to determine the image detection result. This ensures that the system can accurately determine whether a scene is normal based on the user's posture and the proportion of skin color pixels, avoiding the misclassification of normal scenes as containing inappropriate content and ensuring the accuracy of the image detection results.

[0099] S220: When the image detection result indicates that the image frame contains illegal content, the illegal content in the image frame is occluded to obtain the target video to be displayed.

[0100] For example, when the image detection result indicates that the image frame includes the illegal content, the image region corresponding to the illegal content is determined; the image region corresponding to the illegal content in the image frame is occluded to obtain the target video to be displayed.

[0101] For example, when the image detection result indicates that there is illegal content in the image frame, the bounding coordinates of the region corresponding to the target type element are determined based on the detection result of the second evaluation parameter. The bounding coordinates include the horizontal coordinate of the upper left corner, the vertical coordinate of the upper left corner, the width of the region, and the height of the region. Based on the bounding coordinates, the position that needs to be precisely occluded is determined to avoid interfering with the normal content in the image frame.

[0102] For example, when occluding inappropriate content in an image frame, different occlusion methods can be used depending on the type of inappropriate content. For instance, if the inappropriate content is a violent image element, a masking strip can be used. This generates a solid-color masking strip or a cartoon-style masking pattern that coordinates with the video background color. The size of the masking strip precisely matches the border coordinates of the inappropriate content to cover the area where the inappropriate content is located. If the inappropriate content is a pornographic target element, Gaussian blur can be used. A preset blur radius parameter is set, and a Gaussian blur algorithm is applied to the area where the inappropriate content is located, making the inappropriate content unrecognizable while preserving the overall visual continuity of the image.

[0103] For example, after occlusion processing, the occluded single-frame images are stitched together according to the original timing of the video stream, while synchronizing the audio stream data. During the stitching process, the original frame rate and resolution of the video are maintained to ensure smooth playback of the target video and avoid issues such as stuttering and frame loss.

[0104] In one implementation, if the image detection results of a consecutive preset number of image frames all indicate that the image frames contain illegal content, the illegal content in the consecutive preset number of image frames is occluded to obtain the target video to be displayed.

[0105] For example, a fixed-length sliding window is set, which includes a preset number of image frames, each of which corresponds to a target detection result. If all preset number of image frames in the sliding window contain illegal content, then all preset number of image frames in the sliding window are occluded. If only a single image frame in the sliding window contains illegal content, then it is considered a false detection caused by noise or compression screen distortion.

[0106] For example, if the preset number is 3 frames, if the image detection results of 3 consecutive image frames all indicate that the image frames contain illegal content, the illegal content in the 3 consecutive image frames will be occluded; if only a single image frame contains illegal content, and there is no illegal content in the next image frame of that single image frame, then it is determined that the image detection result of that single image frame has an error, and that single image frame will not be occluded.

[0107] In the embodiments of this application, since the presence of illegal content in the video to be displayed usually occurs in consecutive image frames, if only a single frame's image detection result indicates the presence of illegal content, it may be a misjudgment of the single frame's image detection result due to noise or compression artifacts. Therefore, this application ensures that the image detection results for a predetermined number of consecutive image frames all indicate that the image frames contain illegal content, thus occluding the illegal content in consecutive image frames. Compared to occlusion processing based on the image detection results of a single image frame, this method helps filter out transient compression noise and does not immediately trigger occlusion. Furthermore, for illegal content that persists across multiple frames, it can accumulate evidence from multiple frames, ensuring more accurate occlusion.

[0108] S230, displaying the target video to be displayed.

[0109] For example, the target video to be displayed after the occlusion process is output to the vehicle's infotainment screen and played synchronously according to the original playback parameters of the video to be displayed (playback parameters include: frame rate, resolution, and volume); wherein, the vehicle's infotainment screen includes, but is not limited to, playback terminals that support user switching through the vehicle's operating interface, such as the vehicle's central control screen, passenger screen, and rear entertainment screen.

[0110] For example, the target screen of the vehicle can be determined based on the video playback command, and the target video to be displayed can be displayed on the target screen. For instance, if the video playback command triggers a playback command for the central control screen, the target video to be displayed is directly output to the vehicle's central control screen, adapting to the screen ratio of the central control screen and automatically adjusting the screen layout to avoid stretching and distortion; if the user triggers a playback command for the rear entertainment screen, the video stream is pushed to the rear entertainment screen through the vehicle's multi-screen interaction protocol, while supporting independent volume adjustment for the front and rear seats to avoid interfering with the driver's operation.

[0111] Optionally, while displaying the target video to be displayed, the vehicle-mounted system can also display a content protection status indicator (e.g., "Child mode is enabled" or "Inappropriate and sensitive content is obscured") at a preset location on the playback interface. This allows users to intuitively confirm the current video's protection status. Simultaneously, it records playback logs, including information such as the playback terminal, playback duration, and the number of obscured image frames, facilitating subsequent querying and tracing.

[0112] In one implementation, when the vehicle's infotainment system activates child mode, it can detect inappropriate content in the video to be displayed and obscure it if such content is detected. For example, child mode can be activated manually by the user to detect inappropriate content in the video, or it can be automatically activated and activated when a child is detected in the vehicle, thus preventing inappropriate content from negatively impacting the child's physical and mental health.

[0113] In the above embodiments, before displaying the video to be displayed, image frames in the video are detected to obtain image detection results. This ensures that whether the image frames contain illegal content can be detected in advance before the video is displayed. If the image frame contains illegal content, the illegal content is occluded to ensure effective filtering of illegal information in the video stream and to prevent illegal content from affecting the physical and mental health of passengers. Furthermore, this application only precisely occludes illegal content, rather than globally occluding or prohibiting the output of the video to be displayed. This ensures that non-illegal content in the video frames can be displayed normally, avoiding the inability to display the video due to global intervention. Thus, it can effectively filter illegal content while taking into account the user's viewing experience.

[0114] Figure 3 This is a schematic flowchart illustrating another vehicle screen display method provided in this application embodiment.

[0115] For example, Figure 3 The display method 300 shown can be executed by the vehicle's infotainment system, or by the vehicle itself, or by the vehicle's processor or chip.

[0116] like Figure 3As shown, the vehicle screen display method 300 includes S301 to S315. The vehicle screen display methods shown in S301 to S315 will be described in detail below.

[0117] S301, a video playback command was detected in the vehicle.

[0118] For example, a video playback command is a control command to start or continue playing video content, used to trigger the decoding and display process of the video stream to be displayed; the command usually includes information such as the source identifier of the playback content, the playback start time, and the playback window parameters.

[0119] S302, the video to be displayed is captured to obtain the image frames of the video to be displayed.

[0120] For example, the video to be displayed is captured according to a preset capture period to obtain image frames of the video to be displayed; wherein, the preset period can be a fixed value set in advance, or it can be a value dynamically determined according to the performance of the vehicle's in-vehicle infotainment system.

[0121] S303 adjusts the resolution of the image frame to the first resolution.

[0122] For example, the resolution of the image frame is adjusted from the original resolution to a first resolution, which is smaller than the original resolution. By downscaling the image frame and inputting the image with the resolution reduced to the first resolution into the target model, the amount of input data to the model can be directly reduced, reducing computational power consumption from the source. This ensures that the computational cost of model inference is significantly reduced without significantly sacrificing detection accuracy, thus meeting the real-time requirements of in-vehicle systems.

[0123] S304, input the image frame of the first resolution into the target model to obtain the target risk parameters.

[0124] For example, after inputting an image frame of the first resolution into the target model, the first risk parameter of the target model can be directly determined as the target risk parameter; or the first risk parameter output by the target model can be judged, and the target risk parameter can be determined based on the judgment result.

[0125] For example, if the first risk parameter is greater than the first threshold, the first risk parameter is determined as the target risk parameter; if the first risk parameter is less than or equal to the first threshold and the first risk parameter is greater than the second threshold, the resolution of the image frame is adjusted to the preset second resolution; the image frame with the second resolution is input into the target model again, and the value output by the target model is determined as the target risk parameter.

[0126] S305, if the target risk parameter is greater than the first threshold, the image frame is divided into blocks to obtain multiple image blocks.

[0127] For example, if the target risk parameter is greater than the first threshold, the image frame can be evenly divided into multiple image blocks according to a fixed number of rows and columns; or the image frame can be divided into blocks by a sliding window to obtain multiple image blocks.

[0128] Optionally, the number of blocks for segmented processing can be determined based on the performance data of the vehicle system. If the vehicle's computing power is low and system resources are consumed in large quantities, the number of blocks is determined as the first number of blocks. If the vehicle's computing power is high and system resources are consumed in small quantities, the number of blocks is determined as the second number of blocks. The second number of blocks is greater than the first number of blocks to ensure that the detection process of illegal content will not affect the normal operation of the vehicle system.

[0129] S306, determine the first evaluation parameter corresponding to the image frame and the second evaluation parameter corresponding to the image block.

[0130] For example, the first evaluation parameter is obtained by evaluating the image frame as a whole, and the second evaluation parameter is obtained by evaluating the image blocks corresponding to the image frame. The first evaluation parameter refers to the result obtained by evaluating the violation risk of the image frame as a whole using the target model, and the second evaluation parameter refers to the result obtained by detecting the image elements of each image block corresponding to the image frame using the target detector.

[0131] S307 identifies the image semantics of image frames.

[0132] For example, a lightweight deep learning model is used to extract and classify the global content features of the entire frame image, and finally output the semantic type corresponding to the image. The semantic type includes violent semantics and pornographic semantics, etc.

[0133] S308, if the image semantics indicate that the image frame is a violent image, the image detection result is obtained according to the first evaluation parameter and the second evaluation parameter.

[0134] For example, if the image semantics indicate that the image frame is a violent image, and the first evaluation parameter indicates that the image frame has a violation risk and the image block includes image elements of the target type, the image detection result indicates that the image frame includes violation content; if the first evaluation parameter indicates that the image frame does not have a violation risk, or the image block does not include image elements of the target type, the image detection result indicates that the image frame does not include violation content.

[0135] S309, if the image semantics indicate that the image frame is a pornographic image, obtain the weight coefficient based on the proportion of human body posture and skin color pixels.

[0136] For example, if the image semantics indicate that the image frame is a pornographic image, when the skin color pixel count is high (greater than the preset percentage) and the human body posture is an exposed posture, the weight coefficient is determined to be a weight value greater than 1; when the skin color pixel count is high and the human body posture is a fitness or sports posture, the weight coefficient is determined to be a weight value less than 1; when the skin color count is less than the preset percentage and the human body posture is a posture other than the above-mentioned exposed posture, fitness posture and sports posture, the weight coefficient is determined to be 1.

[0137] S310, determine the image detection result based on the weighting coefficient, the first evaluation parameter, and the second evaluation parameter.

[0138] For example, the product of the first evaluation parameter and the weight coefficient is determined as the weighted first evaluation parameter; the product of the second evaluation parameter and the weight coefficient is determined as the weighted second evaluation parameter. If the weighted first evaluation parameter is greater than a first set threshold and the weighted second evaluation parameter is greater than a second set threshold, the image detection result indicates that there is illegal content in the image frame; otherwise, the image detection result indicates that there is no illegal content in the image frame.

[0139] Alternatively, the implementation methods of S301 to S308 can be found in [reference needed]. Figure 2 The relevant description of S210 will not be repeated here.

[0140] S311, Does the image detection result indicate that the image frame contains illegal content? If yes, proceed to S312; if no, proceed to S313.

[0141] For example, determine whether the image detection result indicates that the image frame contains illegal content; if the image detection result indicates that the image frame contains illegal content, determine the image region corresponding to the illegal content; if the image detection result indicates that the image frame does not contain illegal content, then display the image frame.

[0142] S312, Identify the image region corresponding to the illegal content.

[0143] For example, if there is illegal content in the image frame, the bounding coordinates of the region corresponding to the target type element are determined according to the detection result of the second evaluation parameter. The bounding coordinates include the horizontal coordinate of the upper left corner, the vertical coordinate of the upper left corner, the width of the region, and the height of the region. The region corresponding to the bounding coordinates is the image region corresponding to the illegal content.

[0144] S313 displays the image frame.

[0145] For example, if there is no illegal content in the image frame, the image frame is output to the vehicle's screen for display.

[0146] S314, the image area corresponding to the violation content is occluded to obtain the target video to be displayed.

[0147] Alternatively, the implementation methods of S311 to S314 can be found in [reference needed]. Figure 2 The relevant descriptions of the S220 are not repeated here.

[0148] S315, displaying the target video to be displayed.

[0149] For example, the target video to be displayed after the occlusion process is output to the vehicle screen and played synchronously according to the original playback parameters of the video to be displayed (playback parameters include: frame rate, resolution, volume).

[0150] Alternatively, the implementation of S315 can be found in [reference needed]. Figure 2 The relevant descriptions of the S230 are not repeated here.

[0151] In the embodiments of this application, the resolution of the image frame is adjusted to a first resolution, and the image frame at the first resolution is input into the target model. This allows the target model to focus on the global structure and core features of the image, suppressing the impact of high-frequency noise in the high-resolution image on the detection process of the target model, improving the generalization ability of the target model, and reducing the risk of overfitting. Image frames are initially screened based on target risk parameters to determine whether an image frame is suspicious. Only when the target risk parameter is greater than a first threshold (i.e., the image frame is suspicious) is the image frame segmented, ensuring that only suspicious image frames are segmented and detected with higher precision, reducing the computational power consumption of the vehicle system. During detection, image semantics, the first evaluation parameter, and the second evaluation parameter are considered simultaneously to ensure relatively accurate image detection results. When illegal content is detected, the image area containing the illegal content is occluded to prevent the illegal content from adversely affecting the physical and mental health of passengers, thereby improving the user experience.

[0152] The above text combined Figures 1 to 3 The display method for a vehicle screen provided in the embodiments of this application is described in detail below; the following will be combined with Figure 4 and Figure 5 The apparatus embodiments of this application are described in detail below. It should be understood that the apparatus in the embodiments of this application can perform the various methods described in the foregoing embodiments of this application, that is, the specific working processes of the various products described below can be referred to the corresponding processes in the foregoing method embodiments.

[0153] Figure 4 This is a schematic diagram of the structure of a vehicle screen display device provided in an embodiment of this application.

[0154] For example, such as Figure 4 As shown, the display device 400 of the vehicle screen includes: The detection module 410 is used to detect image frames in the video to be displayed in response to a video playback command in the vehicle, and obtain image detection results. The image detection results are used to indicate whether the image frame contains illegal content. The processing module 420 is used to occlude the illegal content in the image frame when the image detection result indicates that the image frame contains illegal content, thereby obtaining the target video to be displayed; and to display the target video to be displayed.

[0155] Optionally, as an embodiment, the detection module 410 is specifically used to: determine the target risk parameter of the image frame; if the target risk parameter is greater than a first threshold, divide the image frame into blocks to obtain at least two image blocks; and obtain the image detection result based on the image frame and the image blocks.

[0156] Optionally, as an embodiment, the detection module 410 is specifically used to: adjust the resolution of the image frame to a preset first resolution, the first resolution being smaller than the original resolution of the image frame; input the image frame at the first resolution into a pre-trained target model to obtain a first risk parameter; and obtain a target risk parameter based on the first risk parameter.

[0157] Optionally, as an embodiment, the detection module 410 is specifically used for: if the first risk parameter is greater than the first threshold, determining the first risk parameter as the target risk parameter; if the first risk parameter is less than or equal to the first threshold and the first risk parameter is greater than the second threshold, adjusting the resolution of the image frame to a preset second resolution; inputting the image frame with the second resolution into the target model to obtain the target risk parameter, wherein the second resolution is less than the original resolution of the image frame and greater than the first resolution.

[0158] Optionally, as an embodiment, the detection module 410 is specifically used to: determine a first evaluation parameter corresponding to the image frame, the first evaluation parameter being used to indicate whether the image frame has a violation risk; determine a second evaluation parameter of the image block corresponding to the image frame, the second evaluation parameter being used to indicate whether the image block includes image elements of the target type; and obtain an image detection result based on the first evaluation parameter and the second evaluation parameter.

[0159] Optionally, as an embodiment, the detection module 410 is specifically used to: identify the image semantics of the image frame; and obtain the image detection result based on the image semantics, the first evaluation parameter, and the second evaluation parameter.

[0160] Optionally, as an embodiment, the detection module 410 is specifically used to: if the image semantics indicate that the image frame is a violent image, and the first evaluation parameter indicates that the image frame has a violation risk and the image block includes image elements of the target type, determine that the image detection result indicates that the image frame includes violation content; and if the first evaluation parameter indicates that the image frame does not have a violation risk, or the image block does not include image elements of the target type, determine that the image detection result indicates that the image frame does not include violation content.

[0161] Optionally, as an embodiment, the detection module 410 is specifically used to: if the image semantics indicate that the image frame is a pornographic image, determine the weight coefficient based on the proportion of skin color pixels in the image frame and the human body posture in the image frame; and determine the image detection result based on the first evaluation parameter, the second evaluation parameter and the weight coefficient.

[0162] Optionally, as an embodiment, the processing module 420 is specifically used to: if the image detection results corresponding to a consecutive preset number of image frames all indicate that the image frames contain illegal content, to perform occlusion processing on the illegal content in the consecutive preset number of image frames, thereby obtaining the target video to be displayed.

[0163] It should be noted that the display device of the aforementioned vehicle screen is embodied in the form of a functional unit. The term "module" here can be implemented in software and / or hardware, without specific limitations.

[0164] For example, a "module" can be a software program, hardware circuit, or a combination of both that implements the above functions. Hardware circuits may include application-specific integrated circuits (ASICs), electronic circuits, processors (e.g., shared processors, proprietary processors, or group processors) and memory for executing one or more software or firmware programs, combined logic circuits, and / or other suitable components that support the described functions.

[0165] Therefore, the units of the various examples described in the embodiments of this application can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0166] Figure 5 This is a schematic diagram of the structure of a vehicle provided in an embodiment of this application.

[0167] For example, vehicle 500 includes processor 510, memory 520 and executable program code 530.

[0168] For example, vehicle 500 includes one or more processors 510 that can support the display method of the vehicle screen in the method embodiment of vehicle 500. Processor 510 can be a general-purpose processor or a special-purpose processor. For example, processor 510 can be a central processing unit (CPU), digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or other programmable logic device, such as discrete gate, transistor logic device, or discrete hardware component.

[0169] For example, processor 510 can be used to control vehicle 500, execute software programs, and process data from the software programs. Vehicle 500 may also include a communication unit for receiving and transmitting signals.

[0170] For example, the vehicle 500 may include one or more memories 520, on which executable program code 530 is stored. The executable program code 530 can be run by the processor 510 to generate instructions, causing the processor 510 to execute the vehicle screen display method described in the above method embodiment according to the instructions.

[0171] Optionally, the memory 520 may also store data. Optionally, the processor 510 may also read data stored in the memory 520, which may be stored at the same memory address as the executable program code 530, or the data may be stored at a different memory address than the executable program code 530.

[0172] For example, the processor 510 and memory 520 can be configured separately or integrated together, for example, integrated on a system-on-a-chip of the terminal device.

[0173] For example, the memory 520 can be used to store related programs of the vehicle screen display method provided in the embodiments of this application, and the processor 510 can be used to call the executable program code 530 stored in the memory 520 when controlling the vehicle to execute the vehicle screen display method of the embodiments of this application; for example, in response to a video playback command in the vehicle, the image frames in the video to be displayed are detected to obtain image detection results, and the image detection results are used to indicate whether the image frame includes illegal content; when the image detection results indicate that the image frame includes illegal content, the illegal content in the image frame is occluded to obtain the target video to be displayed; and the target video to be displayed is displayed.

[0174] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the vehicle screen display method of any of the foregoing embodiments.

[0175] The computer-readable storage medium may include, but is not limited to, any type of disk, including floppy disks, optical disks, Digital Video Discs (DVDs), Compact Disc Read-Only Memory (CD-ROMs), microdrives, and magneto-optical disks, read-only memory (ROMs), random access memory (RAMs), erasable programmable read-only memory (EPROMs), electrically erasable programmable read-only memory (EEPROMs), dynamic random access memory (DRAMs), video random access memory (VRAMs), flash memory devices, magnetic cards or optical cards, nanosystems (including molecular memory ICs), or any type of medium or device suitable for storing instructions and / or data.

[0176] This application also provides a computer program product that, when run on a computer, causes the computer to perform the aforementioned related steps to implement a vehicle screen display method as described in the above embodiments.

[0177] In addition, the vehicle provided in the embodiments of this application may specifically be a chip, component or module, and the vehicle may include a connected processor and a memory; wherein, the memory is used to store instructions, and the processor can call and execute the instructions to cause the chip to execute a vehicle screen display method in the above embodiments.

[0178] The vehicle, computer-readable storage medium, computer program product, or chip provided in this application are all used to execute the corresponding vehicle screen display method provided above. Therefore, the beneficial effects that can be achieved can be referred to the beneficial effects in the corresponding vehicle screen display method provided above, and will not be repeated here.

[0179] Through the above description of the embodiments, those skilled in the art will understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0180] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0181] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A display method for a vehicle screen, characterized in that, The display method includes: In response to a video playback command in the vehicle, image frames in the video to be displayed are detected to obtain image detection results, which are used to indicate whether the image frame contains illegal content; When the image detection result indicates that the image frame contains the illegal content, the illegal content in the image frame is occluded to obtain the target video to be displayed. The target video to be displayed is shown.

2. The display method according to claim 1, characterized in that, The process of detecting image frames in the video to be displayed to obtain image detection results includes: Determine the target risk parameters of the image frame; If the target risk parameter is greater than the first threshold, the image frame is divided into blocks to obtain at least two image blocks; The image detection result is obtained based on the image frame and the image block.

3. The display method according to claim 2, characterized in that, Determining the target risk parameters of the image frame includes: The resolution of the image frame is adjusted to a preset first resolution, which is smaller than the original resolution of the image frame; The first resolution image frame is input into the pre-trained target model to obtain the first risk parameter; The target risk parameter is obtained based on the first risk parameter.

4. The method according to claim 3, characterized in that, The process of obtaining the target risk parameter based on the first risk parameter includes: If the first risk parameter is greater than the first threshold, the first risk parameter is determined as the target risk parameter; If the first risk parameter is less than or equal to the first threshold and the first risk parameter is greater than the second threshold, the resolution of the image frame is adjusted to a preset second resolution; the image frame with the second resolution is input into the target model to obtain the target risk parameter, wherein the second resolution is less than the original resolution of the image frame and greater than the first resolution.

5. The display method according to claim 2, characterized in that, Determining the image detection result based on the image frame and the image patch includes: Determine a first evaluation parameter corresponding to the image frame, the first evaluation parameter being used to indicate whether the image frame has a risk of violation; Determine a second evaluation parameter for the image block corresponding to the image frame, the second evaluation parameter being used to indicate whether the image block includes an image element of the target type; The image detection result is obtained based on the first evaluation parameter and the second evaluation parameter.

6. The method according to claim 5, characterized in that, The display method further includes: Identify the image semantics of the image frame; The process of obtaining the image detection result based on the first evaluation parameter and the second evaluation parameter includes: The image detection result is obtained based on the image semantics, the first evaluation parameter, and the second evaluation parameter.

7. The display method according to claim 6, characterized in that, The process of obtaining the image detection result based on the image semantics, the first evaluation parameter, and the second evaluation parameter includes: If the image semantics indicate that the image frame is a violent image, and the first evaluation parameter indicates that the image frame has a violation risk, and the image block includes image elements of the target type, then the image detection result indicates that the image frame includes the violation content. When the first evaluation parameter indicates that the image frame does not have a violation risk, or the image block does not contain image elements of the target type, the image detection result is determined to indicate that the image frame does not contain the violation content.

8. The display method according to claim 6, characterized in that, The process of obtaining the image detection result based on the image semantics, the first evaluation parameter, and the second evaluation parameter includes: If the image semantics indicate that the image frame is a pornographic image, a weighting coefficient is determined based on the proportion of skin color pixels in the image frame and the human body posture in the image frame; The image detection result is determined based on the first evaluation parameter, the second evaluation parameter, and the weighting coefficient.

9. The display method according to any one of claims 1 to 8, characterized in that, When the image detection result indicates that the image frame contains the illegal content, the illegal content in the image frame is occluded to obtain the target video to be displayed, including: If the image detection results of a consecutive preset number of image frames all indicate that the image frames contain the illegal content, the illegal content in the preset number of image frames is occluded to obtain the target video to be displayed.

10. A vehicle, characterized in that, The vehicles include: Memory, used to store executable program code; A processor for calling and running the executable program code from the memory, causing the vehicle to perform the display method as described in any one of claims 1 to 9.

Images