Vehicle child mode detection method, device and equipment, storage medium and computer program product

Abstract

The invention relates to the technical field of automobiles, in particular to a vehicle child mode detection method, device and equipment, a storage medium and a computer program product. The method comprises the following steps: performing face detection on a back row passenger image based on a preset face detection deep learning model to obtain passenger face information and confidence information; determining a face posture angle according to face key point data in the passenger face information; judging whether a face corresponding to the passenger face information is a front face or not based on the face posture angle and the confidence degree information; if yes, age estimation is carried out on the front face image based on a preset age model, and the preset age model is constructed based on KL offline loss and average difference loss; if the age estimation results are all smaller than the preset child age threshold value, the vehicle is controlled to start the child mode, and the child riding safety is improved.

Description

Technical Field

[0001] The present application relates to the technical field of automobiles, and particularly to a method, device, equipment, storage medium and computer program product for detecting a vehicle child mode. Background Art

[0002] With the rapid development of intelligent vehicle technology, the intelligent and user-friendly design of the vehicle interior environment has gradually become the focus of market attention. Among them, in order to improve the safety and comfort of child passengers in the vehicle, many vehicle models have introduced the child mode function. Existing child detection technologies mainly include pressure sensor detection method, seat belt induction detection method and visual sensor detection method. In the pressure sensor detection method and the seat belt induction detection method, the pressure sensor and the seat belt induction detection method highly depend on the child seat or specific hardware configuration. When a child does not use a special seat or does not fasten the seat belt correctly, the detection function will fail. Most of the visual sensor detection methods only focus on the morphological features of the occupant or simple face recognition technology, but have weak adaptability to the changes in the posture and angle of the occupant, and are easily affected by light, occlusion and complex scenes, resulting in high false detection rate and missed detection rate, and poor safety of children riding in the vehicle. Therefore, how to improve the safety of children riding in the vehicle has become an urgent problem to be solved. Summary of the Invention

[0003] The main purpose of the present application is to provide a method, device, equipment, storage medium and computer program product for detecting a vehicle child mode, aiming to solve the technical problem of how to improve the safety of children riding in the vehicle.

[0004] To achieve the above purpose, the present application provides a method for detecting a vehicle child mode, and the method includes the following steps:

[0005] Based on a preset deep learning model for face detection, perform face detection on the rear row occupant image to obtain occupant face information and confidence information;

[0006] Determine the face pose angle according to the face key point data in the occupant face information;

[0007] Based on the face pose angle and the confidence information, determine whether the face corresponding to the occupant face information is a frontal face;

[0008] If so, based on a preset age model, perform age estimation on the frontal face image, where the preset age model is constructed based on KL offline loss and mean difference loss;

[0009] If the age estimation results are all less than the preset child age threshold, control the vehicle to turn on the child mode.

[0010] In one embodiment, before the step of controlling the vehicle to turn on the child mode if all the age estimation results are less than the preset child age threshold, the following steps are further included:

[0011] If all the age estimation results are greater than the preset child age threshold, control the vehicle to maintain the current mode;

[0012] If there are values greater than the preset child age threshold and values less than the preset child age threshold in the age estimation results at the same time, send a signal suggesting to turn on the child mode to the vehicle control terminal.

[0013] In one embodiment, after the step of controlling the vehicle to turn on the child mode if all the age estimation results are less than the preset child age threshold, the following steps are further included:

[0014] Turn on the preset child safety lock;

[0015] Adjust the in-vehicle environment parameters, where the in-vehicle environment parameters include the air volume of the air conditioner and the volume of the speaker;

[0016] Push the preset child entertainment content to the preset in-vehicle display device and turn on the preset rear row monitoring device.

[0017] In one embodiment, the step of performing face detection on the rear row occupant image based on the preset face detection deep learning model to obtain the occupant face information and the confidence information includes:

[0018] Scan the rear row occupant image based on the preset face detection deep learning model to generate a plurality of rectangular bounding boxes;

[0019] Based on the comparison results of each rectangular bounding box with the preset bounding box confidence threshold, screen the plurality of rectangular bounding boxes to obtain the key bounding boxes;

[0020] Adopt a key point detection algorithm to extract the facial feature points in the key bounding boxes to obtain the occupant face information, and use the confidence corresponding to the facial feature points as the confidence information.

[0021] In one embodiment, the step of determining the face pose angle according to the face key point data in the occupant face information includes:

[0022] Extract the face key point data in the occupant face information, where the face key point data includes the eye corner feature point data, the nose tip feature point data, and the mouth corner feature point data;

[0023] Based on the eye corner feature point data, the nose tip feature point data, and the mouth corner feature point data, adopt the Euler angle algorithm to define the key point coordinates, determine the horizontal distance and the vertical distance, determine the roll angle, determine the pitch angle, and determine the yaw angle;

[0024] Determine the face pose angle according to the key point coordinates, the horizontal distance and the vertical distance, the roll angle, the pitch angle, and the yaw angle.

[0025] In one embodiment, the step of determining whether the face corresponding to the occupant face information is a frontal face based on the face pose angle and the confidence information includes:

[0026] Obtain the confidence of the eye corner feature point data, the confidence of the nose tip feature point data, and the confidence of the mouth corner feature point data respectively;

[0027] Compare the confidence of the eye corner feature point data, the confidence of the nose tip feature point data, and the confidence of the mouth corner feature point data with a preset pose confidence threshold respectively;

[0028] Determine whether the roll angle, the pitch angle, and the yaw angle are all within a preset pose angle range;

[0029] Based on the comparison result and the determination result, determine whether the face corresponding to the occupant face information is a frontal face.

[0030] In addition, to achieve the above object, the present application also proposes a vehicle child mode detection device, which includes:

[0031] A face detection module, configured to perform face detection on the rear row occupant image based on a preset face detection deep learning model to obtain occupant face information and confidence information;

[0032] A pose angle module, configured to determine a face pose angle according to the face key point data in the occupant face information;

[0033] A frontal face judgment module, configured to determine whether the face corresponding to the occupant face information is a frontal face based on the face pose angle and the confidence information;

[0034] An age estimation module, configured to, if so, perform age estimation on the frontal face image based on a preset age model, where the preset age model is constructed based on KL offline loss and mean difference loss;

[0035] A target module, configured to control the vehicle to turn on the child mode if the age estimation results are all less than a preset child age threshold.

[0036] In addition, to achieve the above object, the present application further provides a vehicle child mode detection device, which includes: a memory, a processor, and a vehicle child mode detection program stored on the memory and executable on the processor, and the vehicle child mode detection program is configured to implement the steps of the vehicle child mode detection method as described above.

[0037] In addition, to achieve the above object, the present application further provides a storage medium, on which a vehicle child mode detection program is stored, and when the vehicle child mode detection program is executed by a processor, it implements the steps of the vehicle child mode detection method as described above.

[0038] In addition, to achieve the above object, the present application further provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the steps of the vehicle child mode detection method as described above.

[0039] Based on a preset face detection deep learning model, the present application performs face detection on the rear row occupant image to obtain occupant face information and confidence information; determines the face pose angle according to the face key point data in the occupant face information; based on the face pose angle and the confidence information, determines whether the face corresponding to the occupant face information is a frontal face; if so, based on a preset age model, performs age estimation on the frontal face image, where the preset age model is constructed based on KL offline loss and mean difference loss; if the age estimation results are all less than a preset child age threshold, controls the vehicle to turn on the child mode. The present application performs face detection on the rear row occupant image through a preset face detection deep learning model, calculates the face pose angle using the key point data and combines the confidence information to determine whether it is a frontal face, ensuring the reliability of the input data; performs age estimation on the frontal face image based on a preset age model constructed based on KL offline loss and mean difference loss, and determines whether it is less than the preset child age threshold, so as to control the vehicle to automatically turn on the child mode, improving the safety of children's riding in the vehicle. Description of the Drawings

[0040] Figure 1 It is a schematic flowchart of the first embodiment of the vehicle child mode detection method of the present application;

[0041] Figure 2 It is a schematic sub - flowchart of the second embodiment of the vehicle child mode detection method of the present application;

[0042] Figure 3 It is a schematic sub - flowchart of the third embodiment of the vehicle child mode detection method of the present application;

[0043] Figure 4 It is a child mode monitoring flowchart of an embodiment of the vehicle child mode detection method of the present application;

[0044] Figure 5 It is a schematic diagram of the module structure of the vehicle child mode detection device according to an embodiment of the present application;

[0045] Figure 6 It is a schematic diagram of the device structure of the hardware operating environment involved in the vehicle child mode detection method according to an embodiment of the present application.

[0046] The implementation, functional features, and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. Specific Embodiments

[0047] It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0048] To better understand the technical solution of the present application, the following will be described in detail in combination with the accompanying drawings of the specification and specific embodiments.

[0049] It should be noted that with the rapid development of intelligent vehicle technology, the intelligent and user-friendly design of the vehicle interior environment has gradually become the focus of market attention. Among them, in order to improve the safety and comfort of child passengers in the vehicle, many vehicle models have introduced the child mode function. Existing child detection technologies mainly include the pressure sensor detection method, the seat belt induction detection method, and the vision sensor detection method. In the pressure sensor detection method and the seat belt induction detection method, the pressure sensor and the seat belt induction detection method highly rely on the child seat or specific hardware configuration. When the child does not use the special seat or does not fasten the seat belt correctly, the detection function will fail. Most of the vision sensor detection methods only focus on the occupant morphological features or simple face recognition technology, but have weak adaptability to the changes in the occupant's posture and angle, and are easily affected by light, occlusion, and complex scenes, resulting in high false detection rates and missed detection rates, and poor child riding safety. Therefore, how to improve child riding safety has become an urgent problem to be solved.

[0050] The main solution of the present application is: based on a preset face detection deep learning model, perform face detection on the rear row occupant image to obtain occupant face information and confidence information; determine the face pose angle according to the face key point data in the occupant face information; based on the face pose angle and the confidence information, determine whether the face corresponding to the occupant face information is a frontal face; if so, based on a preset age model, perform age estimation on the frontal face image, where the preset age model is constructed based on the KL offline loss and the mean difference loss; if the age estimation results are all less than the preset child age threshold, control the vehicle to turn on the child mode.

[0051] In this application, a preset deep learning model for face detection is used to detect the faces in the rear occupant images. The face pose angle is calculated using the key point data, and combined with the confidence information to determine whether it is a frontal face, ensuring the reliability of the input data. A preset age model constructed based on the KL offline loss and the mean difference loss estimates the age of the frontal face image and determines whether it is less than the preset child age threshold, thereby controlling the vehicle to automatically turn on the child mode, improving the safety of children riding in the vehicle.

[0052] It should be noted that the execution subject of the method in this embodiment can be a computing service device with data processing, network communication, and program running functions, or the above-mentioned vehicle child mode detection device with the same or similar functions. This embodiment and the following embodiments will be described by taking the vehicle child mode detection device as an example.

[0053] Based on this, the first embodiment of the vehicle child mode detection method of this application is proposed. Please refer to Figure 1 , Figure 1 which is the flowchart of the first embodiment of the vehicle child mode detection method of this application.

[0054] In this embodiment, the vehicle child mode detection method includes the following steps:

[0055] S1: Based on a preset deep learning model for face detection, detect the faces in the rear occupant images to obtain occupant face information and confidence information;

[0056] It should be noted that the preset deep learning model for face detection refers to a model pre-trained based on the deep neural network algorithm, which can automatically detect the face region and related features from the image. This model is trained with a large amount of face data and has the advantages of high recognition accuracy and strong generalization ability. The rear occupant image is the image data of the rear seat area captured by the on-vehicle camera, which is used to analyze the facial features and positions of the occupants. Face detection is a computer vision technology that identifies the position of the face in the image and marks the face region in the form of a rectangular bounding box. The occupant face information is the face-related data extracted from the detection results, including the bounding box coordinates, the positions of key points (such as the corners of the eyes, the tip of the nose, etc.), and the feature attributes (such as gender, age, etc.). The confidence information is the credibility score of the model for the detection result, indicating the reliability of the recognition result, usually ranging from 0 to 1. The higher the confidence, the more reliable the detection result.

[0057] Specifically, the camera installed in the rear row of the vehicle collects the images of the occupants in real time and transmits the image data to the vehicle-mounted controller for analysis. The camera covers the rear seat area to ensure no obstruction within the visible range and maintains the clarity and stability of the images. The collected images are processed such as format conversion, size scaling, color normalization, etc. to adapt to the input requirements of the deep learning model. At the same time, for images with poor lighting conditions, enhancement algorithms are applied to improve the detection effect.

[0058] Furthermore, the preprocessed images are input into a preset deep learning model for face detection for analysis. The model scans the potential face regions in the images, generates multiple rectangular bounding box candidate regions, and calculates the confidence score for each region. According to the set confidence threshold (such as 0.5), the low-confidence detection results are filtered out, and only the high-confidence regions are retained. Further, key point information such as the coordinate positions of the eyes, nose, and mouth is extracted from the selected face regions for subsequent pose analysis and age estimation. Finally, the detected occupant face information (bounding box coordinates and key point data) and its confidence score are output as the input for the subsequent steps.

[0059] Through the preset deep learning model for face detection, high-precision face detection can be achieved in complex environments (such as light changes, occlusions, etc.), reducing false detections and missed detections, and improving the quality and reliability of the basic data for child mode detection. Based on the camera and deep learning algorithms, without relying on pressure sensors or specific child seats, the hardware dependence and system cost are greatly reduced, the applicability and flexibility of the system are enhanced, and it is suitable for various vehicle models and seat configuration scenarios. This step not only outputs the face position information but also includes key points and confidence scores, providing basic data support for subsequent pose angle calculation and age estimation, ensuring the accuracy of the subsequent analysis process. Real-time image analysis can quickly respond to the changes of the occupants, timely judge the presence of children and switch the vehicle mode, improving the safety protection ability and reducing manual operation and waiting time. The deep learning model has the ability to continuously optimize and update, and can improve the detection ability and recognition accuracy for future application scenarios according to new training data and algorithm upgrades, providing further expansion support for intelligent driving and automatic control. Through the above steps, using the visual analysis ability of the deep learning model, the efficient detection and data extraction of the occupant face are ensured, laying a foundation for the subsequent judgment and control of the child mode, with significant advantages of high precision, low cost, and easy expansion.

[0060] S2: Determine the face pose angle according to the face key point data in the occupant face information;

[0061] It should be noted that the facial key point data refers to the significant feature points on the face, such as the coordinate positions of the eye corners, the tip of the nose, the corners of the mouth, etc. These key points are used to analyze facial features and structures, and support the calculation of pose angles and subsequent analysis. The pose angle refers to the rotation angle of the face relative to the camera in three-dimensional space, including the pitch angle, the yaw angle, and the roll angle.

[0062] Specifically, from the occupant's facial information output by the detection model, obtain the key point coordinates such as the eye corners, the tip of the nose, the corners of the mouth, and their confidence levels. Set a confidence level threshold (such as 0.2), and eliminate the key point data with low confidence levels to ensure the calculation accuracy. Convert the key point positions to a normalized coordinate system (such as the Cartesian coordinate system) for subsequent calculations.

[0063] Furthermore, calculate the facial rotation angle based on the horizontal and vertical distances between the left and right eye corner points. If the Roll angle deviates significantly from 0 degrees, it is necessary to perform rotation correction on the facial image to ensure accurate subsequent calculations. Calculate the vertical rotation angle based on the height difference between the eye corner and the tip of the nose positions. Calculate the left-right rotation angle based on the horizontal offset of the tip of the nose relative to the eye corners. Compare the calculated angles with the set threshold (such as ±25 degrees) to determine whether the face meets the requirements of the frontal face pose.

[0064] Through the calculation of pose angles, false detection cases caused by side faces or significantly tilted poses can be excluded, and only frontal face data is retained, improving the accuracy and stability of face recognition. Using the three-dimensional pose analysis algorithm, it can adapt to complex scenarios (such as slightly tilted heads, occlusions, or lighting changes), reduce detection errors affected by environmental factors, and ensure that the vehicle system maintains high-performance detection capabilities under various conditions. Through pose angle correction, it can ensure that the input image to the age estimation model is frontal face data, improve the accuracy and reliability of age estimation, and reduce misjudgment or missed judgment cases. This step not only improves the accuracy of face recognition, but also reduces invalid detections and false alarms through three-dimensional pose judgment and correction functions, optimizes the logic efficiency of child mode switching, and improves the system's response speed and user experience. This step is based on facial key point data, accurately judges the facial direction of the occupant through three-dimensional pose angle calculation (Pitch, Yaw, Roll), ensures that the input image meets the frontal face standard, and provides high-quality data support for subsequent age estimation and child mode triggering, with the advantages of high accuracy, strong robustness, and wide adaptability.

[0065] S3: Based on the facial pose angle and the confidence information, determine whether the face corresponding to the occupant's facial information is a frontal face;

[0066] It should be noted that a frontal face means that the face is facing the camera directly, the pose angle is within the set threshold range (such as ±25 degrees), and the key point confidence level reaches the set standard, meeting the frontal face judgment conditions.

[0067] Specifically, obtain the pose angles (Pitch, Yaw, Roll) and key point confidence information of the occupant's face from the previous steps. Set angle thresholds: Define the range of the front face pose angles. For example: Pitch ∈ [-25°, 25°], Yaw ∈ [-25°, 25°], Roll ∈ [-25°, 25°]. Check item by item whether the pose angles are within the set thresholds. If all three angles meet the threshold conditions, it is considered that the face direction is close to the front face. If any angle exceeds the threshold, it is determined that the face is not a front face, and the detection is ended or the re-detection process is entered.

[0068] Furthermore, check whether the confidence levels of key points such as the corners of the eyes, the tip of the nose, and the corners of the mouth meet the minimum threshold (e.g., 0.2). If the confidence level of the key point is lower than the threshold, even if the pose angle meets the conditions, the data is regarded as invalid, and the face result is re-detected or ignored. Make a comprehensive judgment based on the pose angle and the confidence level. If it is a front face, output and mark it as "front face" for subsequent age estimation. If it is not a front face, output and mark it as "not a front face", and directly exit the analysis or re-detect.

[0069] Through the dual judgment mechanism of pose angle and confidence level, filter out side face, occlusion or mis-detection data, ensure that the image input to the age estimation model is high-quality front face data, thereby improving the accuracy and stability of subsequent analysis. Using pose angle correction and confidence level screening techniques, this step can adapt to complex scenarios (such as light changes, slight head tilts, etc.), reduce misjudgments caused by environmental impacts, and improve the robustness of the system in practical applications. Pose and confidence level filtering provides high-quality input data for age estimation, avoids errors caused by angle deviation or low confidence level, thereby improving the accuracy of the age estimation result, and ensuring more accurate and timely triggering of the child mode. The front face judgment can effectively avoid false alarm situations caused by non-front face poses or fuzzy detections, reduce user intervention, and improve the automation degree and user experience of the system. This step combines pose angle judgment and confidence level screening to ensure the reliability and accuracy of the detection result, provides high-quality input data for subsequent age estimation and child mode triggering, has the advantages of high precision, strong robustness and low false alarm rate, and thus effectively improves the overall performance and safety of the child detection system.

[0070] S4: If so, based on a preset age model, estimate the age of the front face human image, wherein the preset age model is constructed based on KL offline loss and mean difference loss;

[0071] It should be noted that the preset age model is a model constructed based on deep learning algorithms, which is used to analyze the facial image features and estimate the age range or specific age value. This model can learn the mapping relationship between facial features and age, so as to achieve accurate prediction. The KL offline loss is the Kullback-Leibler (KL) divergence loss function, which is used to measure the difference between two probability distributions. In age estimation, the KL loss can compare the age probability distribution predicted by the model with the true age distribution, improving the stability and continuity of the model's age estimation. The mean difference loss is used to measure the mean absolute error between the predicted age value and the true age value, emphasizing the accuracy of the model for specific age values and helping to reduce the impact of extreme errors on the prediction results. The frontal face image refers to the face image that has passed the previous steps of pose judgment and confidence screening, and is confirmed to meet the pose angle threshold and be a frontal face. These images are regarded as high-quality inputs to ensure that the age estimation model can obtain reliable data.

[0072] Specifically, the input frontal face images are subjected to size adjustment, color normalization, and feature alignment processing to ensure that the data format meets the model input requirements. The processed images are input into the deep learning network, and key features (such as facial contours, texture details, and symmetry) are extracted through convolutional layers. According to the pose correction results, the facial alignment state of the input images is further optimized to ensure that the model's attention area is concentrated on facial features.

[0073] Furthermore, the model maps the facial features to an age probability distribution and predicts the likelihood of each age range. By calculating the KL divergence between the model output distribution and the true distribution, the model's prediction ability for age intervals is optimized to ensure that the estimation results highly match the true distribution. The model parameters are further adjusted to minimize the average error between the predicted value and the true age value, improving the accuracy of specific age estimation. The most likely age value and confidence score are output. According to the set threshold (such as under 12 years old), it is judged whether it belongs to a child occupant, providing a basis for triggering the subsequent child mode.

[0074] By combining the KL offline loss and the mean difference loss, the model can not only predict specific age values, but also ensure the performance of age prediction results in terms of continuity and stability, thus adapting to the situation where the boundary between children and adults is blurred. The KL loss emphasizes the matching of age distributions and reduces prediction errors across age groups; the mean difference loss optimizes the accuracy of specific age values, reduces extreme errors, and improves overall robustness. Based on vision algorithms and deep learning models, child identification can be achieved without pressure sensors or special child seats, effectively reducing hardware costs and improving the compatibility and universality of the system. The age estimation results directly drive the child mode judgment and control process without manual intervention, ensuring that the child safety function can be activated in a timely manner when a child occupant is detected, improving the response speed and safety. Based on deep learning algorithms, the model can continuously optimize the prediction accuracy with the accumulation of data, adapt to multiple populations and complex scenarios, and support the continuous upgrade of future vehicle intelligent control systems. This step realizes high-precision age estimation of frontal face images through a deep learning age model based on the KL offline loss and the mean difference loss, providing a reliable basis for the intelligent triggering of the child mode. This method has the advantages of strong robustness, wide application range, and fast response speed, effectively improving the accuracy and practicality of the child detection system, while reducing hardware costs and meeting the development needs of intelligent vehicles.

[0075] S5: If all age estimation results are less than the preset child age threshold, control the vehicle to turn on the child mode.

[0076] It should be noted that the age estimation result is the age value or age range of the occupant calculated by the age model, which is used to determine whether the occupant belongs to a child. The preset child age threshold is a pre-set age boundary value (such as under 12 years old), which is used to distinguish children from adults. When the age estimation result is lower than this threshold, the occupant is regarded as a child. The child mode is a safety and entertainment function mode designed by the vehicle specifically for child occupants, including automatically turning on the child safety lock, adjusting the air conditioner air volume, reducing the volume, pushing child entertainment content, and providing real-time monitoring of the rear occupants, etc.

[0077] Specifically, receive all the occupant age prediction values and confidence information output by the age estimation module. If all occupant ages are less than the preset child age threshold, the system further analyzes whether the child mode trigger condition is met. If any one of the age estimation values is greater than the threshold, the child mode is not directly triggered, but is determined according to the subsequent logic (such as prompting the driver to confirm).

[0078] Further, after confirming that the ages of all occupants are less than the threshold, the system sends a control signal to activate the child mode. The child safety lock is activated to prevent children from accidentally opening the door. The air volume and temperature of the air conditioner are adjusted to a comfortable environment suitable for children. The volume of the speaker is reduced to avoid irritating children with sound. Entertainment content suitable for children to watch or listen to is pushed to the in-vehicle display device. The rear occupant monitoring camera is activated, allowing the driver to observe the status of children in real time through the central control screen, reducing the safety hazards caused by driver distraction.

[0079] Based on the age estimation result, it automatically determines whether to activate the child mode without manual operation, quickly responds to the presence of children, and effectively improves safety and convenience. The child safety lock is automatically turned on to prevent children from accidentally opening the door and causing accidents. Combined with air conditioner and volume adjustment, it creates a comfortable driving environment, which helps to reduce the discomfort and anxiety of children during long-distance driving. The real-time monitoring function allows the driver to observe the status of children in the back row at any time, reducing the need to frequently look back during driving, thereby improving driving concentration and driving safety. The system relies only on cameras and algorithms to implement functions, without the need for special child seats or additional sensors, reducing hardware costs and adapting to different vehicle types and seat configurations. It supports software upgrades and algorithm optimizations, and can adjust the trigger logic and environmental settings of the child mode according to usage scenarios and user requirements to adapt to the development of future intelligent driving technologies. This step analyzes the age estimation result and automatically determines whether to enable the child mode, providing an intelligent, highly real-time, and highly automated child safety protection solution. This method not only improves the safety and comfort of children riding in the car, but also reduces the driver's operation burden, enhances the vehicle's intelligent control ability, and at the same time has the advantages of low cost, easy deployment, and scalability, making it suitable for the wide application of future intelligent vehicles.

[0080] In this embodiment, based on a preset deep learning model for face detection, face detection is performed on the rear occupant image to obtain occupant face information and confidence information; according to the face key point data in the occupant face information, the face pose angle is determined; based on the face pose angle and confidence information, it is determined whether the face corresponding to the occupant face information is a frontal face; if so, based on a preset age model, age estimation is performed on the frontal face image, where the preset age model is constructed based on KL offline loss and mean difference loss; if the age estimation results are all less than the preset child age threshold, the vehicle is controlled to turn on the child mode. In this embodiment, face detection is performed on the rear occupant image through a preset deep learning model for face detection, the face pose angle is calculated using key point data and combined with confidence information to determine whether it is a frontal face, ensuring the reliability of the input data; age estimation is performed on the frontal face image based on a preset age model constructed based on KL offline loss and mean difference loss to determine whether it is less than the preset child age threshold, thereby controlling the vehicle to automatically turn on the child mode and improving the safety of children riding in the car.

[0081] Based on the above first embodiment, a second embodiment of the vehicle child mode detection method of the present application is proposed. Before step S5 in this embodiment, it further includes:

[0082] S5a1: If all of the age estimation results are greater than the preset child age threshold, control the vehicle to maintain the current mode;

[0083] S5a2: If there are both values greater than the preset child age threshold and values less than the preset child age threshold in the age estimation results, send a child mode suggestion activation signal to the vehicle control terminal.

[0084] It should be noted that the current mode refers to the default mode in which the vehicle is currently operating, the state where the child mode is not enabled, including standard air conditioning settings, volume control, and safety settings, which are suitable for adult occupants. The child mode suggestion activation signal refers to when the detection result shows that there are both children and adults riding together, the system sends a suggestion signal to the vehicle control terminal to prompt the driver whether to activate the child mode for the driver to confirm the operation.

[0085] Specifically, receive the age prediction results of all occupants from the age estimation module. According to the age value of each occupant, compare it with the preset child age threshold (such as 12 years old) and classify and process. If all age estimation values are greater than the threshold, it is considered that there are no child occupants in the vehicle, and enter the current mode holding state. If all age estimation values are less than the threshold, directly trigger the child mode (described in the previous steps). If there are both cases where the age estimation results are less than the threshold and greater than the threshold, enter the suggestion mode.

[0086] Furthermore, if all age estimation values are greater than the threshold, do not change the current vehicle settings, maintain the existing air conditioning air volume, volume, and safety lock configuration, and no further operation is required. If the detection result includes both child and adult occupants, the system will generate a child mode suggestion activation signal and send it to the vehicle control terminal. After receiving the prompt, the driver can choose: Activate the child mode: Adjust the vehicle settings according to the needs, such as reducing the air conditioning air volume, reducing the volume, and pushing child entertainment content, etc. Maintain the current mode: Ignore the prompt and continue to use the current environmental settings.

[0087] For different situations of the passenger composition, dynamically determine whether to switch to the child mode, taking into account the needs of both children and adult passengers, and improving the flexibility and adaptability of vehicle environment control. When the ages of all passengers exceed the threshold, the system maintains the current mode, reduces unnecessary operations caused by misjudgment, keeps the vehicle in a normal operating state, and enhances the user experience. In the case of children and adults traveling together, the system provides mode suggestions instead of directly switching, fully respecting the driver's decision-making power while ensuring that the safety needs of children are taken into account. Automatically generate mode suggestion prompt signals, eliminating the need for the driver to manually check the status of rear passengers, reducing distracting operations, and improving driving safety and convenience.

[0088] Please refer to Figure 2 , Figure 2 which is a schematic diagram of a sub-process in the second embodiment of the vehicle child mode detection method of the present application.

[0089] As Figure 2 shown, in this embodiment, after step S5, it further includes:

[0090] S5b1: Turn on the preset child safety lock;

[0091] S5b2: Adjust the in-vehicle environment parameters, where the in-vehicle environment parameters include the air volume of the air conditioner and the speaker volume;

[0092] S5b3: Push the preset child entertainment content to the preset in-vehicle display device and turn on the preset rear row monitoring device.

[0093] It should be noted that the preset child safety lock is a vehicle safety mechanism that, when enabled in the vehicle door lock system, prevents children from accidentally opening the door and ensures the safety of children during driving. The in-vehicle environment parameters refer to the comfort settings controlled by the vehicle system, including the air volume of the air conditioner and the speaker volume, which are used to automatically adjust the environmental conditions according to the needs of children to provide a suitable riding experience.

[0094] Specifically, after determining that it is a child according to the age estimation result, the system automatically activates the child lock function of the door to prevent children from accidentally opening the door and causing safety hazards. At the same time, the operation permission of the electric window is locked to avoid accidental situations caused by children accidentally touching it. The air volume and temperature of the air conditioner are automatically adjusted to a comfortable range suitable for children (such as a lower wind speed and a mild temperature) to avoid affecting the health of children due to too low temperature or too large air volume. The speaker volume is reduced to a range suitable for children to reduce noise stimulation and ensure riding comfort and auditory safety.

[0095] Furthermore, the preset children's videos, cartoons or music can be pushed to the in-vehicle display device and played through the touch screen or voice control to meet the children's entertainment needs. The rear camera is turned on to transmit the real-time image to the driver's central control screen, so that the driver can monitor the status of the children at any time during driving. It supports alarm or prompt function. When abnormal movements of children (such as standing or breaking free from the seat belt) are detected, the system will issue an alarm to remind the driver.

[0096] Automatically activate the child safety lock to effectively prevent children from accidentally opening the door or touching the window button, reduce the risk of accidents, and ensure driving safety. Automatically adjust the air volume and temperature of the air conditioner to prevent children from being cold or overheated, and provide a more comfortable in-car environment. Push children's entertainment content to relieve children's anxiety during long-term rides, provide entertainment and educational functions, and enhance the riding experience. The rear-seat monitoring device allows the driver to observe the child's status in real time, reducing the driver's frequent back-checking due to concerns about the child's condition, thereby improving driving concentration and ensuring driving safety. Automatically trigger multiple control functions without manual settings, reducing the driver's operating burden and improving the level of automation and intelligence. Applicable to different models and seat layouts, without relying on child seats or additional sensors, easy to popularize and promote, and reduce hardware costs.

[0097] Based on the above first embodiment, in this embodiment, step S1 includes:

[0098] S11: Scanning the rear passenger image based on the preset face detection deep learning model to generate a plurality of rectangular bounding boxes;

[0099] S12: based on the comparison result of each rectangular bounding box with a preset bounding box confidence threshold, screening the plurality of rectangular bounding boxes to obtain a key bounding box;

[0100] S13: extracting facial feature points in the key boundary frame using a key point detection algorithm to obtain the occupant's facial information, and using the confidence level corresponding to the facial feature points as the confidence level information.

[0101] It should be noted that the rectangular bounding box is a rectangular box that marks the area where the face is located in the image. It is defined by boundary coordinates and is used to frame the detected face position. The preset bounding box confidence threshold is a pre-set confidence scoring standard, which is used to filter out bounding boxes with higher reliability in the detection results and eliminate bounding boxes with lower confidence or false detection. The key bounding box is a bounding box with higher confidence that is retained after screening, indicating that the model determines it as a candidate box for the face area. The key point detection algorithm is a computer vision algorithm used to extract significant feature points (such as the corners of the eyes, the tip of the nose, the corners of the mouth, etc.) from the face area to provide support for subsequent posture estimation and age prediction.

[0102] Specifically, the rear passenger image is input into a preset face detection deep learning model. The model performs multi-scale scanning on the image to detect areas where faces may exist and generates multiple rectangular bounding boxes. Each bounding box contains position information (such as the coordinates of the upper left corner and the lower right corner) and a confidence score. The confidence scores of the multiple detected bounding boxes are compared with a preset bounding box confidence threshold, and the bounding boxes with confidence higher than the threshold are selected as key bounding boxes. If multiple bounding boxes overlap, the non-maximum suppression (NMS) algorithm is applied to merge the overlapping areas, retaining the optimal bounding box and removing redundant detection results.

[0103] Furthermore, for the key bounding box areas, a key point detection algorithm is used to extract facial feature points, including the corners of the eyes, the tip of the nose, the corners of the mouth, etc. The extracted key point data contains specific coordinate positions and confidence scores. Based on the confidence information calculated by the key point detection algorithm, the reliability of each feature point is further evaluated. The occupant face information including the bounding box coordinates, the key point positions, and the confidence scores is output, providing basic data for subsequent pose analysis and age estimation.

[0104] Generating multiple bounding boxes through the deep learning model and using the confidence threshold to screen the key bounding boxes can effectively filter out false detections and missed detections, ensuring high reliability and high precision of the recognition results. Supporting multi-scale detection and confidence screening, it is applicable to different lighting conditions, complex backgrounds, and partially occluded scenarios, improving the robustness of the detection system. Using the key point detection algorithm to extract facial feature points provides detailed and accurate data input for subsequent pose analysis and age estimation, thereby enhancing the detection and judgment capabilities of the entire system. The deep learning model and the bounding box screening algorithm can quickly process image data, meet the real-time monitoring requirements of the vehicle, and ensure the timely response of the child mode detection system. Relying on the camera and the algorithm can complete the detection without additional sensors or hardware devices, reducing costs and improving compatibility with different vehicle models.

[0105] Based on a preset deep learning model for face detection, this embodiment performs face detection on the rear row occupant image to obtain occupant face information and confidence information; determines the face pose angle according to the face key point data in the occupant face information; based on the face pose angle and the confidence information, determines whether the face corresponding to the occupant face information is a frontal face; if so, performs age estimation on the frontal face image based on a preset age model, where the preset age model is constructed based on KL offline loss and mean difference loss; if the age estimation results are all less than a preset child age threshold, controls the vehicle to turn on the child mode. This embodiment performs face detection on the rear row occupant image through a preset deep learning model for face detection, calculates the face pose angle using the key point data and combines the confidence information to determine whether it is a frontal face, ensuring the reliability of the input data; performs age estimation on the frontal face image using a preset age model constructed based on KL offline loss and mean difference loss, determines whether it is less than the preset child age threshold, and thus controls the vehicle to automatically turn on the child mode, improving the safety of children riding in the vehicle.

[0106] Based on the above second embodiment, a third embodiment of the vehicle child mode detection method of the present application is proposed. Please refer to Figure 3 , Figure 3 which is a schematic diagram of a sub-process in the third embodiment of the vehicle child mode detection method of the present application.

[0107] In this embodiment, step S2 includes:

[0108] S21: Extract the face key point data in the occupant face information, where the face key point data includes eye corner feature point data, nose tip feature point data, and mouth corner feature point data;

[0109] S22: Based on the eye corner feature point data, the nose tip feature point data, and the mouth corner feature point data, use the Euler angle algorithm to define key point coordinates, determine the horizontal distance and the vertical distance, determine the roll angle, determine the pitch angle, and determine the yaw angle;

[0110] S23: Determine the face pose angle according to the key point coordinates, the horizontal distance and the vertical distance, the roll angle, the pitch angle, and the yaw angle.

[0111] It should be noted that the Euler angle algorithm is a mathematical calculation method that describes the object pose through three-dimensional rotation angles, including roll angle (Roll), pitch angle (Pitch), and yaw angle (Yaw), and is used to analyze the rotation state of the face in three-dimensional space. The roll angle represents the rotation angle of the face tilting left and right around the nose bridge axis, such as the angle change when the head is tilted to the side. The pitch angle represents the rotation angle of the face looking up or down. The yaw angle represents the rotation angle of the face turning left or right. The horizontal distance and the vertical distance refer to the straight-line distance between key points and are used to calculate the angle relationship and facial pose correction.

[0112] Specifically, obtain the key point coordinates and confidence information of the eye corners, the tip of the nose, and the corners of the mouth from the output results of the face detection model. Set a key point confidence threshold (such as 0.2), and eliminate the key points with low confidence to ensure data reliability. Map the key point coordinates to a unified coordinate system (such as the Cartesian coordinate system) for subsequent mathematical calculations.

[0113] Furthermore, calculate the horizontal distance and the vertical distance according to the eye corner coordinates. Calculate the roll angle of the face based on the relative positions of the left and right eye corner points. For a face with a Roll offset, perform rotation correction to ensure accurate subsequent calculations. Calculate the pitch angle of the head based on the height difference between the tip of the nose and the center of the two eyes. Calculate the yaw angle of the head based on the horizontal offset between the tip of the nose and the center of the two eyes. Take the calculated Roll, Pitch, and Yaw angles as the results of the face pose angles and pass them to the subsequent module for judgment and processing.

[0114] Precisely calculate the three-dimensional rotation angle through the Euler angle algorithm, which can distinguish the frontal face, profile face, or tilted pose of the face, providing high-quality input data for subsequent age estimation and child mode determination. Support multi-angle face detection. Even when the face is partially tilted or slightly rotated, the pose can still be accurately judged, reducing false positives and false negatives in complex environments. Adjust the image with a large roll angle through the pose correction algorithm to ensure that a standardized face image is used for subsequent algorithm analysis, improving the calculation accuracy and the overall performance of the system. The Euler angle calculation speed is fast, which can process the pose changes of the rear occupants in real time, support the dynamic enabling and monitoring functions of the child mode, and improve the intelligent control level. The output pose angles can not only be used for child mode recognition but also be extended to functions such as driver state monitoring (such as fatigue detection) or expression analysis, with strong scalability and practical value.

[0115] Based on the above second embodiment, in this embodiment, step S3 includes:

[0116] S31: Obtain the confidence of the eye corner feature point data, the confidence of the tip of the nose feature point data, and the confidence of the corner of the mouth feature point data respectively;

[0117] S32: Compare the confidence of the eye corner feature point data, the confidence of the tip of the nose feature point data, and the confidence of the corner of the mouth feature point data with the preset pose confidence threshold respectively;

[0118] S33: Determine whether the roll angle, the pitch angle, and the yaw angle are all within the preset pose angle range;

[0119] S34: Based on the comparison result and the determination result, judge whether the face corresponding to the occupant face information is a frontal face.

[0120] It should be noted that the preset pose confidence threshold is the lowest confidence standard set in advance (such as 0.2 or 0.5), which is used to filter the key point data with low confidence to ensure the quality of the input data. The preset pose angle range is the effective range of the face pose angle set in advance (for example, ±25°), which is used to determine whether the face is close to the frontal face state.

[0121] Specifically, the confidence data of the feature points of the eye corners, the tip of the nose, and the corners of the mouth are extracted from the results of the detection model respectively. The confidence of each feature point is compared with the preset threshold (such as 0.2 or 0.5) one by one. If all the confidences are higher than the threshold, then proceed to the next angle judgment; if the confidence of any feature point is lower than the threshold, it is directly determined that it is not a frontal face, and the judgment process ends.

[0122] Furthermore, according to the roll angle (Roll), pitch angle (Pitch), and yaw angle (Yaw) calculated in the previous steps, compare them with the preset pose angle range (such as ±25°). If all the angles are within the specified range, it is determined that the face pose meets the frontal face condition. If any angle exceeds the range, it is determined that it is not a frontal face, and the analysis ends. When both the confidence and the pose angle conditions are met, it is determined to be a frontal face. If any condition is not met, the result of "not a frontal face" is output, and the data is passed to the re-detection or ignore process. A mark is output for the frontal face result, and the key point coordinates and pose angles are passed to the age estimation module. An error mark is output for the non-frontal face result, and the face data is requested again or the analysis process is terminated.

[0123] Through the double-layer verification mechanism (confidence and pose angle), false detections and low-confidence data are effectively filtered, ensuring the reliability of the data quality entering the age estimation link and improving the final recognition and control accuracy. Using the pose angle judgment and confidence filtering technology, it can meet the face detection requirements in complex environments (such as light changes, partial occlusion, and slight head tilt), reducing the missed detection and false detection rates. When the confidence is low or the angle exceeds the limit, the calculation process is directly terminated, avoiding invalid calculations, reducing system resource consumption, and improving the detection speed and real-time response ability. Combining the confidence and pose judgment results, only the frontal face data that meets the conditions is output, reducing the false alarm situation caused by misidentification in the system, improving the user experience and security. The output frontal face data provides a high-quality basic input for subsequent age estimation, improving the accuracy of triggering the children's mode and ensuring the high credibility of the age judgment result.

[0124] Based on a preset deep learning model for face detection, this embodiment performs face detection on the rear occupant image to obtain occupant face information and confidence information; determines the face pose angle according to the face key point data in the occupant face information; and determines whether the face corresponding to the occupant face information is a frontal face based on the face pose angle and the confidence information. If so, based on a preset age model, age estimation is performed on the frontal face image, where the preset age model is constructed based on KL offline loss and mean difference loss. If the age estimation results are all less than the preset child age threshold, the vehicle is controlled to turn on the child mode. This embodiment performs face detection on the rear occupant image through a preset deep learning model for face detection, calculates the face pose angle using the key point data, and combines the confidence information to determine whether it is a frontal face, ensuring the reliability of the input data; performs age estimation on the frontal face image using a preset age model constructed based on KL offline loss and mean difference loss, and determines whether it is less than the preset child age threshold, thereby controlling the vehicle to automatically turn on the child mode and improving the safety of children riding in the vehicle.

[0125] Exemplarily, to help understand the technical concept or technical principle of the vehicle child mode detection method in the above embodiment, please refer to Figure 4 , Figure 4 which is the flowchart of child mode monitoring in an embodiment of the vehicle child mode detection method of this application.

[0126] As Figure 4 shown, the vehicle child mode detection method includes the following 10 steps.

[0127] S1. Image acquisition: Transmit the picture of the rear occupant monitoring camera associated with the child mode to the vehicle-mounted controller. The installation of the camera needs to meet the condition that the field of view can cover the entire rear occupant area, and at the same time, the camera installation needs to be stable.

[0128] S2. Image preprocessing: mainly includes tasks such as converting the picture format and scaling the picture size, so that the picture meets the requirements for entering the algorithm model.

[0129] S3. Deep learning model occupant face detection: After processing such as model format conversion and model quantization, deploy the trained lightweight deep learning model into the vehicle-mounted controller and call the hardware inference acceleration unit of the vehicle-mounted controller to complete the board-side inference of the model. Input the preprocessed picture into the occupant face detection model, decode the inference result, and obtain the face information of the occupant position in the picture, including the outer contour rectangle coordinate frame of the face and the face confidence.

[0130] S4. Deep learning model occupant face key point detection: The trained lightweight deep learning model is processed through model format conversion, model quantization, etc., and then deployed into the vehicle-mounted controller and the hardware inference acceleration unit of the vehicle-mounted controller is called to complete the board-side inference of the model. The outer contour rectangle coordinate frame obtained in S3 is used to crop the image to obtain the occupant face image. The preprocessed picture is input into the face key point detection model, and the inference result is decoded to obtain a total of 19 occupant face key points including eyes, mouth, nose tip, nose corner points, etc. Each key point contains point position coordinates and point confidence levels.

[0131] S5. Calculate the three angles of the face pose of each occupant face detection result in the right-handed Cartesian coordinate system: pitch angle, yaw angle, and roll angle.

[0132] S6. Through the face correction module, for the face with a deviated roll angle, after calculating the angle through key points and rotating the picture, the roll angle is set to 0. The steps are as follows:

[0133] 1. Calculate the sum of the confidence levels of the inner and outer eye corner points and take the larger value.

[0134] 2. Denote the distance between the two eye corner points as: r = sqrt((x_1 - x_2)^2 + (y_1 - y_2)^2)

[0135] 3. Calculate the length of L as: L = abs(x_1 - x_2), and calculate the length of h: h = abs(y_1 - y_2)

[0136] 4. Calculate the rotation angle and rotate the picture: a = arctan(h / L)

[0137] S7. According to the three calculated face pose angles, determine whether the current face is a frontal face. To improve the accuracy of age estimation, the three pose angles need to satisfy the range of plus or minus 25 degrees, and the key point confidence levels used to calculate the pose angles need to be greater than 0.2. Only then can age estimation be performed.

[0138] S8. For the faces that pass the frontal face judgment, a lightweight deep learning model is used for age estimation. The modeling idea of this model is mainly based on the feature that age is ordered. First, estimate the approximate age range, and then calculate the specific age value according to the proportion of this age range. In the model, KL offline loss and mean difference loss are used to build the age model for age estimation.

[0139] S9. For the age values output in the entire scenario, make a judgment. If all age values are less than 12 years old, enable the child mode. If some age values are less than 12 years old, ask the driver whether to enable the child mode. If all age values are greater than 12 years old or no age values are detected, do nothing.

[0140] S10. After activating the child mode, the controller will actively activate the child lock to prevent accidents caused by children pulling the door. And the rear entertainment screen will automatically push entertainment content related to the child mode, adjust the air conditioner wind speed, and slightly open the window. And the driver can also observe the child's status in real time through the occupant monitoring camera to reduce the driver's distracted driving situation.

[0141] The embodiment of the present application also provides a vehicle child mode detection device. Please refer to Figure 5 , Figure 5 which is the module structure diagram of the vehicle child mode detection device in the embodiment of the present application. The vehicle child mode detection device includes:

[0142] A face detection module 501, configured to perform face detection on the rear row occupant image based on a preset face detection deep learning model to obtain occupant face information and confidence information;

[0143] A pose angle module 502, configured to determine the face pose angle according to the face key point data in the occupant face information;

[0144] A frontal face judgment module 503, configured to judge whether the face corresponding to the occupant face information is a frontal face based on the face pose angle and the confidence information;

[0145] An age estimation module 504, configured to, if so, perform age estimation on the frontal face image based on a preset age model, where the preset age model is constructed based on KL offline loss and mean difference loss;

[0146] A target module 505, configured to, if the age estimation results are all less than a preset child age threshold, control the vehicle to enable the child mode.

[0147] The vehicle child mode detection device provided by the embodiment of the present application adopts the vehicle child mode detection method in the above embodiment, and can solve the technical problem of how to improve the safety of children riding in the vehicle. Compared with the prior art, the beneficial effects of the vehicle child mode detection device provided by the embodiment of the present application are the same as those of the vehicle child mode detection method provided by the above embodiment, and other technical features in the vehicle child mode detection device are the same as the features disclosed in the above embodiment method, and will not be elaborated here.

[0148] The present application provides a vehicle child mode detection device. The vehicle child mode detection device includes: at least one processor; and a memory communicatively connected to the at least one processor. The memory stores instructions executable by the at least one processor. When the instructions are executed by the at least one processor, the at least one processor is enabled to execute the vehicle child mode detection method in the foregoing embodiments.

[0149] Reference is made below to Figure 6 , which shows a schematic structural diagram of a vehicle child mode detection device suitable for implementing the embodiments of the present application. The vehicle child mode detection device in the embodiments of the present application may include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), in-vehicle terminals (such as in-vehicle navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 6 The vehicle child mode detection device shown is merely an example and should not impose any limitation on the functions and scope of use of the embodiments of the present application.

[0150] As Figure 6As shown, the vehicle child mode detection device may include a processing device 1001 (such as a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM: Read Only Memory) 1002 or the program loaded from the storage device 1003 into the random access memory (RAM: Random Access Memory) 1004. In the RAM 1004, various programs and data required for the operation of the vehicle child mode detection device are also stored. The processing device 1001, the ROM 1002, and the RAM 1004 are connected to each other through a bus 1005. The input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the vehicle child mode detection device to communicate with other devices wirelessly or wiredly to exchange data. Although the figure shows a vehicle child mode detection device with various systems, it should be understood that it is not required to implement or have all the shown systems. More or fewer systems can be alternatively implemented or had.

[0151] Specifically, according to the embodiments disclosed in the present application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program codes for executing the methods shown in the flowcharts. In such an embodiment, the computer program can be downloaded and installed from the network through the communication device, or installed from the storage device 1003, or installed from the ROM 1002. When the computer program is executed by the processing device 1001, the above functions defined in the methods of the embodiments disclosed in the present application are executed.

[0152] The vehicle child mode detection device provided by the present application adopts the vehicle child mode detection method in the above embodiments and can solve the technical problem of how to improve the safety of children in the car. Compared with the prior art, the beneficial effects of the vehicle child mode detection device provided by the present application are the same as those of the vehicle child mode detection method provided by the above embodiments, and the other technical features in the vehicle child mode detection device are the same as those disclosed in the method of the previous embodiment, and will not be elaborated here.

[0153] It should be understood that each part disclosed in this application can be implemented by hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in a suitable manner in any one or more embodiments or examples.

[0154] As described above, the above are only specific embodiments of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

[0155] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., computer programs) stored thereon, and the computer-readable program instructions are used to execute the vehicle child mode detection method in the above embodiments.

[0156] The computer-readable storage medium provided by this application can be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination of the above. More specific examples of computer-readable storage media can include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM) or flash memory, optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above. In this embodiment, the computer-readable storage medium can be any tangible medium that contains or stores a program, and this program can be used by or combined with an instruction execution system, device, or device. The program code contained on the computer-readable storage medium can be transmitted by any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0157] The above computer-readable storage medium can be included in the vehicle child mode detection device; it can also exist alone without being assembled into the vehicle child mode detection device.

[0158] The above computer-readable storage medium carries one or more programs, which, when executed by a vehicle child mode detection device, cause the vehicle child mode detection device to: perform face detection on the rear occupant image based on a preset deep learning model for face detection, to obtain occupant face information and confidence information; determine the face pose angle according to the face key point data in the occupant face information; based on the face pose angle and the confidence information, determine whether the face corresponding to the occupant face information is a frontal face; if so, perform age estimation on the frontal face image based on a preset age model, where the preset age model is constructed based on KL offline loss and mean difference loss; if the age estimation results are all less than a preset child age threshold, control the vehicle to turn on the child mode. Computer program code for performing the operations of the present application may be written in one or more programming languages or combinations thereof. The programming languages include object-oriented programming languages such as Java, Smalltalk, C++, and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code may execute entirely on the user's computer, partially on the user's computer, execute as a stand-alone software package, execute partially on the user's computer and partially on a remote computer, or execute entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0159] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code that contains one or more executable instructions for implementing the specified logical function. It should also be noted that, in some alternative implementations, the functions marked in the blocks may occur in a different order than marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, may be implemented by a dedicated hardware-based system for performing the specified functions or operations, or may be implemented by a combination of dedicated hardware and computer instructions.

[0160] The modules involved in the embodiments of the present application can be implemented in software or in hardware. In some cases, the name of the module does not constitute a limitation on the unit itself.

[0161] The readable storage medium provided by the present application is a computer-readable storage medium. The computer-readable storage medium stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned vehicle child mode detection method, and can solve the technical problem of how to improve the safety of children in the car. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present application are the same as those of the vehicle child mode detection method provided by the above embodiments, and will not be elaborated here.

[0162] The embodiments of the present application provide a computer program product, including a computer program, and the steps of the above-mentioned vehicle child mode detection method are implemented when the computer program is executed by a processor.

[0163] The computer program product provided by the present application can solve the technical problem of how to improve the safety of children in the car. Compared with the prior art, the beneficial effects of the computer program product provided by the embodiments of the present application are the same as those of the vehicle child mode detection method provided by the above embodiments, and will not be elaborated here.

[0164] The above are only the preferred embodiments of the present application, and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present application, or directly or indirectly applied to other related technical fields, shall be included in the patent scope of the present application by the same token.

Claims

1. A vehicle child mode detection method, characterized in that: The method comprises: Based on the preset face detection deep learning model, face detection is performed on the rear passenger images to obtain the passenger face information and confidence information; Determining facial posture angles according to facial key point data in the occupant facial information; Based on the facial posture angle and the confidence information, determining whether the face corresponding to the occupant facial information is a frontal face; If yes, age estimation is performed on the frontal face image based on a preset age model, wherein the preset age model is constructed based on KL offline loss and mean difference loss; If the age estimation results are all lower than the preset child age threshold, the vehicle is controlled to turn on the child mode.

2. The method according to claim 1, characterized in that Before the step of controlling the vehicle to start the child mode if the age estimation results are all less than the preset child age threshold, the method further includes: If the age estimation results are all greater than the preset child age threshold, controlling the vehicle to maintain the current mode; If the age estimation result includes values ​​greater than the preset child age threshold and values ​​less than the preset child age threshold, a child mode activation suggestion signal is sent to the vehicle control terminal.

3. The method according to claim 1, characterized in that After the step of controlling the vehicle to turn on the child mode if the age estimation results are all less than the preset child age threshold, the method further includes: Turn on the preset child safety lock; Adjusting in-vehicle environment parameters, wherein the in-vehicle environment parameters include air conditioning air volume and speaker volume; Push preset children's entertainment content to the preset in-vehicle display device and turn on the preset rear-seat monitoring device.

4. The method according to claim 1, characterized in that The step of performing face detection on the rear passenger image based on a preset face detection deep learning model to obtain passenger face information and confidence information includes: Based on the preset face detection deep learning model, scanning the rear passenger image to generate a plurality of rectangular bounding boxes; Based on the comparison result of each rectangular bounding box with a preset bounding box confidence threshold, the plurality of rectangular bounding boxes are screened to obtain a key bounding box; A key point detection algorithm is used to extract facial feature points in the key boundary frame to obtain the occupant's facial information, and the confidence level corresponding to the facial feature points is used as the confidence level information.

5. The method according to claim 1, characterized in that The step of determining the facial posture angle according to the facial key point data in the occupant's facial information comprises: Extracting facial key point data from the passenger's facial information, wherein the facial key point data includes eye corner feature point data, nose tip feature point data, and mouth corner feature point data; Based on the eye corner feature point data, the nose tip feature point data and the mouth corner feature point data, the Euler angle algorithm is used to define key point coordinates, determine horizontal distance and vertical distance, determine roll angle, determine pitch angle and determine yaw angle; The facial posture angle is determined according to the key point coordinates, the horizontal distance and the vertical distance, the roll angle, the pitch angle, and the yaw angle.

6. The method according to claim 5, characterized in that The step of judging whether the face corresponding to the occupant face information is a frontal face based on the face posture angle and the confidence information comprises: Respectively obtaining the confidence of the eye corner feature point data, the confidence of the nose tip feature point data, and the confidence of the mouth corner feature point data; Comparing the confidence of the eye corner feature point data, the confidence of the nose tip feature point data, and the confidence of the mouth corner feature point data with a preset posture confidence threshold value respectively; Determining whether the roll angle, the pitch angle, and the yaw angle are all within a preset attitude angle range; Based on the comparison result and the judgment result, determine whether the face corresponding to the occupant facial information is a frontal face.

7. A vehicle child mode detection device, characterized in that: The device comprises: A face detection module is used to perform face detection on rear passenger images based on a preset face detection deep learning model to obtain passenger face information and confidence information; A posture angle module, used to determine the facial posture angle according to the facial key point data in the occupant's facial information; A frontal face judgment module, used for judging whether the face corresponding to the occupant face information is a frontal face based on the face posture angle and the confidence information; An age estimation module, for estimating the age of a frontal face image based on a preset age model, wherein the preset age model is constructed based on a KL offline loss and a mean difference loss; The target module is used to control the vehicle to turn on the child mode if the age estimation results are all less than the preset child age threshold.

8. A computer device, characterized in that: The device comprises: a memory, a processor, and a vehicle child mode detection program stored in the memory and executable on the processor, wherein the vehicle child mode detection program is configured to implement the steps of the vehicle child mode detection method according to any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium stores a vehicle child mode detection program, and when the vehicle child mode detection program is executed by the processor, the steps of the vehicle child mode detection method according to any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the vehicle child mode detection method according to any one of claims 1 to 6 are implemented.

Images