Automobile seat adjusting model training method and automobile seat adjusting method and device

Abstract

The invention provides an automobile seat adjustment model training method and device and an automobile seat adjustment method and device, and relates to the technical field of intelligent control in order to relieve the technical problem that in the prior art, the prediction precision of the seat position is insufficient. Inputting the video frame image into an automobile seat adjusting model to obtain automobile seat angle data output by the automobile seat adjusting model; and based on the angle data of the automobile seat, the angle of the automobile seat is controlled and adjusted, so that the more accurate self-adaptive adjustment effect of the automobile seat is achieved, and the user experience is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent control, and in particular to a vehicle seat adjustment model training method, a vehicle seat adjustment method and a vehicle seat adjustment device. Background Art

[0002] In automobile design, the current main method is to use a comfort scoring model based on a quadratic equation to predict the comfort score under different seat positions by utilizing the linear or quadratic relationship between seat position and comfort score learned during the training process to determine the optimal seat position. This method has the advantages of being simple to form, easy to understand and implement, computationally efficient, having a small amount of model calculation, and being suitable for real-time or near-real-time application scenarios. However, this method also has certain defects. For example, the comfort scoring model based on a quadratic equation learns the linear or quadratic relationship between the comfort score and seat position during the training process, which may not capture more complex nonlinear relationships, resulting in insufficient prediction accuracy of the seat position, which in turn affects the seat position adjustment effect and user experience. Summary of the Invention

[0003] In view of this, the object of the present invention is to provide a car seat adjustment model training method, a car seat adjustment method and a device to improve the adaptive adjustment effect and user experience of the car seat.

[0004] In a first aspect, the present invention provides a method for training a car seat adjustment model, comprising:

[0005] Acquire a training sample data set; wherein the training sample data set includes a plurality of training sample data; each training sample data includes an original video frame image and human body part label data, human body attribute label data, and seat angle label data corresponding to the original video frame image;

[0006] Based on the training sample data set, a training operation is iteratively performed on the automobile seat adjustment model; wherein the automobile seat adjustment model includes a human body part detection sub-model and a human body attribute prediction sub-model connected in sequence, and a seat angle assessment sub-model connected to the human body part detection sub-model and the human body attribute prediction sub-model respectively; the training operation includes:

[0007] Select target training sample data from the training sample data set;

[0008] Inputting the original video frame images in the target training sample data into the automobile seat adjustment model, so that the automobile seat adjustment model performs feature extraction and human body part detection on the original video frame images through the human body part detection sub-model to obtain human body part detection data, performs feature extraction and human body attribute prediction on the human body part detection data through the human body attribute prediction sub-model to obtain human body attribute prediction data, and performs feature extraction and seat angle evaluation on the human body part detection data and human body attribute prediction data through the seat angle evaluation sub-model to obtain seat angle evaluation data;

[0009] Calculating a first loss value based on the human body part detection data, the human body attribute prediction data, and the seat angle evaluation data, as well as the human body part label data, the human body attribute label data, and the seat angle label data in the target training sample data;

[0010] Based on the first loss value, the first model parameters of the human body part detection sub-model, the human body attribute prediction sub-model and the seat angle evaluation sub-model are updated.

[0011] Optionally, calculating the first loss value based on the human body part detection data, the human body attribute prediction data, and the seat angle evaluation data, as well as the human body part label data, the human body attribute label data, and the seat angle label data in the target training sample data, includes:

[0012] Obtaining a first human body detection loss value based on the human body part detection data and the human body part label data in the target training sample data;

[0013] Based on the human attribute prediction data and the human attribute label data in the target training sample data, an attribute prediction loss value is obtained;

[0014] Based on the seat angle evaluation data and the seat angle label data in the target training sample data, an angle evaluation loss value is obtained;

[0015] A first loss value is obtained based on the first human detection loss value, the attribute prediction loss value and the angle evaluation loss value.

[0016] Optionally, based on the first loss value, updating first model parameters of each of the human body part detection sub-model, the human body attribute prediction sub-model, and the seat angle assessment sub-model includes:

[0017] Determining a first gradient value corresponding to the seat angle evaluation sub-model based on the first loss value and the seat angle evaluation data; updating the first gradient value based on the optimization function, and inputting the updated first gradient value into the seat angle evaluation sub-model to perform backpropagation to update a first model parameter of the seat angle evaluation sub-model;

[0018] Determining a second gradient value corresponding to the human attribute prediction sub-model based on the first loss value and the human attribute prediction data; updating the second gradient value based on the optimization function, inputting the updated second gradient value into the human attribute prediction sub-model to perform back propagation to update the first model parameter of the human attribute prediction sub-model;

[0019] Based on the first loss value and the human body part detection data, the third gradient value corresponding to the human body part detection sub-model is determined; the third gradient value is updated based on the optimization function, and the updated third gradient value is input into the human body part detection sub-model to perform back propagation to update the first model parameters of the human body part detection sub-model.

[0020] Optionally, the car seat adjustment model further includes a user behavior prediction sub-model, which is connected to the human body part detection sub-model; each training sample data further includes user behavior label data corresponding to the original video frame image;

[0021] After the original video frame images in the target training sample data are input into the car seat adjustment model, the following steps are also included:

[0022] The user behavior prediction sub-model is used to extract features from the human body part detection data and predict user behavior to obtain user behavior prediction data;

[0023] Calculating a second loss value based on the human body part detection data and the user behavior prediction data, as well as the human body part label data and the user behavior label data in the target training sample data;

[0024] Based on the second loss value, the second model parameters of the human body part detection sub-model and the first model parameters of the user behavior prediction sub-model are updated.

[0025] Optionally, the human body part detection sub-model includes a multi-scale feature extraction module, a multi-scale feature enhancement module, a multi-scale feature fusion module and a human body part detection module connected in sequence;

[0026] The human body part detection sub-model is used to extract features and detect human body parts on the original video frame image to obtain human body part detection data, including:

[0027] The original video frame image is preprocessed and multi-scale feature extraction is performed by a multi-scale feature extraction module to obtain first feature images of multiple scales;

[0028] Performing feature enhancement processing on the first feature image of each scale by a multi-scale feature enhancement module, thereby obtaining a second feature image corresponding to each first feature image;

[0029] Performing feature fusion processing on the plurality of second feature images through a multi-scale feature fusion module to obtain a third feature image;

[0030] The third feature image is subjected to target recognition by the human body part detection module to obtain human body part detection data.

[0031] Optionally, the multi-scale feature extraction module includes a normalization layer and multiple downsampling layers connected in sequence; the multi-scale feature enhancement module includes multiple reconstruction attention units, each reconstruction attention unit includes a spatial channel reconstruction layer and a two-layer routing attention layer connected in sequence; the multi-scale feature fusion module includes a path aggregation feature fusion layer; the human body part detection module includes a decoupling layer; the multiple downsampling layers are connected to the multiple reconstruction attention units in a one-to-one correspondence; the multiple reconstruction attention units are respectively connected to the path aggregation feature fusion layer; and the path aggregation feature fusion layer is connected to the decoupling layer;

[0032] The human body part detection sub-model is used to extract features and detect human body parts from the original video frame images to obtain human body part detection data, including:

[0033] The original video frame image is normalized by a normalization layer to obtain a normalized video frame image;

[0034] Downsampling the normalized video frame image in different scales in sequence through multiple downsampling layers to obtain first feature images of multiple scales;

[0035] Through the spatial channel reconstruction layer and the double-layer routing attention layer in the multiple reconstruction attention units, the multiple first feature images are respectively subjected to spatial channel reconstruction and key feature interaction processing to obtain second feature images corresponding to the multiple first feature images;

[0036] Performing bidirectional path feature extraction and adaptive feature fusion processing on the plurality of second feature images through a path aggregation feature fusion layer to obtain a third feature image;

[0037] The third feature image is processed through the decoupling layer to classify and locate human body parts to obtain human body part detection data.

[0038] Optionally, the human attribute prediction sub-model includes a plurality of first residual attention modules connected in sequence, a first feature fusion module respectively connected to the plurality of first residual attention modules, and a human attribute prediction module connected to the first feature fusion module;

[0039] The human body attribute prediction sub-model is used to extract features from human body part detection data and predict human body attributes to obtain human body attribute prediction data, including:

[0040] Performing feature extraction on the human body part detection data in sequence through multiple first residual attention modules to obtain first human body part feature images output by the multiple first residual attention modules;

[0041] Performing feature fusion processing on the plurality of first human body part feature images through a first feature fusion module to obtain a first human body feature fusion image;

[0042] The human attribute prediction module performs human attribute prediction on the first human feature fusion image to obtain human attribute prediction data.

[0043] Optionally, each first residual attention module includes at least two first residual attention units connected in sequence, each first residual attention unit includes a first residual layer and a first compressed excitation attention layer jump-connected to the first residual layer; the first feature fusion module includes a first parallelized feature interaction fusion layer; the human attribute prediction module includes multiple fully connected layers;

[0044] The human body attribute prediction sub-model is used to extract features from human body part detection data and predict human body attributes to obtain human body attribute prediction data, including:

[0045] Through the first residual layer and the first compressed excitation attention layer in each first residual attention unit in the multiple first residual attention modules, feature extraction at different semantic levels and feature enhancement across semantic levels are sequentially performed on the human body part detection data to obtain first human body part feature images output by the multiple first residual attention modules;

[0046] Performing parallel feature extraction and cross-level feature fusion processing on the plurality of first human body part feature images through the first parallel feature interactive fusion layer in the first feature fusion module to obtain a first human body feature fusion image;

[0047] The first human feature fusion image is subjected to multi-classification mapping processing through multiple fully connected layers in the human attribute prediction module to obtain human attribute prediction data.

[0048] Optionally, the seat angle assessment sub-model includes a plurality of second residual attention modules connected in sequence, a second feature fusion module connected to the plurality of second residual attention modules respectively, and a seat angle assessment module connected to the second feature fusion module;

[0049] The seat angle evaluation sub-model extracts features from the human body part detection data and human body attribute prediction data, and evaluates the seat angle to obtain seat angle evaluation data, including:

[0050] Performing feature extraction on the human body part detection data in sequence through multiple second residual attention modules to obtain second human body part feature images output by the multiple second residual attention modules;

[0051] Performing feature fusion processing on the plurality of second human body part feature images by a second feature fusion module to obtain a second human body feature fusion image;

[0052] The seat angle evaluation module performs seat angle evaluation based on the human attribute prediction data and the second human feature fusion image to obtain seat angle evaluation data.

[0053] Optionally, each second residual attention module includes at least two second residual attention units connected in sequence, each second residual attention unit includes a second residual layer and a second compressed excitation attention layer jump-connected to the second residual layer; the second feature fusion module includes a second parallelized feature interaction fusion layer; the seat angle evaluation module includes a fully connected layer;

[0054] The seat angle evaluation sub-model extracts features from the human body part detection data and human body attribute prediction data, and evaluates the seat angle to obtain seat angle evaluation data, including:

[0055] Through the second residual layer and the second compressed excitation attention layer in each second residual attention unit in the multiple second residual attention modules, feature extraction at different semantic levels and feature enhancement across semantic levels are sequentially performed on the human body part detection data to obtain second human body part feature images output by the multiple second residual attention modules;

[0056] Performing parallel feature extraction and cross-level feature fusion processing on the plurality of second human body part feature images through the second parallel feature interactive fusion layer in the second feature fusion module to obtain a second human body feature fusion image;

[0057] Through the fully connected layer in the seat angle assessment module, after determining the human body proportion data based on the human body attribute prediction data, the first seat deployment angle is estimated based on the human body proportion data, and the second seat deployment angle is obtained by key point extraction and angle calculation based on the second human body feature fusion image, and the seat angle assessment data is calculated based on the first seat deployment angle and the second seat deployment angle.

[0058] Optionally, the user behavior prediction sub-model includes a plurality of third residual attention modules and user behavior prediction modules connected in sequence;

[0059] The user behavior prediction sub-model extracts features from the human body part detection data and predicts user behavior to obtain user behavior prediction data, including:

[0060] Performing feature extraction on the human body part detection data in sequence through multiple third residual attention modules to obtain third human body part feature images output by the multiple third residual attention modules;

[0061] The user behavior prediction module performs user behavior prediction on the third human body part feature image to obtain user behavior prediction data.

[0062] Optionally, each third residual attention module includes a third residual layer and a third compressed excitation attention layer jump-connected to the third residual layer; the user behavior prediction module includes a fourth residual layer and a fully connected layer jump-connected to the fourth residual layer;

[0063] The user behavior prediction sub-model extracts features from the human body part detection data and predicts user behavior to obtain user behavior prediction data, including:

[0064] Through the third residual layer and the third compressed excitation attention layer in the multiple third residual attention modules, feature extraction at different semantic levels and feature enhancement across semantic levels are sequentially performed on the human body part detection data to obtain third human body part feature images output by the multiple third residual attention modules;

[0065] Through the fourth residual layer and fully connected layer in the user behavior prediction module, the feature images of the third human body parts are subjected to feature extraction and binary classification mapping at different semantic levels to obtain user behavior prediction data.

[0066] Optionally, calculating the second loss value based on the human body part detection data and the user behavior prediction data, as well as the human body part label data and the user behavior label data in the target training sample data, includes:

[0067] Obtaining a second human body detection loss value based on the human body part detection data and the human body part label data in the target training sample data;

[0068] Based on the user behavior prediction data and the user behavior label data in the target training sample data, the behavior prediction loss value is obtained;

[0069] A second loss value is obtained based on the second human detection loss value and the behavior prediction loss value.

[0070] Optionally, updating the second model parameter of the human body part detection sub-model and the first model parameter of the user behavior prediction sub-model based on the second loss value includes:

[0071] Determining a fourth gradient value corresponding to the human body part detection sub-model based on the second loss value and the human body part detection data; updating the fourth gradient value based on the optimization function, and inputting the updated fourth gradient value into the human body attribute prediction sub-model to perform back propagation to update the second model parameter of the human body attribute prediction sub-model;

[0072] Based on the second loss value and the user behavior prediction data, the fifth gradient value corresponding to the user behavior prediction sub-model is determined; the fifth gradient value is updated based on the optimization function, and the updated fifth gradient value is input into the user behavior prediction sub-model to perform back propagation to update the first model parameter of the user behavior prediction sub-model.

[0073] In a second aspect, the present invention provides a method for adjusting a vehicle seat, comprising:

[0074] Acquire video frame images inside the car cabin;

[0075] Inputting the video frame image into the car seat adjustment model to obtain car seat angle data output by the car seat adjustment model; wherein the car seat adjustment model is trained using the above-mentioned car seat adjustment model training method;

[0076] Based on the car seat angle data, control and adjust the angle of the car seat.

[0077] Optionally, the car seat adjustment model further outputs user behavior data; and based on the car seat angle data, controls the angle of the car seat, including:

[0078] Based on user behavior data, when it is determined that the user is in a standing state, the angle of the car seat is controlled and adjusted based on the car seat angle data.

[0079] In a third aspect, the present invention provides a car seat adjustment model training device, comprising:

[0080] A data acquisition module is used to acquire a training sample data set; wherein the training sample data set includes a plurality of training sample data; each training sample data includes an original video frame image and human body part label data, human body attribute label data, and seat angle label data corresponding to the original video frame image;

[0081] A model training module is used to iteratively perform training operations on the car seat adjustment model based on a training sample data set; wherein the car seat adjustment model includes a human body part detection sub-model and a human body attribute prediction sub-model connected in sequence, and a seat angle evaluation sub-model connected to the human body part detection sub-model and the human body attribute prediction sub-model respectively; the training operation includes: selecting target training sample data from the training sample data set; inputting the original video frame image in the target training sample data into the car seat adjustment model, so that the car seat adjustment model extracts features from the original video frame image and detects human parts through the human body part detection sub-model to obtain human body part detection data , perform feature extraction and human attribute prediction on the human body part detection data through the human body attribute prediction sub-model to obtain human body attribute prediction data, and perform feature extraction and seat angle evaluation on the human body part detection data and human body attribute prediction data through the seat angle evaluation sub-model to obtain seat angle evaluation data; calculate a first loss value based on the human body part detection data, human body attribute prediction data and seat angle evaluation data, as well as the human body part label data, human body attribute label data and seat angle label data in the target training sample data; based on the first loss value, update the first model parameters of the human body part detection sub-model, the human body attribute prediction sub-model and the seat angle evaluation sub-model respectively.

[0082] In a fourth aspect, the present invention provides a vehicle seat adjustment device, comprising:

[0083] An acquisition module, used for acquiring video frame images in a car cabin;

[0084] A prediction module, configured to input a video frame image into a car seat adjustment model to obtain car seat angle data output by the car seat adjustment model; wherein the car seat adjustment model is trained using the above-mentioned car seat adjustment model training method;

[0085] The control module is used to control and adjust the angle of the car seat based on the car seat angle data.

[0086] In a fifth aspect, the present invention also provides an electronic device comprising: a processor and a memory; the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, it implements the above-mentioned car seat adjustment model training method or car seat adjustment method.

[0087] In a sixth aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, which implements the above-mentioned automobile seat adjustment model training method or automobile seat adjustment method when the computer program is executed by a processor.

[0088] The embodiments of the present invention provide a car seat adjustment model training method, a car seat adjustment method and a device, which obtain video frame images inside the car cabin; input the video frame images into the car seat adjustment model to obtain car seat angle data output by the car seat adjustment model; and control and adjust the angle of the car seat based on the car seat angle data to achieve a more precise adaptive adjustment effect of the car seat and improve the user experience.

[0089] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0091] Figure 1 A flow chart of a method for training a car seat adjustment model provided by an embodiment of the present invention is shown;

[0092] Figure 2 A flowchart of a method for obtaining a training sample set provided by an embodiment of the present invention is shown;

[0093] Figure 3 A schematic structural diagram of a car seat adjustment model provided by an embodiment of the present invention is shown;

[0094] Figure 4 A schematic structural diagram of a human body part detection sub-model provided in an embodiment of the present invention is shown;

[0095] Figure 5 A schematic diagram of the structure of a human attribute prediction sub-model provided by an embodiment of the present invention is shown;

[0096] Figure 6 A schematic diagram of the structure of a seat angle evaluation sub-model provided by an embodiment of the present invention is shown;

[0097] Figure 7 A schematic diagram showing the structure of a user behavior prediction sub-model provided by an embodiment of the present invention is shown;

[0098] Figure 8 A flow chart showing another method for training a car seat adjustment model provided by an embodiment of the present invention is shown;

[0099] Figure 9A schematic flow chart of a method for adjusting a car seat provided by an embodiment of the present invention is shown;

[0100] Figure 10 A schematic structural diagram of a car seat adjustment model training device provided by an embodiment of the present invention is shown;

[0101] Figure 11 A schematic structural diagram of a vehicle seat adjustment device provided by an embodiment of the present invention is shown;

[0102] Figure 12 A schematic structural diagram of an electronic device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0103] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0104] In order to facilitate those skilled in the art to better understand this application, the technical terms involved in this application are briefly introduced below.

[0105] The original video frame image is used to reflect the image data inside the car cabin. In this application, the video data is obtained through a monitoring device inside the car cabin, and the video data is frame-sliced ​​into a single image to obtain a video frame image.

[0106] Human body part data is data used to reflect the user's position in the car cabin. In this application, human body part detection data includes human body data, face data, upper body data, etc.

[0107] Human attribute data is data used to reflect the user's human attributes. In this application, human attribute prediction data includes weight, height, age, gender, etc.

[0108] The seat angle evaluation data is used to reflect the seat angle data in the car cabin. In this application, the seat angle evaluation data includes the leg extension angle and waist extension angle of the seat.

[0109] User behavior data is used to reflect the user's behavior on the seat. In this application, user behavior data includes standing up and not standing up.

[0110] The car seat adjustment model is a back propagation neural network (BPNN) model that predicts the original video frame image by learning the mapping relationship between the original video frame image and human body part data, human body attribute data, seat angle data and user behavior data. In this application, the car seat adjustment model includes a human body part detection sub-model, a human body attribute prediction sub-model, a seat angle evaluation sub-model and a user behavior prediction sub-model, wherein the human body part detection sub-model is connected to the human body attribute prediction sub-model, and the seat angle evaluation sub-model is respectively connected to the human body part detection sub-model and the human body attribute prediction sub-model; the user behavior prediction sub-model is connected to the human body part detection sub-model.

[0111] After introducing the technical terms involved in this application, the technical solutions provided by this application are described in detail.

[0112] The present application embodiment provides a method for training a car seat adjustment model. Figure 1 As shown, the general process of a car seat adjustment model training method provided in an embodiment of the present application is as follows:

[0113] Step 110: Obtain a training sample data set; wherein the training sample data set includes multiple training sample data; each training sample data includes an original video frame image and human body part label data, human body attribute label data, seat angle label data, and user behavior label data corresponding to the original video frame image.

[0114] In the embodiments of this application, Figure 2 As shown, when obtaining the training sample data set, the following methods can be used but are not limited to:

[0115] The occupancy monitoring system (OMS) in the car cabin collects RGB-IR (RGB-Infrared) video data from the car cabin and performs frame slicing on the video data to obtain a single original video frame image.

[0116] Based on the original video frame image, the human body part label data, human body attribute label data and seat angle label data corresponding to the original video frame image are determined by label annotation;

[0117] A training sample data set is obtained based on original video frame images, human body part label data, human body attribute label data, seat angle label data and user behavior label data.

[0118] Specifically, by performing target detection and user behavior recognition tasks on the RGB-IR original video stream data, a training sample data set is obtained. The specific process is as follows: Figure 2 As shown:

[0119] First, for the object detection task, the RGB-IR video data is frame-sliced ​​and the human-related frames in each frame are annotated to obtain the label data of the human body, face, and upper body. Specifically, a preset annotation model is first used to automatically annotate the human body, face, and upper body in each frame of the video image. Then, the object detection annotation tool is used for manual verification to obtain the label data GT of the human body, face, and upper body.

[0120] Label the weight, height, age, gender, and seat deployment angle in each frame of video image to obtain weight, height, age, and gender label data GT and seat deployment angle label data GT;

[0121] For the user behavior recognition task, the user's behavior in each frame of video image is annotated to obtain user action label data. Specifically, the user's standing up action in each frame of video image is manually annotated to obtain user action label data GT.

[0122] Furthermore, the label data GT is processed. First, the RGB color image and grayscale image in the original video frame image are normalized to obtain a feature image with a single channel value of 0-1; then, the label data GT of the human body, face and upper body are processed, and the VOC format label files in the label data of the human body, face and upper body are converted into the [cx, cy, h, w, cls] format, where cx and cy are the center point coordinates of the target bounding box, h is the height of the target bounding box, w is the width of the target bounding box, and cls is the category position index of the target category. For example, if the category is "upper body", cls is 2, the index starts from 0, and is recorded as the human body part label data GT-DET; weight and height are converted into ... , age and gender label data GT are converted to obtain human attribute label data GT-CLS, for example, gender: male-0, female-1; age: category: 100 categories (i.e. 1-100 years old); weight: 70 categories (i.e. 30-100 kg); height: 60 categories (i.e. 140-200 cm); the seat deployment angle label data GT is normalized to obtain seat angle label data GT-Angle, for example, the seat angle includes the foot deployment angle and the waist deployment angle, where the maximum angle is 121 degrees; the user action label data GT is binary-classified to obtain user behavior label data GT-Behavior, for example, the user's standing up action is converted to 1, and the user's non-standing up action is converted to 0.

[0123] Furthermore, in the embodiment of the present application, the original video frame image is normalized, specifically including the following process:

[0124] Use the cv2.imread function to read the image data in the original video frame image and convert the image data channels from BGR to RGB. At this time, the dimensions of the original video frame image are [B, C, H, W], where C = 3; where B represents the number of samples, C represents the channel data, H represents the height of the image, and W represents the width of the image.

[0125] The converted original video frame image is then processed by normalization. For example, the original RGB value is [230, 100, 20]. After normalization, it is divided by 255 to [0.0784, 0.0824, 0.0980]. Then the average value [0.485, 0.456, 0.406] is subtracted in RGB order, and then divided by the standard deviation [0.229, 0.224, 0.225] to finally get [-1.7754, -1.6681, -1.3687], so that all values ​​have zero mean and unit variance after processing.

[0126] Grayscale image normalization follows a similar process based on RGB color images. However, to ensure that channel C = 3, the grayscale image channel needs to be repeated so that C is converted from 1 to 3. Normalizing the original video frame image improves the convergence and stability of the car seat adjustment model training.

[0127] Step 120: Select target training sample data from the training sample data set.

[0128] Step 130: Input the original video frame image in the target training sample data into the automobile seat adjustment model, so that the automobile seat adjustment model performs feature extraction and human part detection on the original video frame image through the human body part detection sub-model to obtain human body part detection data, performs feature extraction and human body attribute prediction on the human body part detection data through the human body attribute prediction sub-model to obtain human body attribute prediction data, performs feature extraction and seat angle evaluation on the human body part detection data and human body attribute prediction data through the seat angle evaluation sub-model to obtain seat angle evaluation data; and performs feature extraction and user behavior prediction on the human body part detection data through the user behavior prediction sub-model to obtain user behavior prediction data.

[0129] Step 140: Calculate a first loss value based on the human body part detection data, human body attribute prediction data, and seat angle evaluation data, as well as the human body part label data, human body attribute label data, and seat angle label data in the target training sample data; and calculate a second loss value based on the human body part detection data and user behavior prediction data, as well as the human body part label data and user behavior label data in the target training sample data.

[0130] Step 150: Based on the first loss value, update the first model parameters of the human body part detection sub-model, the human body attribute prediction sub-model and the seat angle evaluation sub-model; and based on the second loss value, update the second model parameters of the human body part detection sub-model and the first model parameters of the user behavior prediction sub-model.

[0131] Step 160, determine whether the iterative training termination condition is met; if so, execute step 170; if not, return to step 120; wherein, the iterative training termination condition is that the number of iterations is not less than the number threshold, or the first loss value is not higher than the first loss value threshold, and the second loss value is not higher than the second loss value threshold.

[0132] Step 170: Based on the first model parameters of the human body part detection sub-model, the human body attribute prediction sub-model and the seat angle evaluation sub-model updated during the last iterative training operation, as well as the second model parameters of the human body part detection sub-model and the first model parameters of the user behavior prediction sub-model, a car seat adjustment model is obtained.

[0133] like Figure 3 As shown, the car seat adjustment model structure specifically includes a human body part detection sub-model, a user behavior prediction sub-model, a human body attribute prediction sub-model and a seat angle evaluation sub-model, wherein the human body part detection sub-model is connected to the human body attribute prediction sub-model, the seat angle evaluation sub-model and the user behavior prediction sub-model respectively, and the human body attribute prediction sub-model is connected to the seat angle evaluation sub-model; the human body part detection sub-model is used to extract features and detect human parts of the original video frame image to obtain human body part detection data, the human body attribute prediction sub-model is used to extract features and predict human attributes of the human body part detection data to obtain human body attribute prediction data, the seat angle evaluation sub-model is used to extract features and evaluate the human body part detection data and human attribute prediction data to obtain seat angle evaluation data; and the user behavior prediction sub-model is used to extract features and predict user behavior of the human body part detection data to obtain user behavior prediction data, and based on the seat angle evaluation data and user behavior prediction data, the human body part label data GT_DET, the human attribute label data GT_CLS, the seat angle label data GT_Angle, and the user behavior prediction data GT_Behavior, the loss is calculated to obtain a loss value.

[0134] Furthermore, the human body part detection sub-model introduces a spatial channel layer convolution operation (SCConv) module, which enhances its ability to learn local features, thereby improving fine-grained recognition capabilities and more accurately capturing the detailed features of the target. This can effectively improve detection accuracy and robustness, as well as the expressiveness of features at different scales, especially in complex scenarios. A dual-layer attention mechanism layer (BiFormer) is introduced to achieve efficient computing performance and query-aware adaptability. It can dynamically and sparsely focus on the most relevant feature information to fuse local and global information, improve the accuracy of human body part detection, and enhance robustness and generalization capabilities in complex scenarios. It also enhances the human body part detection sub-model's ability to handle targets of different scales.

[0135] The human attribute prediction sub-model uses a residual network (unifyNet-Backbone) built by the sub-construction, and introduces a compressed excitation attention layer (SELayer) and a multi-convolutional layer (Multi-CNN) to process the image, and then predicts human attributes through multiple fully connected layers (FC);

[0136] The user behavior prediction sub-model uses a residual-like network (ResNet-FC) as its backbone and introduces a compressed attention layer (SELayer). By adaptively readjusting the weights between channels, it can better capture important feature information, thereby improving the model's sensitivity to key targets. A fully connected layer (FC) is added to the last layer to convert network features into user behavior categories.

[0137] When calculating the loss, the human part detection sub-model uses the iou-loss loss to calculate the iou loss between the predicted bounding box and the true bounding box; Obj-loss is used to calculate the loss between the predicted target confidence and the true confidence; cls-loss-det is used to calculate the loss between the classification prediction and the true classification; the loss of the human attribute prediction sub-model is recorded as cls-loss-all, which is the sum of the losses of the four fully connected FC layers; the loss of the seat angle evaluation sub-model is recorded as L1-loss to calculate the loss between the predicted seat angle and the true angle; the loss of the user behavior prediction sub-model is cls-loss-beh; all losses are multiplied by the adaptive weight parameter ada-weights and added together, and the total loss is recorded as loss, and the entire model parameters are jointly optimized.

[0138] In an optional embodiment, as Figure 4As shown in FIG, the human body part detection sub-model includes a multi-scale feature extraction module, a multi-scale feature enhancement module, a multi-scale feature fusion module, and a human body part detection module connected in sequence; wherein, the human body part detection sub-model performs feature extraction and human body part detection on the original video frame image to obtain human body part detection data, including:

[0139] The original video frame image is preprocessed and multi-scale feature extraction is performed through the multi-scale feature extraction module to obtain first feature images of multiple scales; the first feature image of each scale is subjected to feature enhancement processing through the multi-scale feature enhancement module to obtain the second feature image corresponding to each first feature image; the multiple second feature images are subjected to feature fusion processing through the multi-scale feature fusion module to obtain the third feature image; the third feature image is subjected to target recognition through the human body part detection module to obtain human body part detection data.

[0140] Furthermore, the multi-scale feature extraction module includes a normalization layer and multiple downsampling layers connected in sequence; the multi-scale feature enhancement module includes multiple reconstruction attention units, each reconstruction attention unit includes a spatial channel reconstruction layer and a two-layer routing attention layer connected in sequence; the multi-scale feature fusion module includes a path aggregation feature fusion layer; the human body part detection module includes a decoupling layer; the multiple downsampling layers are connected to the multiple reconstruction attention units in a one-to-one correspondence; the multiple reconstruction attention units are respectively connected to the path aggregation feature fusion layer; the path aggregation feature fusion layer is connected to the decoupling layer; wherein, the original video frame image is subjected to feature extraction and human body part detection by the human body part detection sub-model to obtain human body part detection data, including:

[0141] The original video frame image is normalized through the normalization layer to obtain a normalized video frame image; the normalized video frame image is sequentially downsampled at different scales through multiple downsampling layers to obtain first feature images of multiple scales; the spatial channel reconstruction layer and the double-layer routing attention layer in the multiple reconstruction attention units are used to perform spatial channel reconstruction and key feature interaction processing on the multiple first feature images to obtain second feature images corresponding to the multiple first feature images; the path aggregation feature fusion layer is used to perform bidirectional path feature extraction and adaptive feature fusion processing on the multiple second feature images to obtain the third feature image; the decoupling layer is used to perform human body part classification and positioning processing on the third feature image to obtain human body part detection data.

[0142] Specifically, if Figure 4As shown in the figure, the normalization layer (Batch-images) of the human body part detection sub-model normalizes the original video frame image (RGB-IR feature image) to obtain a normalized video frame image; the normalized video frame image is downsampled by the first downsampling layer (Stem-Conv) to obtain the first downsampled image; the first downsampled image is downsampled by the second downsampling layer (Dark2) to obtain the second downsampled image; the second downsampled image is downsampled by the third downsampling layer (Dark3), the fourth downsampling layer (Dark4) and the fifth downsampling layer (Dark5) in sequence. Sampling processing is performed to obtain first feature images stage3, stage4 and stage5 at multiple scales; spatial channel reconstruction and key feature interaction processing are performed on stage3, stage4 and stage5 respectively through multiple spatial channel reconstruction layers SCConv and double-layer routing attention layer BiFormer to obtain multiple second feature images; bidirectional path feature extraction and adaptive feature fusion processing are performed on multiple second feature images through path aggregation feature fusion layer (PAFPN) to obtain third feature image; human body part classification and positioning processing are performed on the third feature image through decoupling layer to obtain human body part detection data.

[0143] In an embodiment of the present application, the human body part detection sub-model adjusts the convolution operation of the human body part detection sub-model through the spatial channel layer, thereby enhancing the learning ability of the human body part detection sub-model for local features, thereby improving the fine-grained recognition ability, and being able to more accurately capture the detailed features of the target, especially in complex scenarios, and can effectively improve the detection accuracy, robustness and the expression ability of features at different scales; through the double-layer attention mechanism layer to achieve efficient computing performance and query perception adaptability, it can dynamically and sparsely focus on the most relevant feature information to fuse local and global information, improve the accuracy of human body part detection, and improve robustness and generalization ability in complex scenarios, and enhance the processing ability of the human body part detection sub-model for targets of different scales.

[0144] In an optional embodiment, as Figure 5 As shown, the human attribute prediction sub-model includes multiple first residual attention modules connected in sequence, first feature fusion modules connected to the multiple first residual attention modules respectively, and a human attribute prediction module connected to the first feature fusion module; wherein, the human attribute prediction sub-model performs feature extraction and human attribute prediction on human part detection data to obtain human attribute prediction data, including:

[0145] The human body part detection data is sequentially subjected to feature extraction through multiple first residual attention modules to obtain first human body part feature images output by multiple first residual attention modules; the multiple first human body feature images are subjected to feature fusion processing through a first feature fusion module to obtain a first human body feature fusion image; the first human body feature fusion image is subjected to human body attribute prediction through a human body attribute prediction module to obtain human body attribute prediction data.

[0146] Furthermore, each first residual attention module includes at least two first residual attention units connected in sequence, each first residual attention unit includes a first residual layer and a first compressed excitation attention layer jump-connected to the first residual layer; the first feature fusion module includes a first parallelized feature interaction fusion layer; the human body attribute prediction module includes multiple fully connected layers; wherein, the human body attribute prediction sub-model is used to perform feature extraction and human body attribute prediction on the human body part detection data to obtain human body attribute prediction data, including:

[0147] Through the first residual layer and the first compressed excitation attention layer in each first residual attention unit in multiple first residual attention modules, the human body part detection data is sequentially subjected to feature extraction at different semantic levels and feature enhancement processing across semantic levels to obtain first human body part feature images output by multiple first residual attention modules; through the first parallelized feature interactive fusion layer in the first feature fusion module, the multiple first human body part feature images are subjected to parallelized feature extraction and cross-level feature fusion processing to obtain a first human feature fusion image; through the multiple fully connected layers in the human attribute prediction module, the first human feature fusion image is subjected to multi-classification mapping processing to obtain human attribute prediction data.

[0148] Specifically, if Figure 5 As shown in the figure, the human attribute prediction sub-model includes three first residual attention modules, a first feature fusion module and four fully connected layers (FC) connected in sequence; wherein, each first residual attention module includes two first residual attention units connected in sequence, each first residual attention unit includes a first residual layer and a first compressed excitation attention layer (SELayer) jump-connected to the first residual layer, and the first residual layer includes two convolutional layers (conv) connected in sequence; the number of convolution kernels in the first first residual attention module is 64, with a size of 3x3, the number of convolution kernels in the second first residual attention module is 128, with a size of 3x3, and the number of convolution kernels in the third first residual attention module is 256, with a size of 3x3; the first feature fusion module uses a multi-convolutional neural network Multi-CNN to perform parallel feature extraction and cross-level feature fusion processing on multiple first human body part feature images to obtain a first human feature fusion image.

[0149] In an optional embodiment, as Figure 6 As shown, the seat angle assessment sub-model includes a plurality of second residual attention modules connected in sequence, a second feature fusion module connected to the plurality of second residual attention modules, and a seat angle assessment module connected to the second feature fusion module; wherein, the seat angle assessment sub-model performs feature extraction and seat angle assessment on human body part detection data and human body attribute prediction data to obtain seat angle assessment data, including:

[0150] The human body part detection data are sequentially subjected to feature extraction by multiple second residual attention modules to obtain second human body part feature images output by multiple second residual attention modules; the multiple second human body part feature images are subjected to feature fusion processing by a second feature fusion module to obtain a second human body feature fusion image; the seat angle is evaluated based on the human body attribute prediction data and the second human body feature fusion image by a seat angle evaluation module to obtain seat angle evaluation data.

[0151] Furthermore, each second residual attention module includes at least two second residual attention units connected in sequence, and each second residual attention unit includes a second residual layer and a second compressed excitation attention layer jump-connected to the second residual layer; the second feature fusion module includes a second parallelized feature interaction fusion layer; the seat angle evaluation module includes a fully connected layer; wherein, the human body part detection data and the human body attribute prediction data are subjected to feature extraction and seat angle evaluation by the seat angle evaluation sub-model to obtain seat angle evaluation data, including: performing different semantic levels on the human body part detection data in sequence through the second residual layer and the second compressed excitation attention layer in each second residual attention unit in multiple second residual attention modules The feature extraction and cross-semantic level feature enhancement processing are performed to obtain the second human body part feature images output by multiple second residual attention modules; the second parallelized feature interactive fusion layer in the second feature fusion module is used to perform parallel feature extraction and cross-level feature fusion processing on the multiple second human body part feature images to obtain the second human body feature fusion image; the fully connected layer in the seat angle evaluation module is used to determine the human body proportion data based on the human body attribute prediction data, and then the first seat deployment angle is estimated based on the human body proportion data, and the key point extraction and angle calculation are performed based on the second human body feature fusion image to obtain the second seat deployment angle, and the seat angle evaluation data is calculated based on the first seat deployment angle and the second seat deployment angle.

[0152] Specifically, if Figure 6As shown in the figure, the seat angle evaluation sub-model includes three second residual attention modules, a second feature fusion module and a fully connected layer (FC) connected in sequence; each second residual attention module includes two second residual attention units connected in sequence, each second residual attention unit includes a second residual layer and a second compressed excitation attention layer (SELayer) jump-connected to the second residual layer, and the second residual layer includes two convolutional layers (conv) connected in sequence; the number of convolution kernels in the first second residual attention module is 64, with a size of 3x3, the number of convolution kernels in the second second residual attention module is 128, with a size of 3x3, and the number of convolution kernels in the third second residual attention module is 256, with a size of 3x3; the second feature fusion module uses a multi-convolutional neural network Multi-CNN to perform feature fusion processing on multiple second human body part feature images to obtain a second human feature fusion image.

[0153] In an optional embodiment, as Figure 7 As shown, the user behavior prediction sub-model includes multiple third residual attention modules and user behavior prediction modules connected in sequence; wherein, the user behavior prediction sub-model performs feature extraction and user behavior prediction on the human body part detection data to obtain user behavior prediction data, including: performing feature extraction on the human body part detection data in sequence through multiple third residual attention modules to obtain third human body part feature images output by multiple third residual attention modules; performing user behavior prediction on the third human body part feature images through the user behavior prediction module to obtain user behavior prediction data.

[0154] Furthermore, each third residual attention module includes a third residual layer and a third compressed excitation attention layer jump-connected to the third residual layer; the user behavior prediction module includes a fourth residual layer and a fully connected layer jump-connected to the fourth residual layer; wherein, the user behavior prediction sub-model performs feature extraction and user behavior prediction on the human body part detection data to obtain user behavior prediction data, including: through the third residual layer and the third compressed excitation attention layer in multiple third residual attention modules, the human body part detection data is sequentially subjected to feature extraction at different semantic levels and feature enhancement processing across semantic levels to obtain third human body part feature images output by multiple third residual attention modules; through the fourth residual layer and the fully connected layer in the user behavior prediction module, the third human body part feature image is subjected to feature extraction and binary classification mapping processing at different semantic levels to obtain user behavior prediction data.

[0155] Specifically, if Figure 7As shown, the user behavior prediction sub-model includes 6 third residual layers connected in sequence, a third compressed excitation attention layer (SELayer) and a fourth residual layer connected to the third residual layer, and a fully connected layer (FC) connected to the fourth residual layer. The third residual layer includes 2 convolutional layers (conv) connected in sequence; the number of convolution kernels in the first third residual layer is 64, with a size of 3x3, the number of convolution kernels in the second and third third residual layers is 128, with a size of 3x3, the number of convolution kernels in the fourth and fifth third residual layers is 256, with a size of 3x3, the number of convolution kernels in the sixth third residual layer is 512, with a size of 3x3, and the number of convolution kernels in the fourth residual layer is 512, with a size of 3x3.

[0156] In an embodiment of the present application, the user behavior prediction sub-model can better capture important feature confidence by adaptively adjusting the weights between channels through the compressed attention layer, thereby improving the sensitivity of the user behavior prediction sub-model to key targets, so that the user behavior prediction sub-model has stronger capabilities in feature selection, especially in complex scenarios. The feature selection is significantly improved; by adding a fully connected layer before the output layer, the grid features are converted into user behavior categories to obtain user behavior prediction data.

[0157] In an optional embodiment, a first loss value is calculated based on human body part detection data, human body attribute prediction data and seat angle evaluation data, as well as human body part label data, human body attribute label data and seat angle label data in target training sample data, including: obtaining a first human body detection loss value based on the human body part detection data and the human body part label data in the target training sample data; obtaining an attribute prediction loss value based on the human body attribute prediction data and the human body attribute label data in the target training sample data; obtaining an angle evaluation loss value based on the seat angle evaluation data and the seat angle label data in the target training sample data; obtaining a first loss value based on the first human body detection loss value, the attribute prediction loss value and the angle evaluation loss value.

[0158] Furthermore, based on the first loss value, the first model parameters of each of the human body part detection sub-model, the human body attribute prediction sub-model and the seat angle evaluation sub-model are updated, including: determining the first gradient value corresponding to the seat angle evaluation sub-model based on the first loss value and the seat angle evaluation data; updating the first gradient value based on the optimization function, inputting the updated first gradient value into the seat angle evaluation sub-model to perform back propagation to update the first model parameters of the seat angle evaluation sub-model; determining the second gradient value corresponding to the human body attribute prediction sub-model based on the first loss value and the human body attribute prediction data; updating the second gradient value based on the optimization function, inputting the updated second gradient value into the human body attribute prediction sub-model to perform back propagation to update the first model parameters of the human body attribute prediction sub-model; determining the third gradient value corresponding to the human body part detection sub-model based on the first loss value and the human body part detection data; updating the third gradient value based on the optimization function, inputting the updated third gradient value into the human body part detection sub-model to perform back propagation to update the first model parameters of the human body part detection sub-model.

[0159] In an optional embodiment, a second loss value is calculated based on the human body part detection data and the user behavior prediction data, as well as the human body part label data and the user behavior label data in the target training sample data, including: obtaining a second human body detection loss value based on the human body part detection data and the human body part label data in the target training sample data; obtaining a behavior prediction loss value based on the user behavior prediction data and the user behavior label data in the target training sample data; and obtaining a third loss value based on the second human body detection loss value and the behavior prediction loss value.

[0160] Further, based on the second loss value, the second model parameters of the human body part detection sub-model and the first model parameters of the user behavior prediction sub-model are updated, including: determining the fourth gradient value corresponding to the human body part detection sub-model based on the second loss value and the human body part detection data; updating the fourth gradient value based on the optimization function, inputting the updated fourth gradient value into the human body attribute prediction sub-model to perform back propagation to update the second model parameters of the human body attribute prediction sub-model; determining the fifth gradient value corresponding to the user behavior prediction sub-model based on the second loss value and the user behavior prediction data; updating the fifth gradient value based on the optimization function, inputting the updated fifth gradient value into the user behavior prediction sub-model to perform back propagation to update the first model parameters of the user behavior prediction sub-model.

[0161] In the embodiments of this application, Figure 8 As shown, the human body part detection sub-model and the human body attribute prediction sub-model constitute a regression model; based on the first loss value, the first model parameters of the human body part detection sub-model, the human body attribute prediction sub-model and the seat angle evaluation sub-model are updated, specifically including:

[0162] Based on the human body part detection data and the human body part label data in the target training sample data, a first human body detection loss value is obtained (result 1); based on the human body attribute prediction data and the human body attribute label data GTS-CLS in the target training sample data, an attribute prediction loss value is obtained (result 2); based on the seat angle evaluation data and the seat angle label data GT-Angle in the target training sample data, an angle evaluation loss value is obtained (result 3); based on the first human body detection loss value, the attribute prediction loss value and the angle evaluation loss value, a first loss value L1 is obtained;

[0163] Based on the first loss value L1 and the first human detection loss value, the attribute prediction loss value, and the angle evaluation loss value, the third gradient value corresponding to the human body part detection sub-model, the second gradient value corresponding to the human body attribute prediction sub-model, and the first gradient value corresponding to the seat angle evaluation sub-model are determined. The expression of the gradient value is:

[0164]

[0165] Where W is the weight and Y is the output value of the model;

[0166] and inputting the first gradient value, the second gradient value, and the third gradient value into the optimizer to update the first gradient value, the second gradient value, and the third gradient value;

[0167] The updated first gradient value and second gradient value are input into the regression model, and the updated third gradient value is input into the seat angle evaluation sub-model. Gradient optimization is performed after back propagation, and the first model parameters of the regression model and the seat angle evaluation sub-model are updated. After the first loss value reaches the first loss value threshold, training is stopped and the weight values ​​of the regression model and the seat angle evaluation sub-model are saved.

[0168] Furthermore, in the embodiment of the present application, based on the second loss value, updating the second model parameter of the human body part detection sub-model and the first model parameter of the user behavior prediction sub-model specifically includes:

[0169] Based on the human body part detection data and the human body part label data GTS-DET in the target training sample data, a second human body detection loss value is obtained; based on the user behavior prediction data and the user behavior label data GTS-Behavior in the target training sample data, a behavior prediction loss value is obtained (Result 4); based on the second human body detection loss value and the behavior prediction loss value, a second loss value L2 is obtained;

[0170] Determine a fourth gradient value corresponding to the human body part detection sub-model and a fifth gradient value corresponding to the user behavior prediction sub-model based on the second loss value L2, the second human body detection loss value, and the behavior prediction loss value;

[0171] and inputting the fourth gradient value and the fifth gradient value into the optimizer to update the fourth gradient value and the fifth gradient value;

[0172] The updated fourth gradient value is input into the human body part detection sub-model, and the updated fifth gradient value is input into the user behavior prediction sub-model. Gradient optimization is performed after back propagation to update the second model parameters of the human body part detection sub-model and the first model parameters of the user behavior prediction model. After the second loss value reaches the second loss value threshold, training is stopped and the weight values ​​of the human body part detection sub-model and the user behavior prediction sub-model are saved.

[0173] In this application, the regression model and the seat angle evaluation model are optimized by the first loss value, and the user behavior prediction model is optimized by the second loss value, so as to achieve overall optimization of the car seat adjustment model and improve the training process of the car seat adjustment model.

[0174] The embodiment of the present application provides a method for adjusting a car seat, which is mainly used to adjust the zero-gravity seat of a car. Figure 9 As shown, the general process of a car seat adjustment method provided in an embodiment of the present application is as follows:

[0175] Step 210: Acquire a video frame image in the vehicle cabin;

[0176] Step 220: Input the video frame image into the car seat adjustment model to obtain car seat angle data output by the car seat adjustment model; wherein the car seat adjustment model is trained using the above-mentioned car seat adjustment model training method;

[0177] Step 230: Based on the car seat angle data, control and adjust the angle of the car seat.

[0178] In an optional embodiment, the car seat adjustment model further outputs user behavior data; and controlling the angle of the car seat based on the car seat angle data includes:

[0179] Based on user behavior data, when it is determined that the user is in a standing state, the angle of the car seat is controlled and adjusted based on the car seat angle data.

[0180] In an embodiment of the present application, video frame images in the car cabin are acquired in real time, and the video frame images are sent to a trained car seat adjustment model, and the car seat adjustment model is used to determine whether the user is in a normal sitting posture. The seat adjustment evaluation submodel in the car seat adjustment model can be used to determine whether the user is in a normal sitting posture. If the current user is in a normal sitting posture and the adjustment function of the seat is not turned on, the user behavior prediction submodel in the car seat adjustment model is in a non-working state, that is, the current car seat does not need to be adjusted; if the current user is in an abnormal sitting posture and the adjustment function of the seat is turned on, the user behavior prediction submodel in the car seat adjustment model is in a working state, and the current car seat is adjusted according to the user behavior prediction data output by the user behavior prediction submodel in combination with the seat adjustment evaluation submodel until the user behavior is in a stopped state or the seat is in a stopped state. The chair angle is restored to a normal sitting position, wherein the user behavior prediction data output by the user behavior prediction sub-model is standing up, then the car seat adjustment model performs feature extraction and human part detection on the original video frame image through the human body part detection sub-model to obtain human body part detection data, performs feature extraction and human body attribute prediction on the human body part detection data through the human body attribute prediction sub-model to obtain human body attribute prediction data, performs feature extraction and seat angle evaluation on the human body part detection data and the human body attribute prediction data through the seat angle evaluation sub-model to obtain current seat angle evaluation data, and then the car seat adjustment model continues to output seat angle evaluation data based on the current user behavior prediction data and the current seat angle evaluation data, until the user behavior prediction data shows that the user is in a normal sitting position or the current seat angle evaluation data is in the state of returning to a normal sitting position, and the car seat adjustment model stops working.

[0181] The present application embodiment provides a car seat adjustment model training device, see Figure 10 As shown, an embodiment of the present application provides a car seat adjustment model training device comprising:

[0182] The data acquisition module 310 is used to acquire a training sample data set; wherein the training sample data set includes a plurality of training sample data; each training sample data includes an original video frame image and human body part label data, human body attribute label data, and seat angle label data corresponding to the original video frame image;

[0183] The model training module 320 is used to iteratively perform training operations on the car seat adjustment model based on the training sample data set; wherein the car seat adjustment model includes a human body part detection sub-model and a human body attribute prediction sub-model connected in sequence, and a seat angle evaluation sub-model connected to the human body part detection sub-model and the human body attribute prediction sub-model respectively; the training operation includes: selecting target training sample data from the training sample data set; inputting the original video frame image in the target training sample data into the car seat adjustment model, so that the car seat adjustment model extracts features and detects human parts on the original video frame image through the human body part detection sub-model to obtain human body part detection data, performing feature extraction and human attribute prediction on the human body part detection data through a human body attribute prediction sub-model to obtain human body attribute prediction data, and performing feature extraction and seat angle evaluation on the human body part detection data and the human body attribute prediction data through a seat angle evaluation sub-model to obtain seat angle evaluation data; calculating a first loss value based on the human body part detection data, the human body attribute prediction data and the seat angle evaluation data, as well as the human body part label data, the human body attribute label data and the seat angle label data in the target training sample data; and updating the first model parameters of the human body part detection sub-model, the human body attribute prediction sub-model and the seat angle evaluation sub-model based on the first loss value.

[0184] The present application provides a vehicle seat adjustment device. Figure 11 As shown, an automobile seat adjustment device provided in an embodiment of the present application includes:

[0185] An acquisition module 410 is used to acquire a video frame image in a vehicle cabin;

[0186] Prediction module 420, configured to input the video frame image into the car seat adjustment model to obtain car seat angle data output by the car seat adjustment model; wherein the car seat adjustment model is trained using the above-mentioned car seat adjustment model training method;

[0187] The control module 430 is used to control and adjust the angle of the car seat based on the car seat angle data.

[0188] The device provided in the embodiment of the present application has the same implementation principle and technical effects as those in the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the device embodiment, reference can be made to the corresponding content in the aforementioned method embodiment.

[0189] like Figure 12As shown, an electronic device 600 provided in an embodiment of the present application includes: a processor 510, a memory 520 and a bus, the memory 520 stores machine-readable instructions executable by the processor 510, when the electronic device is running, the processor 510 and the memory 520 communicate through the bus, and the processor 510 executes the machine-readable instructions to perform the steps of the above-mentioned car seat adjustment training method or car seat adjustment method.

[0190] Specifically, the above-mentioned memory 520 and processor 510 can be general-purpose memory and processor, which are not specifically limited here. When the processor 510 runs the computer program stored in the memory 520, it can execute the above-mentioned car seat adjustment training method or car seat adjustment method.

[0191] The processor 510 may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by an integrated logic circuit of hardware in the processor 510 or by instructions in the form of software. The above-mentioned processor 510 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The various methods, steps, and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in conjunction with the embodiments of the present application can be directly embodied as being executed by a hardware decoding processor, or can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium well-known in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in memory 520, and processor 510 reads information in memory 520 and, in conjunction with its hardware, completes the steps of the above method.

[0192] Corresponding to the above-mentioned car seat adjustment training method or car seat adjustment method, an embodiment of the present application also provides a computer-readable storage medium, which stores machine-executable instructions. When the computer-executable instructions are called and executed by the processor, the computer-executable instructions prompt the processor to execute the steps of the above-mentioned car seat adjustment training method or car seat adjustment method.

[0193] The car seat adjustment training device or car seat adjustment device provided in the embodiment of the present application can be specific hardware on the device or software or firmware installed on the device. The device provided in the embodiment of the present application, its implementation principle and the technical effect produced are the same as those in the aforementioned method embodiment. For the sake of brief description, where the device embodiment is not mentioned, reference can be made to the corresponding content in the aforementioned method embodiment. Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can all refer to the corresponding processes in the aforementioned method embodiment, and will not be repeated here.

[0194] In the embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interface, the indirect coupling or communication connection of the device or unit can be electrical, mechanical or other forms.

[0195] For another example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of code, and a part of the module, program segment or code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of the boxes in the block diagram and / or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0196] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0197] In addition, each functional unit in the embodiments provided in the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0198] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling an electronic device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, and other media that can store program code.

[0199] It should be noted that similar numbers and letters represent similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are only used to distinguish the description and are not to be understood as indicating or implying relative importance.

[0200] Finally, it should be noted that the above embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The scope of protection of the present application is not limited thereto. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above embodiments within the technical scope disclosed in the present application, or perform equivalent replacements for some of the technical features thereof. However, these modifications, changes, or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present application. They should all be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A car seat adjustment model training method, characterized in that: include: Acquire a training sample data set; wherein the training sample data set includes a plurality of training sample data; each of the training sample data includes an original video frame image and human body part label data, human body attribute label data, and seat angle label data corresponding to the original video frame image; Based on the training sample data set, a training operation is iteratively performed on the automobile seat adjustment model; wherein the automobile seat adjustment model includes a human body part detection sub-model and a human body attribute prediction sub-model connected in sequence, and a seat angle assessment sub-model connected to the human body part detection sub-model and the human body attribute prediction sub-model respectively; the training operation includes: Selecting target training sample data from the training sample data set; Inputting the original video frame image in the target training sample data into the automobile seat adjustment model, so that the automobile seat adjustment model performs feature extraction and human part detection on the original video frame image through the human part detection sub-model to obtain human part detection data, performs feature extraction and human attribute prediction on the human part detection data through the human attribute prediction sub-model to obtain human attribute prediction data, and performs feature extraction and seat angle evaluation on the human part detection data and the human attribute prediction data through the seat angle evaluation sub-model to obtain seat angle evaluation data; Calculating a first loss value based on the human body part detection data, the human body attribute prediction data, and the seat angle evaluation data, as well as the human body part label data, the human body attribute label data, and the seat angle label data in the target training sample data; Based on the first loss value, the first model parameters of the human body part detection sub-model, the human body attribute prediction sub-model and the seat angle evaluation sub-model are updated.

2. The car seat adjustment model training method according to claim 1, characterized in that: Calculating a first loss value based on the human body part detection data, the human body attribute prediction data, and the seat angle evaluation data, as well as the human body part label data, the human body attribute label data, and the seat angle label data in the target training sample data, includes: Obtaining a first human body detection loss value based on the human body part detection data and the human body part label data in the target training sample data; Obtaining an attribute prediction loss value based on the human attribute prediction data and the human attribute label data in the target training sample data; Obtaining an angle evaluation loss value based on the seat angle evaluation data and the seat angle label data in the target training sample data; The first loss value is obtained based on the first human detection loss value, the attribute prediction loss value and the angle evaluation loss value.

3. The car seat adjustment model training method according to claim 1, characterized in that: Updating first model parameters of the human body part detection sub-model, the human body attribute prediction sub-model, and the seat angle assessment sub-model based on the first loss value includes: determining a first gradient value corresponding to the seat angle assessment sub-model based on the first loss value and the seat angle assessment data; updating the first gradient value based on an optimization function, and inputting the updated first gradient value into the seat angle assessment sub-model to perform backpropagation to update a first model parameter of the seat angle assessment sub-model; Determining a second gradient value corresponding to the human attribute prediction sub-model based on the first loss value and the human attribute prediction data; updating the second gradient value based on an optimization function, inputting the updated second gradient value into the human attribute prediction sub-model to perform back propagation to update a first model parameter of the human attribute prediction sub-model; Based on the first loss value and the human body part detection data, a third gradient value corresponding to the human body part detection sub-model is determined; the third gradient value is updated based on the optimization function, and the updated third gradient value is input into the human body part detection sub-model to perform back propagation to update the first model parameter of the human body part detection sub-model.

4. The car seat adjustment model training method according to claim 1, characterized in that: The car seat adjustment model also includes a user behavior prediction sub-model, which is connected to the human body part detection sub-model; each of the training sample data also includes user behavior label data corresponding to the original video frame image; After inputting the original video frame image in the target training sample data into the car seat adjustment model, the method further includes: Performing feature extraction and user behavior prediction on the human body part detection data by the user behavior prediction sub-model to obtain user behavior prediction data; Calculating a second loss value based on the human body part detection data and the user behavior prediction data, and the human body part label data and the user behavior label data in the target training sample data; Based on the second loss value, the second model parameters of the human body part detection sub-model and the first model parameters of the user behavior prediction sub-model are updated.

5. The vehicle seat adjustment model training method according to any one of claims 1 to 4, characterized in that: The human body part detection sub-model includes a multi-scale feature extraction module, a multi-scale feature enhancement module, a multi-scale feature fusion module and a human body part detection module connected in sequence; The human body part detection sub-model is used to perform feature extraction and human body part detection on the original video frame image to obtain human body part detection data, including: Preprocessing and multi-scale feature extraction are performed on the original video frame image by the multi-scale feature extraction module to obtain first feature images of multiple scales; performing feature enhancement processing on the first feature image at each scale by the multi-scale feature enhancement module to obtain a second feature image corresponding to each first feature image; Performing feature fusion processing on the plurality of second feature images by the multi-scale feature fusion module to obtain a third feature image; The human body part detection module performs target recognition on the third feature image to obtain the human body part detection data.

6. The car seat adjustment model training method according to claim 5, characterized in that: The multi-scale feature extraction module includes a normalization layer and multiple downsampling layers connected in sequence; the multi-scale feature enhancement module includes multiple reconstruction attention units, each of which includes a spatial channel reconstruction layer and a two-layer routing attention layer connected in sequence; the multi-scale feature fusion module includes a path aggregation feature fusion layer; The human body part detection module includes a decoupling layer; a plurality of the downsampling layers are connected to a plurality of the reconstruction attention units in a one-to-one correspondence; a plurality of the reconstruction attention units are respectively connected to the path aggregation feature fusion layer; the path aggregation feature fusion layer is connected to the decoupling layer; The human body part detection sub-model is used to perform feature extraction and human body part detection on the original video frame image to obtain human body part detection data, including: Normalizing the original video frame image through the normalization layer to obtain a normalized video frame image; Downsampling the normalized video frame image in different scales in sequence through the multiple downsampling layers to obtain the first feature images of multiple scales; Performing spatial channel reconstruction and key feature interaction processing on the plurality of first feature images respectively through the spatial channel reconstruction layer and the double-layer routing attention layer in the plurality of reconstruction attention units to obtain the second feature images corresponding to the plurality of first feature images; Performing bidirectional path feature extraction and adaptive feature fusion processing on the plurality of second feature images through the path aggregation feature fusion layer to obtain the third feature image; The third feature image is subjected to human body part classification and positioning processing through the decoupling layer to obtain the human body part detection data.

7. The vehicle seat adjustment model training method according to any one of claims 1 to 4, characterized in that: The human attribute prediction sub-model includes a plurality of first residual attention modules connected in sequence, a first feature fusion module connected to the plurality of first residual attention modules respectively, and a human attribute prediction module connected to the first feature fusion module; The human body attribute prediction sub-model is used to extract features from the human body part detection data and predict human body attributes to obtain human body attribute prediction data, including: Performing feature extraction on the human body part detection data in sequence through the plurality of first residual attention modules, thereby obtaining first human body part feature images output by the plurality of first residual attention modules; Performing feature fusion processing on the plurality of first human body part feature images by the first feature fusion module to obtain a first human body feature fusion image; The human attribute prediction module performs human attribute prediction on the first human feature fusion image to obtain the human attribute prediction data.

8. The car seat adjustment model training method according to claim 7, characterized in that: Each of the first residual attention modules includes at least two first residual attention units connected in sequence, each of the first residual attention units includes a first residual layer and a first compressed excitation attention layer jump-connected to the first residual layer; the first feature fusion module includes a first parallelized feature interaction fusion layer; the human attribute prediction module includes multiple fully connected layers; The human body attribute prediction sub-model is used to extract features from the human body part detection data and predict human body attributes to obtain human body attribute prediction data, including: performing feature extraction at different semantic levels and feature enhancement across semantic levels on the human body part detection data in sequence through the first residual layer and the first compressed excitation attention layer in each of the first residual attention units in the plurality of first residual attention modules, thereby obtaining the first human body part feature images output by the plurality of first residual attention modules; Performing parallel feature extraction and cross-level feature fusion processing on the plurality of first human body part feature images through the first parallel feature interactive fusion layer in the first feature fusion module to obtain the first human body feature fusion image; The first human feature fusion image is subjected to multi-classification mapping processing through the multiple fully connected layers in the human attribute prediction module to obtain the human attribute prediction data.

9. The vehicle seat adjustment model training method according to any one of claims 1 to 4, characterized in that: The seat angle assessment sub-model includes a plurality of second residual attention modules connected in sequence, a second feature fusion module connected to the plurality of second residual attention modules respectively, and a seat angle assessment module connected to the second feature fusion module; The seat angle evaluation sub-model performs feature extraction and seat angle evaluation on the human body part detection data and the human body attribute prediction data to obtain seat angle evaluation data, including: sequentially extracting features from the human body part detection data using a plurality of the second residual attention modules, thereby obtaining a plurality of second human body part feature images output by the second residual attention modules; Performing feature fusion processing on the plurality of second human body part feature images by the second feature fusion module to obtain a second human body feature fusion image; The seat angle evaluation module performs seat angle evaluation based on the human attribute prediction data and the second human feature fusion image to obtain the seat angle evaluation data.

10. The car seat adjustment model training method according to claim 9, characterized in that: Each of the second residual attention modules includes at least two second residual attention units connected in sequence, each of which includes a second residual layer and a second compressed excitation attention layer jump-connected to the second residual layer; the second feature fusion module includes a second parallelized feature interaction fusion layer; the seat angle evaluation module includes a fully connected layer; The seat angle evaluation sub-model performs feature extraction and seat angle evaluation on the human body part detection data and the human body attribute prediction data to obtain seat angle evaluation data, including: Performing feature extraction at different semantic levels and feature enhancement across semantic levels on the human body part detection data in sequence through the second residual layer and the second compressed excitation attention layer in each second residual attention unit in the plurality of second residual attention modules, thereby obtaining the second human body part feature images output by the plurality of second residual attention modules; Performing parallel feature extraction and cross-level feature fusion processing on the plurality of second human body part feature images through the second parallel feature interactive fusion layer in the second feature fusion module to obtain the second human body feature fusion image; Through the fully connected layer in the seat angle evaluation module, after determining the human body proportion data based on the human body attribute prediction data, the first seat deployment angle is estimated based on the human body proportion data, and key point extraction and angle calculation are performed based on the second human body feature fusion image to obtain the second seat deployment angle, and the seat angle evaluation data is calculated based on the first seat deployment angle and the second seat deployment angle.

11. The car seat adjustment model training method according to claim 4, characterized in that: The user behavior prediction sub-model includes a plurality of third residual attention modules and user behavior prediction modules connected in sequence; The user behavior prediction sub-model is used to extract features from the human body part detection data and predict user behavior to obtain user behavior prediction data, including: sequentially performing feature extraction on the human body part detection data through the plurality of third residual attention modules to obtain third human body part feature images output by the plurality of third residual attention modules; The user behavior prediction module performs user behavior prediction on the third human body part feature image to obtain the user behavior prediction data.

12. The car seat adjustment model training method according to claim 11, characterized in that: Each of the third residual attention modules includes a third residual layer and a third compressed excitation attention layer jump-connected to the third residual layer; the user behavior prediction module includes a fourth residual layer and a fully connected layer jump-connected to the fourth residual layer; The user behavior prediction sub-model is used to extract features from the human body part detection data and predict user behavior to obtain user behavior prediction data, including: Performing feature extraction at different semantic levels and feature enhancement across semantic levels on the human body part detection data in sequence through the third residual layer and the third compressed excitation attention layer in the plurality of third residual attention modules, thereby obtaining the third human body part feature images output by the plurality of third residual attention modules; The fourth residual layer and the fully connected layer in the user behavior prediction module are used to perform feature extraction and binary classification mapping processing at different semantic levels on the third human body part feature image to obtain the user behavior prediction data.

13. The car seat adjustment model training method according to claim 4, characterized in that: Calculating a second loss value based on the human body part detection data and the user behavior prediction data, and the human body part label data and the user behavior label data in the target training sample data, including: Obtaining a second human body detection loss value based on the human body part detection data and the human body part label data in the target training sample data; Obtaining a behavior prediction loss value based on the user behavior prediction data and the user behavior label data in the target training sample data; The second loss value is obtained based on the second human detection loss value and the behavior prediction loss value.

14. The car seat adjustment model training method according to claim 4, characterized in that: Updating the second model parameter of the human body part detection sub-model and the first model parameter of the user behavior prediction sub-model based on the second loss value includes: Determining a fourth gradient value corresponding to the human body part detection sub-model based on the second loss value and the human body part detection data; updating the fourth gradient value based on an optimization function, and inputting the updated fourth gradient value into the human body attribute prediction sub-model to perform back propagation to update the second model parameter of the human body attribute prediction sub-model; Based on the second loss value and the user behavior prediction data, determine the fifth gradient value corresponding to the user behavior prediction sub-model; update the fifth gradient value based on the optimization function, input the updated fifth gradient value into the user behavior prediction sub-model to perform back propagation to update the first model parameter of the user behavior prediction sub-model.

15. A method for adjusting a car seat, characterized in that: include: Acquire video frame images inside the car cabin; Inputting the video frame image into a car seat adjustment model to obtain car seat angle data output by the car seat adjustment model; wherein the car seat adjustment model is trained using the car seat adjustment model training method according to any one of claims 1 to 14; Based on the car seat angle data, the angle of the car seat is controlled and adjusted.

16. The vehicle seat adjustment method according to claim 15, characterized in that: The car seat adjustment model also outputs user behavior data; Controlling the angle of the car seat based on the car seat angle data includes: When it is determined based on the user behavior data that the user is in a standing state, the angle of the car seat is controlled and adjusted based on the car seat angle data.

17. A car seat adjustment model training device, characterized in that: include: A data acquisition module is used to acquire a training sample data set; wherein the training sample data set includes a plurality of training sample data; each training sample data includes an original video frame image and human body part label data, human body attribute label data, and seat angle label data corresponding to the original video frame image; A model training module is used to iteratively perform training operations on the car seat adjustment model based on the training sample data set; wherein the car seat adjustment model includes a human body part detection sub-model and a human body attribute prediction sub-model connected in sequence, and a seat angle evaluation sub-model respectively connected to the human body part detection sub-model and the human body attribute prediction sub-model; the training operation includes: selecting target training sample data from the training sample data set; inputting the original video frame image in the target training sample data into the car seat adjustment model, so that the car seat adjustment model performs feature extraction and human body part detection on the original video frame image through the human body part detection sub-model to obtain human body part detection data, and The human body attribute prediction sub-model performs feature extraction and human body attribute prediction on the human body part detection data to obtain human body attribute prediction data, and the seat angle evaluation sub-model performs feature extraction and seat angle evaluation on the human body part detection data and the human body attribute prediction data to obtain seat angle evaluation data; a first loss value is calculated based on the human body part detection data, the human body attribute prediction data and the seat angle evaluation data, as well as the human body part label data, the human body attribute label data and the seat angle label data in the target training sample data; based on the first loss value, the first model parameters of the human body part detection sub-model, the human body attribute prediction sub-model and the seat angle evaluation sub-model are updated.

18. A car seat adjustment device, characterized in that: include: An acquisition module, used for acquiring video frame images in a car cabin; a prediction module, configured to input the video frame image into a car seat adjustment model to obtain car seat angle data output by the car seat adjustment model; wherein the car seat adjustment model is trained using the car seat adjustment model training method according to any one of claims 1 to 14; The control module is used to control and adjust the angle of the car seat based on the car seat angle data.

19. An electronic device, characterized in that: include: processor and memory; The memory stores a computer program that can be run on the processor, and when the processor executes the computer program, it implements the automobile seat adjustment model training method as described in any one of claims 1 to 14 or the automobile seat adjustment method as described in any one of claims 15 to 16.

20. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the vehicle seat adjustment model training method according to any one of claims 1 to 14 or the vehicle seat adjustment method according to any one of claims 15 to 16.

Images