Safety belt adjusting method and device, electronic equipment, storage medium and program product

Abstract

The embodiment of the invention provides a safety belt adjusting method and device, electronic equipment, a storage medium and a program product. The method comprises the steps that in response to a vehicle starting instruction, multi-mode sensing data representing the posture of a passenger is collected; according to the multi-mode sensing data, the real-time posture of the passenger is determined; generating a safety belt adjusting instruction based on the real-time posture; the safety belt adjusting instruction carries the target height of a safety belt guider; the safety belt adjusting instruction is sent to an executing mechanism; and the executing mechanism moves the safety belt guider to the target height according to the safety belt adjusting instruction. The method is used for achieving the effect of automatically adjusting the safety belt.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a seat belt adjustment method, device, electronic device, storage medium, and program product. Background Technology

[0002] With the continuous development of automotive safety technology and related fields such as smart cockpits, the requirements for occupant protection in vehicles are gradually increasing. In the existing vehicle safety protection system, seat belts remain the primary means of achieving occupant restraint and collision protection.

[0003] In related technologies, the height of the seat belt is usually adjusted by an adjustment mechanism located on the B-pillar of the vehicle body. Occupants can manually adjust the position of the seat belt guide according to their own height or sitting posture so that the seat belt is roughly at the appropriate working height.

[0004] However, this adjustment method relies on the occupant's personal experience and subjective judgment, making it difficult for the occupant to accurately assess the optimal position of the seat belt in different sitting positions, thus affecting its restraint effect. Summary of the Invention

[0005] This application provides a method, apparatus, electronic device, storage medium, and program product to achieve the desired effects.

[0006] In a first aspect, this application provides a seatbelt adjustment method, comprising:

[0007] In response to a vehicle start command, it collects multimodal sensor data representing occupant posture;

[0008] Based on the multimodal sensing data, the real-time attitude of the occupants is determined;

[0009] Based on the real-time attitude, a seatbelt adjustment command is generated; the seatbelt adjustment command carries the target height of the seatbelt guide.

[0010] The seatbelt adjustment command is sent to the actuator; the actuator moves the seatbelt guide to the target height according to the seatbelt adjustment command.

[0011] In one embodiment, the multimodal sensing data includes at least one of image data, seat pressure distribution, and seat posture information; determining the occupant's real-time posture based on the multimodal sensing data includes:

[0012] Feature extraction is performed on the image data, seat pressure distribution, and seat posture information respectively to obtain target fusion features;

[0013] The real-time pose is determined based on the target fusion features.

[0014] In one embodiment, the step of extracting features from the image data, seat pressure distribution data, and seat posture information to obtain target fusion features includes:

[0015] Based on the local receptive field, feature extraction is performed on the image data to obtain image features that characterize the spatial distribution of key parts of the occupant's body; the image features are used to characterize the spatial distribution of key parts of the occupant's body.

[0016] Based on the continuous change characteristics of seat pressure distribution data in the time dimension, time-series features are extracted from the seat pressure distribution data to obtain pressure features; the pressure features characterize the dynamic change trend of the occupant's sitting posture.

[0017] Based on the global correlation between multiple posture parameters in the seat posture information, joint feature extraction is performed on the seat posture information to obtain posture features; the posture features represent the overall posture state of the occupant.

[0018] The image features, pressure features, and pose features are fused to obtain the target fused features.

[0019] In one embodiment, before extracting features from the image data, seat pressure distribution data, and seat posture information to obtain the target fusion features, the method further includes:

[0020] Adjusting the distribution of pixel grayscale in the image data; and / or suppressing high-frequency noise in the image data; and / or

[0021] The seat pressure distribution data is normalized; and / or the seat pressure distribution data is smoothed.

[0022] In one embodiment, a seatbelt adjustment command is generated based on the real-time attitude, including:

[0023] The real-time attitude is input into a preset height mapping model to generate the target height; the height mapping model is pre-established based on the biomechanical analysis results of the seat belt's position under different occupant sitting postures and collision simulation data;

[0024] Based on the target height, the seatbelt adjustment command is generated.

[0025] In one embodiment, the multimodal sensing data includes at least one of seat posture information; the method further includes:

[0026] When the rate of change of backrest angle or shoulder displacement in the seat posture information is detected to meet the adjustment conditions, the real-time posture is corrected; or

[0027] When the adjustment time exceeds a preset time threshold, the real-time attitude is corrected; the adjustment time is the duration for the actuator to move the seat belt guide to the target height.

[0028] Secondly, this application provides a seatbelt adjustment device, comprising:

[0029] The response module is used to collect multimodal sensor data representing occupant posture in response to vehicle start commands;

[0030] The attitude determination module is used to determine the real-time attitude of the occupant based on the multimodal sensing data.

[0031] The instruction generation module is used to generate a seatbelt adjustment instruction based on the real-time attitude; the seatbelt adjustment instruction carries the target height.

[0032] The sending module is used to send the seat belt adjustment command to the actuator; the actuator moves the seat belt guide to the target height according to the seat belt adjustment command.

[0033] Thirdly, embodiments of this application provide an electronic device, including: a memory and a processor;

[0034] The memory stores computer-executed instructions;

[0035] The processor executes computer execution instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0036] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0037] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0038] The seatbelt adjustment method, electronic device, storage medium, and program product provided in this application collect multimodal sensing data characterizing occupant posture, determine the occupant's real-time posture based on the multimodal sensing data, generate a seatbelt adjustment command based on the real-time posture, and send the seatbelt adjustment command to the actuator, which then drives the seatbelt guide to move to the target height. In this way, the height of the seatbelt guide can be adjusted according to changes in occupant posture without manual adjustment by the occupant, thereby improving the seatbelt restraint effect. Attached Figure Description

[0039] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0040] Figure 1 This is a flowchart illustrating a seatbelt adjustment method in one embodiment;

[0041] Figure 2 This is a flowchart illustrating a seatbelt adjustment method in an exemplary embodiment.

[0042] Figure 3 A schematic diagram of the device provided in this application;

[0043] Figure 4 A schematic diagram of the structure of the electronic device provided in this application.

[0044] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0045] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0046] The seatbelt height adjuster is a car safety accessory and an important component of the three-point seatbelt. It typically consists of a guide rail, slider, and cover. This device is installed on the upper inner side of the B-pillar or in the center pillar trim panel. Through its assembly and the eyelet assembly, it allows for vertical adjustment of the seatbelt anchor point.

[0047] After the occupant is seated, they can press or pull the unlocking mechanism on the height adjuster to move the slider up and down along the guide rail. Once the target position is reached, the unlocking mechanism is released, thus completing the manual adjustment of the seatbelt height. In this way, the occupant can initially set the effective height of the seatbelt according to their own height or sitting posture.

[0048] However, the above adjustment methods rely on the occupant's personal experience and subjective judgment, making it difficult for them to accurately assess the optimal position of the seatbelt in different seating postures. Furthermore, in actual use, some occupants do not actively adjust the seatbelt height after getting into the vehicle, or only set it once upon initial seating. When the occupant's posture changes during vehicle movement, the actual position of the seatbelt on the occupant's body is prone to shift, making it difficult to maintain it within the ideal force zone, thus affecting the seatbelt's restraint effect on the occupant.

[0049] Therefore, this application proposes a method for automatically adjusting a seat belt.

[0050] In one embodiment, a seatbelt adjustment method is provided, such as Figure 1 As shown, the seatbelt adjustment method includes:

[0051] Step 102: In response to the vehicle start command, collect multimodal sensor data representing the occupant's posture.

[0052] When the vehicle receives a start command and enters the working state, it triggers the occupant posture acquisition process, simultaneously acquiring multimodal sensor data to characterize occupant posture. This multimodal sensor data may include occupant image data, seat pressure distribution data, seat posture information, and vehicle operating status data.

[0053] Optionally, a multimodal sensor array can be installed on the vehicle to collect multimodal data on occupant posture.

[0054] For example, a multimodal sensor array may include a vision sensor, a seat pressure sensor, and a seat posture sensor.

[0055] Step 104: Determine the real-time attitude of the occupants based on the multimodal sensing data.

[0056] Optionally, the corresponding posture characteristics of the occupant can be determined separately for each type of data in the multimodal sensing data. For example, the occupant's body contour or key part position can be determined by image recognition based on occupant image data, the occupant's force distribution characteristics can be determined based on seat pressure distribution data, and the seat posture state can be determined based on seat posture information. The real-time posture of the occupant can then be determined by comprehensively considering the above posture characteristics.

[0057] Optionally, feature extraction and fusion processing can be performed on the multimodal sensing data to obtain fused features that characterize the occupant's posture state, and posture recognition can be performed based on the fused features to determine the occupant's real-time posture.

[0058] Step 106: Generate seat belt adjustment commands based on real-time attitude; the seat belt adjustment commands carry the target height.

[0059] Optionally, a target height for the seatbelt guide can be generated based on the real-time attitude, and a seatbelt adjustment command can be generated based on the target height.

[0060] Optionally, the real-time attitude can be input into a preset height mapping model to generate the target height of the seat belt guide, and a seat belt adjustment command can be generated based on the target height.

[0061] Optionally, the target position of the seat belt on the occupant's body can be calculated based on the information representing the spatial position of key parts of the occupant's body in real-time posture, and the target height of the seat belt guide can be determined based on the target position.

[0062] For example, the height position of the occupant's shoulder in the vehicle coordinate system can be determined based on the spatial position of the key points representing the occupant's shoulder in the real-time posture, and this height position can be used as the target action position of the seat belt on the occupant's body; then, the target height of the seat belt guide can be calculated based on the relative height relationship between the target action position and the installation reference position of the seat belt guide on the vehicle body.

[0063] Optionally, the initial height can be corrected based on real-time attitude changes to obtain the target height.

[0064] For example, the initial height of the seat belt guide can be determined when the vehicle starts or when the occupant sits down, and the real-time posture changes relative to the initial posture can be continuously monitored during subsequent driving. When the occupant's torso is detected to be leaning forward or backward, causing a change in the shoulder position, the initial height can be adjusted upward or downward according to the magnitude of the change, thereby obtaining the target height.

[0065] Step 108: Send the seat belt adjustment command to the actuator; the actuator moves the seat belt guide to the target height according to the seat belt adjustment command.

[0066] Optionally, the actuator includes a seatbelt guide ring height adjustment motor and a guide rail mechanism that cooperates with the height adjustment motor; wherein the height adjustment motor is used to provide driving force, and the guide rail mechanism is used to limit the movement direction of the seatbelt guide.

[0067] Optionally, after receiving the seat belt adjustment command, the actuator controls the height adjustment motor to rotate by a preset angle according to the target height carried in the adjustment command, so that the seat belt guide moves along the guide rail mechanism; when the seat belt guide is detected to have reached the target height, the rotation of the height adjustment motor is stopped.

[0068] In this embodiment, when the actuator receives the seat belt adjustment command, it parses the target height carried in the seat belt adjustment command, and controls the height adjustment motor to rotate by a corresponding angle according to the difference between the target height and the current height, thereby driving the seat belt guide to move along the guide rail mechanism; when the seat belt guide reaches the target height, the rotation of the height adjustment motor stops.

[0069] For example, the height adjustment motor can be a stepper motor with a step angle of 5.625° / 64, a reduction ratio of 1:64, and a maximum output torque of approximately 0.3 Nm, which can achieve millimeter-level positioning accuracy of the seat belt guide in the vertical direction, for example, the positioning accuracy can be controlled within ±0.5 mm.

[0070] For example, the guide rail mechanism can be an aluminum alloy guide rail with a length of, for example, 200 mm and a surface roughness of, for example, Ra 0.8 μm, to reduce the frictional resistance during the movement of the guide. In this embodiment, the sliding resistance of the guide on the guide rail is no greater than 0.5 N, and the maximum adjustment speed of the seat belt guide can reach, for example, 50 mm / s, thereby improving the response speed while ensuring the smoothness of the adjustment.

[0071] In this embodiment, multimodal sensing data characterizing occupant posture is collected, and the occupant's real-time posture is determined based on this data. A seatbelt adjustment command is then generated based on the real-time posture and sent to the actuator, which drives the seatbelt guide to move to the target height. This method allows for adjustment of the seatbelt guide height based on changes in occupant posture without manual adjustment by the occupant, thereby improving the seatbelt restraint effect.

[0072] Furthermore, in one embodiment, the multimodal sensor array includes a vision sensor, a seat pressure sensor, and a seat posture sensor, for acquiring multimodal sensing data characterizing occupant posture from different dimensions.

[0073] The vision sensor includes two global shutter cameras. The main camera, located on the top of the cockpit, has an output resolution of 1920×1080, a frame rate of 30fps, and a field of view of 120°, used to capture images of the occupant's upper body. The auxiliary camera, located inside the B-pillar, has an output resolution of 1280×720, a frame rate of 60fps, and a field of view of 90°, used to focus on capturing image information of the occupant's shoulder area. The vision sensor has a dynamic range of 130dB and is capable of high dynamic range imaging to adapt to application scenarios with rapid changes in light and dark, such as when the vehicle is entering or exiting tunnels.

[0074] The seat pressure sensor is a flexible thin-film pressure sensor with a range of 0–100 kg, a sensitivity of 0.5 mV / N, and a sampling frequency of 100 Hz. It is distributed in the seat cushion and backrest areas, with 16 pressure sensing units in the seat cushion area and 8 pressure sensing units in the backrest area, forming a 24-point pressure matrix for collecting seat pressure distribution data between the occupant and the seat.

[0075] The seat attitude sensor is a nine-axis inertial measurement unit (IMU) that integrates a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer. It has a sampling frequency of 200Hz and is used to measure the seat slide rail displacement and backrest tilt angle. The measurement accuracy of the seat slide rail displacement is ±1mm, and the measurement accuracy of the backrest tilt angle is ±0.5°.

[0076] In this embodiment, the data characteristics of the multimodal sensing data are as follows: Image data is output in YUV420 format after ISP preprocessing, with the main camera's single frame data size being approximately 2.1 MB and the auxiliary camera's single frame data size being approximately 0.8 MB. The image data is transmitted to the vehicle domain controller for processing via the MIPI-CSI2 interface. The vehicle domain controller has a computing power of approximately 50 TOPS and supports TensorRT acceleration. The seat pressure distribution data is a 24-channel analog signal, which is converted into a digital signal by a 12-bit analog-to-digital converter (ADC) and output in the format [P1, P2, ..., P24]. The unit of each pressure value is Newtons (N), and the single frame data size is 48 bytes. The seat attitude information is output by the IMU, including Euler angle information (roll, pitch, and yaw, in °) and displacement information (X / Y / Z directions, in mm), with a single frame data size of 32 bytes.

[0077] Furthermore, in one embodiment, the multimodal sensing data includes at least one of image data, seat pressure distribution, and seat posture information; determining the occupant's real-time posture based on the multimodal sensing data includes: extracting features from the image data, seat pressure distribution, and seat posture information respectively to obtain target fusion features; and determining the real-time posture based on the target fusion features.

[0078] Since different types of multimodal sensing data differ in their information representation formats and emphasis on occupant posture, in this embodiment, feature extraction is performed on the multimodal sensing data separately, and the extracted features are fused to obtain target fused features. Then, posture recognition is performed based on the target fused features to determine the real-time posture of the occupant.

[0079] Optionally, the occupant's attitude state vector can be constructed based on the target fusion features, and the attitude state vector can be used as the occupant's real-time attitude.

[0080] Furthermore, the target fusion features can be normalized and combined to construct an attitude state vector that includes feature components such as the occupant's torso tilt direction and shoulder height position.

[0081] Optionally, the target fusion features can be input into a pre-trained posture recognition model to obtain the occupant's real-time posture. For example, the posture recognition model may include multiple feature extraction branches and a fusion recognition module. Each feature extraction branch processes different types of features, and the fusion recognition module fuses the outputs of each feature extraction branch and outputs the occupant's real-time posture based on the fusion result.

[0082] In the above embodiments, by extracting features from image data, seat pressure distribution and seat posture information respectively, the target fusion features can be accurately obtained.

[0083] In one embodiment, the above-mentioned feature extraction of image data, seat pressure distribution data, and seat posture information to obtain target fusion features includes: extracting features from image data based on local receptive fields to obtain image features representing the spatial distribution of key parts of the occupant's body; the image features are used to represent the spatial distribution of key parts of the occupant's body; extracting temporal features from seat pressure distribution data based on the continuous change characteristics of seat pressure distribution data in the time dimension to obtain pressure features; the pressure features represent the dynamic change trend of the occupant's sitting posture; extracting joint features from seat posture information based on the global correlation between multiple posture parameters in seat posture information to obtain posture features; the posture features represent the overall posture state of the occupant; and fusing image features, pressure features, and posture features to obtain target fusion features.

[0084] In this embodiment, the focus is primarily on the upper body region of the occupant. Therefore, during image feature extraction, local receptive fields are set to encode the regional features related to the seatbelt's position. Here, a local receptive field refers to the perception and encoding of local areas within the image during feature extraction.

[0085] Optionally, by setting local receptive fields, local regions in the image are perceived step by step, local features such as edges and contours within the local regions are encoded, and the receptive range is gradually expanded as the network layers increase, so as to form an image structural feature representation from local to global; and the image features are compressed by pooling operations to obtain image feature vectors.

[0086] Alternatively, a lightweight convolutional neural network (CNN) can be used to extract features from the image data.

[0087] For example, the MobileNetV3-Small network structure can be used to extract features from the main camera image and the auxiliary camera image. After processing by the MobileNetV3-Small network, an image feature map with 1280 channels and a spatial resolution of 7×7 is output. Subsequently, global average pooling is performed on the image feature map to converge the features in the spatial dimension, resulting in a 1280-dimensional image feature vector. This image feature vector serves as the image feature input for subsequent multimodal feature fusion.

[0088] Since seat pressure distribution data can reflect the force state between the occupant and the seat, and changes in occupant posture usually manifest as continuous changes in pressure distribution over time, in this embodiment, time-series feature extraction is performed on the seat pressure distribution data to obtain pressure features that characterize the dynamic changes in occupant posture.

[0089] Optionally, initial pressure data of the occupant can be collected by multiple pressure sensing units disposed on the seat surface. The initial pressure data is a 24-channel analog signal. After the analog signal is converted into a digital signal by a 12-bit analog-to-digital converter (ADC), seat pressure distribution data is formed. For example, the seat pressure distribution data can be represented as P1, P2, ..., P24, where P1 to P24 are... 24 These represent the pressure values ​​collected by each pressure sensing unit, in Newtons (N), with a single frame data size of approximately 48 bytes.

[0090] For example, seat pressure distribution data collected at multiple consecutive time points can be used as time-series input. A Long Short-Term Memory (LSTM) network can be introduced to model the pressure distribution data to extract the temporal features formed by the change of pressure distribution over time. Through the memory units of the LSTM network, historical pressure distribution information is retained, and the current pressure change trend is encoded, thereby obtaining a pressure feature vector that can reflect the dynamic changes in the occupant's sitting posture.

[0091] Since seat posture information typically includes multiple interrelated posture parameters, such as seat rail displacement, backrest tilt angle, and corresponding posture changes, there are overall constraints between these posture parameters. Processing any single posture parameter alone is insufficient to accurately reflect the occupant's overall posture state. Therefore, in this embodiment, based on the global correlation of seat posture information, joint feature extraction is performed on multiple posture parameters to obtain posture features that characterize the overall posture state of the seat.

[0092] Optionally, the IMU Euler angle information (including roll angle, pitch angle and yaw angle) and the corresponding displacement information obtained by the seat attitude sensor can be input as attitude parameters into the Transformer encoder to model the global correlation between attitude parameters.

[0093] For example, the Transformer encoder can employ an 8-head self-attention mechanism and include a 2-layer encoding structure. By jointly modeling multi-dimensional pose parameters, it outputs a 128-dimensional pose feature vector that characterizes the overall pose state of the seat. The pose feature vector serves as the pose feature input for subsequent multi-modal feature fusion.

[0094] A gated fusion mechanism can be used to weight and fuse image features, pressure features, and pose features to dynamically adjust the contribution of different modal features. For example, the target fusion features can be generated as follows:

[0095] Formula (1)

[0096] in, Representing image features, Indicates pressure characteristics, W represents the pose feature; W1, W2, and W3 are learnable weight matrices with a dimension of 1280×128, used for linear mapping of different modal features; b is the bias term; σ() is the Sigmoid activation function, whose output is used as the gating weights to dynamically adjust the contribution of each modal feature in the fusion process according to the effectiveness of different modal features.

[0097] In the above embodiments, image data, seat pressure distribution, and seat posture information are fused into target fusion features through different methods.

[0098] Furthermore, before extracting features from the image data, seat pressure distribution data, and seat posture information to obtain the target fusion features, the process also includes: adjusting the distribution of pixel grayscale in the image data; and / or suppressing high-frequency noise in the image data; and / or normalizing the seat pressure distribution data; and / or smoothing the seat pressure distribution data.

[0099] By performing grayscale distribution adjustment processing on image data, such as histogram equalization or adaptive histogram equalization, the distribution of pixel grayscale in the image can be remapped, making the pixel values ​​more evenly distributed within the grayscale range.

[0100] Optionally, high-frequency noise in the image data can be suppressed, for example by performing Gaussian filtering, bilateral filtering or other low-pass filtering operations on the image data, to reduce high-frequency interference components introduced by sensor noise, illumination jitter or image acquisition error.

[0101] Optionally, the seat pressure distribution data can be normalized. For example, the pressure values ​​collected by each pressure channel can be scaled proportionally according to the range of the pressure sensor or the preset maximum pressure value, so that the normalized pressure data falls into a uniform value range.

[0102] Optionally, the seat pressure distribution data can be smoothed, for example by using sliding window filtering or time averaging filtering to smooth the pressure data at multiple consecutive time points, so as to reduce high-frequency fluctuations caused by instantaneous jitter or measurement errors.

[0103] In an exemplary embodiment, multimodal sensing data is input into an attitude recognition model, which extracts features from image data, seat pressure distribution, and seat attitude information to obtain target fusion features.

[0104] Optionally, the pose recognition model can employ a multi-branch feature extraction structure. For example, the first feature branch is used for image processing, the second feature branch is used for processing seat pressure distribution data, and the third feature branch is used for processing seat pose information.

[0105] The feature extraction of image data, seat pressure distribution and seat posture information by each feature branch can be referred to the description in the embodiment.

[0106] After obtaining image features, pressure features, and pose features, Representing image features, Indicates pressure characteristics, This indicates that pose features are input into the fusion module for feature fusion processing. For example, a gating fusion mechanism can be used to dynamically adjust the contribution of different modal features to generate the target fused features. It outputs the occupant's posture recognition results based on the target fusion features.

[0107] Optionally, the pose recognition model is pre-trained. First, a multimodal dataset for model training is constructed. This dataset is based on multimodal sensor data collected from 2000 volunteers, whose BMI ranges from 18 to 35, covering different body types and various sitting postures. The total size of the dataset is approximately 1 million frames. In this embodiment, the dataset is divided according to a preset ratio: 80% of the data is used as the training set, 15% as the validation set, and 5% as the test set, for model training, parameter tuning, and performance evaluation.

[0108] During model training, a joint loss function is used to optimize the pose recognition model. This joint loss function includes both a classification loss for pose classification and a regression loss for key part location prediction. The joint loss function can be expressed as follows:

[0109] Formula (2)

[0110] in, =0.7, The sitting posture classification loss is calculated using Softmax and cross-entropy loss functions, which are used to constrain the model's ability to recognize the occupant's sitting posture category. The shoulder position regression loss is used. Optionally, the mean squared error loss can be used to constrain the accuracy of the model's prediction of the occupant's shoulder spatial position.

[0111] Furthermore, during model training, the AdamW optimizer is used to update the model parameters, with the initial learning rate set to 1×10⁻⁶. -4 The learning rate is dynamically adjusted using a cosine annealing scheduling strategy. During training, a small batch training method with a batch size of 64 is used to iterate and update the model for, for example, 50 training cycles until the model's performance metrics on the validation set converge, thereby obtaining a pose recognition model for online inference.

[0112] In the above embodiments, different types of data were preprocessed to make the subsequent generation of more accurate target heights possible.

[0113] In one embodiment, the above-mentioned generation of seat belt adjustment instructions based on real-time attitude includes: inputting the real-time attitude into a pre-generated height mapping model to generate a target height; the height mapping model is pre-established based on the biomechanical analysis results of the seat belt's position under different occupant sitting postures and collision simulation data; and generating seat belt adjustment instructions based on the target height.

[0114] The height mapping model refers to a model used to describe the correspondence between the real-time posture of an occupant and the optimal effective height of the seat belt. This model is established based on the biomechanical characteristics of the seat belt in the shoulder area under different occupant seating postures and simulation data from vehicle collision scenarios.

[0115] Optionally, when the real-time attitude is input into the height mapping model, the height mapping model can determine the target position of the seat belt on the occupant's body based on the parameter information representing the occupant's sitting posture in the real-time attitude, and further convert the target position into the target height of the seat belt guide.

[0116] Optionally, the effectiveness of seat belts under different seating postures can be analyzed through biomechanical experiments. For example, a Hybrid III 50th percentile dummy is used to conduct a frontal collision test at a speed of 56 km / h under different backrest angles (e.g., 0°–30°) and different seat rail positions (front / middle / rear). During the test, pressure distribution data of the seat belt in the dummy's shoulder area is collected, and the height range of the seat belt under the condition of optimal restraint without significant compression is determined based on the pressure magnitude and distribution.

[0117] After completing the biomechanical experiments, the experimental data were modeled. For example, the relationship between the target height of the seatbelt and the backrest tilt angle can be fitted as a quadratic function model, with the following form:

[0118] Formula (3)

[0119] in, The backrest tilt angle is expressed in degrees; a, b, and c are parameters obtained by fitting the experimental data using the least squares method, and the goodness of fit R of the experimental data is given. 2 It is approximately 0.95.

[0120] In a typical application scenario, the parameters can be set to a = 0.02, b = −1.5, c = 850, and the target height. The unit is millimeters, and this parameter combination is applicable to occupants with a height range of 160–185cm.

[0121] Optionally, after determining the target height, a seatbelt adjustment command is generated based on the target height. For example, the target height can be compared with the current actual height of the seatbelt guide, the height difference between the two can be calculated, and the direction and amount of movement of the seatbelt guide can be determined based on the height difference.

[0122] For example, the seatbelt adjustment command may include at least one control parameter, which indicates the direction of movement, distance of movement, and speed of movement of the actuator. Specifically, when the target height is higher than the current height, an adjustment command is generated to drive the seatbelt guide upward; when the target height is lower than the current height, an adjustment command is generated to drive the seatbelt guide downward.

[0123] Furthermore, the height difference can be converted into a control quantity for the actuator. For example, based on the step angle, reduction ratio, and guide rail transmission parameters of the adjusting motor in the actuator, the height difference is converted into the corresponding number of motor rotation steps or rotation angle, and the number of rotation steps or rotation angle is encapsulated as a seat belt adjustment command and sent to the actuator.

[0124] Optionally, when generating seatbelt adjustment commands, the height difference can be limited or segmented to avoid discomfort to occupants caused by a large, one-time adjustment. For example, the target height adjustment process can be divided into multiple small-step adjustment processes, with time intervals set between adjacent adjustment processes, thereby achieving smooth adjustment of the seatbelt height.

[0125] After the actuator completes the adjustment, it can determine whether the seat belt guide has reached the target height based on the height sensor or motor position feedback information. When it is confirmed that the target height has been reached, the adjustment command is stopped, thus completing one seat belt height adjustment process.

[0126] In the above embodiments, by inputting the real-time posture into a height mapping model pre-established based on the biomechanical analysis results of the seat belt's position under different sitting postures and collision simulation data to generate a target height, and generating a seat belt adjustment command based on the target height, the uncertainty caused by relying solely on fixed gear positions or occupant experience for adjustment can be avoided, enabling the seat belt height to be adaptively adjusted according to the occupant's real-time sitting posture.

[0127] Since occupants may experience posture changes during vehicle operation due to adjustments in seating position or changes in road conditions, this embodiment introduces a real-time posture correction mechanism based on posture changes.

[0128] Furthermore, in one embodiment, the aforementioned multimodal sensing data includes at least one of seat posture information; the method further includes: when the backrest angle change rate or shoulder displacement in the seat posture information is detected to meet the adjustment conditions, correcting the real-time posture; or when the adjustment time exceeds a preset time threshold, correcting the real-time posture; the adjustment time is the duration for the actuator to move the seat belt guide to the target height.

[0129] In this embodiment, the backtilt angle change rate or shoulder displacement is used as the trigger condition for correcting the real-time attitude.

[0130] Optionally, for example, when a change rate of backrest tilt angle greater than 2° / s is detected, or a shoulder displacement greater than 5mm is detected and lasts for at least 3 frames, it is determined that the occupant's posture has changed significantly and the real-time posture needs to be corrected.

[0131] Optionally, when the time from the start of the adjustment to its completion is less than or equal to 1.5 seconds, a reassessment of the real-time attitude can also be triggered to avoid adjustment deviations caused by attitude changes during the adjustment process.

[0132] Optionally, after the correction condition is triggered, the multimodal sensing data at the current moment is reacquired, and the attitude recognition process is re-executed based on the multimodal sensing data to update the real-time attitude and generate a seat belt adjustment command.

[0133] Optionally, based on the original real-time attitude, the real-time attitude is incrementally corrected according to the detected change in backrest tilt angle or shoulder displacement, thereby obtaining the corrected real-time attitude, and a seat belt adjustment command is generated based on the corrected real-time attitude.

[0134] Optionally, the detected attitude change can be mapped to an incremental vector in the feature space, and the incremental vector can be superimposed or weighted with the original fused features according to a preset weight to achieve smooth correction of the real-time attitude.

[0135] In the above embodiments, the target height can be generated based on the real-time posture of the occupants.

[0136] In one exemplary embodiment, combined with Figure 2 As shown, Figure 2 This is a seatbelt adjustment method in one embodiment.

[0137] Step 202: In response to the vehicle start command, collect multimodal sensing data representing the occupant's posture; the multimodal sensing data includes at least one of image data, seat pressure distribution, and seat posture information.

[0138] Step 204: Adjust the distribution of pixel grayscale in the image data; and / or suppress high-frequency noise in the image data; and / or normalize the seat pressure distribution data; and / or smooth the seat pressure distribution data.

[0139] Step 206: Extract features from image data, seat pressure distribution and seat posture information respectively to obtain target fusion features; and determine real-time posture based on target fusion features.

[0140] Step 208: Input the real-time attitude into the preset height mapping model to generate the target height; and generate the seat belt adjustment command based on the target height.

[0141] Step 210: Send the seat belt adjustment command to the actuator; the actuator moves the seat belt guide to the target height according to the seat belt adjustment command.

[0142] Step 212: When the backrest angle change rate or shoulder displacement in the seat posture information meets the adjustment conditions, correct the real-time posture; or when the adjustment time exceeds the preset time threshold, correct the real-time posture; the adjustment time is the duration for the actuator to move the seat belt guide to the target height.

[0143] Step 214: Generate a seatbelt adjustment command based on the corrected real-time attitude and send the seatbelt adjustment command to the actuator.

[0144] In the above embodiments, the occupant's real-time posture is identified using multimodal sensing data, and a target seatbelt height is generated based on the posture fusion results, enabling adaptive adjustment of the seatbelt height according to changes in the occupant's sitting posture. Simultaneously, a posture change detection and correction mechanism is introduced during the adjustment process to avoid adjustment deviations caused by posture changes, thereby improving the accuracy and continuity of seatbelt adjustment and the occupant protection effect.

[0145] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0146] Based on the same inventive concept, this application also provides a seat belt adjusting device for implementing the seat belt adjusting method described above. The solution provided by this seat belt adjusting device is similar to the solution described in the seat belt adjusting method above. Therefore, the specific limitations in one or more device embodiments provided below can be found in the limitations of the seat belt adjusting method described above, and will not be repeated here.

[0147] In one embodiment, such as Figure 3 As shown, a state determination device 300 is provided, comprising:

[0148] The response module is used to collect multimodal sensor data representing occupant posture in response to vehicle start commands.

[0149] The attitude determination module is used to determine the real-time attitude of the occupants based on multimodal sensor data.

[0150] The instruction generation module is used to generate seat belt adjustment instructions based on real-time attitude; the seat belt adjustment instructions carry the target height.

[0151] The sending module is used to send seat belt adjustment commands to the actuator; the actuator moves the seat belt guide to the target height according to the seat belt adjustment commands.

[0152] In one embodiment, the multimodal sensing data includes at least one of image data, seat pressure distribution, and seat posture information; the posture determination module includes:

[0153] The feature extraction unit is used to extract features from image data, seat pressure distribution, and seat posture information to obtain target fusion features.

[0154] The determination unit is used to determine the real-time pose based on the target fusion features.

[0155] In one embodiment, the feature extraction unit includes:

[0156] The first extraction subunit is used to extract features from image data based on the local receptive field to obtain image features that characterize the spatial distribution of key parts of the occupant's body; the image features are used to characterize the spatial distribution of key parts of the occupant's body.

[0157] The second extraction subunit is used to extract time-series features from the seat pressure distribution data based on the continuous change characteristics of the seat pressure distribution data in the time dimension, so as to obtain pressure features; the pressure features characterize the dynamic change trend of the occupant's sitting posture.

[0158] The third extraction subunit is used to perform joint feature extraction on the seat posture information based on the global correlation between multiple posture parameters in the seat posture information to obtain posture features; the posture features represent the overall posture state of the occupant.

[0159] The feature fusion subunit is used to fuse image features, pressure features, and pose features to obtain target fusion features.

[0160] In one embodiment, the attitude determination module further includes:

[0161] A grayscale adjustment unit is used to adjust the distribution of pixel grayscale in image data; and / or

[0162] A noise processing unit is used to suppress high-frequency noise in image data; and / or

[0163] Normalization units are used to normalize seat pressure distribution data; and / or

[0164] The smoothing unit is used to smooth the seat pressure distribution data.

[0165] In one embodiment, the instruction generation module includes:

[0166] The mapping unit is used to input the real-time attitude into the preset height mapping model to generate the target height. The height mapping model is pre-established based on the biomechanical analysis results of the seat belt action position under different occupant sitting postures and collision simulation data.

[0167] The adjustment command unit is used to generate seat belt adjustment commands based on the target height.

[0168] In one embodiment, the above-mentioned apparatus further includes:

[0169] The first posture correction module is used to correct the real-time posture when the rate of change of backrest angle or shoulder displacement in the seat posture information meets the adjustment conditions; or

[0170] The second attitude correction module is used to correct the real-time attitude when the adjustment time exceeds a preset time threshold; the adjustment time is the duration for the actuator to move the seat belt guide to the target height; the corrected real-time attitude is used to generate seat belt adjustment commands.

[0171] Each module in the above-mentioned device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0172] Figure 4 A schematic diagram of the structure of the electronic device provided in this application. Figure 4 As shown, the electronic device 400 provided in this embodiment includes at least one processor 401 and a memory 402. Optionally, the device 400 further includes a communication component 403. The processor 401, memory 402, and communication component 403 are connected via a bus 404.

[0173] In a specific implementation, at least one processor 401 executes computer execution instructions stored in memory 402, causing at least one processor 401 to perform the above-described method.

[0174] The specific implementation process of processor 401 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0175] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0176] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0177] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0178] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0179] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0180] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0181] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0182] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0183] The units described as separate components may or may not be physically separate. The 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 the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0184] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0185] If a function is implemented as 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 this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0186] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0187] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for adjusting a seatbelt, characterized in that, include: In response to a vehicle start command, it collects multimodal sensor data representing occupant posture; Based on the multimodal sensing data, the real-time attitude of the occupants is determined; Based on the real-time attitude, a seatbelt adjustment command is generated; the seatbelt adjustment command carries the target height of the seatbelt guide. The seatbelt adjustment command is sent to the actuator; the actuator moves the seatbelt guide to the target height according to the seatbelt adjustment command.

2. The method according to claim 1, characterized in that, The multimodal sensing data includes at least one of image data, seat pressure distribution, and seat posture information; determining the occupant's real-time posture based on the multimodal sensing data includes: Feature extraction is performed on the image data, seat pressure distribution, and seat posture information respectively to obtain target fusion features; The real-time pose is determined based on the target fusion features.

3. The method according to claim 2, characterized in that, The step involves extracting features from the image data, seat pressure distribution data, and seat posture information to obtain target fusion features, including: Based on the local receptive field, feature extraction is performed on the image data to obtain image features that characterize the spatial distribution of key parts of the occupant's body; the image features are used to characterize the spatial distribution of key parts of the occupant's body. Based on the continuous change characteristics of seat pressure distribution data in the time dimension, time-series features are extracted from the seat pressure distribution data to obtain pressure features; the pressure features characterize the dynamic change trend of the occupant's sitting posture. Based on the global correlation between multiple posture parameters in the seat posture information, joint feature extraction is performed on the seat posture information to obtain posture features; the posture features represent the overall posture state of the occupant. The image features, pressure features, and pose features are fused to obtain the target fused features.

4. The method according to claim 2, characterized in that, Before extracting features from the image data, seat pressure distribution data, and seat posture information to obtain the target fusion features, the process further includes: Adjusting the distribution of pixel grayscale in the image data; and / or suppressing high-frequency noise in the image data; and / or The seat pressure distribution data is normalized; and / or the seat pressure distribution data is smoothed.

5. The method according to claim 1, characterized in that, Based on the real-time attitude, a seatbelt adjustment command is generated, including: The real-time attitude is input into a pre-generated height mapping model to generate the target height; the height mapping model is pre-established based on the biomechanical analysis results of the seat belt's position under different occupant sitting postures and collision simulation data; Based on the target height, the seatbelt adjustment command is generated.

6. The method according to claim 1, characterized in that, The multimodal sensing data includes at least one of the seat posture information; the method further includes: When the rate of change of backrest angle or shoulder displacement in the seat posture information is detected to meet the adjustment conditions, the real-time posture is corrected; or When the adjustment time exceeds a preset time threshold, the real-time attitude is corrected; the adjustment time is the duration for the actuator to move the seat belt guide to the target height; the corrected real-time attitude is used to generate the seat belt adjustment command.

7. A seatbelt adjustment device, characterized in that, include: The response module is used to collect multimodal sensor data representing occupant posture in response to vehicle start commands; The attitude determination module is used to determine the real-time attitude of the occupant based on the multimodal sensing data. The instruction generation module is used to generate a seatbelt adjustment instruction based on the real-time attitude; the seatbelt adjustment instruction carries the target height. The sending module is used to send the seat belt adjustment command to the actuator; the actuator moves the seat belt guide to the target height according to the seat belt adjustment command.

8. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images