A passenger posture active protection method, system, device and medium

Abstract

The application relates to a passenger posture active protection method, system, device and medium, the method comprising collecting vehicle multi-source perception data, and preprocessing the vehicle multi-source perception data to obtain characteristic parameters; the characteristic parameters are input into a pre-trained collision risk assessment model for analysis, and a collision risk assessment result is output; when a collision risk level index reaches a preset trigger threshold, an unavoidable collision state is determined, and a posture adjustment process is entered; based on a collision type and the characteristic parameters, matching is performed in a preset protection strategy library to obtain a corresponding posture response strategy; the body size parameters and the current sitting posture state of a passenger are acquired to construct a passenger state model; a current adaptive target posture is generated in combination with the posture response strategy and the passenger state model, and a posture adjustment operation is performed on the passenger based on the target posture. The application has the effect of improving the safety of passengers.

Description

Technical Field

[0001] This application relates to the technical field of automotive safety control, and in particular to a method, system, device, and medium for active protection of occupant posture. Background Technology

[0002] Currently, with the rapid development of autonomous driving and smart cockpit technologies, vehicles can improve occupant safety by sensing environmental information and taking proactive measures when facing potential collision risks. For example, some vehicles already have auxiliary functions such as collision warning and automatic braking, and can even provide preliminary protection for occupants before a collision by using seat belt pretensioning and seat adjustment.

[0003] However, existing protection strategies often focus on the vehicle's dynamic response or the triggering of passive safety devices, with insufficient consideration for the individual characteristics and real-time posture of the occupants. In actual collisions, the severity of occupant injury depends not only on the impact intensity but also on their body posture at the moment of impact. If the occupant is in a non-standard sitting posture such as tilting, leaning forward, or twisting, the protective effect of traditional seat belts and airbags will be significantly reduced.

[0004] The existing technical solutions mentioned above have the following drawbacks: existing solutions usually rely on a single risk threshold to trigger preset actions, and cannot provide personalized and dynamic response control for different collision scenarios, different occupant characteristics and attitude states, so there is room for improvement. Summary of the Invention

[0005] To enhance occupant safety, this application provides an active occupant posture protection method, system, device, and medium.

[0006] The above-mentioned objective of this application is achieved through the following technical solution:

[0007] An active occupant posture protection method, the active occupant posture protection method comprising:

[0008] Collect vehicle multi-source perception data and preprocess the vehicle multi-source perception data to obtain feature parameters;

[0009] The feature parameters are input into a pre-trained collision risk assessment model for analysis, and the collision risk assessment results are output, including collision risk level indicators and collision type.

[0010] When the collision risk level index reaches the preset trigger threshold, it is determined to be an unavoidable collision state and enters the attitude adjustment process.

[0011] Based on the collision type and the feature parameters, a matching is performed in a preset protection strategy library to obtain a corresponding attitude response strategy. The attitude response strategy includes a preset target attitude candidate set and attitude response priority.

[0012] Obtain the occupant's physical parameters and current sitting posture to construct an occupant state model;

[0013] The attitude response strategy and the occupant state model are combined to generate the current suitable target attitude, and the occupant attitude adjustment operation is performed based on the target attitude.

[0014] By adopting the above technical solutions, and through the collection and preprocessing of multi-source vehicle perception data, multi-dimensional perception information such as forward targets, lateral environment, and traffic network can be comprehensively acquired, improving the overall perception capability of complex traffic scenarios. This provides a high-precision and comprehensive input foundation for subsequent collision risk identification. By inputting feature parameters into a pre-trained collision risk assessment model for analysis, the collision level and type can be quickly and accurately determined, thereby improving the system's timeliness and predictive ability for dangerous situations. By initiating an attitude adjustment process after the collision risk level reaches the trigger threshold, an active protection mechanism can be activated before a collision occurs, effectively gaining attitude adjustment time and improving the timeliness of occupant protection response. By matching attitude response strategies and combining them with occupant state models to generate the currently suitable target attitude, dynamic matching of protection strategies and individual differences can be achieved, thereby enhancing the effectiveness and personalization of protective measures and ultimately reducing the risk of occupant injury.

[0015] In one example, this application can be further configured such that the collection of multi-source perception data of the vehicle specifically includes:

[0016] Obtain the relative distance, relative velocity, and relative acceleration of the target object ahead;

[0017] Acquire the identification information of the target object ahead, the trajectory data of the lane ahead, and the information of the side obstacles;

[0018] Acquire vehicle network information, including emergency braking events of the vehicle in front, road anomaly warnings, or construction warnings;

[0019] The relative distance, relative speed, relative acceleration, identification information of the forward target, the trajectory data of the lane ahead, the information of the lateral obstacles, and the vehicle network information are combined to obtain the vehicle multi-source perception data.

[0020] By adopting the above technical solutions, the relative distance, relative speed, and relative acceleration of the target ahead can be obtained, enabling real-time assessment of the relative dynamic relationship with the vehicle ahead, thereby improving the accuracy of collision time judgment; by obtaining target recognition information, lane trajectory, and lateral obstacle information, a comprehensive understanding of the target ahead and the environmental structure can be achieved, thereby improving the environmental adaptability of risk assessment; by obtaining vehicle network event information, external traffic warning information can be introduced for prediction, thereby enhancing the system's ability to respond to emergencies; by summarizing the above data to construct multi-source perception input, a rich perception information map can be built, thereby improving the expressive power and contextual understanding ability of the model's input dimensions.

[0021] In one example, this application can be further configured such that: the preprocessing of the vehicle multi-source perception data to obtain feature parameters specifically includes:

[0022] Calculate the collision time interval based on the relative velocity and the relative distance;

[0023] A determination result is obtained by fusing and analyzing the identification information of the target object ahead with the trajectory data of the lane ahead.

[0024] The collision time interval, the determination result, the relative distance, the relative speed, the relative acceleration, the identification information, the lateral obstacle information, and the vehicle network information are encoded and summarized to form the feature parameters.

[0025] By adopting the above technical solutions, the collision time distance can be calculated based on relative speed and relative distance, thereby quantifying the temporal risk between the model and the target ahead, thus enhancing the model's sensitivity to emergency situations. By fusing identification information and lane trajectory data to generate judgment results, the ability to analyze target behavior and identify deviation risks can be improved, thereby increasing the reliability of collision prediction. By encoding and summarizing multidimensional information to form feature parameters, the standardization and efficient expression of input data can be achieved, thereby optimizing the model's analysis efficiency and generalization ability.

[0026] In one example, this application can be further configured such that: the occupant posture active protection method also includes:

[0027] A collision risk assessment model is constructed using a supervised learning algorithm based on deep neural networks. The collision risk assessment model structure includes a feature encoding layer, a feature fusion layer, and a risk determination output layer.

[0028] Historical driving data and accident case data of various collision types are collected to construct a training dataset. The training dataset includes the relative distance, relative speed, relative acceleration, collision time, target object identification type, lane position status, lateral obstacle information, and vehicle network event information between the vehicle and the target object.

[0029] Based on the collision consequence information recorded in the accident cases, each training sample in the training dataset is labeled with a corresponding collision risk level label and collision type label;

[0030] The multidimensional perceptual features of each group of samples in the training dataset are used as the model training input, and the collision risk level label and collision type label corresponding to each group of samples are used as the supervised training target to train the collision risk assessment model, thereby obtaining the pre-trained collision risk assessment model.

[0031] By adopting the above technical solutions, a collision risk assessment model is constructed using a supervised learning algorithm based on deep neural networks. This model incorporates a feature encoding layer, a fusion layer, and an output layer, enabling nonlinear expression and deep semantic extraction of multi-source heterogeneous features. This enhances the model's ability to understand and judge complex traffic conditions. By collecting historical driving data and accident case data of various collision types to construct a training dataset, the model training is ensured to cover a wide range of scenarios and collision conditions, thus improving its applicability and risk identification capabilities. Introducing collision consequence information to label risk levels and types aligns the model output with real-world consequences, improving the accuracy and interpretability of model training. Using multi-dimensional perceptual features as supervised input for model training improves the accuracy and robustness of the collision prediction model, thereby enhancing the overall reliability of the protection system.

[0032] In one example, this application can be further configured such that, before inputting the feature parameters into a pre-trained collision risk assessment model for analysis and outputting a collision risk assessment result, wherein the collision risk assessment result includes a collision risk level index and a collision type, the occupant attitude active protection method further includes:

[0033] Obtain the current vehicle's braking capacity assessment information, which includes braking response time, braking system health status, current vehicle load status, and the coefficient of friction of the road surface.

[0034] The pre-trained collision risk assessment model is dynamically adjusted based on the braking capability assessment information. The dynamic adjustment includes adaptive setting of the risk level determination threshold, selection of model substructure, or adjustment of feature weight factors.

[0035] By adopting the above technical solution, information such as the current vehicle's braking response time, braking system status, load condition, and adhesion coefficient can be obtained to assess the vehicle's braking capability in real time, thereby providing actual control capability constraints for collision risk assessment. By dynamically adjusting the collision risk assessment model based on braking capability, adaptive optimization of risk threshold setting, model substructure, and feature weight factors can be achieved, thereby improving the model's response accuracy and scenario adaptability under different vehicle conditions, and enhancing the robustness and practicality of the protection system.

[0036] In one example, this application can be further configured as follows: obtaining the occupant's physical parameters and current sitting posture, and constructing an occupant state model, specifically includes:

[0037] Obtain the occupant's physical parameters, including height, weight, shoulder width, hip width, and spine length;

[0038] The current sitting posture of the occupant is obtained, including head position, back tilt angle, pelvic rotation angle and lower limb support angle.

[0039] Based on the physical parameters and the sitting posture, an occupant state model is constructed. The occupant state model represents the current occupant's posture space occupancy and motion constraint characteristics in the vehicle seat.

[0040] By adopting the above technical solutions, by acquiring the occupant's physical parameters such as height, weight, and shoulder width, a static descriptive feature reflecting the individual's physical structure can be constructed, thus providing basic constraints for posture adaptation. By collecting the current sitting posture, such as back tilt angle and pelvic rotation angle, the actual posture of the occupant in the seat can be dynamically characterized, thus providing a more realistic and accurate input basis. By fusing physical parameters and sitting posture to construct an occupant state model, the spatial distribution of occupant posture and restricted movement areas can be comprehensively described, thus providing precise parameter support for target posture selection and adjustment path planning, and improving the individual adaptability of protection strategies.

[0041] In one example, this application can be further configured such that: the generation of the currently adapted target posture by combining the attitude response strategy and the occupant state model specifically includes:

[0042] Based on the posture space occupancy pattern and motion constraint characteristics represented in the occupant state model, the target posture candidate set is screened to determine the executable target posture subset.

[0043] Based on the executable target posture subset, combined with the posture response priority, the collision type and occupant comfort requirements, an optimization algorithm is executed to determine the currently suitable target posture. The occupant comfort requirements are set based on the motion constraint features and preset ergonomic threshold judgment rules.

[0044] By adopting the above technical solution, the target posture candidate set can be screened based on the spatial occupancy pattern and motion constraint characteristics in the occupant state model, thus eliminating posture schemes that do not conform to the actual sitting posture constraints, thereby ensuring the feasibility and safety of posture adjustment. By combining posture response priority, collision type and occupant comfort requirements to execute the optimization algorithm, a dynamic trade-off can be achieved between response efficiency and physiological tolerance, thereby determining the target posture that has both protective effect and meets ergonomic requirements, improving the overall performance of posture adjustment and user experience.

[0045] The second objective of this invention is achieved through the following technical solution:

[0046] An active occupant posture protection system, the active occupant posture protection system comprising:

[0047] The multi-source sensing acquisition module is used to acquire multi-source sensing data of the vehicle and preprocess the multi-source sensing data of the vehicle to obtain feature parameters;

[0048] The collision risk assessment module is used to input the feature parameters into a pre-trained collision risk assessment model for analysis and output collision risk assessment results, which include collision risk level indicators and collision types.

[0049] The collision trigger determination module is used to determine an unavoidable collision state when the collision risk level index reaches a preset trigger threshold, and then enter the attitude adjustment process.

[0050] The attitude strategy matching module is used to perform matching in a preset protection strategy library based on the collision type and the feature parameters to obtain the corresponding attitude response strategy. The attitude response strategy includes a preset target attitude candidate set and attitude response priority.

[0051] The occupant state modeling module is used to obtain the occupant's physical parameters and current sitting posture, and to construct an occupant state model.

[0052] The attitude generation and execution module is used to combine the attitude response strategy and the occupant state model to generate the currently adapted target attitude, and to perform attitude adjustment operations on the occupant based on the target attitude.

[0053] By adopting the above technical solutions, and through the collection and preprocessing of multi-source vehicle perception data, multi-dimensional perception information such as forward targets, lateral environment, and traffic network can be comprehensively acquired, improving the overall perception capability of complex traffic scenarios. This provides a high-precision and comprehensive input foundation for subsequent collision risk identification. By inputting feature parameters into a pre-trained collision risk assessment model for analysis, the collision level and type can be quickly and accurately determined, thereby improving the system's timeliness and predictive ability for dangerous situations. By initiating an attitude adjustment process after the collision risk level reaches the trigger threshold, an active protection mechanism can be activated before a collision occurs, effectively gaining attitude adjustment time and improving the timeliness of occupant protection response. By matching attitude response strategies and combining them with occupant state models to generate the currently suitable target attitude, dynamic matching of protection strategies and individual differences can be achieved, thereby enhancing the effectiveness and personalization of protective measures and ultimately reducing the risk of occupant injury.

[0054] The above-mentioned objective three of this application is achieved through the following technical solution:

[0055] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described active occupant attitude protection method.

[0056] The fourth objective of this application is achieved through the following technical solution:

[0057] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described active occupant attitude protection method.

[0058] In summary, this application includes the following beneficial technical effects:

[0059] 1. By collecting and preprocessing multi-source vehicle perception data, multi-dimensional perception information such as forward targets, lateral environment, and traffic network can be comprehensively acquired, improving the overall perception capability of complex traffic scenarios and providing a high-precision, comprehensive input foundation for subsequent collision risk identification. By inputting feature parameters into a pre-trained collision risk assessment model for analysis, the collision level and type can be quickly and accurately determined, thereby improving the system's timeliness and predictive ability for dangerous situations. By entering the attitude adjustment process after the collision risk level reaches the trigger threshold, the active protection mechanism can be activated before a collision occurs, effectively gaining attitude adjustment time and improving the timeliness of occupant protection response. By matching attitude response strategies and combining them with occupant state models to generate the currently suitable target attitude, dynamic matching of protection strategies and individual differences can be achieved, thereby enhancing the effectiveness and personalization of protection measures and ultimately reducing the risk of occupant injury.

[0060] 2. By acquiring information such as the current vehicle's braking response time, braking system status, load condition, and adhesion coefficient, the braking capability of the current vehicle can be assessed in real time, thus providing actual control capability constraints for collision risk assessment; by dynamically adjusting the collision risk assessment model based on braking capability, adaptive optimization of risk threshold setting, model substructure, and feature weight factors can be achieved, thereby improving the model's response accuracy and scenario adaptability under different vehicle conditions, and enhancing the robustness and practicality of the protection system;

[0061] 3. By screening the candidate set of target postures based on the spatial occupancy pattern and motion constraint features in the occupant state model, posture schemes that do not conform to the actual sitting posture constraints can be eliminated, thereby ensuring the feasibility and safety of posture adjustment; by combining posture response priority, collision type and occupant comfort requirements to execute the optimization algorithm, a dynamic trade-off can be achieved between response efficiency and physiological tolerance, thereby determining the target posture that has both protective effect and meets ergonomic requirements, improving the overall performance of posture adjustment and user experience. Attached Figure Description

[0062] Figure 1 This is a flowchart of an active occupant posture protection method according to an embodiment of this application;

[0063] Figure 2 This is a schematic diagram of an active occupant posture protection system according to one embodiment of this application;

[0064] Figure 3 This is a schematic diagram of a device according to one embodiment of this application. Detailed Implementation

[0065] The present application will be further described in detail below with reference to the accompanying drawings.

[0066] In one embodiment, such as Figure 1 As shown, this application discloses an active occupant posture protection method, which specifically includes the following steps:

[0067] S10: Collect multi-source perception data of vehicles and preprocess the multi-source perception data of vehicles to obtain feature parameters.

[0068] Specifically, multi-source perception data from millimeter-wave radar, cameras, and vehicle-to-everything (V2X) communication modules are continuously acquired. This includes the relative distance and speed of objects ahead, target identification categories, lane trajectory information, lateral obstacle status, and road event prompts. All heterogeneous data are uniformly converted into a timestamp-aligned format and preprocessed using normalization, noise filtering, and temporal completion methods. The processed data is then encoded using feature extraction algorithms to construct a set of feature parameter vectors for subsequent model analysis. For example, when the vehicle approaches a stationary obstacle at a speed of 60 km / h, the radar and camera each return a distance value of 2.5 m. After fusion, the target is identified as a stationary object, and an initial feature vector of "forward static collision risk" is generated.

[0069] S20: Input the feature parameters into the pre-trained collision risk assessment model for analysis, and output the collision risk assessment results, which include the collision risk level index and the collision type.

[0070] Specifically, the system invokes a pre-built collision risk assessment model, receives the feature parameter vector generated in the current period as model input, and calculates the collision risk level score of the current driving scenario based on the internally constructed perception feature map and risk level classifier. It also infers the possible collision type by combining the pattern recognition results, such as frontal rear-end collision, side scrape, or lateral collision. At the same time, it generates a structured risk assessment result and stores it in a short-term memory cache to support the subsequent response strategy matching process. For example, when the target is identified as a "low-speed moving vehicle in front" and the distance between the vehicles is less than 1.5m, the model returns a medium-to-high level rear-end collision risk score and marks it as a "low-speed rear-end collision" type.

[0071] S30: When the collision risk level index reaches the preset trigger threshold, it is determined to be an unavoidable collision state and the attitude adjustment process is initiated.

[0072] Specifically, after obtaining the collision risk assessment results, the collision risk level index output by the model is determined in real time. The current risk level is compared with the preset risk trigger threshold. When the risk level reaches or exceeds the corresponding threshold, it is determined that the current vehicle operation state has entered the unavoidable collision zone. The control logic switches the system operation state from the conventional monitoring mode to the active protection mode. At the same time, the conventional driving assistance control logic is frozen, and the execution channel corresponding to the attitude adjustment process is activated to ensure that the occupant protection-related actions are completed first within the remaining available time window. For example, when the predicted collision time distance is less than the preset safety threshold and the deceleration is insufficient to avoid the collision, the unavoidable collision state determination is triggered and the attitude intervention stage is entered.

[0073] S40: Based on the collision type and feature parameters, a matching is performed in the preset protection strategy library to obtain the corresponding attitude response strategy. The attitude response strategy includes a preset target attitude candidate set and attitude response priority.

[0074] Specifically, based on the collision type label and current feature parameters returned by the collision risk assessment model, the corresponding rule set is called from the predefined posture response strategy library. The parameter similarity between the current situation and various typical scenarios in the library is compared, and the strategy entry with the highest matching degree is extracted. This strategy entry contains a candidate set composed of multiple preset target postures and a response priority ranking structure for different collision modes. This strategy structure is then encapsulated as the input parameter for the next posture generation process. For example, when the "side collision left door" type is identified, the candidate postures returned by the strategy library include "upper body right rotation + abdomen tuck + leg tightening" as the main posture combination, with the priority ranking being right rotation > bending over > leg tightening.

[0075] S50: Obtain the occupant's physical parameters and current sitting posture to construct an occupant state model.

[0076] Specifically, after entering the posture adjustment process, the occupant's physical parameters are obtained through in-vehicle sensors or pre-stored user information. Combined with seat sensors, posture detection units, or historical sitting posture data, the current occupant's sitting posture status information is obtained. The physical parameters and the sitting posture status are fused and modeled to depict the occupant's body size distribution, posture angle, and joint movement constraint range at the current moment. This forms an occupant status model that reflects the occupant's space occupancy and adjustable range of motion. For example, for occupants who are tall and have a large seat recline angle, their adjustable forward tilt will be limited to avoid the risk of secondary injury.

[0077] S60: Combines the attitude response strategy and the occupant state model to generate the current suitable target attitude, and performs attitude adjustment operations on the occupant based on the target attitude.

[0078] Specifically, after receiving the target posture candidate set and occupant state model from the posture response strategy, the system calculates the adaptation score between each candidate posture and the current occupant state based on the posture matching algorithm, selects the target posture with the best physical feasibility and comfort, and then generates an adjustment instruction set to control the seat angle, motor push rod and restraint device and other components to quickly complete the target posture adjustment process within the allowable time range, so as to optimize the force posture at the time of collision and reduce the risk of occupant injury to the greatest extent. For example, if an occupant is overweight and sits too far back, when it is determined to be a frontal collision, the system will select "10 degrees of straight back forward tilt + 5 degrees of head forward tilt + shoulder strap limit reinforcement" as the target posture and complete the posture adjustment operation within 200ms.

[0079] By collecting and preprocessing multi-source vehicle perception data, multi-dimensional perception information such as forward targets, lateral environment, and traffic network can be comprehensively acquired, improving the overall perception capability of complex traffic scenarios. This provides a high-precision and comprehensive input foundation for subsequent collision risk identification. By inputting feature parameters into a pre-trained collision risk assessment model for analysis, the collision level and type can be quickly and accurately determined, thereby improving the system's timeliness and predictive ability for hazardous conditions. By initiating an attitude adjustment process after the collision risk level reaches a trigger threshold, an active protection mechanism can be activated before a collision occurs, effectively gaining attitude adjustment time and improving the timeliness of occupant protection response. By matching attitude response strategies and combining them with occupant state models to generate the currently suitable target attitude, dynamic matching of protection strategies and individual differences can be achieved, thereby enhancing the effectiveness and personalization of protective measures and ultimately reducing the risk of occupant injury.

[0080] In one embodiment, step S10, namely collecting multi-source perception data of the vehicle, specifically includes:

[0081] S11: Obtain the relative distance, relative velocity, and relative acceleration of the target object ahead.

[0082] Specifically, forward-facing radar, laser ranging equipment, or image depth sensing modules are used to identify and lock the position of a target object in front of the current lane and continuously track its trajectory. The spatial distance information between the current vehicle and the target object is extracted in real time to calculate the relative distance. At the same time, the relative speed of the target object is estimated by combining the rate of change of the target object's position over time, and its relative acceleration is further derived based on the speed change trend. This forms the basic motion characteristic parameters for subsequent collision risk analysis. For example, when the target object is a vehicle traveling in the same lane and decelerates rapidly, its relative acceleration will increase significantly, indicating a potential rear-end collision risk.

[0083] S12: Obtain identification information of objects ahead, trajectory data of the lane ahead, and information on obstacles to the side.

[0084] Specifically, the vehicle-mounted camera image recognition module, LiDAR mapping engine, and high-precision map matching module are used to fuse and solve the images and depth data ahead to extract the type identification, size features, and behavioral status information of the target objects. At the same time, the lane line extraction algorithm is executed to geometrically fit the road ahead to obtain the lane trajectory curvature and extension direction. Based on the lateral LiDAR point cloud density and occlusion analysis, information on possible obstacle areas on both sides of the road is extracted. This constructs forward target object information, lane geometry information, and lateral spatial risk layout with multi-dimensional semantic attributes. For example, when a large truck is driving in the center of the lane ahead and there are hard guardrails on both sides, the lateral variable space is extremely small.

[0085] S13: Obtain vehicle network information, including emergency braking events of the vehicle in front, road abnormality warnings, or construction warnings.

[0086] Specifically, the system receives real-time information broadcasts from nearby vehicles and traffic management facilities through V2V (vehicle-to-vehicle) and V2I (vehicle-to-infrastructure) communication modules. Based on vehicle identification codes and geographical locations, it performs spatiotemporal matching and extracts event data from the matched information, including emergency braking signals triggered by vehicles ahead, road construction area markers pushed by traffic management platforms, and abnormal weather warnings. This type of data is injected into the perception and decision-making process as potential risk information in the current environment that cannot be directly perceived by sensors. For example, if a road within 300 meters ahead is temporarily closed due to sudden flooding, this information can be perceived in advance via V2X communication and trigger an early warning mechanism.

[0087] S14: The relative distance, relative speed, relative acceleration, identification information of forward objects, trajectory data of the lane ahead, information of lateral obstacles, and vehicle network information are summarized to obtain multi-source perception data of the vehicle.

[0088] Specifically, the acquired multi-source perception information is structurally integrated, and the relative distance, relative speed, and relative acceleration are fused and calibrated synchronously according to the timestamp. Combined with the identification category and behavior state of the forward target, the geometric trajectory and passable range of the current lane are matched. At the same time, combined with the potential obstacle risks in the lateral space and the auxiliary information from the Internet of Vehicles, a unified multi-dimensional feature data structure is constructed and cached in the perception data pool for subsequent modeling and risk analysis. For example, the multi-source perception data at a certain moment can be uniformly represented as: the vehicle in front is a sedan, 20 meters away, decelerating relative to the road, the road is a curve, there is an obstacle on the right, and the vehicle in front has triggered an emergency braking event.

[0089] In one embodiment, step S10 involves preprocessing the vehicle multi-source perception data to obtain feature parameters, specifically including:

[0090] S15: Calculate the collision time based on relative speed and relative distance.

[0091] Specifically, based on relative speed and relative distance, a time-to-collision calculation formula is executed, dividing the relative distance by the absolute value of the relative speed. A threshold range is set when the relative speed is close to zero to avoid meaningless infinite results. At the same time, a dynamic safety factor correction term is introduced to deal with sudden changes during acceleration and deceleration. The estimated time required for a collision with the target object in front under the current motion trend is obtained, which serves as a key indicator of the degree of approach of potential risks. For example, when the relative distance is 15 meters and the relative speed is -5 meters / second, the calculated collision time is 3 seconds. If this value is less than the preset safety time, it is marked as a high-risk state.

[0092] S16: Based on the identification information of the target object ahead and the trajectory data of the lane ahead, a judgment result is obtained through fusion analysis.

[0093] Specifically, the target object's identification type, such as passenger cars, trucks, motorcycles, and pedestrians, is modeled with its location. Combined with information such as curvature changes, lane numbers, and traffic space boundaries in the trajectory data of the lane ahead, fusion logic is executed to determine the dynamic interaction relationship between the target object and the current vehicle. The predicted trajectory is calculated to see if it intersects with the current vehicle through projection mapping. If it does, it is further determined whether the target object enters the main vehicle's forward path. At the same time, it is assessed whether the target object will change lanes, decelerate, or remain stationary within the future time window. Finally, a target behavior judgment result is generated for risk assessment. For example, if the identified target is a pedestrian located at a curved intersection and the current predicted trajectory has a high degree of intersection, it is marked as an interference object that requires special intervention.

[0094] S17: Encode and summarize the collision time distance, judgment result, relative distance, relative speed, relative acceleration, identification information, lateral obstacle information and vehicle network information to form feature parameters.

[0095] Specifically, the raw perception data and calculated indicators are encoded and compressed according to a fixed data structure template, including normalization, time series synchronization, and labeling. The collision time distance and judgment result are used as key variables in the time-series prediction features, and are vectorized and concatenated with other parameters such as relative distance, relative speed, and relative acceleration. Combined with recognition information, lateral obstacle spatial information, and vehicle network event encoding data, a multi-dimensional feature vector structure containing semantic and dynamic information is formed as the input unit for subsequent model analysis. For example, the feature vector can be constructed in the form of [0.25,1,15.3,-5.0,-0.7,2,1,0], corresponding to the collision time distance, judgment result label, relative distance, speed, acceleration, target type, lateral obstacle presence, and emergency braking event marker, respectively, thereby achieving unified semantic fusion of multi-source data.

[0096] In one embodiment, the active occupant posture protection method further includes:

[0097] S201: A collision risk assessment model is constructed using a supervised learning algorithm based on deep neural networks. The collision risk assessment model structure includes a feature encoding layer, a feature fusion layer, and a risk judgment output layer.

[0098] Specifically, the supervised learning structure is constructed by calling the deep neural network modeling function to learn the input features. First, the feature encoding layer is set to perform low-dimensional transformation and vectorization of the input perceptual data. Then, a multi-head attention mechanism and residual connection structure are introduced in the feature fusion layer to improve the interactive expression ability between multi-source features. The fused output is then sent to the risk judgment output layer to perform classification prediction operation, which is used to output the collision risk level and collision type label. The feature encoding layer adopts a fully connected structure with ReLU activation, the feature fusion layer adopts a stacked structure to fuse spatial location information and semantic information, and the risk judgment layer normalizes the output based on the softmax function. For example, in a certain training process, the perceptual input includes data of a car with a speed of 70km / h and a car 15m away. After the model is trained, it can output a medium-to-high risk level and predict a rear-end collision.

[0099] S202: Collect historical driving data and accident case data of various collision types to build a training dataset. The training dataset includes the relative distance, relative speed, relative acceleration, collision time distance, target object identification type, lane position status, lateral obstacle information, and vehicle network event information between the vehicle and the target object.

[0100] Specifically, by accessing driving behavior databases and accident scene log files, typical collision event samples occurring under different road types and traffic densities are extracted, including urban straight-going rear-end collisions, high-speed lane-changing collisions, and side scrapes when meeting oncoming traffic in narrow lanes. For each sample, the spatial distance, speed change trend, and acceleration data between the vehicle and the target are extracted, and the collision time distance is further calculated based on historical trajectories. At the same time, the vehicle type of the target ahead, such as a sedan, truck, or two-wheeled vehicle, is identified, and the lane information of the vehicle and the location and type of lateral obstacles are extracted. Combined with vehicle network broadcast data, events such as sudden braking of the vehicle in front or road construction events are recorded synchronously to form a training dataset with diverse scenarios and rich features. For example, a certain road condition sample includes joint information such as the vehicle's speed of 90km / h, the target vehicle's acceleration of -3.2m / s², the distance between the two vehicles of 22m, lane narrowing, and construction warning events.

[0101] S203: Based on the collision consequence information recorded in accident cases, label each training sample in the training dataset with the corresponding collision risk level label and collision type label.

[0102] Specifically, by analyzing the accident consequence field in historical accident data, risk levels are labeled for known collision samples in the training dataset. The collision consequence information includes factors such as vehicle damage level, personal injury level, collision angle, and collision area. Multiple indicators are combined to set risk level label ranges such as low risk (minor scratches), medium risk (moderate vehicle damage), and high risk (serious collision or personal injury). At the same time, collision type labels such as rear-end collision, side collision, and oncoming collision are labeled according to the collision location and vehicle behavior. A consistency review mechanism is introduced in the labeling process to ensure accuracy. For example, if a training sample is involved in a left-side collision at a highway ramp exit, causing the vehicle's airbags to deploy, it can be labeled as high-risk and a side collision type.

[0103] S204: Use the multidimensional perceptual features of each group of samples in the training dataset as the model training input, and use the collision risk level label and collision type label corresponding to each group of samples as the supervised training target to train the collision risk assessment model and obtain the pre-trained collision risk assessment model.

[0104] Specifically, by constructing a tensor input structure, the multidimensional perceptual features of each set of samples are loaded as input variables into the model training interface. At the same time, the corresponding collision risk level label and collision type label are used as supervision targets. The model parameters are optimized by minimizing the cross-entropy loss function and the L2 regularization term. An iterative backpropagation algorithm is used to adjust the network weights. After each epoch of training, the accuracy and loss value changes are evaluated on the validation set, and an early stopping mechanism is executed to prevent overfitting. For example, when the training set contains a batch of high-risk samples in high-speed scene rear-end collisions, the model can gradually learn to recognize the risk increase trend under the combination of high speed + short distance + negative acceleration. Finally, a set of pre-trained model weight parameters with good generalization ability in real-world scenarios is obtained.

[0105] In one embodiment, before step S20, i.e. before inputting the feature parameters into a pre-trained collision risk assessment model for analysis and outputting collision risk assessment results, which include collision risk level indicators and collision type, the occupant attitude active protection method further includes:

[0106] S205: Obtain current vehicle braking capacity assessment information, including braking response time, braking system health status, current vehicle load status, and road surface adhesion coefficient.

[0107] Specifically, by collecting response time sequence data after the vehicle control command is issued, the time interval between the triggering of the braking command and the wheel-end braking response is calculated to obtain the braking response time. Then, the health status of the braking system is evaluated by reading the status information related to the master cylinder, vacuum booster, hydraulic lines and brake calipers. Furthermore, the current load status of the vehicle is obtained by combining signals such as on-board weighing module or electric drive current feedback. At the same time, the adhesion coefficient of the road surface is estimated based on the wheel-end speed change, tire pressure status, wheel slippage or road recognition model results. For example, in icy and snowy environments, the risk of reduced adhesion coefficient is judged and marked by tire slip ratio and wheel lock-up frequency. Thus, an evaluation dataset for comprehensively describing the current braking performance of the vehicle is formed for subsequent dynamic adjustment of the model.

[0108] S206: Dynamically adjust the pre-trained collision risk assessment model based on braking capability assessment information. The dynamic adjustment includes adaptive setting of risk level judgment threshold, selection of model substructure, or adjustment of feature weight factors.

[0109] Specifically, after obtaining braking capacity assessment information, the assessment threshold in the original collision risk model is adjusted according to the current vehicle response capability and the environment. For example, when the vehicle is detected to be fully loaded and the road surface adhesion coefficient is lower than the threshold, the collision critical index value required for high-risk output in the model is actively reduced to trigger the protection process in advance. Based on the current assessment results, a sub-network structure branch adapted to severe working conditions is selected in the model architecture to enhance the judgment accuracy. At the same time, the feature factors of the model input are dynamically weighted to improve the reliability of the decision. For example, the calculation weight of the relative speed factor is increased in high-speed driving conditions, and the weight of the correlation factor between vehicle load and response time is increased in low-adhesion scenarios to ensure that the risk output results are closer to the current actual braking capability, thereby improving the accuracy and timeliness of collision recognition.

[0110] In one embodiment, step S50, which involves acquiring the occupant's physical parameters and current sitting posture, and constructing an occupant state model, specifically includes:

[0111] S51: Obtain the occupant's physical parameters, including height, weight, shoulder width, hip width, and spine length.

[0112] Specifically, occupant physical parameters are obtained by calling occupant health record databases, in-vehicle camera image recognition, or seat-embedded sensor arrays. Height can be obtained by measuring the relative height of the seat headrest in conjunction with stereo vision, weight can be obtained by detecting the pressure sensor under the seat, shoulder width and hip width can be estimated by image recognition and seat pressure distribution, and spine length can be estimated based on back contact surface sensor array and skeletal modeling algorithm. For example, in practical applications, the driver's seat area can be identified by real-time images combined with depth measurement to determine that an occupant has a shoulder width of 45cm, a hip width of 41cm, and a weight of 78kg. This data is then used in conjunction with downstream modeling logic to provide a structural basis for subsequent posture response adaptation.

[0113] S52: Obtain the occupant's current sitting posture, which includes head position, back tilt angle, pelvic rotation angle, and lower limb support angle.

[0114] Specifically, by decoding data from multi-area flexible pressure sensors inside the seat, and combining this with three-dimensional position sensors and camera modules at the front seat headrests, the system extracts the occupant's current sitting posture parameters. The head position is represented by three-dimensional coordinates to indicate the current headrest position, the back tilt angle is based on real-time angle changes measured by a backrest bending sensor array, and the pelvic rotation angle is obtained through a posture detection unit at the bottom of the seat. The lower limb support angle is calculated by combining the leg support pressure area and the sinking amplitude of the front edge of the seat cushion. The entire sitting posture acquisition process can be continuously executed during vehicle operation and the sitting posture cache is updated periodically. For example, when the occupant's back tilt angle is detected to be 110° and the lower limb support angle is 25°, it can be determined that the occupant is in a semi-resting state, thus providing accurate sitting posture input for subsequent posture response matching.

[0115] S53: Construct an occupant state model based on body parameters and sitting posture. The occupant state model represents the current occupant's posture space occupancy and motion constraint characteristics in the vehicle seat.

[0116] Specifically, by combining the acquired physical parameters and sitting posture, an occupant state model is constructed through spatial geometric modeling and constraint analysis algorithms. First, the relative spatial position relationship between the occupant and the seat is calibrated in three-dimensional space. Then, based on the biomechanical database and the boundary conditions of the vehicle's internal structure, a motion range boundary matrix is ​​established. For example, in a scenario where an occupant is 183cm tall, has a back tilt angle of 125 degrees, a hip width of 40cm, and is currently sitting on the right, it can be deduced that the occupant has a state of "larger pelvic torque + smaller back support surface" during lateral adjustment. This state model further outputs the maximum allowable displacement range, minimum safe support angle, and potentially restricted parts of the shoulders and head, providing a precise adaptation basis for subsequent target posture selection and response execution.

[0117] In one embodiment, step S60, which combines the attitude response strategy and the occupant state model to generate the currently adapted target attitude, specifically includes:

[0118] S61: Based on the posture space occupancy pattern and motion constraint characteristics represented in the crew state model, the target posture candidate set is screened to determine the executable target posture subset.

[0119] Specifically, based on the occupant state model's output of the 3D posture space occupancy boundary and motion constraint parameters, a feasibility verification process is performed on each posture scheme in the target posture candidate set. This includes spatial compatibility judgment, posture transition smoothness check, and local force safety prediction. If a target posture has a head tilt angle exceeding the head and neck movement limit in the model or requires the occupant's lower limbs to quickly retract in a short time to complete the adjustment action, then the posture will be marked as unexecutable and removed from the candidate set. Finally, all posture schemes that meet the spatial layout constraints and motion tolerance range are retained as the target posture subset. For example, in a high-risk frontal collision scenario, the original candidate set contains 8 forward tilt avoidance strategies and 3 lateral retraction strategies. After screening, only 4 postures with the head and pelvis moving forward simultaneously and the legs slightly retracted are retained to meet the body matching conditions and action execution feasibility, and are used as the input set for the next optimization algorithm.

[0120] S62: Based on the executable target posture subset, combined with posture response priority, collision type and occupant comfort requirements, an optimization algorithm is executed to determine the current suitable target posture. Occupant comfort requirements are set based on motion constraint features and preset ergonomic threshold judgment rules.

[0121] Specifically, after obtaining a subset of target postures, a comprehensive scoring optimization algorithm is invoked to score each posture scheme in multiple dimensions. The scoring indicators include the degree of matching with the current collision type, the strategy level score under the response priority rule, and the occupant comfort assessment score. The comfort assessment is judged by comparing the motion constraint indicators extracted from the occupant state model with the ergonomic constraint rules built into the system. For example, different levels of comfort factors are set for the back force angle, the degree of spinal curvature, and the lower limb posture transformation speed. Finally, all scores are weighted and fused to generate an adaptation index and sorted. The posture with the highest comprehensive adaptation degree is selected as the current adaptation target posture. For example, in a scenario where a "side-front low-speed collision" is detected, the optimization algorithm scores the three candidate postures as 86.3, 91.7, and 78.5, respectively. Since the second posture performs best in terms of back support stability and response time, it is finally determined as the final execution object and passed to the subsequent adjustment module.

[0122] Furthermore, after executing the posture adjustment operation corresponding to the target posture, the system continuously collects the occupant's posture feedback data and vehicle dynamic status information to determine whether the posture deviation and time delay between the current occupant posture and the target posture meet the dynamic stability requirements and posture recovery time limit requirements. If it is detected that the actual executed posture is not completed within the preset time window or there is a posture deviation caused by secondary vehicle impact, braking slippage, or occupant's autonomous reaction, a secondary posture adjustment mechanism is triggered. On the basis of the original posture response strategy, an incremental posture adjustment command is executed to further optimize the control strategy of the restraint device for the head, back, or lower limbs. For example, during the first adjustment, the occupant only completed the upper body forward leaning avoidance action, but the pelvic rotation angle was abnormal due to road bumps. The system will automatically trigger the fine adjustment action of the lower limb support angle to ensure that the overall posture returns to the target stable range, thereby improving the response closed-loop capability and dynamic adaptation effect of posture control.

[0123] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0124] In one embodiment, an active occupant posture protection system is provided, which corresponds one-to-one with the active occupant posture protection method described in the above embodiments. For example... Figure 2 As shown, this occupant attitude active protection system includes a multi-source sensing and acquisition module, a collision risk assessment module, a collision trigger determination module, an attitude strategy matching module, an occupant state modeling module, and an attitude generation and execution module. Detailed descriptions of each functional module are as follows:

[0125] The multi-source perception acquisition module is used to collect multi-source perception data of the vehicle and preprocess the multi-source perception data of the vehicle to obtain feature parameters;

[0126] The collision risk assessment module is used to input feature parameters into a pre-trained collision risk assessment model for analysis and output collision risk assessment results, which include collision risk level indicators and collision type.

[0127] The collision trigger determination module is used to determine an unavoidable collision state when the collision risk level index reaches the preset trigger threshold, and then enter the attitude adjustment process.

[0128] The attitude strategy matching module is used to match the collision type and feature parameters in a preset protection strategy library to obtain the corresponding attitude response strategy. The attitude response strategy includes a preset target attitude candidate set and attitude response priority.

[0129] The occupant state modeling module is used to obtain the occupant's physical parameters and current sitting posture, and to construct an occupant state model.

[0130] The attitude generation and execution module is used to combine the attitude response strategy and the crew state model to generate the current suitable target attitude, and to perform attitude adjustment operations on the crew based on the target attitude.

[0131] Optionally, the multi-source sensing acquisition module includes:

[0132] The forward target perception submodule is used to obtain the relative distance, relative velocity, and relative acceleration of the target object in front;

[0133] The environmental information acquisition submodule is used to acquire identification information of objects ahead, lane trajectory data ahead, and information on side obstacles;

[0134] The vehicle network information collection submodule is used to acquire vehicle network information, including emergency braking events of the vehicle in front, road abnormality warnings, or construction warnings.

[0135] The multi-source information fusion submodule is used to summarize relative distance, relative speed, relative acceleration, identification information of forward targets, forward lane trajectory data, lateral obstacle information and vehicle network information to obtain vehicle multi-source perception data;

[0136] The collision time-distance calculation submodule is used to calculate the collision time-distance based on relative speed and relative distance.

[0137] The target behavior determination submodule is used to perform fusion analysis based on the identification information of the target object ahead and the trajectory data of the lane ahead to obtain the determination result;

[0138] The feature encoding submodule is used to encode and summarize the collision time distance, judgment result, relative distance, relative speed, relative acceleration, recognition information, lateral obstacle information and vehicle network information to form feature parameters.

[0139] Optionally, this occupant attitude active protection system also includes:

[0140] The risk assessment model construction module is used to construct a collision risk assessment model using a supervised learning algorithm based on deep neural networks. The collision risk assessment model structure includes a feature encoding layer, a feature fusion layer, and a risk judgment output layer.

[0141] The training data construction module is used to collect historical driving data and accident case data of various collision types to build a training dataset. The training dataset includes the relative distance, relative speed, relative acceleration, collision time distance, target object recognition type, lane position status, lateral obstacle information, and vehicle network event information between the vehicle and the target object.

[0142] The sample labeling module is used to label each training sample in the training dataset with the corresponding collision risk level label and collision type label based on the collision consequence information recorded in accident cases.

[0143] The model training module is used to take the multidimensional perceptual features of each group of samples in the training dataset as the model training input, and take the collision risk level label and collision type label corresponding to each group of samples as the supervised training target to train the collision risk assessment model and obtain the pre-trained collision risk assessment model.

[0144] The braking capacity assessment module is used to obtain the current vehicle's braking capacity assessment information, which includes braking response time, braking system health status, current vehicle load status, and the coefficient of friction of the road surface.

[0145] The model dynamic adjustment module is used to dynamically adjust the pre-trained collision risk assessment model based on braking capability assessment information. The dynamic adjustment includes adaptive setting of risk level judgment threshold, selection of model substructure, or adjustment of feature weight factors.

[0146] Optional, the occupant state modeling module includes:

[0147] The physical parameter acquisition submodule is used to acquire the physical parameters of the occupants, including height, weight, shoulder width, hip width, and spine length.

[0148] The posture state acquisition submodule is used to acquire the current sitting posture state of the occupant, which includes head position, back tilt angle, pelvic rotation angle and lower limb support angle.

[0149] The occupant state modeling submodule is used to construct an occupant state model based on body parameters and sitting posture. The occupant state model represents the current occupant's posture space occupancy and motion constraint characteristics in the vehicle seat.

[0150] Optionally, the pose generation and execution module includes:

[0151] The attitude feasibility screening submodule is used to screen the target attitude candidate set based on the attitude space occupancy pattern and motion constraint characteristics represented in the crew state model, and determine the executable target attitude subset.

[0152] The attitude optimization decision submodule is used to determine the current suitable target attitude by executing an optimization algorithm based on an executable subset of target attitudes, combined with attitude response priority, collision type and occupant comfort requirements. Occupant comfort requirements are set based on motion constraint features and preset ergonomic threshold judgment rules.

[0153] For specific limitations regarding an active occupant attitude protection system, please refer to the limitations regarding an active occupant attitude protection method described above, which will not be repeated here. Each module in the aforementioned active occupant attitude protection system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0154] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 3 As shown. The computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements an active occupant posture protection method.

[0155] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps:

[0156] Collect multi-source perception data of vehicles and preprocess the multi-source perception data to obtain feature parameters;

[0157] The feature parameters are input into a pre-trained collision risk assessment model for analysis, and the collision risk assessment results are output, including collision risk level indicators and collision type.

[0158] When the collision risk level index reaches the preset trigger threshold, it is determined to be an unavoidable collision state and enters the attitude adjustment process.

[0159] Based on collision type and feature parameters, a matching is performed in a preset protection strategy library to obtain the corresponding attitude response strategy. The attitude response strategy includes a preset target attitude candidate set and attitude response priority.

[0160] Obtain the occupant's physical parameters and current sitting posture to construct an occupant state model;

[0161] The current target posture is generated by combining the posture response strategy and the occupant state model, and the posture adjustment operation is performed on the occupant based on the target posture.

[0162] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0163] Collect multi-source perception data of vehicles and preprocess the multi-source perception data to obtain feature parameters;

[0164] The feature parameters are input into a pre-trained collision risk assessment model for analysis, and the collision risk assessment results are output, including collision risk level indicators and collision type.

[0165] When the collision risk level index reaches the preset trigger threshold, it is determined to be an unavoidable collision state and enters the attitude adjustment process.

[0166] Based on collision type and feature parameters, a matching is performed in a preset protection strategy library to obtain the corresponding attitude response strategy. The attitude response strategy includes a preset target attitude candidate set and attitude response priority.

[0167] Obtain the occupant's physical parameters and current sitting posture to construct an occupant state model;

[0168] The current target posture is generated by combining the posture response strategy and the occupant state model, and the posture adjustment operation is performed on the occupant based on the target posture.

[0169] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0170] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above.

[0171] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for active protection of occupant posture, characterized in that, The active occupant posture protection method includes: Collect vehicle multi-source perception data and preprocess the vehicle multi-source perception data to obtain feature parameters; The feature parameters are input into a pre-trained collision risk assessment model for analysis, and the collision risk assessment results are output, including collision risk level indicators and collision type. When the collision risk level index reaches the preset trigger threshold, it is determined to be an unavoidable collision state and enters the attitude adjustment process. Based on the collision type and the feature parameters, a matching is performed in a preset protection strategy library to obtain a corresponding attitude response strategy. The attitude response strategy includes a preset target attitude candidate set and attitude response priority. Obtain the occupant's physical parameters and current sitting posture to construct an occupant state model; The attitude response strategy and the occupant state model are combined to generate the current suitable target attitude, and the occupant attitude adjustment operation is performed based on the target attitude.

2. The active occupant posture protection method according to claim 1, characterized in that, The collected multi-source perception data of the vehicle specifically includes: Obtain the relative distance, relative velocity, and relative acceleration of the target object ahead; Acquire the identification information of the target object ahead, the trajectory data of the lane ahead, and the information of the side obstacles; Acquire vehicle network information, including emergency braking events of the vehicle in front, road anomaly warnings, or construction warnings; The relative distance, relative speed, relative acceleration, identification information of the forward target, the trajectory data of the lane ahead, the information of the lateral obstacles, and the vehicle network information are combined to obtain the vehicle multi-source perception data.

3. The active occupant posture protection method according to claim 2, characterized in that, The step of preprocessing the vehicle multi-source perception data to obtain feature parameters specifically includes: Calculate the collision time interval based on the relative velocity and the relative distance; A determination result is obtained by fusing and analyzing the identification information of the target object ahead with the trajectory data of the lane ahead. The collision time interval, the determination result, the relative distance, the relative speed, the relative acceleration, the identification information, the lateral obstacle information, and the vehicle network information are encoded and summarized to form the feature parameters.

4. The active occupant posture protection method according to claim 1, characterized in that, The aforementioned active occupant posture protection method also includes: A collision risk assessment model is constructed using a supervised learning algorithm based on deep neural networks. The collision risk assessment model structure includes a feature encoding layer, a feature fusion layer, and a risk determination output layer. Historical driving data and accident case data of various collision types are collected to construct a training dataset. The training dataset includes the relative distance, relative speed, relative acceleration, collision time, target object identification type, lane position status, lateral obstacle information, and vehicle network event information between the vehicle and the target object. Based on the collision consequence information recorded in the accident cases, each training sample in the training dataset is labeled with a corresponding collision risk level label and collision type label; The multidimensional perceptual features of each group of samples in the training dataset are used as the model training input, and the collision risk level label and collision type label corresponding to each group of samples are used as the supervised training target to train the collision risk assessment model, thereby obtaining the pre-trained collision risk assessment model.

5. The active occupant posture protection method according to claim 1, characterized in that, Before inputting the feature parameters into a pre-trained collision risk assessment model for analysis and outputting a collision risk assessment result, which includes a collision risk level index and a collision type, the occupant attitude active protection method further includes: Obtain the current vehicle's braking capacity assessment information, which includes braking response time, braking system health status, current vehicle load status, and the coefficient of friction of the road surface. The pre-trained collision risk assessment model is dynamically adjusted based on the braking capability assessment information. The dynamic adjustment includes adaptive setting of the risk level determination threshold, selection of model substructure, or adjustment of feature weight factors.

6. The active occupant posture protection method according to claim 1, characterized in that, The process of acquiring the occupant's physical parameters and current sitting posture to construct an occupant state model specifically includes: Obtain the occupant's physical parameters, including height, weight, shoulder width, hip width, and spine length; The current sitting posture of the occupant is obtained, including head position, back tilt angle, pelvic rotation angle and lower limb support angle. Based on the physical parameters and the sitting posture, an occupant state model is constructed. The occupant state model represents the current occupant's posture space occupancy and motion constraint characteristics in the vehicle seat.

7. The active occupant posture protection method according to claim 6, characterized in that, The step of generating the currently adapted target posture by combining the posture response strategy and the occupant state model specifically includes: Based on the posture space occupancy pattern and motion constraint characteristics represented in the occupant state model, the target posture candidate set is screened to determine the executable target posture subset. Based on the executable target posture subset, combined with the posture response priority, the collision type and occupant comfort requirements, an optimization algorithm is executed to determine the currently suitable target posture. The occupant comfort requirements are set based on the motion constraint features and preset ergonomic threshold judgment rules.

8. An active occupant posture protection system, characterized in that, The occupant posture active protection system includes: The multi-source sensing acquisition module is used to acquire multi-source sensing data of the vehicle and preprocess the multi-source sensing data of the vehicle to obtain feature parameters; The collision risk assessment module is used to input the feature parameters into a pre-trained collision risk assessment model for analysis and output collision risk assessment results, which include collision risk level indicators and collision types. The collision trigger determination module is used to determine an unavoidable collision state when the collision risk level index reaches a preset trigger threshold, and then enter the attitude adjustment process. The attitude strategy matching module is used to perform matching in a preset protection strategy library based on the collision type and the feature parameters to obtain the corresponding attitude response strategy. The attitude response strategy includes a preset target attitude candidate set and attitude response priority. The occupant state modeling module is used to obtain the occupant's physical parameters and current sitting posture, and to construct an occupant state model. The attitude generation and execution module is used to combine the attitude response strategy and the occupant state model to generate the currently adapted target attitude, and to perform attitude adjustment operations on the occupant based on the target attitude.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the active occupant posture protection method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the active occupant posture protection method as described in any one of claims 1 to 7.

Images