Vehicle driving state estimation method and system in centralized driving mode

Abstract

The invention provides a vehicle driving state estimation method and system in a centralized driving mode, and relates to the technical field of electric automobiles, and the method comprises the following steps: determining wheel dynamic parameters based on sensor detection information and a seven-degree-of-freedom dynamic model related to wheels, and combining a preset Dugoff tire model to obtain initial tire force parameters; a vehicle-mounted infrared camera is used for obtaining road surface image information, the road surface image information is analyzed through a deep convolutional neural network based on semantic segmentation and a vehicle dynamics inversion method, an adaptive fusion mechanism is introduced for adjustment, and a target road surface adhesion coefficient is determined; correcting the initial tire force parameter by using the target road adhesion coefficient to obtain a target tire force parameter; inputting the current sensor detection information and the target tire force parameter into a self-adaptive unscented Kalman filter to obtain vehicle driving state information; the vehicle driving state information comprises the longitudinal speed, the transverse speed, the yaw velocity and the side slip angle of each tire.

Description

Technical Field

[0001] This invention relates to the field of electric vehicle technology, and in particular to a method and system for estimating vehicle driving state under a centralized drive mode. Background Technology

[0002] In real-world road environments, vehicles often operate under various conditions, including alternating acceleration, braking, and steering maneuvers, as well as sudden changes in road surface adhesion coefficients. Under these conditions, the vehicle's longitudinal and lateral dynamic responses change rapidly. If the vehicle's true driving state cannot be estimated in a timely and accurate manner, vehicle stability control systems such as ESP, ABS, and active steering control will struggle to respond effectively, leading to safety risks such as sideslip, fishtailing, or increased braking distance. Centralized drive structures, by centrally arranging motors and distributing driving and braking forces to each wheel via a transfer case, offer advantages such as compact structure, centralized energy management, and flexible control. However, in centralized drive systems, vehicle sensors are typically mounted in the center of the vehicle body, making it impossible to directly measure the driving or braking forces of each wheel. Furthermore, the coordinated control of the transfer case and multiple motors increases the complexity of vehicle dynamic coupling, making methods based on a single model or single sensor information insufficient for accurate estimation requirements in high-dynamic scenarios. Traditional dynamic models heavily rely on precise parameters, making tire force calculations prone to errors when road surface adhesion changes abruptly. Single-vision methods degrade in performance under weather conditions such as rain, fog, or strong light, while inertial sensors suffer from drift issues and lack adaptive mechanisms to handle time-varying noise. Therefore, existing centralized drive vehicle state estimation methods have significant limitations in terms of model accuracy, noise adaptability, and environmental perception capabilities, making it difficult to meet the higher requirements for accuracy, real-time performance, and robustness in estimating vehicle driving states under complex dynamic conditions. Summary of the Invention

[0003] In view of this, the present invention proposes a method and system for estimating vehicle driving state under a centralized driving mode.

[0004] The technical solution of this invention is implemented as follows: The first aspect of this invention provides a method for estimating the vehicle's driving state under a centralized driving mode, comprising:

[0005] Wheel dynamic parameters are determined based on sensor detection information and a seven-degree-of-freedom dynamic model related to the wheel. Initial tire force parameters are obtained by combining the preset Dugoff tire model. The sensor detection information includes the longitudinal acceleration, lateral acceleration, yaw rate, front wheel steering angle, and wheel speed of each wheel. The wheel dynamic parameters include the tire longitudinal slip ratio, tire slip angle, vertical load of each wheel, and wheel center velocity. The initial tire force parameters include initial longitudinal tire force parameters and initial lateral tire force parameters.

[0006] Road surface image information is acquired using an on-board infrared camera. The road surface image information is then analyzed using a deep convolutional neural network based on semantic segmentation and a vehicle dynamics inversion method, respectively. An adaptive fusion mechanism is introduced for adjustment to determine the target road surface adhesion coefficient.

[0007] The initial tire force parameters are corrected using the target road surface adhesion coefficient to obtain the target tire force parameters; the target tire force parameters include target longitudinal tire force parameters and target lateral tire force parameters.

[0008] The current sensor detection information and the target tire force parameters are input into an adaptive unscented Kalman filter to obtain vehicle driving state information; the vehicle driving state information includes the longitudinal velocity, lateral velocity, yaw rate and center of gravity sideslip angle of each tire.

[0009] Based on the above technical solutions, preferably, before determining the wheel dynamic parameters based on sensor detection information and a seven-degree-of-freedom dynamic model related to the wheel, and combining this with a preset Dugoff tire model to obtain the initial tire force parameters, the method includes:

[0010] Multiple sets of historical tire force parameters and corresponding tire slip ratios and tire side slip angles are obtained; the historical tire force parameters include historical lateral force parameters and historical longitudinal force parameters.

[0011] A first nonlinear factor is determined based on the mapping relationship between the historical lateral force parameters and the tire slip ratio, and a second nonlinear factor is determined based on the mapping relationship between the historical longitudinal force parameters and the tire slip angle.

[0012] The traditional Dugoff tire model is optimized using the first nonlinear factor and the second nonlinear factor to obtain the preset Dugoff tire model.

[0013] Based on the above technical solutions, preferably, the determination of wheel dynamic parameters based on sensor detection information and a seven-degree-of-freedom dynamic model related to the wheel, combined with a preset Dugoff tire model, to obtain initial tire force parameters includes:

[0014] Based on the sensor detection information, a seven-degree-of-freedom vehicle dynamics model is constructed, including longitudinal, lateral, and yaw motions of the vehicle body and rotation of the four wheels, and the wheel dynamics parameters are determined.

[0015] The wheel dynamics parameters are input into the preset Dugoff tire model to obtain the initial tire force parameters.

[0016] Based on the above technical solutions, preferably, the step of acquiring road surface image information using an onboard infrared camera, analyzing the road surface image information using a deep convolutional neural network based on semantic segmentation and a vehicle dynamics inversion method, and introducing an adaptive fusion mechanism for adjustment to determine the target road surface adhesion coefficient includes:

[0017] The preprocessed road surface image information is input into the deep semantic segmentation network SegFormer to extract the target road surface area image; the preprocessing includes image denoising, contrast stretching and histogram equalization, and resolution scaling and normalization using non-local means (NL-means).

[0018] The target road surface area image is used to identify multiple road surface types and determine the classification probability of different road surface types.

[0019] The calibration adhesion coefficient of the road surface corresponding to the maximum classification probability is determined as the first road surface adhesion coefficient.

[0020] Based on the above technical solutions, preferably, the step of acquiring road surface image information using an onboard infrared camera, analyzing the road surface image information using a deep convolutional neural network based on semantic segmentation and a vehicle dynamics inversion method, and introducing an adaptive fusion mechanism for adjustment to determine the target road surface adhesion coefficient includes:

[0021] The sensor detection information and the initial tire force parameters are input into an adaptive unscented Kalman filter to obtain the second road surface adhesion coefficient.

[0022] The first road surface adhesion coefficient and the second road surface adhesion coefficient are adaptively fused based on a preset visual classification confidence threshold and the maximum classification probability to determine the target road surface adhesion coefficient.

[0023] Based on the above technical solutions, preferably, the step of adaptively fusing the first road surface adhesion coefficient and the second road surface adhesion coefficient based on a preset visual classification confidence threshold and the maximum classification probability to determine the target road surface adhesion coefficient includes:

[0024] If the preset visual classification confidence threshold is not greater than the maximum classification probability, and the second road surface adhesion coefficient is within the effective range of the first road surface adhesion coefficient, the average value of the first road surface adhesion coefficient and the second road surface adhesion coefficient is determined as the target road surface adhesion coefficient; if the second road surface adhesion coefficient is not within the effective range of the first road surface adhesion coefficient, the first road surface adhesion coefficient is determined as the target road surface adhesion coefficient.

[0025] If the preset visual classification confidence threshold is greater than the maximum classification probability, the second road surface adhesion coefficient is determined as the target road surface adhesion coefficient.

[0026] Based on the above technical solutions, preferably, the step of correcting the initial tire force parameters using the target road surface adhesion coefficient to obtain the target tire force parameters includes:

[0027] The preset Dugoff tire model is optimized using the target road surface adhesion coefficient to obtain the target Dugoff tire model;

[0028] The initial longitudinal tire force parameters and the initial lateral tire force parameters are input into the target Dugoff tire model for updating, thereby obtaining the target longitudinal tire force parameters and the target lateral tire force parameters.

[0029] More preferably, a second aspect of the present invention provides a vehicle driving state estimation system under a centralized driving mode, comprising: a parameter acquisition module, an image analysis module, a parameter correction module, and an information determination module; wherein,

[0030] The parameter acquisition module is configured to determine wheel dynamic parameters based on sensor detection information and a seven-degree-of-freedom dynamic model related to the wheel, and to obtain initial tire force parameters by combining a preset Dugoff tire model. The sensor detection information includes the longitudinal acceleration, lateral acceleration, yaw rate, front wheel steering angle, and wheel speed of each wheel. The wheel dynamic parameters include the tire longitudinal slip ratio, tire slip angle, vertical load of each wheel, and wheel center velocity. The initial tire force parameters include initial longitudinal tire force parameters and initial lateral tire force parameters.

[0031] The image analysis module is configured to acquire road surface image information using an on-board infrared camera, analyze the road surface image information using a deep convolutional neural network based on semantic segmentation and a vehicle dynamics inversion method, and introduce an adaptive fusion mechanism for adjustment to determine the target road surface adhesion coefficient.

[0032] The parameter correction module is configured to correct the initial tire force parameters using the target road surface adhesion coefficient to obtain target tire force parameters; the target tire force parameters include target longitudinal tire force parameters and target lateral tire force parameters;

[0033] The information determination module is configured to input the current sensor detection information and the target tire force parameters into an adaptive unscented Kalman filter to obtain vehicle driving state information; the vehicle driving state information includes the longitudinal velocity, lateral velocity, yaw rate and center of gravity sideslip angle of each tire.

[0034] More preferably, a third aspect of the present invention provides an electronic device, including a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the vehicle driving state estimation method under the centralized driving mode described in the first aspect.

[0035] More preferably, a fourth aspect of the present invention provides a computer storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the vehicle driving state estimation method under the centralized driving mode described in the first aspect.

[0036] The vehicle driving state estimation method and system under centralized drive mode of the present invention have the following advantages over the prior art:

[0037] 1. By combining deep learning and dynamic computation with a multi-source information fusion method, and infrared imaging compensation, all-weather monitoring is achieved. An adaptive fusion mechanism is introduced to dynamically adjust the visual / dynamic weights to optimize the road surface adhesion coefficient. While ensuring the estimation convergence speed, the accuracy of adhesion coefficient estimation under abrupt working conditions is significantly improved, thereby improving the estimation accuracy of vehicle driving state and enhancing the adaptability and robustness of overall vehicle state estimation in complex environments.

[0038] 2. An improved Dugoff tire model is used to calculate tire force. By mining the true nonlinear mapping relationship between tire force, slip ratio, and sideslip angle through historical data, a first nonlinear factor and a second nonlinear factor are introduced to enable the model to dynamically describe the saturation characteristics of tire force. This corrects the nonlinear attenuation of lateral force when the sideslip angle is large and compensates for the coupling effect of longitudinal force when the sideslip angle exists. It achieves accurate nonlinear correction of the traditional Dugoff tire model and significantly improves the accuracy and robustness of tire force estimation under complex working conditions.

[0039] 3. By using the dual condition judgment of visual classification confidence threshold and effective range of road adhesion coefficient, adaptive fusion and robust decision-making of multi-source road adhesion coefficient estimation results are realized, which significantly improves the accuracy, reliability and adaptability of target road adhesion coefficient estimation, avoids the dominance of a single method under extreme working conditions, and achieves complementary advantages. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 A flowchart illustrating a vehicle driving state estimation method under a centralized driving mode provided in an embodiment of the present invention;

[0042] Figure 2 This is a schematic diagram of a vehicle driving state estimation system under a centralized driving mode provided in an embodiment of the present invention;

[0043] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0044] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0045] In some embodiments, such as Figure 1 As shown, Figure 1 This is a flowchart illustrating a vehicle driving state estimation method under a centralized drive mode provided by an embodiment of the present invention; the vehicle driving state estimation method under a centralized drive mode provided by the present invention includes:

[0046] S110: Based on sensor detection information and a seven-degree-of-freedom dynamic model related to the wheel, determine the wheel dynamic parameters, and combine them with the preset Dugoff tire model to obtain the initial tire force parameters; the sensor detection information includes the longitudinal acceleration, lateral acceleration, yaw rate, front wheel steering angle, and wheel speed of each wheel; the wheel dynamic parameters include the tire longitudinal slip ratio, tire slip angle, vertical load of each wheel, and wheel center velocity; the initial tire force parameters include the initial longitudinal tire force parameters and the initial lateral tire force parameters.

[0047] In this example, onboard sensors of a centrally driven vehicle are used to acquire sensor data such as longitudinal acceleration, lateral acceleration, yaw rate, front wheel steering angle, and wheel speed. A seven-degree-of-freedom vehicle dynamics model is established, encompassing the vehicle's longitudinal, lateral, and yaw motions, as well as the rotation of the four wheels. Based on this dynamics model, wheel dynamic parameters such as tire longitudinal slip ratio, tire sideslip angle, vertical load on each wheel, and wheel center velocity are calculated. These calculated wheel dynamic parameters are then input into a pre-defined Dugoff tire model for calculation, yielding initial tire force parameters.

[0048] In some embodiments, before determining wheel dynamic parameters based on sensor detection information and a seven-degree-of-freedom dynamic model related to the wheel in S110, and combining this with a preset Dugoff tire model to obtain initial tire force parameters, the method includes:

[0049] Acquire multiple sets of historical tire force parameters and corresponding tire slip ratios and tire side slip angles; historical tire force parameters include historical lateral force parameters and historical longitudinal force parameters.

[0050] The first nonlinear factor is determined based on the mapping relationship between historical lateral force parameters and tire slip ratio, and the second nonlinear factor is determined based on the mapping relationship between the historical longitudinal force parameters and tire slip angle.

[0051] The traditional Dugoff tire model is optimized using the first and second nonlinear factors to obtain the preset Dugoff tire model.

[0052] The formulas for calculating the lateral and longitudinal forces in the default Dugoff tire model are as follows:

[0053]

[0054] Here, λ and the function f(λ) can be described as:

[0055]

[0056] G S =aS 2 -bS+c

[0057] G α = dtan(α) + e;

[0058] In the formula, F z C represents the longitudinal force of the tire. x This represents the longitudinal slip stiffness of the tire, where S is the tire slip ratio and C is the longitudinal slip ratio. y The lateral stiffness is represented by α, the tire slip angle by α, μ by μ, the road adhesion coefficient by μ, λ by λ, ε by ε, and G by ε. s G α This represents the scaling factor, and a, b, c, d, and e are adjustment factors that can be adjusted based on actual test data of different tires. x This represents the vehicle's longitudinal speed.

[0059] In some embodiments, S110, wheel dynamic parameters are determined based on sensor detection information and a seven-degree-of-freedom dynamic model related to the wheel, and initial tire force parameters are obtained by combining a preset Dugoff tire model, including:

[0060] Based on sensor detection information, a seven-degree-of-freedom vehicle dynamics model is constructed, including longitudinal, lateral, and yaw motions of the vehicle body and rotation of the four wheels, and the wheel dynamics parameters are determined.

[0061] Input the wheel dynamics parameters into the preset Dugoff tire model to obtain the initial tire force parameters.

[0062] The expressions for the vertical loads of each wheel are as follows:

[0063]

[0064] In the formula, F z11 F z12 F z21 F z22 The longitudinal forces on the front left wheel, front right wheel, rear left wheel, and rear right wheel are respectively, in m. a For tire weight, m b Where L is the vehicle weight, H is the wheelbase, and a is the vehicle's center of gravity height. xmean Let a be the longitudinal acceleration of the vehicle. ymean Let be the lateral acceleration of the vehicle, b be the distance from the center of mass to the rear axle, a be the distance from the center of mass to the front axle, and t be the lateral acceleration of the vehicle. r The rear wheel track is t. f denoted as the front wheel track, and h as the acceleration due to gravity.

[0065] The expression for the center velocity of the wheel is as follows:

[0066]

[0067] In the formula, v x11 v x12 v x21 v x22 Let be the wheel center velocities of the front left wheel, front right wheel, rear left wheel, and rear right wheel, respectively; 'a' be the distance from the center of mass to the front axle; 'δ' be the steering wheel angle; and 'ω' be the wheel center velocity. r v is the yaw rate of the vehicle. x With v y These represent the vehicle's longitudinal speed and lateral speed, respectively.

[0068] The formula for calculating the longitudinal slip ratio of a tire is as follows:

[0069]

[0070] In the formula, s is the tire longitudinal slip ratio, R is the effective wheel radius, and ω ij V represents the rotational speed of each wheel. xij The longitudinal velocity of the rotation center of each wheel.

[0071] The formula for calculating the tire slip angle is as follows:

[0072] α11 =δ-arctan((v y +aw r ) / (v x -0.5t f w r ))

[0073] α 12 =δ-arctan((v y +aw r ) / (v x +0.5t f w r ))

[0074] α 21 =arctan((-v y +bw r ) / (v x -0.5t r w r ))

[0075] α 22 =arctan((-v y +bw r ) / (v x +0.5t r w r ));

[0076] In the formula, α 11 α 12 α 21 α 22 These are the tire slip angles for the front left wheel, front right wheel, rear left wheel, and rear right wheel, respectively; δ is the steering wheel angle of the front wheel; v x With v y Let a and b represent the vehicle's longitudinal and lateral velocities, respectively, and a and b be the distances from the center of mass to the front and rear axles, respectively. ω r Let t be the yaw rate of the vehicle. r The rear wheel track is t. f This refers to the track width of the front wheels.

[0077] S120 uses an onboard infrared camera to acquire road surface image information. It then analyzes the road surface image information using a deep convolutional neural network based on semantic segmentation and a vehicle dynamics inversion method, and introduces an adaptive fusion mechanism for adjustment to determine the target road surface adhesion coefficient.

[0078] By using an onboard infrared camera installed at the front of the vehicle to acquire road surface image information, it can stably collect infrared thermal images in complex environments such as daytime, nighttime, rain, snow, fog, and haze. The infrared image acquisition cycle is synchronized with the timestamp of the vehicle control system to ensure data consistency, which is used for subsequent estimation and control loop.

[0079] In some embodiments, S120, road surface image information is acquired using an onboard infrared camera, and the road surface image information is analyzed using a deep convolutional neural network based on semantic segmentation and a vehicle dynamics inversion method, respectively, and an adaptive fusion mechanism is introduced for adjustment to determine the target road surface adhesion coefficient, including:

[0080] The preprocessed road surface image information is input into the deep semantic segmentation network SegFormer to extract the target road surface area image; the preprocessing includes image denoising using non-local means (NL-means), contrast stretching and histogram equalization, and resolution scaling and normalization.

[0081] The ResNet50 deep convolutional neural network was used to identify multiple road surface types in the target road surface area image and to determine the classification probability of different road surface types.

[0082] The calibration adhesion coefficient of the road surface corresponding to the maximum classification probability is determined as the first road surface adhesion coefficient.

[0083] In this example, the preprocessed road surface image information is input into the deep semantic segmentation network SegFormer. Non-road areas, such as pedestrians, obstacles, and trees, are removed, retaining only the target road surface area image, i.e., the image of the road surface ROI (Region of Interest). The road surface ROI image is then input into the deep convolutional neural network ResNet50 for multi-type road surface recognition, outputting the classification probability corresponding to different road surface types.

[0084] P vis = [p1, p2, ..., p n ];

[0085] Where, p i This represents the probability that the road surface belongs to the i-th category (such as dry asphalt, slippery, icy, snowy, muddy, etc.).

[0086] Get the road surface type index corresponding to the highest classification probability:

[0087]

[0088] According to the i-th * Calibration coefficient of road surface Determine the first adhesion coefficient estimate corresponding to the visual method:

[0089]

[0090] In some embodiments, S120, road surface image information is acquired using an onboard infrared camera, and the road surface image information is analyzed using a deep convolutional neural network based on semantic segmentation and a vehicle dynamics inversion method, respectively, and an adaptive fusion mechanism is introduced for adjustment to determine the target road surface adhesion coefficient, including:

[0091] The sensor detection information and initial tire force parameters are input into an adaptive unscented Kalman filter to obtain the second road adhesion coefficient.

[0092] The first road surface adhesion coefficient and the second road surface adhesion coefficient are adaptively fused based on a preset visual classification confidence threshold and the maximum classification probability to determine the target road surface adhesion coefficient.

[0093] In this embodiment, sensor detection information and initial tire force parameters are input into an adaptive unscented Kalman filter (AUKF) to obtain the second road adhesion coefficient μ. dyn The AUKF-based road adhesion coefficient estimation algorithm uses the current road adhesion coefficient μ of each wheel. i The state variable is x = (μ1, μ2, μ3, μ4). T With longitudinal and lateral acceleration a x a y The signal quantity that can be collected by sensors such as yaw rate γ can be used as the observed variable z = (a x a y ,γ) τ The initial tire force parameter F calculated in the above steps x0 F y0 As the input, the AUKF sampling time is defined as Δt. After discretizing the system, a state-space equation is established, and the second road adhesion coefficient μ of the road surface at the current time is output through the AUKF algorithm. dyn Based on this, an adaptive fusion mechanism of visual and dynamic information is used to calculate and fuse the final road surface adhesion coefficient, thereby determining the target road surface adhesion coefficient.

[0094] In some embodiments, the target road surface adhesion coefficient is determined by adaptively fusing the first road surface adhesion coefficient and the second road surface adhesion coefficient based on a preset visual classification confidence threshold and the maximum classification probability, including:

[0095] If the preset visual classification confidence threshold is not greater than the maximum classification probability, and the second road surface adhesion coefficient is within the effective range of the first road surface adhesion coefficient, the average of the first road surface adhesion coefficient and the second road surface adhesion coefficient is determined as the target road surface adhesion coefficient; if the second road surface adhesion coefficient is not within the effective range of the first road surface adhesion coefficient, the first road surface adhesion coefficient is determined as the target road surface adhesion coefficient.

[0096] If the preset visual classification confidence threshold is greater than the maximum classification probability, the second road surface adhesion coefficient is determined as the target road surface adhesion coefficient.

[0097] In this embodiment, a visual classification confidence threshold P0 is set; based on the maximum classification probability p max With road surface type i * Judgment: If p max ≥P0, meaning high confidence in visual classification: query road surface type i * Effective range of adhesion coefficient like If the two are considered to be the same, their average is taken: Otherwise, the dynamic estimation is considered abnormal, and only visual estimation is used: μ fused =μ vis If p max <P0, meaning insufficient visual confidence, visual estimation is discarded, and only dynamic estimation is used: μ fused =μ dyn That is, using an adaptive fusion mechanism of visual and dynamic information to determine the final road surface adhesion coefficient μ. fused The calculation.

[0098] S130, the initial tire force parameters are corrected using the target road surface adhesion coefficient to obtain the target tire force parameters; the target tire force parameters include the target longitudinal tire force parameters and the target lateral tire force parameters.

[0099] In some embodiments, S130, the initial tire force parameters are corrected using the target road surface adhesion coefficient to obtain the target tire force parameters, including:

[0100] The target Dugoff tire model is obtained by optimizing the preset Dugoff tire model using the target road surface adhesion coefficient.

[0101] The initial longitudinal tire force parameters and initial lateral tire force parameters are input into the target Dugoff tire model for updating, resulting in the target longitudinal tire force parameters and target lateral tire force parameters.

[0102] The final fusion adhesion coefficient μ fused Feedback is sent to the preset Dugoff tire model to update the longitudinal force F. x Lateral force F y The calculation is performed; within the control cycle, the next round of dynamic estimation is carried out based on the updated Dugoff tire model to construct a complete closed-loop path, thereby improving control accuracy and system robustness.

[0103] S140 inputs the current sensor detection information and target tire force parameters into the adaptive unscented Kalman filter to obtain vehicle driving status information; the vehicle driving status information includes the longitudinal velocity, lateral velocity, yaw rate and center of gravity sideslip angle of each tire.

[0104] By inputting the current sensor detection information and target tire force parameters into the AUKF vehicle driving state estimator, the final vehicle driving state estimation result can be obtained. The specific process is as follows:

[0105] The input information for the state estimator includes at least: longitudinal acceleration, lateral acceleration, and yaw rate detected by sensors; and tire longitudinal / lateral forces corrected by road adhesion coefficient feedback. These inputs are then fed into the AUKF vehicle driving state estimator: the estimator's state vector is: x = [v x v y w r β] T Where: v x v is the longitudinal velocity of the vehicle. y w is the lateral velocity of the vehicle. r Let β be the yaw rate of the vehicle around its center of mass, and β be the sideslip angle of the vehicle's center of mass. The input vector is u = [F...]. xij ,F yij ,δ] 9×1 The observed variable is z = [a x a y w r ] 3×1 Based on the seven-degree-of-freedom vehicle dynamics model and measurement information, the state equations and measurement equations required for estimation can be established as follows:

[0106] x(k+1)=f(x(k),u(k))+w(k)

[0107] z(k) = h(x(k), u(k)) + v(k);

[0108] In the formula, x(k) is the state variable of the system at time k, z(k) is the observed value of the system at time k, u(k) is the control input at time k, w(k) and v(k) are the process noise and measurement noise, respectively, and f(·) and h(·) are the nonlinear state transition and observation functions, respectively.

[0109] The overall algorithm logic of the AUKF state estimator is as follows: First, the system is initialized, and the state vector estimation is initialized. The covariance matrix is ​​P0, the process noise covariance matrix Q0 and the observation noise covariance matrix R0 are initialized, and the unscented transformation parameters (such as α, β, κ) are set. Then, the unscented transformation is used to generate Sigma points, and further update and predict the state and covariance. Each predicted Sigma point is mapped to the observation space. The predicted observation value and observation covariance are calculated and fused to update the Kalman gain. The state and covariance matrices are updated, the forgetting factor is determined, the measurement noise is calculated based on the Sage-Husa algorithm, and the positive definiteness of the noise is corrected. The observation noise is updated. If sampling is not finished, the above algorithm process is repeated to obtain real-time state data. The specific algorithm steps are as follows:

[0110] Estimate based on the current state Covariance P k-1 Construct 2n+1 Sigma points centered at the center:

[0111]

[0112] Perform Sigma point weight calculation:

[0113]

[0114] Substituting each Sigma point into the system state transition function for nonlinear propagation, we obtain the predicted Sigma point:

[0115]

[0116] Calculate the weighted average predicted state:

[0117]

[0118] Calculate the predicted covariance:

[0119]

[0120] Predicted observation vector time update:

[0121]

[0122] Map each predicted Sigma point to the observation space:

[0123]

[0124] Calculate the observed predicted values ​​and the observed covariance:

[0125]

[0126] Calculate the Kalman gain:

[0127]

[0128] Update state and covariance:

[0129]

[0130] Perform adaptive updates to the measurement noise covariance while ensuring the positive definiteness of the measurement noise covariance matrix:

[0131]

[0132] d k+1 = (1-b) / (1-b) k+1 )

[0133]

[0134] In the formula: Q k Let r be the covariance matrix of w(k). k and R k The mean matrix and covariance matrix of the measurement noise are respectively. Let be the residual sequence matrix of the observed variables at time k+1; b is the forgetting factor, which is generally taken as 0.95 < b < 0.99.

[0135] The output is the final estimated vehicle driving state at the current moment:

[0136]

[0137] The estimation results will be fed back into the road surface adhesion coefficient inversion and subsequent controller design, forming a closed-loop path of perception-estimation-control. It should be noted that the overall algorithm logic of the AUKF state estimator is existing technology; details can be found in relevant published literature.

[0138] In some embodiments, please refer to Figure 2 , Figure 2 This is a schematic diagram of a vehicle driving state estimation system under a centralized drive mode, provided in an embodiment of the present invention. The present invention provides a vehicle driving state estimation system 200 under a centralized drive mode, comprising: a parameter acquisition module 210, an image analysis module 220, a parameter correction module 230, and an information determination module 240; wherein...

[0139] The parameter acquisition module 210 is configured to determine wheel dynamic parameters based on sensor detection information and a seven-degree-of-freedom dynamic model related to the wheel, and obtain initial tire force parameters by combining a preset Dugoff tire model. The sensor detection information includes the longitudinal acceleration, lateral acceleration, yaw rate, front wheel steering angle, and wheel speed of each wheel. The wheel dynamic parameters include the tire longitudinal slip ratio, tire slip angle, vertical load of each wheel, and wheel center velocity. The initial tire force parameters include initial longitudinal tire force parameters and initial lateral tire force parameters.

[0140] The image analysis module 220 is configured to acquire road surface image information using an on-board infrared camera, analyze the road surface image information using a deep convolutional neural network based on semantic segmentation and a vehicle dynamics inversion method, and introduce an adaptive fusion mechanism for adjustment to determine the target road surface adhesion coefficient.

[0141] The parameter correction module 230 is configured to correct the initial tire force parameters using the target road surface adhesion coefficient to obtain the target tire force parameters; the target tire force parameters include the target longitudinal tire force parameters and the target lateral tire force parameters;

[0142] The information determination module 240 is configured to input the current sensor detection information and the target tire force parameters into an adaptive unscented Kalman filter to obtain vehicle driving state information; the vehicle driving state information includes the longitudinal velocity, lateral velocity, yaw rate and center of gravity sideslip angle of each tire.

[0143] In some embodiments, the vehicle driving state estimation system 200 under centralized driving mode includes an optimization module; the optimization module is specifically configured as follows:

[0144] Acquire multiple sets of historical tire force parameters and corresponding tire slip ratios and tire side slip angles; historical tire force parameters include historical lateral force parameters and historical longitudinal force parameters.

[0145] The first nonlinear factor is determined based on the mapping relationship between historical lateral force parameters and tire slip ratio, and the second nonlinear factor is determined based on the mapping relationship between the historical longitudinal force parameters and tire slip angle.

[0146] The traditional Dugoff tire model is optimized using the first and second nonlinear factors to obtain the preset Dugoff tire model.

[0147] In some embodiments, the parameter acquisition module 210 is specifically configured as follows:

[0148] Based on sensor detection information, a seven-degree-of-freedom vehicle dynamics model is constructed, including longitudinal, lateral, and yaw motions of the vehicle body and rotation of the four wheels, and the wheel dynamics parameters are determined.

[0149] Input the wheel dynamics parameters into the preset Dugoff tire model to obtain the initial tire force parameters.

[0150] In some embodiments, the image analysis module 220 is specifically configured as follows:

[0151] The preprocessed road surface image information is input into the deep semantic segmentation network SegFormer to extract the target road surface area image; the preprocessing includes image denoising using non-local means (NL-means), contrast stretching and histogram equalization, and resolution scaling and normalization.

[0152] The ResNet50 deep convolutional neural network was used to identify multiple road surface types in the target road surface area image and to determine the classification probability of different road surface types.

[0153] The calibration adhesion coefficient of the road surface corresponding to the maximum classification probability is determined as the first road surface adhesion coefficient.

[0154] In some embodiments, the image analysis module 220 is specifically configured as follows:

[0155] The sensor detection information and initial tire force parameters are input into an adaptive unscented Kalman filter to obtain the second road adhesion coefficient.

[0156] The first road surface adhesion coefficient and the second road surface adhesion coefficient are adaptively fused based on a preset visual classification confidence threshold and the maximum classification probability to determine the target road surface adhesion coefficient.

[0157] In some embodiments, the image analysis module 220 is further configured as follows:

[0158] If the preset visual classification confidence threshold is not greater than the maximum classification probability, and the second road surface adhesion coefficient is within the effective range of the first road surface adhesion coefficient, the average of the first road surface adhesion coefficient and the second road surface adhesion coefficient is determined as the target road surface adhesion coefficient; if the second road surface adhesion coefficient is not within the effective range of the first road surface adhesion coefficient, the first road surface adhesion coefficient is determined as the target road surface adhesion coefficient.

[0159] If the preset visual classification confidence threshold is greater than the maximum classification probability, the second road surface adhesion coefficient is determined as the target road surface adhesion coefficient.

[0160] In some embodiments, the parameter correction module 230 is specifically configured as follows:

[0161] The target Dugoff tire model is obtained by optimizing the preset Dugoff tire model using the target road surface adhesion coefficient.

[0162] The initial longitudinal tire force parameters and initial lateral tire force parameters are input into the target Dugoff tire model for updating, resulting in the target longitudinal tire force parameters and target lateral tire force parameters.

[0163] It should be noted that the vehicle driving state estimation system under centralized driving mode provided in this application embodiment and the vehicle driving state estimation method under centralized driving mode provided in this application embodiment are based on the same application concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned vehicle driving state estimation method under centralized driving mode, and the repeated parts will not be described again.

[0164] In some embodiments, please refer to Figure 3 , Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device 300 provided in this application includes a processor 310 and a memory 320; the memory 320 stores a computer program, wherein the computer program, when executed by the processor, implements the aforementioned vehicle driving state estimation method under a centralized driving mode.

[0165] Specifically, processor 310 may include, for example, a general-purpose microprocessor, an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. Processor 310 may also include onboard memory for caching purposes. Processor 310 may be a single processing unit or multiple processing units for performing different actions of the method flow according to embodiments of this application.

[0166] The memory 320 may be any medium capable of containing, storing, transmitting, propagating, or transmitting instructions. For example, the memory 320 may include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, apparatuses, or propagation media. Specific examples of the memory 320 include: magnetic storage devices such as magnetic tape or hard disk drives (HDDs); optical storage devices such as optical discs (CD-ROMs); and may also be random access memory (RAM) or flash memory; and / or wired / wireless communication links.

[0167] This application also provides a computer-readable medium storing a computer program that, when executed by a processor, implements the vehicle driving state estimation method under the centralized drive mode described above. This computer-readable medium may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into that device / apparatus / system. The aforementioned computer-readable medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0168] According to embodiments of this application, a computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wired, optical fiber, radio frequency signals, etc., or any suitable combination thereof.

[0169] Those skilled in the art will understand that the features described in the various embodiments and / or claims of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments and / or claims of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application. Therefore, the scope of this application should not be limited to the above embodiments, but should be defined not only by the appended claims, but also by their equivalents. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should be included within the protection scope of this invention.

Claims

1. A method for estimating vehicle driving state under a centralized drive system, characterized in that, include: Wheel dynamic parameters are determined based on sensor detection information and a seven-degree-of-freedom dynamic model related to the wheel. Initial tire force parameters are obtained by combining the preset Dugoff tire model. The sensor detection information includes the longitudinal acceleration, lateral acceleration, yaw rate, front wheel steering angle, and wheel speed of each wheel. The wheel dynamic parameters include the tire longitudinal slip ratio, tire slip angle, vertical load of each wheel, and wheel center velocity. The initial tire force parameters include initial longitudinal tire force parameters and initial lateral tire force parameters. Road surface image information is acquired using an on-board infrared camera. The road surface image information is then analyzed using a deep convolutional neural network based on semantic segmentation and a vehicle dynamics inversion method, respectively. An adaptive fusion mechanism is introduced for adjustment to determine the target road surface adhesion coefficient. The initial tire force parameters are corrected using the target road surface adhesion coefficient to obtain the target tire force parameters; The target tire force parameters include target longitudinal tire force parameters and target lateral tire force parameters; The current sensor detection information and the target tire force parameters are input into an adaptive unscented Kalman filter to obtain vehicle driving state information; the vehicle driving state information includes the longitudinal velocity, lateral velocity, yaw rate and center of gravity sideslip angle of each tire.

2. The vehicle driving state estimation method under centralized drive mode as described in claim 1, characterized in that, Before determining wheel dynamic parameters based on sensor detection information and a wheel-related seven-degree-of-freedom dynamic model, and combining this with a preset Dugoff tire model to obtain initial tire force parameters, the method includes: Multiple sets of historical tire force parameters and corresponding tire slip ratios and tire side slip angles are obtained; the historical tire force parameters include historical lateral force parameters and historical longitudinal force parameters. A first nonlinear factor is determined based on the mapping relationship between the historical lateral force parameters and the tire slip ratio, and a second nonlinear factor is determined based on the mapping relationship between the historical longitudinal force parameters and the tire slip angle. The traditional Dugoff tire model is optimized using the first nonlinear factor and the second nonlinear factor to obtain the preset Dugoff tire model.

3. The vehicle driving state estimation method under centralized drive mode as described in claim 2, characterized in that, The wheel dynamic parameters are determined based on sensor detection information and a seven-degree-of-freedom dynamic model related to the wheel. Combined with a preset Dugoff tire model, initial tire force parameters are obtained, including: Based on the sensor detection information, a seven-degree-of-freedom vehicle dynamics model is constructed, including longitudinal, lateral, and yaw motions of the vehicle body and rotation of the four wheels, and the wheel dynamics parameters are determined. The wheel dynamics parameters are input into the preset Dugoff tire model to obtain the initial tire force parameters.

4. The vehicle driving state estimation method under centralized drive mode as described in claim 1, characterized in that, The method involves acquiring road surface image information using an onboard infrared camera, analyzing the road surface image information using a deep convolutional neural network based on semantic segmentation and a vehicle dynamics inversion method, and introducing an adaptive fusion mechanism for adjustment to determine the target road surface adhesion coefficient, including: The preprocessed road surface image information is input into the deep semantic segmentation network SegFormer to extract the target road surface area image; the preprocessing includes image denoising, contrast stretching and histogram equalization, and resolution scaling and normalization using non-local means (NL-means). The target road surface area image is used to identify multiple road surface types and determine the classification probability of different road surface types. The calibration adhesion coefficient of the road surface corresponding to the maximum classification probability is determined as the first road surface adhesion coefficient.

5. The vehicle driving state estimation method under centralized drive mode as described in claim 4, characterized in that, The method involves acquiring road surface image information using an onboard infrared camera, analyzing the road surface image information using a deep convolutional neural network based on semantic segmentation and a vehicle dynamics inversion method, and introducing an adaptive fusion mechanism for adjustment to determine the target road surface adhesion coefficient, including: The sensor detection information and the initial tire force parameters are input into an adaptive unscented Kalman filter to obtain the second road surface adhesion coefficient. The first road surface adhesion coefficient and the second road surface adhesion coefficient are adaptively fused based on a preset visual classification confidence threshold and the maximum classification probability to determine the target road surface adhesion coefficient.

6. The vehicle driving state estimation method under centralized drive mode as described in claim 5, characterized in that, The adaptive fusion of the first road surface adhesion coefficient and the second road surface adhesion coefficient based on a preset visual classification confidence threshold and the maximum classification probability to determine the target road surface adhesion coefficient includes: If the preset visual classification confidence threshold is not greater than the maximum classification probability, and the second road surface adhesion coefficient is within the effective range of the first road surface adhesion coefficient, the average value of the first road surface adhesion coefficient and the second road surface adhesion coefficient is determined as the target road surface adhesion coefficient; if the second road surface adhesion coefficient is not within the effective range of the first road surface adhesion coefficient, the first road surface adhesion coefficient is determined as the target road surface adhesion coefficient. If the preset visual classification confidence threshold is greater than the maximum classification probability, the second road surface adhesion coefficient is determined as the target road surface adhesion coefficient.

7. The vehicle driving state estimation method under centralized drive mode as described in claim 2, characterized in that, The step of correcting the initial tire force parameters using the target road surface adhesion coefficient to obtain the target tire force parameters includes: The preset Dugoff tire model is optimized using the target road surface adhesion coefficient to obtain the target Dugoff tire model; The initial longitudinal tire force parameters and the initial lateral tire force parameters are input into the target Dugoff tire model for updating, thereby obtaining the target longitudinal tire force parameters and the target lateral tire force parameters.

8. A vehicle driving state estimation system under a centralized drive mode, characterized in that, include: The module comprises a parameter acquisition module, an image analysis module, a parameter correction module, and an information determination module; among which, The parameter acquisition module is configured to determine wheel dynamic parameters based on sensor detection information and a seven-degree-of-freedom dynamic model related to the wheel, and to obtain initial tire force parameters by combining a preset Dugoff tire model. The sensor detection information includes the longitudinal acceleration, lateral acceleration, yaw rate, front wheel steering angle, and wheel speed of each wheel. The wheel dynamic parameters include the tire longitudinal slip ratio, tire slip angle, vertical load of each wheel, and wheel center velocity. The initial tire force parameters include initial longitudinal tire force parameters and initial lateral tire force parameters. The image analysis module is configured to acquire road surface image information using an on-board infrared camera, analyze the road surface image information using a deep convolutional neural network based on semantic segmentation and a vehicle dynamics inversion method, and introduce an adaptive fusion mechanism for adjustment to determine the target road surface adhesion coefficient. The parameter correction module is configured to correct the initial tire force parameters using the target road surface adhesion coefficient to obtain target tire force parameters; the target tire force parameters include target longitudinal tire force parameters and target lateral tire force parameters; The information determination module is configured to input the current sensor detection information and the target tire force parameters into an adaptive unscented Kalman filter to obtain vehicle driving state information; the vehicle driving state information includes the longitudinal velocity, lateral velocity, yaw rate and center of gravity sideslip angle of each tire.

9. An electronic device comprising a processor and a memory; said memory having a storage for a computer program, wherein, When the computer program is executed by the processor, it implements the vehicle driving state estimation method under the centralized driving mode as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, It stores a computer program, wherein when the computer program is executed by a processor, it implements the vehicle driving state estimation method under a centralized driving mode as described in any one of claims 1 to 7.

Images