阅读说明：本技术 一种视觉与动力学融合的路面附着系数估计方法 (Vision and dynamics fused road adhesion coefficient estimation method ) 是由 熊璐 金达 冷搏 杨兴 于洋 关佚卓 余卓平 于 2020-05-26 设计创作，主要内容包括：本发明涉及一种视觉与动力学融合的路面附着系数估计方法,包括以下步骤：S1：获取车辆行驶过程中前方的路面图像；S2：将路面图像训练完成的路面分类模型,得到前方路面类型；S3：根据前方路面类型和路面类型与路面附着系数映射关系,获取路面附着系数视觉估计值θ<Sub>ximage</Sub>；S4：获取车辆行驶过程中轮胎的动力学信息；S5：利用路面附着系数-轮胎纵向力估计器,获取当前路面峰值附着系数估计值θ<Sub>x</Sub>；S6：结合路面附着系数视觉估计值θ<Sub>ximage</Sub>和当前路面峰值附着系数估计值θ<Sub>x</Sub>,利用模糊推理规则得到最终路面附着系数估计值,与现有技术相比,本发明具有估计精度高、实时性好、鲁棒性强等优点。(The invention relates to a road adhesion coefficient estimation method integrating vision and dynamics, which comprises the following steps of: s1: acquiring a road surface image in front of a vehicle in the driving process; s2: obtaining a front road type by using a road classification model after the road image training is finished; s3: obtaining a road adhesion coefficient vision estimation value theta according to the front road surface type and the mapping relation between the road surface type and the road adhesion coefficient ximage (ii) a S4: acquiring dynamic information of a tire in the running process of a vehicle; s5: obtaining the current road surface peak value adhesion coefficient estimated value theta by utilizing the road surface adhesion coefficient-tire longitudinal force estimator x (ii) a S6: visual estimation value theta combined with road adhesion coefficient ximage And the current road surface peak value adhesion coefficient estimated value theta x Compared with the prior art, the method has the advantages of high estimation precision, good real-time performance, strong robustness and the like.)

1. A road adhesion coefficient estimation method integrating vision and dynamics is characterized by comprising the following steps of:

s1: acquiring a road surface image in front of a vehicle in the driving process;

s2: inputting the road image into the trained road classification model to obtain the front road type;

s3: obtaining a road adhesion coefficient vision estimation value theta according to the front road surface type and the mapping relation between the road surface type and the road adhesion coefficient_ximage；

S4: acquiring dynamic information of a tire in the running process of a vehicle;

s5: tire using road surface adhesion coefficientA longitudinal force estimator for obtaining the current road surface peak value adhesion coefficient estimated value theta_x；

S6: visual estimation value theta combined with road adhesion coefficient_ximageAnd the current road surface peak value adhesion coefficient estimated value theta_xAnd obtaining a final road adhesion coefficient estimation value by using a fuzzy inference rule.

2. The vision and dynamics-fused road adhesion coefficient estimation method according to claim 1, wherein the training process of the road classification model specifically comprises:

s21: acquiring road surface images under different weather conditions to obtain road surface images comprising different road surface types;

s22: marking the road surface type of the obtained road surface image;

s23: and (5) performing end-to-end training on the deep convolution neural network by using the marked road surface image to obtain a trained road surface classification model.

3. The method of claim 2, wherein the road surface types include dry road surface, wet road surface and snow and ice road surface, the dry road surface corresponds to dry weather road surface, the wet road surface corresponds to rainy road surface, and the snow and ice road surface corresponds to snow and ice road surface.

4. The vision and dynamics integrated road adhesion coefficient estimation method according to claim 3, wherein the road type and road adhesion coefficient mapping relationship is specifically:

the road surface adhesion coefficient of the dry road surface is 0.85;

the road surface adhesion coefficient of the wet and slippery road surface is 0.6;

the road surface adhesion coefficient of the ice and snow road surface is 0.15.

5. A visual and dynamic fusion road surface adhesion coefficient estimation method according to claim 2, wherein the deep convolutional neural network is based on a deep learning framework of Pytorch, tensoflow or Caffe.

6. The vision and dynamics fused road adhesion coefficient estimation method according to claim 1, wherein the road adhesion coefficient-tire longitudinal force estimator specifically comprises:

s51: building a tire-vehicle dynamic model;

s52: and establishing a road adhesion coefficient-tire longitudinal force estimator based on a disturbance observation theory.

7. The vision and dynamics integrated road adhesion coefficient estimation method according to claim 6, wherein the tire-vehicle dynamics model is:

where ω represents the wheel angular velocity, R represents the wheel radius, T_mIndicating the driving/braking torque, mu, acting on the wheel_x(θ_xλ) represents the coefficient of adhesion of the current tire to the ground, F_zIndicating the vertical load to which the wheel is subjected, I_ωRepresenting the moment of inertia of the wheel, lambda representing the wheel slip ratio, v_xRepresenting the longitudinal speed at the centre of the wheel, theta_xRepresenting the peak value adhesion coefficient of the current dynamic road surface;

the single-wheel dynamic model specifically comprises the following steps:

in the formula, theta represents the peak value adhesion coefficient of the road surface, namely the mu-lambda curve is the mostThe peak value adhesion coefficient of the corresponding road surface of the high point; λ represents a wheel slip ratio, c₁Representing the cornering stiffness of the tyre, i.e. the slope of the μ - λ curve at the origin, c₂、c₃、c₄Respectively, control parameters of the curve descending segment.

8. A vision and dynamics fused road adhesion coefficient estimation method according to claim 7, wherein the expression of the road adhesion coefficient-tire longitudinal force estimator is:

in the formula (I), the compound is shown in the specification,

9. A vision and dynamics fused road adhesion coefficient estimation according to claim 1The method is characterized in that the input of the fuzzy inference rule is a road adhesion coefficient vision estimation value theta_ximageAnd road surface adhesion coefficient visual estimation value theta_ximageAnd the current road surface peak value adhesion coefficient estimated value theta_xDifference value of (a) | θ_ximage-θ_xAnd the output is longitudinal force estimator gain K, road adhesion coefficient estimator gain gamma and an excitation threshold value for triggering the fusion estimator, and corresponding intervals are divided according to S, M, B (corresponding to small, medium and large) fuzzy membership degrees.

10. A visual and dynamic fusion road adhesion coefficient estimation method as claimed in claim 9, wherein the visual road adhesion coefficient estimation value θ is_ximageIs set to [0.15,0.85 ]]Visual estimation value theta of road adhesion coefficient_ximageAnd the current road surface peak value adhesion coefficient estimated value theta_xDifference of | θ of_ximage-θ_xArgument field of | (0.05, 1.0)]The argument of the gain K of the longitudinal force estimator is set to [3,30 ]]The argument region of the gain gamma of the road adhesion coefficient estimator is set to [3,30 ]]The argument field of the excitation threshold of the trigger fusion estimator is set to [0.03, 0.15%]。

Technical Field

The invention relates to the field of electric automobile control, in particular to a road adhesion coefficient estimation method integrating vision and dynamics.

Background

The road adhesion coefficient is an important parameter in vehicle system dynamics control, the control quality is greatly influenced by the estimation accuracy, and if the current road adhesion coefficient can be accurately obtained, the number of traffic accidents in rainy and snowy days can be greatly reduced.

Current research methods can be broadly divided into two categories: one is to estimate road adhesion coefficients based on a vehicle dynamics estimator, mainly taking into account the excitation of the tires by the road. The advantage of this method is that the hardware required for identification is relatively easy to satisfy, and is already installed in most cars, and the accuracy of identification is high. However, in order to obtain an accurate recognition result, a sufficient excitation condition is required, that is, the recognition effect is significant only when the utilization coefficient of the tire road is close to the peak adhesion coefficient, which makes the vehicle liable to be unstable. Another method is to measure the road adhesion coefficient directly. This type of research typically requires the use of other sensors, such as cameras and lidar. The advantage of this approach is that the detection range of recognition is wide and predictive. However, most sensors used are expensive and susceptible to environmental influences.

Disclosure of Invention

The invention aims to overcome the defects of the prior art and provide a visual and dynamic integrated road adhesion coefficient estimation method which is high in estimation precision, good in real-time performance and strong in robustness.

The purpose of the invention can be realized by the following technical scheme:

the invention provides a road adhesion coefficient estimation method integrating vision and dynamics, which comprises the following steps of:

s1: acquiring a road surface image in front of a vehicle in the driving process;

s2: inputting the road image into the trained road classification model to obtain the front road type;

s3: obtaining a road adhesion coefficient vision estimation value theta according to the front road surface type and the mapping relation between the road surface type and the road adhesion coefficient_ximage；

S4: acquiring dynamic information of a tire in the running process of a vehicle;

s5: obtaining the current road surface peak value adhesion coefficient estimated value theta by utilizing the road surface adhesion coefficient-tire longitudinal force estimator_x；

S6: visual estimation value theta combined with road adhesion coefficient_ximageAnd the current road surface peak value adhesion coefficient estimated value theta_xAnd obtaining a final road adhesion coefficient estimation value by using a fuzzy inference rule.

Further, the training process of the road surface classification model specifically includes:

s21: acquiring road surface images under different weather conditions to obtain road surface images comprising different road surface types;

s22: marking the road surface type of the obtained road surface image;

s23: and (5) performing end-to-end training on the deep convolution neural network by using the marked road surface image to obtain a trained road surface classification model.

Further preferably, the road surface types include a dry road surface, a wet road surface and an ice and snow road surface, the dry road surface corresponds to a road surface in dry weather, the wet road surface corresponds to a road surface in rainy weather, and the ice and snow road surface corresponds to a road surface in ice and snow weather. The invention considers three road surface types including most road surface conditions possibly encountered in the driving process, and can ensure the accuracy of the estimated value of the road surface adhesion coefficient.

More preferably, the mapping relationship between the road surface type and the road surface adhesion coefficient is specifically as follows:

the road surface adhesion coefficient of the dry road surface is 0.85;

the road surface adhesion coefficient of the wet and slippery road surface is 0.6;

the road surface adhesion coefficient of the ice and snow road surface is 0.15.

And correspondingly setting different pavement adhesion coefficients according to the characteristics of different pavement types.

Further preferably, the deep learning framework based on the deep convolutional neural network is Pytorch, tenserflow or Caffe.

Further, the step of establishing the road adhesion coefficient-tire longitudinal force estimator specifically comprises:

s51: building a tire-vehicle dynamic model;

s52: and establishing a road adhesion coefficient-tire longitudinal force estimator based on a disturbance observation theory.

Information on slip ratio and wheel drive torque is readily available and therefore an estimator for estimating road adhesion coefficient by longitudinal force is relatively easy to build.

Furthermore, the tire-vehicle dynamics model is as follows:

where ω represents the wheel angular velocity, R represents the wheel radius, T_mIndicating the driving/braking torque, mu, acting on the wheel_x(θ_xλ) represents the coefficient of adhesion of the current tire to the ground, F_zIndicating the vertical load to which the wheel is subjected, I_ωRepresenting the moment of inertia of the wheel, lambda representing the wheel slip ratio, v_xRepresenting the longitudinal speed at the centre of the wheel, theta_xRepresenting the peak adhesion coefficient of the current dynamic road surface.

The single-wheel dynamic model specifically comprises the following steps:

in the formula, theta tableThe peak value adhesion coefficient of the road surface is characterized, namely the peak value adhesion coefficient of the road surface corresponding to the highest point of the mu-lambda curve; λ represents a wheel slip ratio, c₁Representing the cornering stiffness of the tyre, i.e. the slope of the μ - λ curve at the origin, c₂、c₃、c₄Respectively, control parameters of the curve descending segment.

The expression of the road adhesion coefficient-tire longitudinal force estimator is as follows:

in the formula (I), the compound is shown in the specification,

representing an estimate of the longitudinal force of the tyre, mu_x(θ_xLambda) represents the current tire ground utilization adhesion coefficient, K represents the longitudinal force estimator gain,

represents the road surface peak adhesion coefficient calculated according to the current longitudinal force and the wheel slip ratio,

the method is an estimated value of the peak value adhesion coefficient of the road surface, gamma represents the gain of a road adhesion coefficient estimator, y is an intermediate variable in the operation process, and has no actual physical meaning.

Further, the input of the fuzzy inference rule is a road adhesion coefficient vision estimation value theta_ximageAnd road surface adhesion coefficient visual estimation value theta_ximageAnd the current road surface peak value adhesion coefficient estimated value theta_xDifference value of (a) | θ_ximage-θ_xI, output as longitudinal force estimator gainK. The road adhesion coefficient estimator gain γ and the excitation threshold that triggers the fusion estimator, and divides the corresponding intervals according to S, M, B (corresponding to small, medium, and large) fuzzy membership.

Further preferably, the visual estimation value theta of the road adhesion coefficient_ximageIs set to [0.15,0.85 ]]Visual estimation value theta of road adhesion coefficient_ximageAnd the current road surface peak value adhesion coefficient estimated value theta_xDifference of | θ of_ximage-θ_xArgument field of | (0.05, 1.0)]The argument of the gain K of the longitudinal force estimator is set to [3,30 ]]The argument region of the gain gamma of the road adhesion coefficient estimator is set to [3,30 ]]The argument field of the excitation threshold of the trigger fusion estimator is set to [0.03, 0.15%]. The following table is the fuzzy logic inference rule in the present invention:

compared with the prior art, the invention has the following advantages:

1) the road adhesion coefficient estimation algorithm obtains a road adhesion coefficient vision estimation value theta according to the road type through image processing_ximageOn the basis, the estimation result is further corrected through a dynamic algorithm, and compared with the traditional sensor estimation method, the estimation precision is higher;

2) the road adhesion coefficient estimation algorithm can simultaneously call road image information and vehicle state information, and has higher convergence speed and good real-time performance compared with the traditional dynamics estimation method;

3) the road adhesion coefficient estimation algorithm adopts a visual and dynamic fusion mode, and can maintain the estimated value at an empirical value through the road adhesion coefficient visual estimation value under the working condition of insufficient tire force excitation so as to realize rapid convergence when the next excitation comes, and the robustness is strong.

4) The method estimates the current road surface peak value adhesion coefficient through the longitudinal force, the slip ratio and the information of the wheel driving torque are easy to obtain, and a road surface adhesion coefficient-tire longitudinal force estimator is easy to establish and convenient to implement;

5) the invention fuses the visual estimation value of the road adhesion coefficient and the current peak adhesion coefficient estimation value of the road through fuzzy reasoning, can better fit the nonlinear characteristics of the system, simultaneously realizes smooth switching of control, avoids the occurrence of buffeting phenomena and improves the stability.

Drawings

FIG. 1 is a schematic flow diagram of the process of the present invention;

FIG. 2 is a schematic flow chart of an embodiment of the present invention;

FIG. 3 is a schematic diagram of a single-wheel kinematic model;

fig. 4 is a schematic diagram of a mapping relationship between the road surface type and the road surface adhesion coefficient.

Detailed Description

The invention is described in detail below with reference to the figures and specific embodiments. It is to be understood that the embodiments described are only a few embodiments of the present invention, and not all embodiments. All other embodiments, which can be obtained by a person skilled in the art without any inventive step based on the embodiments of the present invention, shall fall within the scope of protection of the present invention.