阅读说明：本技术 基于相机传感器路面信息识别的主动悬架预瞄控制方法 (Active suspension pre-aiming control method based on camera sensor road surface information identification ) 是由 陈志勇 ***强 于远彬 于 2020-07-15 设计创作，主要内容包括：本发明公开了一种基于相机传感器路面识别的主动悬架预瞄控制方法,包括以下步骤：通过相机传感器获取前方路面预瞄信息；对前方路面预瞄信息进行数据处理,获得前方路面不平度值,根据不平度划分准则确定路面不平度等级；接收前方路面不平度等级指令,并建立基于前方路面预瞄信息的主动悬架模型；设计多性能指标约束的主动悬架H<Sub>∞</Sub>最优控制器；使用多目标约束优化算法获取H<Sub>∞</Sub>最优控制器中的权重值,以实现悬架多种性能指标的性能平衡。本发明利用相机传感器获取路面预瞄信息,结合H<Sub>∞</Sub>最优控制,改善了主动悬架阻尼参数调整过程中的时滞现象,实现了主动悬架的多性能指标均衡控制。(The invention discloses an active suspension pre-aiming control method based on camera sensor road surface identification, which comprises the following steps: acquiring the pre-aiming information of the front road surface through a camera sensor; carrying out data processing on the pre-aiming information of the front road surface to obtain an unevenness value of the front road surface, and determining the grade of the unevenness of the road surface according to an unevenness dividing criterion; receiving a front road surface unevenness grade instruction, and establishing an active suspension model based on front road surface preview information; design of active suspension H with multiple performance index constraints ∞ An optimal controller; obtaining H using a multi-objective constrained optimization algorithm ∞ And optimizing the weight value in the controller to realize the performance balance of various performance indexes of the suspension. The invention utilizes the camera sensor to acquire the road surface preview information in combination with H ∞ Optimal control is achieved, the time lag phenomenon in the process of adjusting the damping parameters of the active suspension is improved, and multi-performance index balance control of the active suspension is achieved.)

1. An active suspension pre-aiming control method based on camera sensor road surface identification is characterized by comprising the following steps:

acquiring front road surface preview information through a camera sensor;

secondly, performing data processing on the front road surface preview information to obtain a front road surface unevenness value, and determining a road surface unevenness grade according to an unevenness dividing criterion;

step three, receiving the front road surface unevenness grade instruction determined in the step two, and establishing an active suspension model based on front road surface preview information;

step four, designing an active suspension H with multiple performance index constraints_∞An optimal controller;

step five, obtaining H by using a multi-objective constraint optimization algorithm_∞And optimizing the weight value in the controller to realize the performance balance of multiple suspension performance indexes.

2. The active suspension pre-aiming control method based on the camera sensor road surface recognition is characterized in that in the first step, before the collection of the front road surface pre-aiming information, the camera sensor is calibrated and calibrated to ensure the accuracy of the collected road surface information.

3. The active suspension pre-aiming control method based on the camera sensor road surface recognition is characterized in that in the second step, the data processing of the front road surface pre-aiming information comprises the following processes:

performing noise reduction processing on an image acquired by a camera sensor by adopting a bilateral filter algorithm;

using an image splicing algorithm based on features to reconstruct a scene, and reducing overlapped contents in the same scene in the front and back continuous frame images: detecting key feature points of two RGB images of adjacent road parts by an SURF algorithm; then matching the feature points between the images by using a k-nearest neighbor algorithm, and removing abnormal values by using a RANSAC algorithm; and finally, estimating a homography transformation matrix between the two images according to the matched feature points, and splicing the continuous frame images by using homography transformation to obtain a new fusion image.

4. The active suspension pre-aiming control method based on camera sensor road surface recognition as claimed in claim 1, wherein in the second step, the calculation of the IRI value is performed after the image processing of the pre-aiming information of the front road surface collected by the camera sensor, so as to obtain the front road surface unevenness value, and the road surface unevenness grade is determined according to a preset formula, including the following processes:

transferring the depth data output by the camera sensor depth camera to an RGB image;

acquiring an aligned RGB image, importing the aligned RGB image into ProVAL software, and selecting a half car model to calculate an IRI value;

determining the grade value alpha of the road surface unevenness according to the grade division rule in the international standard ISO 8608_r。

5. The active suspension pre-aiming control method based on camera sensor pavement recognition is characterized in that the step of transferring the depth data output by the camera sensor depth camera to an RGB image comprises the following steps:

firstly, the depth data acquired by the camera sensor is converted from image coordinates to world coordinates, and the calculation process is as follows:

Z_IR＝Z_D

in the formula (X)_IR,Y_IR,Z_IR) World coordinates of three-dimensional points on the camera sensor; (u)_D,v_D) Pixel coordinates of the depth image;is the distortion center coordinate of the camera sensor plane;is the focal length of the sensor portion of the IR camera; z_DRepresenting depth values, obtained by a camera sensor; intrinsic parameters of remaining camera sensors

then, converting the depth coordinate information of the camera sensor into RGB image coordinates, wherein a calculation formula is as follows;

wherein (X)_RGB,Y_RGB,Z_RGB) World coordinates relative to the RGB camera sensor; (X)_IR,Y_IR,Z_IR) World coordinates relative to the IR camera sensor; the rotation matrix R and the translation matrix t are external parameters obtained through calibration and estimation of a camera sensor;

finally, the world coordinate (X) is determined according to_RGB,Y_RGB,Z_RGB) Pixel coordinates (u) mapped to RBG image_RGB,v_RGB) The above step (1);

wherein the content of the first and second substances,are all intrinsic parameters of the RGB camera sensor obtained by camera sensor calibration.

6. The active suspension pre-aiming control method based on the camera sensor road surface recognition is characterized in that the step three of establishing an active suspension model based on the front road surface pre-aiming information comprises the following processes:

the front suspension and the rear suspension adopt Newton's second law to obtain a suspension model dynamic equation as follows:

wherein M is the sprung mass of the vehicle body, I is the moment of inertia of the vehicle body relative to the center of mass, and M_f、m_rFor unsprung masses, x, of the front and rear wheels_sIs the absolute displacement of the vehicle body, x_f、x_rRespectively, the unsprung mass displacement of the front and rear wheels, theta is the pitch angle, k_sf、k_srFor the rigidity of the front and rear suspensions, c_f、c_rIntrinsic damping coefficient, k, of front and rear suspensions, respectively_tf、k_trStiffness of front and rear tires, u_f、u_rIs according to H_∞Controlling the damping force generated by the strategy, wherein a and b are respectively the front and rear wheelbases, and L is the pre-aiming distance;

the state variables that define the suspension system are:

the disturbance inputs are:

wherein w (t) is the front wheel input; w (t-tau) is the rear wheel input; τ ═ (a + b)/v is the time delay between the front and rear wheels;

control input is U ═ U_f,u_r]^T；

Rewriting the system equation into the form of a state equation:

X＝AX+BU+D₁w(t)+D₂w(t-τ)

N_vthe number of state variables of the model;

7. the active suspension pre-aiming control method based on camera sensor road surface recognition is characterized in that the step four is used for designing an active suspension H with multiple performance index constraints_∞The optimal controller comprises the following processes:

considering system control targets as system sprung mass vertical acceleration, system pitch angle acceleration, front and rear suspension dynamic travel and front and rear wheel tire deformation, determining a system performance objective function as follows:

in the formula, T is control cycle time;J₃＝E[(z_s+aθ-z_f)²]；J₄＝E[(z_s-bθ-z_r)²]；J₅＝E[(z_f-h_f)²]；J₆＝E[(z_r-h_r)²]； is according to H_∞Optimally controlling the generated suspension damping force;

the above formula is rewritten as follows:

in the formula, the performance index is restricted; q ═ C_yu ^TC_yuIs a symmetric semi-positive definite matrix; f ═ D_yu ^TD_yu、N＝C_yu ^TD_yuIs a positive definite matrix; c_yu＝C_yu1C_yu2；

According to H_∞Control law, assuming that the measurement vector ψ is:

ψ＝C_mx+ξ

wherein, C_mA conversion matrix of state and measurement is adopted, and xi is a measurement error;

to minimize the performance indicator function, the system control input U is solved for:

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

in the formula, t_pIs the preview time; c_a、C_bControlling gain for feedback and feedforward; wherein, C_a＝F^-1(N^T+B^TS)；C_a＝-F^-1N^T(ii) a S is a steady state solution obtained by the Riccati equation:

S(A-BF^-1N^T)+(A-BF^-1N^T)^TS-S(BF^-1B^T-^-2DD^T)S+(Q-NF^-1N^T)＝0

estimation of system state vectorComprises the following steps:

wherein L is H_∞Gain vector of optimal controller: l ═ PC_m ^TR^-1(ii) a R is a positive definite matrix of measurement errors; p is the covariance matrix of the estimation error:

Technical Field

The invention relates to parameter control of an active suspension, in particular to a pre-aiming control method of the active suspension based on camera sensor road surface identification.

Background

When the vehicle runs, the excitation provided by the uneven road surface may cause the vehicle to bump, so that the vehicle generates vertical vibration, and the smoothness and comfort of the vehicle are affected. In order to improve the ride comfort of the vehicle, it is necessary to provide a new method for improving the suspension performance. With the continuous development of machine vision, the traditional 2D camera sensor has been gradually transformed into a 3D camera sensor, which can not only represent two-dimensional plane information but also represent three-dimensional stereo coordinate information in the photographed image, such as: kinect, ZED 2KStereo Camera, and Bumble Bee, among others. If the stereoscopic vision sensor can be applied to road surface unevenness detection, pre-aiming information is provided for suspension control, and the suspension is preset with parameters to complete the vibration reduction function, so that the suspension performance is greatly improved.

The road heading information may allow the vehicle suspension to adjust to the most appropriate setting before receiving the road excitation input. For example, when the camera sensor scans a concave-convex road surface, the suspension controller is adjusted to be in a soft damping state in advance to absorb road surface excitation so as to reduce vehicle body vibration and improve comfort, and is timely switched to be in a hard damping state after passing through the concave-convex road surface so as to inhibit subsequent oscillation of a vehicle body.

The existing research of active suspension control through preview information only stays at the theoretical level of wheelbase preview control. In the document [ Madau D P, Khaykin B l. continuous variable semi-active suspension system using centrally located road surface rate and accelerometer sensors: US 2003 ], road surface information fed back from a front axle is applied to the adjustment of a rear axle suspension according to a travel time difference between the front axle and the rear axle. This method is generally time-lag and only works after the front wheels of the vehicle have run for a period of time on the road, but the driver has experienced vibrations from the undulations. In the literature [ study on a laser radar-based road unevenness reconstruction method ] road height information in front of a vehicle is obtained by a laser radar, but the method is not generally implemented due to the cost and computational complexity of the laser radar.

Therefore, the active suspension pre-aiming control method based on camera sensor road surface identification is designed and developed, not only can the road surface pre-aiming information be accurately obtained, the time lag phenomenon in the active suspension control process is improved, but also the implementation cost of the method is reduced due to the use of the low-cost camera sensor.

Disclosure of Invention

Aiming at the technical problems, the invention provides an active suspension pre-aiming control method based on camera sensor road surface identification, which can realize road surface information identification at low cost and is easy to realize active suspension pre-aiming control.

Acquiring road surface pre-aiming information according to a Kinect camera sensor, calculating road surface unevenness, adding the unevenness information into a suspension system model, and establishing a suspension system dynamic model based on the vehicle front pre-aiming control; secondly, a suspension controller is designed to realize optimization of multiple performance indexes of a suspension model, improve a time lag phenomenon in a suspension parameter adjusting process and reduce adverse effects caused by road excitation.

In order to realize the functions, the invention adopts the technical scheme that:

an active suspension pre-aiming control method based on camera sensor road surface identification comprises the following steps:

acquiring front road surface preview information through a camera sensor;

secondly, performing data processing on the front road surface preview information to obtain a front road surface unevenness value, and determining a road surface unevenness grade according to an unevenness dividing criterion;

step three, receiving the front road surface unevenness grade instruction determined in the step two, and establishing an active suspension model based on front road surface preview information;

step four, designing an active suspension H with multiple performance index constraints_∞An optimal controller;

step five, obtaining H by using a multi-objective constraint optimization algorithm_∞And optimizing the weight value in the controller to realize the performance balance of various performance indexes of the suspension.

Further, in the first step, before the front road surface preview information is collected, the camera sensor is calibrated and calibrated first, so that the accuracy of the collected road surface information is ensured.

Further, in the second step, the data processing of the forward road preview information includes the following processes:

performing noise reduction processing on an image acquired by a camera sensor by adopting a bilateral filter algorithm;

using an image splicing algorithm based on features to reconstruct a scene, and reducing overlapped contents in the same scene in the front and back continuous frame images: detecting key feature points of two RGB images of adjacent road parts by an SURF algorithm; then matching the feature points between the images by using a k-nearest neighbor algorithm, and removing abnormal values by using a RANSAC algorithm; and finally, estimating a homography transformation matrix between the two images according to the matched feature points, and splicing the continuous frame images by using homography transformation to obtain a new fusion image.

Further, according to claim 1, in the second step, after the image processing, the method for active suspension pre-aiming control based on the camera sensor road surface recognition is performed, the calculation of the IRI value is performed on the pre-aiming information of the front road surface collected by the camera sensor, so as to obtain the value of the front road surface unevenness, and the road surface unevenness grade is determined according to a preset formula, which includes the following steps:

transferring the depth data output by the camera sensor depth camera to an RGB image;

acquiring an aligned RGB image (3D point cloud data containing a longitudinal contour of a road surface), importing the RGB image into ProVAL software, and selecting a semi-vehicle model to calculate an IRI value;

determining the grade value alpha of the road surface unevenness according to the grade division rule in the international standard ISO 8608_r。

Further, the transferring the depth data output by the camera sensor depth camera to the RGB image comprises the following steps:

firstly, the depth data acquired by the camera sensor is converted from image coordinates to world coordinates, and the calculation process is as follows:

Z_IR＝Z_D

in the formula (X)_IR,Y_IR,Z_IR) World coordinates of three-dimensional points on the camera sensor; (u)_D,v_D) Pixel coordinates of the depth image;

is the distortion center coordinate of the camera sensor plane;

is the focal length of the sensor portion of the IR camera; z_DRepresenting depth values, obtained by a camera sensor; intrinsic parameters of remaining camera sensorsObtaining through camera sensor calibration;

then, converting the depth coordinate information of the camera sensor into RGB image coordinates, wherein a calculation formula is as follows;

wherein (X)_RGB,Y_RGB,Z_RGB) World coordinates relative to the RGB camera sensor; (X)_IR,Y_IR,Z_IR) World coordinates relative to the IR camera sensor; the rotation matrix R and the translation matrix t are external parameters obtained through calibration and estimation of a camera sensor;

finally, the world coordinate (X) is determined according to_RGB,Y_RGB,Z_RGB) Pixel coordinates (u) mapped to RBG image_RGB,v_RGB) The above step (1);

wherein the content of the first and second substances,

are all intrinsic parameters of the RGB camera sensor obtained by camera sensor calibration.

Further, the step three of establishing the active suspension model based on the front road aiming information comprises the following processes:

the front suspension and the rear suspension adopt Newton's second law to obtain a suspension model dynamic equation as follows:

wherein M is the sprung mass of the vehicle body, I is the moment of inertia of the vehicle body relative to the center of mass, and M_f、m_rFor unsprung masses, x, of the front and rear wheels_sIs the absolute displacement of the vehicle body, x_f、x_rRespectively, the unsprung mass displacement of the front and rear wheels, theta is the pitch angle, k_sf、k_srFor the rigidity of the front and rear suspensions, c_f、c_rIntrinsic damping coefficient, k, of front and rear suspensions, respectively_tf、k_trStiffness of front and rear tires, u_f、u_rIs according to H_∞Controlling the damping force generated by the strategy, wherein a and b are respectively the front and rear wheelbases, and L is the pre-aiming distance;

the state variables that define the suspension system are:

the disturbance inputs are:

wherein w (t) is the front wheel input; w (t-tau) is the rear wheel input; τ ═ (a + b)/v is the time delay between the front and rear wheels;

control input is U ═ U_f,u_r]^T；

Rewriting the system equation into the form of a state equation:

X＝AX+BU+D₁w(t)+D₂w(t-τ)

N_vthe number of state variables of the model;

further, the step four designs the active suspension H with multiple performance index constraints_∞The optimal controller comprises the following processes:

considering system control targets as system sprung mass vertical acceleration, system pitch angle acceleration, front and rear suspension dynamic travel and front and rear wheel tire deformation, determining a system performance objective function as follows:

in the formula, T is control cycle time;J₃＝E[(z_s+aθ-z_f)²]；J₄＝E[(z_s-bθ-z_r)²]；J₅＝E[(z_f-h_f)²]；J₆＝E[(z_r-h_r)²]；

is according to H_∞Optimally controlling the generated suspension damping force;

the above formula is rewritten as follows:

in the formula, the performance index is restricted; q ═ C_yu ^TC_yuIs a symmetric semi-positive definite matrix; f ═ D_yu ^TD_yu、N＝C_yu ^TD_yuIs a positive definite matrix; c_yu＝C_yu1C_yu2；

Using H_∞Control, assume that the measurement vector ψ is:

ψ＝C_mx+ξ

wherein, C_mA conversion matrix of state and measurement is adopted, and xi is a measurement error;

to minimize the performance indicator function, the system control input U is solved for:

in the formula (I), the compound is shown in the specification,for the best estimation vector of the system state, r (t) is a vector containing the preview information, including the pre-axial and wheelbase preview information, and the formula is as follows:

in the formula, t_pIs the preview time; c_a、C_bControlling gain for feedback and feedforward; wherein, C_a＝F^-1(N^T+B^TS)；C_a＝-F^{- 1}N^T(ii) a S is a steady state solution obtained by the Riccati equation:

S(A-BF^-1N^T)+(A-BF^-1N^T)^TS-S(BF^-1B^T-^-2DD^T)S+(Q-NF^-1N^T)＝0

estimation of system state vectorComprises the following steps:

wherein L is H_∞Gain vector of optimal controller: l ═ PC_m ^TR^-1(ii) a R is a positive definite matrix of measurement errors; p is the covariance matrix of the estimation error:

the invention has the following beneficial effects:

1) a Kinect camera sensor with low cost is selected as a pavement information acquisition sensor, so that the development cost of the method is reduced while the acquisition precision is ensured, and the method is easy to realize;

2) the method comprises the steps that front road information is used as input of pre-aiming control, an active suspension pre-aiming control model based on the front road information is established, road interference can be prevented in advance, and a suspension system is adjusted to the most appropriate control parameters in advance before a vehicle inputs road excitation response;

3) designing H constrained by multiple performance indicators_∞The predictive controller generates a feedforward term using the front road surface information on the basis of feedback control to form more effective control input.

Drawings

Fig. 1 is a flow chart of an active suspension pre-aiming control method based on camera sensor road surface recognition.

Fig. 2 is a schematic view of a camera sensor mounting position.

Fig. 3 is an active suspension semi-vehicle model.

Fig. 4 is a vehicle body acceleration transfer characteristic curve.

Fig. 5 is a tire dynamic deformation transfer characteristic curve.

Fig. 6 is a process diagram of the present invention.

Detailed Description

The invention is described in further detail below with reference to the following figures and examples: