Road surface three-dimensional information acquisition method based on vehicle front preview and vehicle body self-adjusting system

Abstract

The invention belongs to the technical field of automobile assistance, and particularly relates to a road surface three-dimensional information obtaining method based on automobile front preview and an automobile body self-adjusting system. The method comprises the following steps: acquiring left image data and right image data of left and right road surfaces of a vehicle, and correcting the left image data and the right image data by setting acquisition parameters; constructing a disparity map of the pavement image; combining the disparity map with set acquisition parameters to obtain a depth map of the pavement image, and obtaining pixel point information of the pavement image according to the obtained depth map of the pavement image; pixel point information is extracted and converted into three-dimensional point cloud information, a pavement fitting plane is obtained through a least square method, and a three-dimensional point cloud model of an actual pavement is constructed. By acquiring the road surface unevenness in the wheel driving route in the three-dimensional point cloud model and combining with the vehicle condition data, the displacement output quantity of the suspension device can be acquired in advance, and the problem that in the prior art, acquisition of three-dimensional information lags behind, so that the suspension device cannot accurately adjust the vehicle body is solved.

Description

Technical Field

[0001] The invention belongs to the technical field of flow measurement, and in particular relates to a method for acquiring three-dimensional road surface information based on vehicle front preview and a vehicle body self-adjustment system. Background Art

[0002] During heavy-load transportation, when a vehicle is traveling on a complex road surface, especially on a road with speed bumps or potholes, the unevenness of the road surface will cause the suspension device to vibrate up and down, resulting in the vehicle body vibrating at different amplitudes. This vibration will not only make the driver feel uncomfortable, but will also cause the goods transported on the vehicle to collide with the vehicle body. At the same time, when the vehicle vibration frequency is large or the vibration amplitude is high, it will have a great impact on the life of the vehicle.

[0003] At present, the methods for suspension devices to obtain road information are mainly divided into indirect recognition and direct recognition. Indirect recognition models the vehicle dynamics parameters and uses the relationship between vehicle response and road excitation to infer road excitation, such as passive suspension and semi-active suspension. This method does not require additional sensors, but the size depends on the accuracy of the suspension model and requires the establishment of accurate dynamic equations. At the same time, it can only identify the road conditions in the current tire contact area, and cannot enable the controller to adjust the suspension actuator in real time to adapt to the road conditions, resulting in signal lag problems. This type of problem has a good control effect on flat roads, but if there are sudden excitations such as potholes and speed bumps on the road, the control effect is poor due to signal lag. Direct recognition uses on-board sensors such as laser radar and visual sensors to directly perceive the road conditions, and can obtain road information in front of the vehicle in advance. Among them, laser radar generates a three-dimensional model of the target by scanning, but it is weak in obtaining road color and texture information. Its accuracy is difficult to guarantee in road recognition in complex environments.

[0004] Therefore, a road surface three-dimensional information extraction technology is needed that can obtain road surface information in advance and has a high recognition ability for road surface texture, so as to realize active control of the suspension device and thus reduce the vibration amplitude and frequency of the vehicle body. Summary of the invention

[0005] The purpose of the present invention is to overcome the defects in the prior art that the vehicle body size cannot be adjusted in time according to the three-dimensional information of the road surface, resulting in the vehicle body being damaged due to the collision of the cargo with the vehicle body due to the uneven road surface, poor control effect, and difficulty in ensuring accuracy, and to provide a method for obtaining three-dimensional road surface information based on front vehicle preview and a vehicle body self-adjustment system.

[0006] The technical solution adopted by the present invention to solve its technical problem is:

[0007] As a first aspect, a method for obtaining three-dimensional road surface information based on vehicle front preview includes the following contents:

[0008] Acquire left image data and right image data of the left and right roads of the vehicle, and correct the left image data and the right image data by setting acquisition parameters;

[0009] Constructing a disparity map of the road image, that is, taking the corrected left image data and right image data as input, and using an improved stereo matching network to generate a disparity map;

[0010] The disparity map is combined with the set acquisition parameters to obtain a road surface image depth map, and pixel point information of the road surface image is obtained based on the obtained road surface image depth map;

[0011] The pixel point information is extracted and converted into three-dimensional point cloud information. RANSAC is used to filter the mismatched points. The road surface fitting plane is obtained through the least squares method to construct a three-dimensional point cloud model of the actual road surface.

[0012] Specifically, when in use, the road surface slope information is obtained according to the constructed road surface three-dimensional point cloud model, and the displacement output corresponding to the suspension device is obtained by combining the road surface slope information and the vehicle condition data;

[0013] The vehicle condition data includes: vehicle speed data, vehicle weight data, and corresponding suspension device size information.

[0014] Specifically, filtering mismatched points by RANSAC includes the following:

[0015] Convert the disparity map obtained based on the improved stereo matching network into a depth map, and extract the pixel information in the depth map;

[0016] Randomly select three pixel points to determine whether they are on the same straight line. If so, reselect them. If not, calculate the plane equation of the plane where the three points are located.

[0017] The distances from different pixel points to the fitting plane are calculated, and a distance threshold is set. The pixel points exceeding the threshold are filtered out to obtain the filtered three-dimensional point cloud data.

[0018] Specifically, the three-dimensional point cloud data filtered by RANSAC is combined with the acquisition coordinate system to construct three-dimensional point cloud data based on the actual coordinate system;

[0019] Using the least squares method, construct a three-dimensional point cloud dataset n;

[0020] According to the fitting plane formula, the pavement fitting plane equation is constructed.

[0021] Specifically, the improved stereo matching network combines a depth perception hierarchical module in the stereo matching network, generates a disparity map by importing the corrected left and right image data into the improved stereo matching network, compares the generated disparity map with the real disparity label obtained based on the lidar, trains the network, and the training is completed to form a stereo matching network.

[0022] Specifically, the improved stereo matching network generates a disparity map including the following contents:

[0023] A method for constructing a cost volume that preserves the group correlation of an improved stereo matching network, dividing left features and right features into multiple groups along the channel dimension;

[0024] Calculate the correlation map between each group and obtain multiple matching cost proposals;

[0025] Pack the matching cost proposals into cost volumes;

[0026] The depth-aware hierarchical module uses three cascaded MBConv blocks to process the early feature maps of the left image data, producing a feature map F that encodes relative depth and hierarchical information. D ;

[0027] Feature F generated by the depth perception module D Adjust the channel weights of the cost volume to enhance the matching accuracy of texture levels and depth-related areas; that is, use the sigmoid function to derive the full-resolution depth map to calculate its depth-aware channel attention weight A C , using the depth-aware channel attention weight A C Adjust each channel of the cost volume;

[0028] The depth perception channel attention weight A C With the feature map F D The obtained 4D difference volume is filtered to obtain a high-precision disparity map.

[0029] Specifically, the depth-aware channel attention weight A C The calculation formula is as follows:

[0030]

[0031] In the formula, F D It is the feature map of hierarchical information.

[0032] As a second aspect, a vehicle body self-adjustment system using the above-mentioned method for acquiring three-dimensional road surface information based on vehicle front preview comprises:

[0033] A binocular vision module is installed on the vehicle suspension device and is used to obtain left image data and right image data of the left and right roads of the vehicle in real time;

[0034] A binocular correction unit, used to correct the original image collected by the binocular vision module;

[0035] Binocular matching unit, based on the stereo matching network improved by the depth perception hierarchical module, obtains a high-precision road disparity map by matching the corrected left and right images;

[0036] The depth information unit obtains the disparity map through the binocular stereo matching algorithm and the road depth map through the camera internal parameters, and extracts the pixel information of the image through the depth map;

[0037] The 3D point cloud unit converts the pixel information extracted from the depth map into 3D point cloud data;

[0038] The intelligent control module obtains the three-dimensional point cloud data to calculate the road slope information, calculates the displacement output corresponding to the suspension device in combination with the vehicle condition data, and controls and adjusts the vehicle body displacement.

[0039] Specifically, the intelligent control module includes:

[0040] A vehicle speed detection unit, used to detect the current speed of the vehicle in real time;

[0041] A weight detection unit, used to detect the current mass of the vehicle in real time;

[0042] The three-dimensional analysis unit obtains the three-dimensional point cloud data in the three-dimensional point cloud unit, the current vehicle speed in the vehicle speed detection unit, the current mass in the weight detection unit and the size of the suspension device, obtains the road surface roughness of the road surface area that the wheels corresponding to the suspension device will pass through in the future, and obtains the slope data of the road surface through the road surface fitting plane.

[0043] Specifically, the intelligent control module also includes a data analysis unit and a control analysis unit;

[0044] The data analysis unit acquires the three-dimensional point cloud data in the three-dimensional point cloud unit, the current vehicle speed, the vehicle weight, and the road surface roughness and slope data obtained by the three-dimensional analysis unit, performs analysis and processing, and calculates the required extension and retraction amount of the actuator;

[0045] The control analysis unit obtains the extension and contraction amount calculated by the data analysis unit, transmits a control command to the actuator, controls the action of the actuator, and adjusts the displacement of the vehicle body.

[0046] The beneficial effects of the method for acquiring three-dimensional road surface information based on vehicle front preview and the vehicle body self-adjustment system of the present invention are:

[0047] The present invention processes the acquired left image data and right image data of the left and right sides of the vehicle through an improved stereo matching network, generates a disparity map of the corresponding road surface, and converts the pixel point information in the depth map into three-dimensional point cloud information by extracting the pixel point information in the depth map. The RANSAC is first used to filter the mismatched points, and then the road surface fitting plane is obtained through the least squares method to construct a three-dimensional point cloud model of the actual road surface. By obtaining the road surface unevenness in the wheel driving route in the three-dimensional point cloud model and combining it with the vehicle condition data, the displacement output of the suspension device can be obtained in advance, thereby solving the problem in the prior art that the three-dimensional information is obtained lagging, thereby making it impossible for the suspension device to accurately adjust the vehicle body.

[0048] The present invention adopts a binocular vision module for non-contact information acquisition, which can obtain future road information in advance without relying on a single sensor; the left and right images taken by the binocular camera are processed by an improved stereo matching network (i.e., GwcNet network) to generate a disparity map of the corresponding road surface, and a depth map is generated based on the camera internal parameters; by extracting the pixel point information in the depth map, it is converted into three-dimensional point cloud information, first using RANSAC to filter the mismatched points, and then obtaining the road surface fitting plane through the least squares method to construct a three-dimensional point cloud model of the actual road surface; by obtaining the road surface unevenness in the wheel driving route in the three-dimensional point cloud model and combining it with the vehicle's own data, the output displacement of the future suspension device is obtained. The present invention uses a binocular vision module, a binocular correction unit, a binocular matching unit, a three-dimensional point cloud unit, and an intelligent control module to obtain the displacement that the suspension device needs to output in the future, which can effectively realize the active control of the suspension device, thereby reducing the vibration of the vehicle body, not only improving the comfort of the driver, but also reducing the collision between the cargo and the vehicle body, ensuring the integrity of the cargo, and improving the vehicle life to a certain extent. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0050] Figure 1 It is a flowchart of the steps of a method for acquiring three-dimensional road surface information based on vehicle front preview according to an embodiment of the present invention.

[0051] Figure 2 This is an installation diagram of a binocular camera and a vehicle suspension device in an embodiment of the present invention.

[0052] Figure 3 Schematic diagram of a binocular matching network in an embodiment of the present invention.

[0053] Figure 4 Schematic diagram of a depth perception hierarchy module in an embodiment of the present invention.

[0054] Figure 5 Schematic diagram of binocular imaging principle in an embodiment of the present invention.

[0055] Figure 6 It is a schematic structural diagram of a vehicle body self-adjusting system in an embodiment of the present invention.

[0056] Figure 7 It is a flow chart of generating a three-dimensional point cloud model in an embodiment of the present invention.

[0057] Figure 8 It is a corresponding diagram of the pothole road surface and the generated three-dimensional point cloud model in the embodiment of the present invention.

[0058] Fig. 9 This is a corresponding diagram between the smooth road surface and the generated three-dimensional point cloud model in the embodiment of the present invention.

[0059] Fig.10 It is a corresponding diagram of the speed bump road surface and the generated three-dimensional point cloud model in the embodiment of the present invention.

[0060] In the figure: 1. Binocular vision module, 2. Binocular correction unit, 3. Binocular matching unit, 4. Depth information unit, 5. Three-dimensional point cloud unit, 6. Intelligent control module, 61. Vehicle speed detection unit, 62. Weight detection unit, 63. Three-dimensional analysis unit, 64. Data analysis unit, 65. Control analysis unit, 7. Actuator. DETAILED DESCRIPTION

[0061] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0062] like Figure 1 The specific embodiment of the method for obtaining three-dimensional road surface information based on vehicle front preview of the present invention shown in the figure includes the following contents:

[0063] Acquire left image data and right image data of the left and right roads of the vehicle, and correct the left image data and the right image data by setting acquisition parameters;

[0064] Constructing a disparity map of the road image, that is, taking the corrected left image data and right image data as input, and using an improved stereo matching network to generate a disparity map;

[0065] The disparity map is combined with the set acquisition parameters to obtain a road surface image depth map, and pixel point information of the road surface image is obtained based on the obtained road surface image depth map;

[0066] The pixel point information is extracted and converted into three-dimensional point cloud information. RANSAC is used to filter the mismatched points. The road surface fitting plane is obtained through the least squares method to construct a three-dimensional point cloud model of the actual road surface.

[0067] It is further explained that, when in use, the road slope information is obtained according to the constructed road surface three-dimensional point cloud model, and the displacement output corresponding to the suspension device is obtained by combining the road slope information and the vehicle condition data. The vehicle condition data in this embodiment includes: vehicle speed data, vehicle weight data, and corresponding suspension device size information.

[0068] See also Figure 2 The present invention processes the acquired left image data and right image data of the left and right sides of the vehicle through an improved stereo matching network to generate a disparity map of the corresponding road surface, and converts the pixel point information in the depth map into three-dimensional point cloud information by extracting it. The RANSAC is first used to filter the mismatched points, and then the road surface fitting plane is obtained through the least squares method to construct a three-dimensional point cloud model of the actual road surface. By obtaining the road surface unevenness in the wheel driving route in the three-dimensional point cloud model and combining it with the vehicle condition data, the displacement output of the suspension device can be obtained in advance, thereby solving the problem of the prior art that the three-dimensional information is obtained lagging behind, thereby making it impossible for the suspension device to accurately adjust the vehicle body.

[0069] The improved stereo matching network combines a depth perception hierarchical module in a stereo matching network (i.e., the GwcNet network). The disparity map is generated by importing the corrected left and right images into the improved stereo matching network. The generated disparity map is compared with the real disparity label obtained based on the lidar, and the network is trained. After the training is completed, it becomes a stereo matching network.

[0070] The improved stereo matching network generates a disparity map including the following:

[0071] Step 1: A method for constructing a cost volume that retains the group correlation of the improved stereo matching network, dividing the left features and the right features into multiple groups along the channel dimension;

[0072] Step 2: Calculate the correlation mapping between each group and obtain multiple matching cost proposals;

[0073] Step 3: Package the matching cost proposals into a cost volume;

[0074] Step 4: The depth-aware hierarchical module uses three cascaded MBConv blocks to process the early feature maps of the left image data, generating a feature map F that encodes relative depth and hierarchical information. D ;

[0075] Step 5: Use the feature F generated by the depth perception module D Adjust the channel weights of the cost volume to enhance the matching accuracy of texture levels and depth-related areas; that is, use the sigmoid function to derive the full-resolution depth map to calculate its depth-aware channel attention weight A C , using the depth-aware channel attention weight A CAdjust each channel of the cost volume;

[0076] Step 6: Set the depth perception channel attention weight A C With the feature map F D The obtained 4D difference volume is filtered to obtain a high-precision disparity map.

[0077] like Figure 3 As shown, the binocular stereo matching network in the present invention adopts group correlation and channel splicing to form a joint cost body and simplify the 3D aggregation network. At the same time, the feature extraction of the left image is performed through the depth perception hierarchy module, so that it not only contains texture and structure information but also contains depth context information, which is conducive to more accurate stereo matching, thereby better processing the texture hierarchy in the image.

[0078] The depth perception hierarchical module structure in the present invention is as follows: Figure 4 As shown in the figure, when performing image feature extraction, the left image is feature processed, and the low-order feature map is processed by three cascaded MBConv blocks to generate a feature map FD with encoded relative depth and hierarchical information. The depth estimation feature map is regarded as a depth volume, and the full-resolution depth map is derived using the sigmoid function to calculate its depth-aware channel attention weight AC, which is filtered with the 4D difference volume obtained from the feature map FD, so that the output feature map contains not only texture and structural information, but also depth context information, which is conducive to more accurate stereo matching and thus better processing of texture hierarchy in the image.

[0079] In the depth perception hierarchical attention module of this embodiment, the feature image can be used as a depth volume, each channel represents the corresponding depth probability, and the normalized depth prediction ZP calculation formula is:

[0080]

[0081] Where F′D is the interpolated full-resolution feature map and the relative depth index of ZC channel c.

[0082] Among them, the depth perception channel attention weight AC formula is as follows:

[0083] In the formula, F D It is the feature map of hierarchical information.

[0084] By combining the depth perception hierarchical module with the GwcNet network, the binocular stereo matching network can enhance its recognition of texture structure when performing feature extraction, making it more capable of recognizing road images with no texture or weak texture, thereby generating high-quality disparity maps.

[0085] Before running the improved stereo matching network, the stereo matching network is first trained on the RSRD dataset, which contains a large number of road images and real road labels obtained through lidar scanning. By comparing the disparity map generated by the binocular matching network with the real road labels obtained based on lidar scanning, the superiority of the binocular stereo matching network can be seen.

[0086] The specific test results of the binocular matching network in the RSRD data set are as follows:

[0087] Model EPE >1px(%) >2px(%) GwcNet+DH 0.186 0.420 0.143

[0088] The EPE in the table is the average error between the predicted disparity map and the true disparity map, and its calculation formula is:

[0089]

[0090] Where N is the total number of pixels in the image; Ω is the entire image area; d i est is the disparity value predicted for a certain point i in the image; d i gt is the true disparity value of a point i in the image.

[0091] By comparing the generated disparity map with the real disparity label based on RSRD, it can be seen that the disparity map generated by the binocular matching algorithm of the present invention has a small pixel error and high matching accuracy. The generated disparity map has high reliability, and the data contained therein has high reliability, which can be used for subsequent depth information extraction, generation of three-dimensional point cloud information, and extraction of three-dimensional data in the image.

[0092] A vehicle body self-adjustment system based on the above-mentioned method for acquiring three-dimensional road surface information based on vehicle front preview, such as Figure 5As shown, it includes a binocular vision module 1, a binocular correction unit 2, a binocular matching unit 3, a depth information unit 4, a three-dimensional point cloud unit 5 and an intelligent control module 6, wherein the binocular vision module 1 uses a binocular camera, which is installed on the vehicle suspension device, and is used to obtain the left image data and the right image data of the left and right roads of the vehicle in real time. The binocular correction unit 2 is used to correct the original image collected by the binocular vision module 1. The binocular matching unit 3 is based on the GwcNet network improved by the depth perception hierarchy module, and obtains a high-precision road disparity map by matching the corrected left and right images. The depth information unit 4 obtains the disparity map through the binocular stereo matching algorithm, obtains the road depth map through the camera internal parameters, and extracts the pixel information of the image through the depth map. The three-dimensional point cloud unit 5 converts the pixel information extracted from the depth map into three-dimensional point cloud data. The intelligent control module 6 obtains the three-dimensional point cloud data to calculate the road slope information, calculates the displacement output corresponding to the suspension device in combination with the vehicle condition data, and controls and adjusts the vehicle body displacement. It should be noted that the above-mentioned set acquisition parameters are the internal parameters of the binocular vision module 1.

[0093] The specific principle is as follows: the depth information unit 4 in this embodiment is used to extract pixel information in the disparity map obtained based on the binocular matching algorithm and calculate the depth information of the road surface, such as Figure 6 As shown, O L is the optical center of the left camera, O R is the optical center of the right camera, p is the imaging point of the road surface on the left camera, and p' is the imaging point of the road surface on the right camera.

[0094] D is the distance between p and p', and the calculation formula of D is:

[0095] D=B-(X L -X R );

[0096] Where B is the binocular camera baseline, i.e., the center distance between the left and right cameras, and x L is the imaging point p of the object relative to the optical center O of the left camera L Imaging offset, x R is the imaging point p' of the object relative to the optical center O of the left camera R The imaging offset.

[0097] According to the principle of similar triangles:

[0098] Where Z is the generated depth map and f is the focal length of the camera.

[0099] We can get:

[0100]

[0101] In the above formula, the camera focal length f and the binocular camera center distance B can be obtained by calibrating the binocular camera.

[0102] Then Figure 6 As shown, after obtaining the depth map of the road surface image through the depth information unit 4, the spatial coordinates of the pixel points are extracted through the internal parameters of the binocular camera, and then the initial image of the three-dimensional point cloud of the road surface is obtained. Since there are a large number of pixel points in the depth map, there may be mismatched three-dimensional data when extracting the spatial coordinates. RANSAC is used to filter the initial three-dimensional point cloud data. By setting a threshold, some mismatched points are filtered to improve the accuracy of the subsequent fitting plane.

[0103] RANSAC (i.e., random sampling consistency) can effectively fit the fitting function under the existence of a noise model. The core of this method is to divide the points into "inside points" and "outside points". In a data set containing "outside points", a continuous iteration method is used to find the optimal parameter model. Points that do not conform to the optimal model are defined as "outside points". In the three-dimensional point cloud of the road surface, the mismatched points in the generated three-dimensional point cloud data are filtered out by specifying a threshold.

[0104] In actual road conditions, in addition to road roughness, road slope is also one of the factors that affect suspension performance. Figure 2 In the process, after removing the mismatched three-dimensional point cloud data, the road surface plane equation can be effectively fitted by adopting the least squares method, and the road surface slope can be obtained through the road surface plane equation. It is further explained that the least squares method in this embodiment finds the best function match for the data by minimizing the sum of squares of errors. In the problem of fitting a plane, the least squares method can be used to determine the plane that a set of data points are closest to, that is, to find a plane so that the sum of the distances of all data points to this plane is minimized. In the least squares method of fitting a plane, there is a set of points in three-dimensional space. By finding a plane equation, the sum of the vertical distances of all points to this plane is minimized. The expression of the plane equation in this embodiment is:

[0105] Ax+By+Cz+D=0(C≠0);

[0106] In the formula, (x, y, z) indicates that the point on the plane satisfies the above plane equation. is the normal vector of the plane, which determines the direction of the plane.

[0107] Transforming the above formula, we can get:

[0108]

[0109] Among them

[0110]

[0111] It can be seen that the least squares matrix corresponding to the plane needs to satisfy the following conditions for fitting the plane:

[0112]

[0113] Where n is the discrete point in the disparity map, which minimizes the sum of squares of the fitting error S; α0, α1, and α2 are all unknown numbers, that is, S is satisfied for α k The partial derivative of (k=0,1,2) is the smallest, that is, there is the following relationship:

[0114]

[0115] Finally, we can get:

[0116]

[0117] The fitting plane equation of the target road surface can be obtained, and the actual slope of the road surface can be obtained by fitting the plane equation.

[0118] According to the above method, a 3D point cloud model identical to the target road surface can be generated based on the built binocular matching network. Figures 7 to 9 As shown, three-dimensional point cloud models of the pothole road surface, flat road surface and speed bump road surface are generated respectively.

[0119] The main function of the intelligent control module 6 is to adjust the actuator output in real time according to the vehicle condition to ensure that the vehicle body is always in a relatively stable state. Figure 5 The intelligent control module 6 in this embodiment includes a vehicle speed detection unit 61 for real-time detection of the current speed of the vehicle, a weight detection unit 62 for real-time detection of the current mass of the vehicle, a three-dimensional analysis unit 63, a data analysis unit 64, and a control analysis unit 65. The three-dimensional analysis unit 63 obtains the three-dimensional point cloud data in the three-dimensional point cloud unit 5, the current vehicle speed in the vehicle speed detection unit 61, the current mass in the weight detection unit 62, and the size of the suspension device, and obtains the road surface roughness of the road surface area that the wheel corresponding to the suspension device will pass through in the future, and obtains the road surface slope data through the road surface fitting plane. The data analysis unit 64 obtains the three-dimensional point cloud data in the three-dimensional point cloud unit 5, the current vehicle speed, the vehicle weight, and the road surface roughness and slope data obtained by the three-dimensional analysis unit 63, performs analysis and processing, and calculates the required extension amount of the actuator 7. The control analysis unit 65 obtains the extension amount calculated by the data analysis unit 64, transmits a control command to the actuator 7, controls the action of the actuator 7, and adjusts the displacement of the vehicle body.

[0120] In this embodiment, when the vehicle is driving normally, the faster the speed, the more obvious the vibration of the vehicle body under the same road conditions. Therefore, a speed detection unit 61 is added to the intelligent control module 6 in this embodiment to detect the current speed of the vehicle in real time and output the detection data. When the vehicle load is greater, the vibration of the vehicle body is smaller under the same road conditions. Therefore, a weight detection unit 62 is added in this embodiment to detect the vehicle mass and output it.

[0121] like Fig.10 As shown, in this embodiment, a binocular camera is used to obtain the road surface image in front of the vehicle, where m u With m t Between is the suspension actuator, x r is the road roughness, k1 is the tire elastic stiffness, m u is the unsprung mass, x u is the unsprung displacement, k2 is the hydraulic cylinder stiffness, c is the hydraulic cylinder damping, m t is the sprung mass, and xt is the sprung displacement. r , combining the vehicle speed with the road roughness, calculating the actual displacement required by the hydraulic cylinder, and adjusting the output displacement of the actuator 7, that is, x u , to ensure that the vehicle body is in a relatively stable state, that is, x t The change of is 0. It should be understood that the adjustment mechanism in this embodiment is a hydraulic system or an actuator or other structure, and no absolute limitation is made here.

[0122] After the data analysis unit 64 in the intelligent control module 6 analyzes and processes the data, the analyzed data is transmitted to the actuator (i.e., the actuator or hydraulic cylinder or electric cylinder) to control the output of the actuator and adjust the displacement of the vehicle body. By adjusting the output displacement of the actuator, the suspension part x can be adjusted. u The displacement of x is adjusted by u The relationship between xr, so as to ensure the body Figure 1 x in t Always in a relatively stable state.

[0123] The present invention adopts a binocular vision module 1 for non-contact information acquisition, and can acquire future road surface information in advance without relying on a single sensor; the left and right images taken by the binocular camera are processed by an improved GwcNet network to generate a disparity map of the corresponding road surface, and a depth map is generated according to the camera internal parameters; by extracting the pixel point information in the depth map, it is converted into three-dimensional point cloud information, firstly RANSAC is used to filter the mismatched points, and then the road surface fitting plane is obtained by the least square method to construct a three-dimensional point cloud model of the actual road surface; by obtaining the road surface unevenness in the wheel driving route in the three-dimensional point cloud model and combining the vehicle's own data, the output displacement of the future suspension device is obtained. The present invention uses a binocular vision module 1, a binocular correction unit 2, a binocular matching unit 3, a three-dimensional point cloud unit 5, and an intelligent control module 6 to obtain the displacement that the suspension device needs to output in the future, which can effectively realize the active control of the suspension device, thereby reducing the vibration of the vehicle body, not only improving the comfort of the driver, but also reducing the collision between the cargo and the vehicle body, ensuring the integrity of the cargo, and improving the vehicle life to a certain extent.

[0124] It should be understood that the specific embodiments described above are only used to explain the present invention, and are not used to limit the present invention. Obvious changes or modifications derived from the spirit of the present invention are still within the protection scope of the present invention.

Claims

1. A method for acquiring three-dimensional road surface information based on vehicle-front preview, characterized in that: Includes the following: Acquire left image data and right image data of the left and right roads of the vehicle, and correct the left image data and the right image data by setting acquisition parameters; Constructing a disparity map of the road image, that is, taking the corrected left image data and right image data as input, and using an improved stereo matching network to generate a disparity map; The disparity map is combined with the set acquisition parameters to obtain a road surface image depth map, and pixel point information of the road surface image is obtained based on the obtained road surface image depth map; The pixel point information is extracted and converted into three-dimensional point cloud information. RANSAC is used to filter the mismatched points. The road surface fitting plane is obtained through the least squares method to construct a three-dimensional point cloud model of the actual road surface.

2. The method for acquiring three-dimensional road surface information based on vehicle front preview according to claim 1, characterized in that: When in use, the road surface slope information is obtained according to the constructed road surface three-dimensional point cloud model, and the displacement output corresponding to the suspension device is obtained by combining the road surface slope information and the vehicle condition data; The vehicle condition data includes: vehicle speed data, vehicle weight data, and corresponding suspension device size information.

3. The method for acquiring three-dimensional road surface information based on vehicle-front preview according to claim 2, characterized in that: The use of RANSAC to filter mismatched points includes the following: Convert the disparity map obtained based on the improved stereo matching network into a depth map, and extract the pixel information in the depth map; Randomly select three pixel points to determine whether they are on the same straight line. If so, reselect them. If not, calculate the plane equation of the plane where the three points are located. The distances from different pixel points to the fitting plane are calculated, and a distance threshold is set. The pixel points exceeding the threshold are filtered out to obtain the filtered three-dimensional point cloud data.

4. The three-dimensional road surface information based on vehicle front preview according to claim 3 comprises: The three-dimensional point cloud data filtered by RANSAC is combined with the acquisition coordinate system to construct the three-dimensional point cloud data based on the actual coordinate system; Using the least squares method, construct a three-dimensional point cloud dataset n; According to the fitting plane formula, the pavement fitting plane equation is constructed.

5. The method for acquiring three-dimensional road surface information based on vehicle front preview according to claim 1, characterized in that: The improved stereo matching network combines a depth perception hierarchical module in the stereo matching network, generates a disparity map by importing the corrected left and right images into the improved stereo matching network, compares the generated disparity map with the real disparity label obtained based on the laser radar, trains the network, and the training is completed to form a stereo matching network.

6. The method for acquiring three-dimensional road surface information based on vehicle front preview according to claim 5, characterized in that: The improved stereo matching network generates a disparity map including the following contents: A method for constructing a cost volume that preserves the group correlation of an improved stereo matching network, dividing left features and right features into multiple groups along the channel dimension; Calculate the correlation map between each group and obtain multiple matching cost proposals; Pack the matching cost proposals into cost volumes; The depth-aware hierarchical module uses three cascaded MBConv blocks to process the early feature maps of the left image data, producing a feature map F that encodes relative depth and hierarchical information. D ; Feature F generated by the depth perception module D Adjust the channel weights of the cost volume to enhance the matching accuracy of texture levels and depth-related areas; that is, use the sigmoid function to derive the full-resolution depth map to calculate its depth-aware channel attention weight A C , using the depth-aware channel attention weight A C Adjust each channel of the cost volume; The depth perception channel attention weight A C With the feature map F D The obtained 4D difference volume is filtered to obtain a high-precision disparity map.

7. The method for acquiring three-dimensional road surface information based on vehicle front preview according to claim 5, characterized in that: The depth-aware channel attention weight A C The calculation formula is as follows: In the formula, F D It is the feature map of hierarchical information.

8. A vehicle body self-adjustment system using the method for acquiring three-dimensional road surface information based on vehicle front preview as described in any one of claims 1 to 7, characterized in that: include: A binocular vision module (1) is installed on the vehicle suspension device and is used to obtain left image data and right image data of the left and right roads of the vehicle in real time; A binocular correction unit (2) is used to correct the original image collected by the binocular vision module (1); The binocular matching unit (3) is based on a stereo matching network improved by a depth perception hierarchical module, and obtains a high-precision road disparity map by matching the corrected left and right images; A depth information unit (4) obtains a disparity map through a binocular stereo matching algorithm, obtains a road depth map through camera internal parameters, and extracts pixel information of the image through the depth map; A three-dimensional point cloud unit (5) converts pixel information extracted from the depth map into three-dimensional point cloud data; The intelligent control module (6) obtains the three-dimensional point cloud data to calculate the road slope information, calculates the displacement output corresponding to the suspension device in combination with the vehicle condition data, and controls and adjusts the displacement of the vehicle body.

9. The vehicle body self-adjusting system according to claim 8, characterized in that: The intelligent control module (6) comprises: A vehicle speed detection unit (61), used for detecting the current speed of the vehicle in real time; A weight detection unit (62), used for detecting the current mass of the vehicle in real time; The three-dimensional analysis unit (63) obtains the three-dimensional point cloud data in the three-dimensional point cloud unit (5), the current vehicle speed in the vehicle speed detection unit (61), the current mass in the weight detection unit (62) and the size of the suspension device, obtains the road surface roughness of the road surface area that the wheel corresponding to the suspension device will pass through in the future, and obtains the road surface slope data through the road surface fitting plane.

10. The vehicle body self-adjusting system according to claim 9, characterized in that: The intelligent control module (6) further comprises a data analysis unit (64) and a control analysis unit (65); The data analysis unit (64) acquires the three-dimensional point cloud data in the three-dimensional point cloud unit (5), the current vehicle speed, the vehicle weight, and the road surface roughness and slope data acquired by the three-dimensional analysis unit (63), performs analysis and processing, and calculates the required extension and contraction amount of the actuator (7); The control analysis unit (65) obtains the extension and contraction amount calculated by the data analysis unit (64), transmits a control command to the actuator (7), controls the action of the actuator (7), and adjusts the displacement of the vehicle body.

Images