Non-contact vehicle wheel load identification method and device based on elliptical constraint Hertz deformation quantization model

Abstract

The present invention discloses a non-contact vehicle wheel load identification method and device based on an elliptical constrained Hertzian deformation quantification model. The method comprises: collecting tire images of a moving vehicle and identifying and segmenting regions of interest; identifying the tire-ground contact line, rim outer contour, and tire sidewall text symbols within the region of interest using a machine vision algorithm; and obtaining key parameters of the tire's geometric structure and mechanical properties from the tire sidewall text symbols using a combined OCR-SQL recognition and retrieval method; and then utilizing the tire elliptical constrained Hertzian deformation quantification model to determine the actual tire-road contact area and calculate the vehicle's single-wheel load. The tire elliptical constrained Hertzian deformation quantification model proposed in this invention builds on traditional Hertzian contact theory by further considering the initial flatness of the tire cross-section and the influence of width constraints when the tire deforms under load, significantly improving the accuracy and reliability of load identification results.

Description

Technical Field

[0001] The present invention relates to the field of highway vehicle load identification, and in particular to a non-contact vehicle wheel load identification method and device based on an elliptical constraint Hertz deformation quantization model. Background Art

[0002] With the continuous increase in highway operating mileage and traffic volume, vehicle-in-motion weighing (WIM) technology is widely used in road and bridge engineering, facilitating traffic load identification and monitoring, abnormal load warnings, and the accumulation of research data. Among them, non-contact WIM technology based on machine vision offers numerous advantages, including low cost, easy installation, strong durability, and portability. In recent years, numerous scholars have conducted extensive research on non-contact WIM technology from the perspectives of principles, systems, and engineering applications.

[0003] Current non-contact WIM technologies generally use machine vision to identify tire deformation characteristics and manufacturing parameters, and then establish a relationship between tire deformation and external loads through a mechanical model, thereby achieving load identification. However, existing technologies generally use a relatively rough tire-road contact mechanical model, resulting in low accuracy and high uncertainty in deformation quantification and load identification models. For example, the classic tire pressure balance model directly treats tire inflation pressure as contact pressure, but in reality, the two are inconsistent and ignore the uneven pressure distribution in the tire-road contact area. The radial compression model establishes the tire deformation stiffness as a key parameter to directly establish the relationship between load and deformation obtained by visual identification. However, obtaining the tire deformation stiffness requires extensive testing and fitting empirical laws, which has poor applicability. The classic Hertz contact theory provides an analytical solution for the pressure and deformation distribution when an elastic sphere / torus contacts a rigid ground. However, in actual engineering, tires in normal use cannot be simplified to torus surfaces, and due to their special structure, tires cannot deform completely according to the laws given by the theory. The above tire mechanical contact models all have the problem of over-simplification of actual conditions or reliance on experience, which leads to poor reliability of the corresponding non-contact WIM technology. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the present invention proposes a non-contact vehicle wheel load identification method and device based on an elliptical constrained Hertz deformation quantization model. The specific technical solution is as follows:

[0005] A non-contact vehicle wheel load identification method based on an elliptical constrained Hertz deformation quantization model comprises the following steps:

[0006] Step 1: Obtain a video of a moving vehicle's tires and input it into a pre-trained convolutional neural network for real-time image recognition to generate initial wheel and rim detection frames. When the initial wheel and rim detection frames enter the image sensor's frame range, perform image capture. Utilize a tire multi-region-of-interest automatic generation method to extract three regions of interest (ROIs) from the initial wheel and rim detection frames: the tire-ground contact area, the rim area, and the tire sidewall text marking area.

[0007] Step 2: Using a machine vision algorithm, the tire-ground contact line segment and the rim outer contour line are respectively identified from the tire-ground contact area and the rim area in Step 1; the tire sidewall text symbols are identified from the tire sidewall text mark area using an OCR-SQL recognition and retrieval method, and various tire parameters including the rim calibration diameter are obtained from the tire sidewall text symbols;

[0008] Step 3: Using the tire spatial pose estimation and contact patch quantification method, the internal parameters of the image sensor that captured the tire video and the rim outer contour line and rim calibration diameter obtained in step 2 are used to calculate the spatial coordinates of the rim center in the optical center coordinate system of the image sensor, and obtain the angle between the rim plane and the imaging plane. Subsequently, the tire-ground contact line segment obtained in step 2 is mapped from the pixel coordinate system to the optical center coordinate system of the image sensor to obtain the actual length L of the tire-ground contact line segment.

[0009] Step 4: Based on the Hertz contact theory, considering the initial flatness characteristics of the tire cross-section and the width constraint of the tire when it deforms under load, a tire elliptical constraint Hertz deformation quantification model is constructed. Based on the tire elliptical constraint Hertz deformation quantification model, the single wheel load of the non-contact vehicle is estimated.

[0010] Furthermore, in step 1, the tire multiple interest region automatic generation method is used to extract three interest regions, namely, the tire-ground contact area, the rim area, and the tire sidewall text marking area, from the wheel initial detection frame and the rim initial detection frame, specifically including:

[0011] (1) Use a pre-trained convolutional neural network to detect all wheels and rims in the tire video image in real time, mark each detected object with an initial detection box, and give the corresponding category probability;

[0012] (2) Delete images that have any of the following features: ① The wheel initial detection frame cannot be formed, that is, the wheel is not completely in the picture or the picture is blurred; ② The probability of being identified as a wheel or rim is lower than a pre-set threshold; ③ The relative position of the wheel initial detection frame and the rim initial detection frame is incorrect;

[0013] (3) The wheel initial detection frame and rim initial detection frame of the filtered image are linearly geometrically transformed to obtain three regions of interest containing appropriate redundant areas, namely, tire-road contact ROI, rim ROI, and tire sidewall text ROI.

[0014] Furthermore, in step 2, a machine vision algorithm is used to identify the tire-ground contact line segment and the rim outer contour line from the three regions of interest in step 1, specifically including:

[0015] (1) Grayscale value transformation and adaptive Wiener filtering are used to enhance ROI images and remove redundant color information;

[0016] (2) Perform global adaptive threshold segmentation on the obtained ROI enhanced grayscale image to obtain several foreground and background regions;

[0017] (3) Multiple morphological transformations are applied to the binary image to eliminate the interference of small areas and concentrate the target elements in the foreground. The foreground area is further screened by specific geometric features, and the pixel coordinates in the foreground area are extracted. The least squares method is used for curve fitting to obtain the tire-ground contact line segment and the rim outer contour line.

[0018] Furthermore, in step 2, the tire sidewall text symbols are identified from the tire sidewall text mark area using an OCR-SQL recognition and retrieval combined method, and various parameters of the tire, including the rim calibration diameter, are obtained from the tire sidewall text symbols, specifically including:

[0019] (1) performing polar coordinate transformation and image enhancement on the text marking area on the tire sidewall, converting the annular area into a rectangular strip grayscale image with a high sharpness value to improve the accuracy of text character recognition; the origin and range of the polar coordinate transformation are determined based on the outer contour line of the rim;

[0020] (2) performing optical character recognition on the rectangular strip grayscale image to identify a tire specification mark character string and tire brand information having a fixed and unified format;

[0021] (3) According to the tire specification mark character string, search in the tire specification database to obtain various parameters of the tire; the tire specification database is established according to the current national standard truck tire specification table, passenger car tire specification table and rim calibration size table; the various parameters of the tire include the new tire design outer diameter D t 、New tire design section width w s and standard inflation pressure p st , Rim calibration diameter D n , tire aspect ratio γ s ;

[0022] (4) According to the tire specifications obtained in step (3), the corresponding tread ratio ζ is retrieved from the tread database w , and thus calculate the tread ratio ζ w Compared with the new tire design section width w s The product of the tire tread width w is obtained. t The tread database includes tire brand, tire structure type, applicable special model, nominal section width and tread ratio ζ w These five parameters.

[0023] Furthermore, the step three includes the following sub-steps:

[0024] S301: Define the pixel coordinate system as the origin O IM Located in the upper left corner of the image, and x IM The positive direction of the axis is horizontal to the right, y IM The positive direction is a horizontally downward plane rectangular coordinate system; the optical center coordinate system of the image sensor is the origin O p Located at the optical center of the camera, and x p The positive direction of the axis is parallel to the road surface and points to the right. p The positive direction of the axis is perpendicular to the road surface and downward, p A spatial rectangular coordinate system with its axis parallel to the road surface and pointing directly in front of the image sensor;

[0025] S302: Calculating the semi-major axis length a of the ellipse of the rim outer contour line IM , semi-minor axis length b IM And the coordinates of the ellipse center (p IM ,q IM );

[0026] S303: According to the principles of photographic geometry, the rim center coordinates (x rc ,y rc ,z rc ) and the angle θ between the rim plane and the image plane;

[0027]

[0028]

[0029] Among them, κ x and κ y is the horizontal and vertical pixel size of the image sensor, f0 is the principal distance of the image sensor lens, c x with c y is the plane coordinate of the intersection of the optical axis and the image plane in the image coordinate system;

[0030] S304: Map the image coordinates of the two endpoints of the tire-ground contact line segment to the optical center coordinate system of the image sensor using the following formula, and calculate the tire-road contact line segment length L:

[0031]

[0032] Among them, (x IM ,y IM ) is the coordinate before mapping, (x, y, z) is the coordinate point after mapping;

[0033] The tire-road contact line segment length L is the Euclidean distance between the two end points of the tire-ground contact line segment in the optical center coordinate system after mapping.

[0034] Furthermore, the step 4 includes the following sub-steps:

[0035] S401: The tire contact coefficient λ is derived from Hertz contact theory:

[0036]

[0037] Among them, D t Design outer diameter for new tire; w s Design section width for new tire; γ s is the tire aspect ratio;

[0038] S402: Calculate the minor semi-axis b of the ellipse in the tire-road contact area according to the following formula h Virtual solution:

[0039]

[0040] in, is the width of the tire-ground contact area within the normal use range, and is the tire tread width w t Half of ; L is the length of the tire-road contact line segment;

[0041] S403: Calculate the single wheel load according to the following formula:

[0042] when hour,

[0043] when hour,

[0044] in,

[0045] E * =πξ t ξ re p re

[0046]

[0047] Among them, E * is the equivalent contact elastic modulus of tire-road, ξ t is the vehicle tire type correction factor, which is determined by the special vehicle type according to the national standard; re is the tire inflation correction factor, ξ st is the empirical correction value when the tire pressure is equal to the standard pressure, p re is the actual tire pressure of the wheel on the same side as the video capture device, p st This is the standard inflation pressure.

[0048] Furthermore, when the tire aspect ratio γ s If data is missing, calculate the new tire design section height h using the following formula s , and then approximate the aspect ratio γ s :

[0049] h s =(D t -D n ) / 2

[0050] γ s =h s / w s ;

[0051] Among them, D n Mark the diameter of the rim.

[0052] A non-contact vehicle wheel load identification device for implementing a non-contact vehicle wheel load identification method, comprising a video acquisition device installed on one side of a lane and a computing device deploying a pre-trained convolutional neural network;

[0053] The video acquisition device is used to collect tire videos of a moving vehicle in real time and send the tire videos of the moving vehicle to the computing device via wireless transmission;

[0054] The computing device is used to receive the tire video of the moving vehicle and the actual tire pressure p of the wheel located on the same side as the video acquisition device, which is sent by the tire pressure monitoring system of the moving vehicle through the vehicle-road cooperative sensing device. re The computing device obtains three regions of interest (ROIs) including the tire-ground contact area, the rim area, and the tire sidewall text marking area through a built-in automatic generation method for multiple regions of interest of the tire. Secondly, based on a machine vision algorithm, the tire-ground contact line segment, the rim outer contour line, and the tire sidewall text symbol are identified from the three regions of interest. Thirdly, through the tire spatial posture estimation and contact print quantification method, the angle between the rim plane and the imaging plane and the actual length L of the tire-ground contact line segment are obtained. Finally, based on the elliptical constrained Hertz deformation quantification model, the load borne by a single tire is solved.

[0055] Furthermore, the video acquisition device is installed on a straight road section, and the imaging plane of the camera is parallel to the lane and perpendicular to the road surface; the distance between the camera and the ground should be sufficient for the camera to fully capture tire images of various types of traffic.

[0056] The beneficial effects of the present invention are as follows:

[0057] (1) The device and method of the present invention can obtain tire deformation characteristics through machine vision, obtain tire TPMS inflation pressure through a vehicle-road cooperative network, obtain tire-related geometry and attribute parameters through a combination of optical character recognition and database retrieval, and finally calculate the single-wheel load using the tire elliptical constraint Hertz deformation quantification model. The device is simple, easy to install and move, and has low construction and maintenance costs, and can achieve efficient and fully automated non-contact vehicle tire load identification.

[0058] (2) The tire elliptical constrained Hertz deformation quantification model proposed in the present invention takes into account the influence of the flatness of the tire cross-sectional shape and the deformation characteristics of the tire thickening area on the basis of the classical Hertz contact model, and obtains a more realistic tire-ground contact area shape and pressure distribution pattern, thereby improving the accuracy of load estimation and the reliability and stability of measurement results. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 This is a schematic diagram of a non-contact vehicle tire elliptical constraint Hertzian deformation quantification and load identification device and method provided by the present invention.

[0060] Figure 2 Figure 1 is a schematic diagram of the proposed elliptical constrained Hertzian deformation quantification model. (a) illustrates the process of developing the elliptical constrained Hertzian deformation quantification model for tires. (b) is a schematic diagram of the virtual state, where the pressure distribution is a mesh surface. (c) is a schematic diagram of the virtual-real state equivalence relationship, where the real-state pressure distribution is the intersection of the two planes and the outer mesh surface.

[0061] Figure 3 Comparison of load identification effects of non-contact WIM systems based on different tire-ground contact models, where (a) is the test result of light-duty tires and (b) is the test result of heavy-duty tires. DETAILED DESCRIPTION

[0062] The present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments, and the purpose and effects of the present invention will become more apparent. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0063] like Figure 1As shown, the contactless vehicle wheel-load identification device of the present invention includes a video acquisition device installed on one side of the lane, a vehicle-road cooperative perception device, and a computing device that deploys a pre-trained convolutional neural network.

[0064] The video capture device is installed on a straight road, with the camera's imaging plane parallel to the lane and perpendicular to the road surface. The distance between the camera and the ground should be sufficient to fully capture tire images of all types of traffic. The video capture device is used to capture real-time tire video from moving vehicles and transmit this video wirelessly to the computing device.

[0065] The computing device is used to receive the tire video of the moving vehicle and the actual tire pressure p of the wheel on the same side as the video acquisition device sent by the tire pressure monitoring system of the moving vehicle through the vehicle-road cooperative sensing device. re The computing device obtains three regions of interest (ROIs) including the tire-ground contact area, the rim area, and the tire sidewall text marking area through the built-in tire multi-ROI automatic generation method. Secondly, based on the machine vision algorithm, the tire-ground contact line segment, the rim outer contour line, and the tire sidewall text symbol are identified from the three regions of interest. Thirdly, through the tire spatial pose estimation and contact print quantification method, the angle between the rim plane and the imaging plane and the actual length L of the tire-ground contact line segment are obtained. Finally, based on the elliptical constrained Hertz deformation quantification model, the load borne by a single tire is solved.

[0066] like Figure 1 As shown, the non-contact vehicle wheel load identification method based on the elliptical constraint Hertz deformation quantization model of the present invention includes the following steps:

[0067] Step 1: Obtain the vehicle's tire video and input it into a pre-trained convolutional neural network (such as a pre-trained YoloV8 model) for real-time image recognition to generate the wheel initial detection frame and the rim initial detection frame; when the wheel initial detection frame and the rim initial detection frame enter the image sensor frame range (preferably within the central 30% range of the camera frame), perform image capture; use the tire multi-interest region automatic generation method to extract three regions of interest (ROIs) from the wheel initial detection frame and the rim initial detection frame: the tire-ground contact area, the rim area, and the tire sidewall text marking area; at the same time, start the communication mechanism between the vehicle-road cooperative network and the vehicle tire pressure monitoring system (TPMS) to obtain the actual tire pressure p of the wheel on the same side as the video acquisition device. re .

[0068] Among them, the automatic generation method of tire multiple interest regions is used to extract three types of interest regions from the initial wheel detection frame and the initial rim detection frame: the tire-ground contact area, the rim area, and the tire sidewall text marking area. Specifically, they include:

[0069] (1) Use a pre-trained convolutional neural network to detect all wheels and rims in the tire video image in real time, mark each detected object with an initial detection box, and give the corresponding category probability;

[0070] (2) Delete images that have any of the following features: ① The initial wheel detection frame cannot be formed, that is, the wheel is not completely in the picture or the picture is blurred; ② The probability of being identified as a wheel or rim is lower than the pre-set threshold; ③ The relative position of the initial wheel detection frame and the initial rim detection frame is incorrect, for example, there is an intersection of the frame lines, etc.

[0071] (3) The wheel initial detection frame and rim initial detection frame of the filtered image are linearly geometrically transformed to obtain three regions of interest containing appropriate redundant areas, namely, tire-road contact ROI, rim ROI, and tire sidewall text ROI.

[0072] Step 2: Using a machine vision algorithm, identify the tire-ground contact line segment and the rim outer contour line from the tire-ground contact area and rim area in step 1 respectively; identify the tire sidewall text symbol from the tire sidewall text mark area through the OCR-SQL recognition and retrieval joint method, and obtain various tire parameters from the tire sidewall text symbol, including standard inflation pressure, cross-sectional geometry parameters, rim parameters, tread parameters and other information.

[0073] Using a machine vision algorithm, the tire-ground contact line segment and the rim outer contour line are identified from the tire-ground contact area and the rim area in step 1, respectively. A specific implementation method is as follows:

[0074] (1) Grayscale value transformation and adaptive Wiener filtering are used to enhance ROI images and remove redundant color information;

[0075] (2) Perform global adaptive threshold segmentation on the obtained ROI enhanced grayscale image to obtain several foreground and background regions;

[0076] (3) A variety of morphological transformations are applied to the binary image, including dilation, erosion, opening, closing, etc., to eliminate the interference of small areas and concentrate the target elements in the foreground. The foreground area is further screened by specific geometric features, and the pixel coordinates in the foreground area are extracted. The least squares method is used for curve fitting to obtain the tire-ground contact line segment and the rim outer contour line.

[0077] In step 2, the tire sidewall text symbols are identified from the tire sidewall text mark area using an OCR-SQL recognition and retrieval combined method, and various tire parameters are obtained from the tire sidewall text symbols, which specifically includes the following steps:

[0078] (1) Perform polar coordinate transformation and image enhancement on the text marking area on the tire sidewall, converting the circular area into a rectangular strip grayscale image with a high sharpness value to improve the accuracy of text character recognition. The origin and range of the polar coordinate transformation are determined based on the outer contour line of the rim;

[0079] (2) Optical character recognition (OCR) is performed on the rectangular strip grayscale image to identify the tire specification mark string and tire brand information in a fixed and unified format. Tire specification marks generally consist of specific capital letters, numbers, and slashes ( / ), and have a high degree of recognition and a high probability of successful recognition. Brand information generally consists of large-scale artistic fonts and trademark graphics, and is also easy to identify for common and regular tire brands.

[0080] (3) According to the tire specification mark character string, a search is performed in the tire specification database to obtain various parameters of the tire. The tire specification database is established based on the truck tire specification table, passenger car tire specification table and rim calibration size table of the current national tire standards. A specific implementation method is: the truck tire specification table is established based on the "Heavy Truck Tire Specifications, Dimensions, Air Pressure and Load (GB / T 2977-2016)", the passenger car tire specification table is established based on the "Passenger Car Tire Specifications, Dimensions, Air Pressure and Load (GB / T2978-2014)", and the rim calibration size table is established based on the "Truck and Bus Wheel Rim Specification Series (GB / T 31961-2015)" and the "Passenger Car Wheel Rim Specification Series (GB / T 3487-2015)".

[0081] According to the tire specification mark string, the tire specification database is searched. The specific search method is as follows: the tire structure type is obtained by analyzing the first occurrence of "R", "D" or "-" parameters in the string; the number before the first occurrence of "R", "D" or "-" is the nominal section width and nominal aspect ratio γ s ; Treat multiple capital letters (such as "PR", "ST", "MPT", etc.) that appear consecutively in the string as special vehicle model codes to obtain the special vehicle type; Based on the above tire structure type, nominal section width, and special vehicle type, query the new tire design outer diameter D from the truck / passenger tire specification table. t 、New tire design section width w s and standard inflation pressure p st The number after the first appearance of "R", "D" or "-" is the nominal diameter code of the rim. The rim calibration diameter D is retrieved from the rim calibration size table based on this parameter, tire structure type, and special vehicle type. n .

[0082] (4) According to the tire specifications obtained in step (3), the corresponding tread ratio ζ is retrieved from the tread databasew , and thus calculate the tread ratio ζ w Compared with the new tire design section width w s The product of the tire tread width w is obtained. t The tread database is established through research. The brand-tread database includes common brands, tire structure types, applicable special models, nominal section width and tread ratio. w Since there is a possibility that the tire brand cannot be effectively identified and the specific processing and manufacturing parameters of all tire brands cannot be obtained, there is a brand-tread database to obtain the tread ratio. w The research shows that the design and manufacturing of vehicle tires generally use 0.6-0.9 as the ζ w If the brand-tread database is missing information, then take ζ w =0.75 as an approximate estimate, then we have

[0083] Step 3: Using the tire spatial pose estimation and contact patch quantification method, the internal parameters of the image sensor used to capture the tire video and the rim outer contour and rim calibration diameter obtained in Step 2 are used to calculate the spatial coordinates of the rim center in the optical center coordinate system of the image sensor, and the angle between the rim plane and the imaging plane is obtained. Subsequently, the tire-ground contact line segment obtained in Step 2 is mapped from the pixel coordinate system to the optical center coordinate system of the image sensor to obtain the actual length L of the tire-ground contact line segment. Step 3 specifically includes the following sub-steps:

[0084] S301: Define the pixel coordinate system as the origin O IM Located in the upper left corner of the image, and x IM The positive direction of the axis is horizontal to the right, y IM The positive direction is the horizontal downward plane rectangular coordinate system; the optical center coordinate system of the image sensor is the origin O p Located at the optical center of the camera, and x p The positive direction of the axis is parallel to the road surface and points to the right. p The positive direction of the axis is perpendicular to the road surface and downward, p A spatial rectangular coordinate system with its axis parallel to the road surface and pointing directly in front of the image sensor;

[0085] S302: Analyze the outer contour of the elliptical rim. The major axis of the ellipse is y IM Axis parallel, minor axis is x IM Axis parallel, calculate the major semi-axis length a of the ellipse of the rim outer contour line IM , semi-minor axis length b IM And the coordinates of the ellipse center (p IM ,q IM );

[0086] S303: According to the principles of photographic geometry, the rim center coordinates (x rc ,y rc ,z rc ) and the angle θ between the rim plane and the image plane;

[0087]

[0088] Among them, κ x and κ y is the horizontal and vertical pixel size of the image sensor, f0 is the principal distance of the image sensor lens, c x with c y is the plane coordinate of the intersection of the optical axis and the image plane in the image coordinate system;

[0089] S304: The above steps obtain the spatial position of the rim plane in the camera optical center coordinate system. The image coordinates of the two endpoints of the tire-ground contact line segment are mapped to the optical center coordinate system of the image sensor using the following formula, and the length L of the tire-road contact line segment is calculated:

[0090]

[0091] Among them, (x IM ,y IM ) is the coordinate before mapping, (x, y, z) is the coordinate point after mapping;

[0092] The tire-road contact line segment length L is the Euclidean distance between the two end points of the tire-ground contact line segment in the optical center coordinate system after mapping.

[0093] Step 4: Based on the Hertz contact theory, considering the initial flat characteristics of the tire cross-section and the width constraint of the tire when it deforms under load, a tire elliptical constraint Hertz deformation quantification model is constructed. Based on the tire elliptical constraint Hertz deformation quantification model, the single wheel load of the non-contact vehicle is estimated.

[0094] According to the classical Hertz contact theory, an elastic sphere with an initial radius of R0 is placed on a horizontal rigid ground. When an external load F acts on it, the sphere-ground contact area is a radius of r c The pressure distribution is as follows:

[0095]

[0096] (x c ,y c ) is the coordinate of a point in the contact area, x c Points to the direction of the ball's roll, y cThe axis is perpendicular to the rolling direction. The maximum central pressure p0 is obtained through the geometric equations and physical equations in elastic mechanics, and the pressure distribution in the contact area is integrated to obtain the relationship between the external load and the contact deformation:

[0097]

[0098] In reality, a tire cannot be considered a sphere. Its surface is close to a torus with an elliptical cross section. The outer radius of the longitudinal section is R1, and the transverse section is a semi-axis with a width direction of w. s 、The height direction semi-axis is h s Therefore, the classical Hertz contact theory needs to be corrected for the cross-section ellipse eccentricity. s >h s , and define the aspect ratio γ s =h s / w s To reflect its flatness. At this time, the tire-ground contact area is also an ellipse, with its major semi-axis a h , the minor semi-axis is b h According to the Hertz contact theory, the grounding coefficient λ is defined as:

[0099]

[0100] The cross section has the largest curvature radius within the contact area between the tire and the road, so the largest curvature radius of the ellipse is taken to represent the cross-sectional characteristics:

[0101]

[0102] Hertz contact theory defines the grounding coefficient λ:

[0103]

[0104] The equivalent Gaussian curvature radius of the above elliptical cross-section ring is:

[0105]

[0106] Based on this result, the spherical contact equation of the classical Hertz contact theory is modified to obtain:

[0107]

[0108] In addition, the actual cross-section of the tire is not a regular oval, but the outer tread that contacts the ground is thickened to prevent risks such as tire blowouts and punctures by foreign objects. The thickening of the outer tread will increase local stiffness. As the load increases, the width of the tire-ground contact area will be constrained by the thickened tread. Assuming that the thickened tread area is fully grounded, the width of the ground contact area will no longer increase regardless of how the external load changes. Here, the "upper limit of the tire-ground contact width after the load increases is" is defined as follows: The reasons for this limitation are explained.

[0109] Therefore, the mechanical model also needs to be modified based on the width constraint assumption. When , the subsequent load increment will be balanced by the increase in the longitudinal length of the contact area and the increase in the pressure distribution. If there is no contact width constraint, the subsequent load increment will be balanced by the increase in the longitudinal length and width of the contact area and the increase in the pressure distribution. However, the change pattern of the three can be directly solved by the Hertz contact theory with the correction of the cross-sectional ellipse eccentricity.

[0110] Based on this, we can first set an unconstrained virtual state, where the tire-road contact area is an ellipse and the outside of the pressure distribution graph is a semi-ellipsoidal surface Ω * , and at this time Through the Hertz contact theory corrected by the eccentricity of the cross-section ellipse, it is easy to find the elliptical contact area. Additional load F borne by the scope add :

[0111]

[0112] The real state includes the tread width constraint. It can be considered that the pressure distribution graph in the real state is composed of a semi-ellipsoidal surface Ω and a pair of constraint planes. The intercepted graph, ellipsoid Ω and Ω * The projections on the road plane are similar. The lengths of the major and minor semi-axes of the former on the road plane are both α times that of the latter. Obviously, α>1. The real state pressure graph lacks the above-mentioned F add part, but increased by Ω, Ω * 、 The part enclosed by four planes / curved surfaces. Integrating this part gives the incremental load F. * :

[0113]

[0114] Since the actual loads borne by the virtual state and the real state are exactly the same, F * =F add , combining the above two equations, we can solve the geometric scaling factor α:

[0115]

[0116] By the definition of the geometric scaling factor α, 2αa h This is the length of the major axis of the tire-road contact area under real conditions. This length should be equal to the actual length L obtained by the tire-ground contact line segment, so it can be solved a h with b h , as the above Ω * The semi-major axis and semi-minor axis of the ellipsoidal surface projected on the road surface:

[0117]

[0118] Here exists The possibility of this indicates that the external load is too small, so the thickened tread area is not fully grounded and does not produce a restraining effect. In this case, according to the classical Hertz contact theory, the wheel load can be directly calculated:

[0119]

[0120] when When the tire has deformed due to the tread width constraint, the wheel load should be solved according to the Hertz contact theory modified by the above method (i.e., the Hertz deformation quantification model based on the tire elliptical constraint as described in the present invention):

[0121]

[0122] Therefore, step 4 specifically includes the following sub-steps:

[0123] S401: The tire contact coefficient λ is derived from Hertz contact theory:

[0124]

[0125] Among them, D t Design outer diameter for new tire; w s Design section width for new tire; γ s The aspect ratio of the tire is generally obtained directly from the recognized tire specification tag string. This parameter reflects the initial flat characteristics of the tire section, that is, the degree of eccentricity of the tire's elliptical section when unloaded. If the aspect ratio data is missing, the design section height of the new tire can be calculated using the following method, and then the aspect ratio can be approximately calculated:

[0126] h s =(D t -D c ) / 2,γ s =h s / w s

[0127] Among them, D n Mark the diameter of the rim.

[0128] S402: Calculate the minor semi-axis b of the ellipse in the tire-road contact area according to the following formula: h Virtual solution:

[0129]

[0130]

[0131] in, is the width of the tire-ground contact area within the normal use range, and is the tire tread width w t half,

[0132] L is the length of the tire-road contact segment;

[0133] S403: Calculate the single wheel load according to the following formula:

[0134] when hour,

[0135] when hour,

[0136] in,

[0137] E * =πξ t ξ re p re

[0138]

[0139] Among them, E * is the equivalent contact elastic modulus of tire-road, ξ t ξ is the vehicle tire type correction factor, which is determined by the special vehicle type according to the national standard. The specific value can be found in Appendix 1. re is the tire inflation correction factor, ξ st It is the empirical correction value when the tire pressure is equal to the standard pressure. It is 1.50 for heavy trucks and buses, and 1.85 for conventional cars and light trucks. re is the actual tire pressure of the wheel on the same side as the video capture device, p st This is the standard inflation pressure.

[0140] Table 1 Tire type correction factor (ξ t )surface

[0141]

[0142] Example:

[0143] This example is only used to more clearly illustrate the technical solution of the present invention and to demonstrate the actual test results of the technical solution, and it cannot be used to limit the scope of protection of the present invention.

[0144] In this example, 6.5-16 light truck tires and 11.00-20 heavy truck tires were selected as test objects. A loading fixture was designed to mount the tires on a hydraulic press in normal operating mode. The loading device sequentially transferred the load from the axle and then the rim to the tire. During the loading process, the bottom of the tire would contact and deform with a horizontal support. To closely simulate the actual operating environment of the WIM system, the support was constructed of a 100mm thick asphalt concrete slab. A load-sensing material was placed on the surface of the horizontal support to determine the tire-road contact area. A simple and low-cost implementation approach involves sequentially placing a layer of white paper, carbon paper, and then white paper. The contact area is determined by the blue mark left by the carbon paper on the white paper. During the test, both tires were inflated to the standard tire pressure. A series of test conditions were designed using the actively applied external load as a variable. See Appendix 2 for details. The standard tire pressure for light truck tires is 0.39 MPa, and the standard tire pressure for heavy truck tires is 0.60 MPa.

[0145] Table 2 Tire loading test conditions of the embodiment

[0146]

[0147] The objectives of the tests conducted in this embodiment are summarized as follows: to compare the wheel load calculation results of the proposed elliptical constrained Hertz deformation quantification model and the cross-section width and height corrected Hertz contact model with those of the traditional tire pressure balance model, radial compression model, and classical torus Hertz contact theory, and to evaluate the accuracy of each model. During these tests, the tire pressure was set to the standard pressure, a known quantity; the tire-road contact area and contact length were directly obtained from load-bearing area sensing materials or through the machine vision method described in this invention; and the tire geometric parameters and cross-sectional characteristics were directly obtained by obtaining the corresponding tire brand and consulting a specification database, or by direct on-site measurement at a specified inflation pressure. Therefore, the data obtained meet the requirements for calculating vehicle wheel loads using various tire-road contact mechanics models. The model calculation results can then be compared with the actual applied wheel load to verify the accuracy of each model.

[0148] Figure 3Figures (a) and (b) show the prediction results of the five models for the light truck tires and heavy truck tires under different external loads. For both tires, the classic torus Hertz contact model has the most significant error, at 17.8% and 34.2%, respectively. The tire pressure balance model's estimation error is slightly smaller than that of the classic torus Hertz contact model (average errors of -8.6% and -19.9% ​​for the two tires, respectively), but overall, the estimated value tends to be lower with increasing load. The radial compression model and the elliptical constraint Hertz deformation quantification model provide relatively accurate prediction results. The elliptical constraint Hertz deformation quantification model proposed in this invention has an average calculation error of -5.9% for light truck tires (with an error range of -9.0% to -0.9%) and an average calculation error of -1.5% for heavy truck tires (with an error range of -6.0% to 1.8%), making it the best performing model among the five models. The American Society for Testing and Materials (ASTM) E1318-09 2017 specification stipulates an accuracy of ±25% for wheel load identification in a Class I WIM system. Based on the test results above, the WIM system wheel load estimation based on the elliptical constrained Hertz deformation quantification model can meet the ASTM standard requirements.

[0149] Those skilled in the art will understand that the foregoing descriptions are merely preferred embodiments of the invention and are not intended to limit the invention. Although the invention has been described in detail with reference to the foregoing examples, those skilled in the art will still be able to modify the technical solutions described in the foregoing examples or substitute equivalents for some of the technical features therein. Any modifications, equivalent substitutions, etc. made within the spirit and principles of the invention shall be included within the scope of protection of the invention.

Claims

1. A non-contact vehicle wheel load identification method based on an elliptical constrained Hertzian deformation quantization model, characterized in that: The method comprises the following steps: Step 1: Obtain a tire video of a moving vehicle and input it into a pre-trained convolutional neural network for real-time image recognition to generate initial wheel detection frames and rim detection frames. When the initial wheel detection frames and rim detection frames enter the image sensor frame range, image capture is performed. Extracting three regions of interest (ROIs) from the wheel initial detection frame and the rim initial detection frame using a tire multi-ROI automatic generation method: a tire-ground contact area, a rim area, and a tire sidewall text marking area; Step 2: Using a machine vision algorithm, the tire-ground contact line segment and the rim outer contour line are respectively identified from the tire-ground contact area and the rim area in Step 1; the tire sidewall text symbols are identified from the tire sidewall text mark area using an OCR-SQL recognition and retrieval method, and various tire parameters including the rim calibration diameter are obtained from the tire sidewall text symbols; Step 3: Using the tire spatial pose estimation and contact patch quantification method, the internal parameters of the image sensor that captured the tire video and the rim outer contour line and rim calibration diameter obtained in step 2 are used to calculate the spatial coordinates of the rim center in the optical center coordinate system of the image sensor, and obtain the angle between the rim plane and the imaging plane. Subsequently, the tire-ground contact line segment obtained in step 2 is mapped from the pixel coordinate system to the optical center coordinate system of the image sensor to obtain the actual length L of the tire-ground contact line segment. Step 4: Based on the Hertz contact theory, considering the initial flatness characteristics of the tire cross-section and the width constraint of the tire when it deforms under load, a tire elliptical constraint Hertz deformation quantification model is constructed. Based on the tire elliptical constraint Hertz deformation quantification model, the single wheel load of the non-contact vehicle is estimated.

2. The non-contact vehicle wheel load identification method based on the elliptical constraint Hertz deformation quantization model according to claim 1 is characterized in that: In the step 1, the tire multi-region-of-interest automatic generation method is used to extract three regions of interest (ROIs) from the wheel initial detection frame and the rim initial detection frame, namely, the tire-ground contact area, the rim area, and the tire sidewall text marking area. Specifically, the method includes: (1) Use a pre-trained convolutional neural network to detect all wheels and rims in the tire video image in real time, mark each detected object with an initial detection box, and give the corresponding category probability; (2) Delete images that have any of the following features: ① The wheel initial detection frame cannot be formed, that is, the wheel is not completely in the picture or the picture is blurred; ② The probability of being identified as a wheel or rim is lower than a pre-set threshold; ③ The relative position of the wheel initial detection frame and the rim initial detection frame is incorrect; (3) The wheel initial detection frame and rim initial detection frame of the filtered image are linearly geometrically transformed to obtain three regions of interest containing appropriate redundant areas, namely, tire-road contact ROI, rim ROI, and tire sidewall text ROI.

3. The non-contact vehicle wheel load identification method based on the elliptical constraint Hertz deformation quantization model according to claim 1 is characterized in that: In the second step, a machine vision algorithm is used to identify the tire-ground contact line segment and the rim outer contour line from the three regions of interest in the first step, specifically including: (1) Grayscale value transformation and adaptive Wiener filtering are used to enhance ROI images and remove redundant color information; (2) Perform global adaptive threshold segmentation on the obtained ROI enhanced grayscale image to obtain several foreground and background regions; (3) Multiple morphological transformations are applied to the binary image to eliminate the interference of small areas and concentrate the target elements in the foreground. The foreground area is further screened by specific geometric features, and the pixel coordinates in the foreground area are extracted. The least squares method is used for curve fitting to obtain the tire-ground contact line segment and the rim outer contour line.

4. The non-contact vehicle wheel load identification method based on the elliptical constraint Hertz deformation quantization model according to claim 1 is characterized in that: In the second step, the tire sidewall text symbols are identified from the tire sidewall text mark area by using the OCR-SQL recognition and retrieval combined method, and various parameters of the tire, including the rim calibration diameter, are obtained from the tire sidewall text symbols, specifically including: (1) performing polar coordinate transformation and image enhancement on the text marking area on the tire sidewall, converting the annular area into a rectangular strip grayscale image with a high sharpness value to improve the accuracy of text character recognition; the origin and range of the polar coordinate transformation are determined based on the outer contour line of the rim; (2) performing optical character recognition on the rectangular strip grayscale image to identify a tire specification mark character string and tire brand information having a fixed and unified format; (3) According to the tire specification mark character string, search in the tire specification database to obtain various parameters of the tire; the tire specification database is established according to the current national standard truck tire specification table, passenger car tire specification table and rim calibration size table; the various parameters of the tire include the new tire design outer diameter D t 、New tire design section width w s and standard inflation pressure p st , Rim calibration diameter D n , tire aspect ratio γ s ; (4) According to the tire specifications obtained in step (3), the corresponding tread ratio ζ is retrieved from the tread database w , and thus calculate the tread ratio ζ w Compared with the new tire design section width w s The product of the tire tread width w is obtained. t The tread database includes tire brand, tire structure type, applicable special model, nominal section width and tread ratio ζ w These five parameters.

5. The non-contact vehicle wheel load identification method based on the elliptical constraint Hertz deformation quantization model according to claim 1 is characterized in that: The step three includes the following sub-steps: S301: Define the pixel coordinate system as the origin O IM Located in the upper left corner of the image, and x IM The positive direction of the axis is horizontal to the right, y IM The positive direction is a horizontally downward plane rectangular coordinate system; the optical center coordinate system of the image sensor is the origin O p Located at the optical center of the camera, and x p The positive direction of the axis is parallel to the road surface and points to the right. p The positive direction of the axis is perpendicular to the road surface and downward, p A spatial rectangular coordinate system with its axis parallel to the road surface and pointing directly in front of the image sensor; S302: Calculating the semi-major axis length a of the ellipse of the rim outer contour line IM , semi-minor axis length b IM And the coordinates of the ellipse center (p IM ,q IM ); S303: According to the principles of photographic geometry, the rim center coordinates (x rc ,y rc ,z rc ) and the angle θ between the rim plane and the image plane; Among them, κ x and κ y is the horizontal and vertical pixel size of the image sensor, f0 is the principal distance of the image sensor lens, c x with c y is the plane coordinate of the intersection of the optical axis and the image plane in the image coordinate system; S304: Map the image coordinates of the two endpoints of the tire-ground contact line segment to the optical center coordinate system of the image sensor using the following formula, and calculate the tire-road contact line segment length L: Among them, (x IM ,y IM ) is the coordinate before mapping, (x, y, z) is the coordinate point after mapping; The tire-road contact line segment length L is the Euclidean distance between the two end points of the tire-ground contact line segment in the optical center coordinate system after mapping.

6. The non-contact vehicle wheel load identification method based on the elliptical constraint Hertz deformation quantization model according to claim 1 is characterized in that: The step 4 includes the following sub-steps: S401: The tire contact coefficient λ is derived from Hertz contact theory: Among them, D t Design outer diameter for new tire; w s Design section width for new tire; γ s is the tire aspect ratio; S402: Calculate the minor semi-axis b of the ellipse in the tire-road contact area according to the following formula h Virtual solution: in, is the width of the tire-ground contact area within the normal use range, and is the tire tread width w t Half of ; L is the length of the tire-road contact line segment; S403: Calculate the single wheel load according to the following formula: when hour, when hour, in, E * =πξ t x re p re Among them, E * is the equivalent contact elastic modulus of tire-road, ξ t is the vehicle tire type correction factor, which is determined by the special vehicle type according to the national standard; re is the tire inflation correction factor, ξ st is the empirical correction value when the tire pressure is equal to the standard pressure, p re is the actual tire pressure of the wheel on the same side as the video capture device, p st This is the standard inflation pressure.

7. The non-contact vehicle wheel load identification method based on the elliptical constraint Hertz deformation quantization model according to claim 6 is characterized in that: When the tire aspect ratio γ s If data is missing, calculate the new tire design section height h using the following formula s , and then approximate the aspect ratio γ s : h s =(D t -D n ) / 2 c s =h s / w s ; Among them, D n Mark the diameter of the rim.

8. A non-contact vehicle wheel load identification device for implementing the non-contact vehicle wheel load identification method according to any one of claims 1 to 7, characterized in that: It includes video acquisition equipment installed on the side of the lane and computing equipment that deploys pre-trained convolutional neural networks; The video acquisition device is used to collect tire videos of a moving vehicle in real time and send the tire videos of the moving vehicle to the computing device via wireless transmission; The computing device is used to receive the tire video of the moving vehicle and the actual tire pressure p of the wheel located on the same side as the video acquisition device, which is sent by the tire pressure monitoring system of the moving vehicle through the vehicle-road cooperative sensing device. re The computing device obtains three regions of interest (ROIs) including the tire-ground contact area, the rim area, and the tire sidewall text marking area through a built-in automatic generation method for multiple regions of interest of the tire. Secondly, based on a machine vision algorithm, the tire-ground contact line segment, the rim outer contour line, and the tire sidewall text symbol are identified from the three regions of interest. Thirdly, through the tire spatial posture estimation and contact print quantification method, the angle between the rim plane and the imaging plane and the actual length L of the tire-ground contact line segment are obtained. Finally, based on the elliptical constrained Hertz deformation quantification model, the load borne by a single tire is solved.

9. The non-contact vehicle wheel identification device according to claim 8, characterized in that: The video acquisition device is installed on a straight road section, and the imaging plane of the camera is parallel to the lane and perpendicular to the road surface; the distance between the camera and the ground should be sufficient for the camera to fully capture tire images of various types of traffic.

Images