Road vehicle load contact-non-contact multi-source decision fusion recognition system and method

Abstract

The invention discloses a road vehicle load contact-non-contact multi-source decision fusion recognition system and method. A non-contact tire load recognition system is used for collecting tire images, calculating a non-contact tire load recognition priori value and establishing a priori uncertainty quantitative model; tire pressure sensing is performed through a contact type tire load recognition system, a contact type tire load recognition priori value is calculated, and a priori uncertainty quantitative model is established; by utilizing a contact-non-contact multi-source information decision fusion and self-adaptive updating algorithm, uncertainty optimization type decision fusion is carried out on the two single-source load identification results containing the uncertainty, and self-adaptive updating of the initial uncertainty is realized through parameter reverse estimation; and a posterior identification result which takes vehicle type differences into consideration and has minimum uncertainty is obtained. According to the invention, vision and piezoelectric measurement information are fused in a decision-making level, and the accuracy and stability of an identification result are improved.

Description

Technical Field

[0001] The invention relates to the field of highway vehicle load identification, and in particular to a highway vehicle load contact-non-contact multi-source decision fusion identification system and method. Background Art

[0002] In recent years, in order to obtain and accumulate highway traffic load data and monitor and warn of abnormal situations such as vehicle overloading, vehicle dynamic weighing (WIM) technology has been widely used in road and bridge engineering. Many scholars at home and abroad have also carried out a lot of theoretical and application research on WIM technology. The existing WIM technology can be divided into two categories in principle, namely contact WIM technology based on piezoelectric sensing and non-contact WIM technology based on image recognition.

[0003] Contact WIM technology calculates vehicle load through tactile signals output by piezoelectric sensors. However, it is difficult to accurately obtain tire width and contact area using only piezoelectric sensors, and it is difficult to obtain the relative position of each piezoelectric sensor in contact with the tire, resulting in significant errors in tire load calculation. Although the above problems can be improved by increasing the area and density of piezoelectric sensor layout, this will significantly increase the difficulty and cost of installation, collection, debugging, and maintenance, making it difficult to apply in practice.

[0004] Non-contact WIM technology will estimate vehicle load by identifying tire deformation characteristics through machine vision methods. Existing non-contact WIM technology generally requires obtaining the key parameter of tire pressure to establish the relationship between deformation and wheel load, and tire pressure is generally obtained through the vehicle's own tire pressure monitoring system (TPMS) or by extracting text markings on the tire sidewall. On the one hand, there are still a large number of heavy vehicles that are not equipped with TPMS at this stage, and there are technical and authority difficulties in obtaining real-time data for vehicles that have installed TPMS systems; on the other hand, extracting text markings on the tire sidewall through optical character recognition (OCR) technology can only obtain the standard tire pressure or maximum safe tire pressure of the tire. If the standard tire pressure is directly used instead of the actual tire pressure for load estimation, it will cause errors that are difficult to assess.

[0005] Therefore, the current mainstream contact / non-contact WIM technologies often adopt a single data source and perception strategy, which leads to technical difficulties in accurately estimating vehicle loads and cannot meet the needs of vehicle load data collection and application. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention proposes a road vehicle load contact-non-contact multi-source decision fusion identification system and method, and the specific technical scheme is as follows:

[0007] A road vehicle load contact-non-contact multi-source decision fusion identification system, including a contact sensor array, a data acquisition device, a video acquisition device and a data processing device;

[0008] The contact sensor array includes a plurality of contact sensor units, each of which contains a piezoelectric sheet and a temperature sensor; the temperature sensor includes a thermistor and a measurement module thereof; the contact sensor array is embedded in the target lane and is used to collect the piezoelectric signal generated by the passing of the vehicle and the internal temperature of each contact sensor unit in the sensor array;

[0009] The camera of the video acquisition device is installed on one side of the target lane, and the image plane of the camera is perpendicular to the ground and parallel to the forward direction of the lane, and the optical axis of the image sensor is at the same position in the longitudinal direction of the road as the contact sensing unit embedded in the target lane; the video acquisition device is used to capture complete tire images of various vehicles;

[0010] The data acquisition device is connected to each contact sensor unit to perform signal conditioning and analog-to-digital conversion on the electrical signals collected by the contact sensor units;

[0011] The data processing device is electrically connected to the data acquisition device and the video acquisition device, and calculates the pressure measurement value acting on each contact sensing unit according to the electrical signal collected by each contact sensing unit; the video acquisition device is used to capture the complete tire images of various vehicles, obtain tire specification information and deformation characteristics, and then calculate the non-contact tire load identification priori value and establish a non-contact priori uncertainty quantification model; according to the contact pressure obtained by each contact sensing unit and the above-mentioned tire deformation characteristics, the contact load identification priori value is calculated and a contact priori uncertainty quantification model is established; and the contact-non-contact multi-source information decision fusion and adaptive update algorithm are used to perform uncertainty optimization decision fusion on two single-source load identification results containing uncertainty, and the initial uncertainty is adaptively updated through parameter inverse estimation, so as to obtain a posteriori identification result with minimized uncertainty that takes into account vehicle type differences.

[0012] Furthermore, the data processing device calculates the internal temperature of each contact sensing unit based on the electrical signal generated by the measurement module of the temperature sensor of each contact sensing unit; calculates the stress change rate acting on each unit based on the electrical signal generated by the piezoelectric sheet of each contact sensing unit, and corrects the stress change rate calculation result by the actual measured internal temperature of the unit; finally, the modified stress change rate is integrated with respect to time to obtain the pressure measurement value acting on each contact sensing unit.

[0013] A road vehicle load contact-non-contact multi-source decision fusion identification method is implemented based on a road vehicle load contact-non-contact multi-source decision fusion identification system, and specifically includes the following steps:

[0014] Step 1: using machine vision methods to identify tire-road contact line segments, rim outer contour lines, tire specification marks and cross-sectional parameters from tire images; at the same time, converting electrical signals generated by contact sensing units into stress rate signals and temperature signals, calculating the relationship between stress change rate and voltage according to the material constitutive structure, structural dimensions, boundary conditions and temperature sensitivity characteristics of the contact sensing units, and obtaining the contact pressure change law of each contact sensing unit through integration;

[0015] Step 2: Calculate the wheel spatial position based on the rim outer contour line and rim outer diameter parameters obtained in step 1, and then obtain the length L of the tire-road contact line segment;

[0016] Step 3: Calculate the non-contact tire load identification prior value through the elliptical footprint assumption and tire pressure balance model, and estimate the uncertainty with the uncertainty of the actual tire inflation pressure as the main source of error;

[0017] Use a contact sensor array and data acquisition equipment embedded in the road surface to obtain the piezoelectric signal generated by the passing of the vehicle and the internal temperature of each contact sensor unit in the sensor array; calculate the stress change rate acting on each contact sensor unit based on the electrical signal generated by the piezoelectric sheet of each contact sensor unit, and correct the stress change rate calculation result by measuring the internal temperature of each contact sensor unit; integrate the stress change rate with respect to time to obtain the pressure measurement value acting on each contact sensor unit; calculate the contact tire load identification prior value and perform uncertainty estimation based on the pressure obtained by the contact tire load identification system and the contact area obtained by the non-contact tire load identification system;

[0018] Step 4: Through the contact-non-contact multi-source information decision fusion and adaptive update algorithm, the uncertainty results of non-contact load identification and contact load identification are optimized and fused to obtain a posteriori results with minimized uncertainty. The initial error distribution is adaptively and dynamically updated based on the identification results, and finally a posteriori identification results with minimized uncertainty that take into account vehicle type differences are obtained.

[0019] Furthermore, in the step 1, a machine vision method is used to identify the tire-road contact line segment, the rim outer contour line, the tire specification mark and the cross-sectional parameters from the tire image, specifically including:

[0020] All wheels and rims in the collected tire video are input into the pre-trained convolutional neural network, and each detected object is recorded as an initial detection box and a category probability;

[0021] The initial detection frame size, category probability and wheel-rim detection frame position relationship in the image are used as the judgment criteria to clean the initial detection frame data, and the wheel detection frame and rim detection frame in the cleaned effective image are linearly geometrically transformed to obtain the tire-road contact ROI, rim ROI and tire sidewall text ROI;

[0022] Finally, the tire-ground contact ROI is input into the tire deformation visual recognition framework to identify the tire-road contact area and contact line segment; the rim ROI is input into the rim outer contour visual recognition process to obtain the rim outer contour line; and the various tire parameters are obtained from the tire sidewall text ROI using the method of optical character recognition and standard database query linkage.

[0023] Furthermore, in the step 3, for non-contact load identification, the following steps are included:

[0024] (a) Calculate the load standard value of each tire for non-contact load identification using the tire pressure balance model:

[0025] F vst,i =p st,i ·ξ G B i L i

[0026] Among them, F vst,i represents the load standard value of the ith tire in non-contact load identification, p st,i represents the standard inflation pressure of the ith tire, B i is the tread width of the ith tire, L i is the length of the tire-road contact line segment of the ith tire, ξ G is the footprint shape factor, defined as the footprint area and its circumscribed rectangular area B i L i The ratio of ξ G =(4+π) / 8;

[0027] (b) Taking the uncertainty of the actual tire inflation pressure as the main error, a priori uncertainty quantification model for non-contact tire load identification is established:

[0028]

[0029]

[0030] in, represents the prior visual load recognition result of the i-th tire, w - is the uncertainty distribution of the prior visual measurement of the i-th tire; μ in and They are the ratio of the actual inflation pressure of road vehicles to the standard inflation pressure p in / p st The mean and variance of the normal distribution it follows; represents the posterior visual load recognition result of the i-th tire, which is obtained by fusion and updating in subsequent steps;

[0031] For contact load identification, the following steps are involved:

[0032] (a) Calculate the standard value of contact wheel load identification for each tire:

[0033] F pst,i =p mp,i ●4ξ C,i B i L i

[0034] Among them, F pst,i represents the contact wheel load identification standard value of the i-th tire, p mp,i is the average pressure measurement value of each contact sensor unit on the i-th tire, m is the number of contact sensor units with obvious piezoelectric response when the i-th tire passes over the contact sensor array; p i is the pressure measurement value of each contact sensor unit with obvious piezoelectric response; ξ C is the area reduction rate, which is defined as the ratio of the actual contact area of ​​the tire tread to the tire footprint area;

[0035] (b) Taking the uncertainty of tire-road footprint shape as the main error body, a priori uncertainty quantification model for contact tire load identification is established:

[0036]

[0037] in, represents the priori result of the hysteresis load identification of the i-th tire, v - is the uncertainty distribution of the a priori piezoelectric measurement, represents the posterior result of piezoelectric load recognition, which needs to be obtained through subsequent fusion and update; μ G and are the mean and variance of the normal distribution obeyed by the imprint shape coefficient.

[0038] Furthermore, the step 4 specifically includes the following sub-steps:

[0039] S4.1: Calculate the fusion gain coefficient K G,i :

[0040]

[0041] S4.2: Calculate a posteriori results for contact-non-contact combined loads

[0042]

[0043]

[0044] in, are the posterior results of the i-th tire respectively. The mean and variance of the normal distribution it follows;

[0045] S4.3: Use the tire pressure balance model to reversely estimate the actual tire inflation pressure and adaptively update the uncertainty quantification model for non-contact load identification.

[0046] Furthermore, the S4.3 includes the following sub-steps:

[0047] S4.3.1: Using the a posteriori result obtained in step S4.2, the actual tire inflation pressure is inversely solved through the tire pressure balance model described in step 3 to obtain the inverse estimate of the actual tire inflation pressure:

[0048]

[0049] S4.3.2: For tires that have completed data collection and calculation, distinguish the type of vehicle to which the tires belong based on the tire specification marks obtained in step 1. When the number of collected tires for cars, light trucks, and heavy trucks exceeds the respective set thresholds M c 、M l 、M h After that, the pressure ratio The calculation results are divided into three groups: Passenger car tire pressure ratio series Light truck tire pressure ratio series Heavy-duty truck tire pressure ratio series Solve the means and standard deviations of the above three sequences respectively, and perform normal distribution tests to obtain the updated tire pressure ratio distributions of the three types of vehicles:

[0050]

[0051] The pressure ratio p in the prior uncertainty quantification model for non-contact tire load identification is replaced by three types of updated pressure ratio distributions. in / p stThe normal distribution obeyed by it is used to complete the classification and update of the uncertainty quantification model of non-contact tire load identification.

[0052] The beneficial effects of the present invention are as follows:

[0053] (1) The contact-non-contact multi-source decision-making fusion identification system and method for highway vehicle loads of the present invention combines contact piezoelectric sensing means and non-contact machine vision sensing means, and integrates and complements piezoelectric information and image information at the decision-making level. By establishing a priori uncertainty quantification models for the two methods and using uncertainty fusion methods to calibrate results and update errors, it not only overcomes the difficulty of accurately obtaining tire width and tire-road contact area through piezoelectric sensing in the prior art, but also solves the problem that the actual tire pressure of the vehicle cannot be obtained through machine vision methods alone in the prior art, and significantly improves the reliability of the identification results compared to the prior art;

[0054] (2) In the method of the present invention, a parameter inverse estimation method is used to establish a relationship between the contact pressure that can be directly obtained and the tire internal pressure that is difficult to directly obtain, thereby achieving an update of the initial pressure ratio error distribution. The updated pressure ratio error distribution takes into account the differences in the inflation laws of different vehicle types, further reducing the possibility of overestimating or underestimating the tire load due to a large gap between the a priori uncertainty quantification model and the actual situation. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 The present invention is a flowchart of a method for contact-non-contact multi-source decision fusion identification of highway vehicle loads according to an embodiment of the present invention.

[0056] Figure 2 Schematic diagram of the contact-non-contact multi-source decision fusion identification system for road vehicle loads according to an embodiment of the present invention, wherein (a) is a bird's-eye view of the system, (b) is an interaction diagram between the system and the outside world; and (c) is a schematic diagram of the data fusion foot bone.

[0057] Figure 3 Schematic diagram of the principle of the monocular mapping algorithm for the tire spatial posture and tire-road contact area provided by the present invention. DETAILED DESCRIPTION

[0058] The present invention will be described in detail below based on the accompanying drawings and preferred embodiments, and the purpose and effects of the present invention will become more clear. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0059] like Figure 1 As shown, the contact-non-contact multi-source decision fusion identification system for highway vehicle loads of the present invention includes a contact sensor array, a data acquisition device, a video acquisition device and a data processing device.

[0060] 1. Contact sensor array

[0061] The contact sensor array includes multiple contact sensor units, each of which contains a piezoelectric sheet and a temperature sensor, and the temperature sensor includes a thermistor and its measurement module. The contact sensor array is embedded in the target lane and is used to collect the piezoelectric signal generated by the passing vehicle and the internal temperature of each contact sensor unit in the sensor array.

[0062] In this embodiment, the piezoelectric sheet is a circular PZT-5H piezoelectric sheet, and the temperature sensor is a PT1000 platinum thermal resistor temperature sensor. Figure 1 As shown, multiple contact sensor units are buried in the target lane. The burial site should be located on a straight section of the road, and the burial position should be within 1 / 3 of the lane line on one side or both sides. When burying, it is necessary to groove the asphalt or concrete layer of the road surface. After installing the contact sensor unit and completing the wiring, the road surface should be leveled and the upper surface of the sensor unit should be flush with the road surface.

[0063] Before installing the contact sensor unit, it should first be temperature pre-calibrated. By measuring the piezoelectric response at different internal temperatures of the contact sensor unit, the temperature calibration coefficient λ(T) is calculated.

[0064] 2. Data acquisition equipment

[0065] The data acquisition device is connected to each contact sensor unit through a shielded wire, and performs signal conditioning and analog-to-digital conversion on the electrical signals collected by the contact sensor unit.

[0066] 3. Video capture equipment

[0067] The video acquisition equipment is installed on one side of the target lane. During installation, ensure that the image plane is perpendicular to the ground and parallel to the direction of the lane, and that the optical axis of the image sensor and the center of the contact sensor array embedded in the road surface are in the same longitudinal position of the road. The camera installation height can be adjusted according to the distance from the lane, with the goal of capturing the complete tire image of various vehicles. The camera should know the following parameters in advance: the principal distance f0 (the vertical distance from the optical center to the image plane), the plane coordinates c of the intersection of the optical axis and the image plane in the image coordinate system x With c y , the pixel size of the image sensor in the horizontal / vertical direction κ x With κ y .

[0068] 4. Data processing equipment

[0069] The data processing equipment is generally an industrial computer. The data processing equipment is electrically connected to the data acquisition equipment and the video acquisition equipment, and the internal temperature of each contact sensing unit is calculated based on the electrical signal collected by the measurement module of the temperature sensor of each contact sensing unit; the stress change rate acting on each unit is calculated based on the electrical signal generated by the piezoelectric sheet of each contact sensing unit, and the calculated result of the stress change rate is corrected by measuring the internal temperature of the unit; finally, the stress change rate is integrated with respect to time to obtain the pressure measurement value acting on each contact sensing unit. The data processing equipment also obtains tire specification information and deformation characteristics based on the complete tire images of various vehicles taken by the video acquisition equipment, and then calculates the non-contact tire load identification prior value and establishes a first priori uncertainty quantification model; based on the contact pressure obtained by each contact sensing unit and the above-mentioned tire deformation characteristics, the contact load identification prior value is calculated and a second priori uncertainty quantification model is established; and the contact-non-contact multi-source information decision fusion and adaptive update algorithm are used to perform uncertainty optimization decision fusion on the two single-source load identification results containing uncertainty, and the initial uncertainty is adaptively updated through parameter inverse estimation, so as to obtain a posteriori identification result that takes into account vehicle type differences and has minimized uncertainty.

[0070] like Figure 2 As shown, the contact-non-contact multi-source decision fusion identification method for highway vehicle loads of the present invention comprises the following steps:

[0071] Step 1: Use machine vision methods to identify the tire-road contact line segment, rim outer contour line, tire specification mark and cross-sectional parameters from the tire image; at the same time, convert the electrical signal generated by the contact sensor unit into a stress rate signal and a temperature signal, calculate the relationship between the stress change rate and the voltage according to the material constitutive structure, structural dimensions, boundary conditions and temperature sensitivity characteristics of the contact sensor unit, and obtain the contact pressure change law of each contact sensor unit through integration.

[0072] (1) For contact load identification

[0073] The stress change rate acting on the piezoelectric piece is calculated from the piezoelectric signal as follows:

[0074]

[0075] Among them, d 33 (T) Reaction material constitutive property, which is the third row and third column component in the piezoelectric strain constant tensor at temperature T; T is the piezoelectric sheet temperature or the internal temperature of the contact sensor unit, and T0 is the reference temperature when calibrating the piezoelectric sheet parameters; A p is the load action area of ​​the piezoelectric film in the 3-direction; λ(T) is the calibration coefficient expression obtained after the above temperature calibration; is the stress change rate in the 3-direction, Um is the voltage across the piezoelectric film, R s It is the external resistance introduced into the measurement circuit.

[0076] (2) For non-contact vehicle load identification

[0077] (2.1) The camera of the video acquisition device remains in working condition and transmits the video signal to the industrial control computer. The pre-trained YOLOv8n network is deployed on the industrial control computer to detect all wheels and rims in the video captured by the camera in real time, and record each detected target as an initial detection box and a category probability.

[0078] When the initial detection frame is located in the middle of the picture, multiple frames of images are continuously captured for subsequent analysis; when the initial tire detection frame appears completely in the picture, data collection of the contact sensor array is triggered at the same time.

[0079] (2.2) For the captured images, the initial detection frames are cleaned based on the size of the detection frames, the class probability, and the positional relationship between the wheel and rim detection frames in the image, and images with low class probabilities of wheels that are not completely within the image range are eliminated. The wheel detection frames and rim detection frames in the cleaned valid images are linearly geometrically transformed to obtain three regions of interest, namely, tire-road contact ROI, rim ROI, and tire sidewall text ROI.

[0080] (2.3) Use machine vision algorithms to further analyze and process the tire-road contact ROI, rim ROI, and tire sidewall text ROI. Specifically:

[0081] The tire-ground contact ROI is input into a tire deformation visual recognition framework including grayscale value transformation, adaptive filtering, threshold segmentation, morphological operation, contact interval recognition method, and geometric primitive fitting to identify the tire-road contact interval and contact line segment;

[0082] The wheel rim ROI is input into the wheel rim outer contour visual recognition process including threshold segmentation, morphological operation, regional feature screening, and fitting point sampling to obtain the wheel rim outer contour line;

[0083] The optical character recognition (OCR) and standard database query linkage method are used to obtain parameters such as rim outer diameter, tire design outer diameter, new tire design section width, and tread width from the tire sidewall text ROI.

[0084] When extracting tire specification parameters from the tire sidewall text ROI, it is very difficult to identify all the required information at once due to the tire's own contamination or wear, and some text is printed too densely or written in artistic fonts. Therefore, the tire specification mark containing the most information is identified first, and then the retrieval system is designed based on the specification information table provided by the current national standards "Heavy-duty vehicle tire specifications, dimensions, pressure and load (GB / T 2977-2016)" and "Passenger car tire specifications, dimensions, pressure and load (GB / T 2978-2014)" to query related parameters, and improve the effective recognition rate of tire parameters through the complementary recognition-query method.

[0085] Among them, due to the limitation of information recognition effectiveness and database size, the parameter of tread width B (the thickened part on the tire cross section that directly contacts the road surface, the width of which is the tread width) may not be directly identified or queried. Studies have shown that the design and manufacture of vehicle tires generally use 0.6-0.9 times the design section width of a new tire as the recommended value range of tread width, so 0.75 times the design section width of a new tire is taken as the approximate estimated value of B.

[0086] In recent years, the fields of neural networks and machine vision have developed rapidly. There are many relatively mature commercial / open source solutions for the above-mentioned algorithms and processes for geometric feature recognition and optical character recognition, which will not be elaborated here.

[0087] Step 2: Calculate the wheel spatial position based on the rim outer contour line and rim outer diameter parameters;

[0088] In the image processing and feature recognition process in step 1, an image coordinate system describing the pixel position is established, that is, the upper left corner of the image is the origin and the horizontal right is x. IM The positive direction of the axis, with vertical downward as y IM Positive direction, length unit is pixel. When calculating the spatial pose, in order to describe the spatial position of the wheel, a three-dimensional camera coordinate system is established, which takes the optical center of the camera as the origin, the x-axis is parallel to the lane, the y-axis is perpendicular to the road surface, and the z-axis points to the object being measured.

[0089] The outer contour line of the wheel rim obtained in step 1 is an ellipse with a major axis parallel to the vertical direction and a minor axis parallel to the horizontal direction in the image coordinate system. The length a of the major semi-axis of the ellipse in the image coordinate system is extracted. IM , short semi-axis length b IM And the coordinates of the center of the ellipse (p IM ,q IM ). Extract the actual rim radius R from the tire sidewall text ROI obtained in step S2 r . Based on the principle of photo geometry, the coordinates of the wheel rim center N (x rc ,yrc ,z rc ) and the angle θ between the rim plane and the image plane, as Figure 3 shown.

[0090] In the camera coordinate system, any point (x, y, z) on the rim boundary satisfies:

[0091]

[0092] And the rim plane is perpendicular to the road surface, then any point (x, y, z) on the rim plane should also satisfy:

[0093] (xx rc )sinθ+(zz rc )cosθ=0

[0094] According to the perspective relationship, if the point (x IM ,y IM ) corresponds to the point (x, y, z) on the rim plane in the camera coordinate system. Since the line connecting a set of corresponding points must pass through the optical center of the camera, the following relationship is satisfied:

[0095]

[0096] Combining the above three equations, we can see that the rim boundary is mapped into an ellipse in the image coordinate system, and its equation is:

[0097]

[0098] From this we calculate:

[0099]

[0100]

[0101] In step 1, the tire-road contact line segment has been identified and the image coordinates of the two end points C have been obtained. 1,IM (x IM,1 ,y IM,1 ), C 2,IM (x IM,2 ,y IM,2 ), then the wheel rim plane equation and the straight line C1O and C2O equations can be combined, as Figure 3 As shown in the figure, through the monocular mapping algorithm of the tire-road contact area, the coordinates of the two end points of the tire-road contact line segment in the camera coordinate system are solved as follows:

[0102]

[0103] Where i = 1, 2. The obtained x ext,i =(x ext,i ,yext,i ,z ext,i ) is the spatial coordinates of the two end points of the tire-road contact line segment, then the actual tire-road contact length is: L = ||x ext,1 -x ext,2 ||2.

[0104] Since multiple valid images are obtained for a single tire object in step 1, this step can repeatedly calculate multiple tire-road actual contact lengths with similar values, and take the arithmetic mean as the final result.

[0105] The tire spatial posture and tire-road contact area monocular mapping algorithm in the method of the present invention can solve the spatial posture only through the image collected by the monocular camera, reducing the complexity of the hardware system. And compared with the scaling factor method commonly used in the existing non-contact WIM technology, the proposed algorithm is applicable to the situation where the tire is parallel to the imaging plane or at a certain angle, expanding the application scope of the existing technology.

[0106] Step 3: Calculate the non-contact tire load identification prior value through the elliptical footprint assumption and tire pressure balance model, and estimate the uncertainty with the uncertainty of the actual tire inflation pressure as the main source of error;

[0107] A contact sensor array and data acquisition equipment embedded in the road surface are used to obtain the piezoelectric signal generated by the passing of the vehicle and the internal temperature of each contact sensor unit in the sensor array; the stress change rate acting on each contact sensor unit is calculated based on the electrical signal generated by the piezoelectric sheet of each contact sensor unit, and the stress change rate calculation result is corrected by measuring the internal temperature of each contact sensor unit; the stress change rate is integrated with respect to time to obtain the pressure measurement value acting on each contact sensor unit; based on the pressure obtained by the contact tire load identification system and the contact area obtained by the non-contact tire load identification system, the contact tire load identification prior value is calculated and the uncertainty is estimated.

[0108] (1) For non-contact load identification

[0109] a. Calculate the standard load value of each tire for non-contact load identification through the tire pressure balance model:

[0110] F vst,i =p st,i ● G,i B i L i

[0111] Among them, F vst,i represents the load standard value of the ith tire in non-contact load identification, p st,i represents the standard inflation pressure of the ith tire, B iis the tread width of the i-th tire, L i is the length L of the tire-road contact line segment of the i-th tire, both of which can be directly obtained from the tire sidewall text mark ROI in step 1; ξ G is the footprint shape factor, which is defined as the ratio of the footprint area to its circumscribed rectangular area BL. According to the elliptical footprint area assumption, ξ G =(4+π) / 8.

[0112] b. Taking the uncertainty of the actual tire inflation pressure as the main error body, a priori uncertainty quantification model for non-contact tire load identification is established:

[0113]

[0114] in, represents the visual load recognition prior result of the i-th tire, w - is the uncertainty distribution of the prior visual measurement of the i-th tire; μ in and They are the ratio of the actual inflation pressure of road vehicles to the standard inflation pressure p in / p st The mean and variance of the normal distribution obeyed, in this embodiment, their initial values ​​are 1.10 and 0.05 respectively; It represents the posterior visual load recognition result of the i-th tire, which is obtained by fusion and update in subsequent steps.

[0115] (2) For contact load identification

[0116] a. Calculate the standard value of contact wheel load identification for each tire:

[0117] F pst,i =p mp,i ·4ξ C,i B i L i

[0118] Among them, F pst,i represents the contact wheel load identification standard value of the i-th tire, p mp,i is the average pressure measurement value of each contact sensor unit on the i-th tire, m is the number of contact sensor units with obvious piezoelectric response when the i-th tire passes over the contact sensor array; p i is the pressure measurement value of each contact sensor unit with obvious piezoelectric response; ξ C is the area reduction rate, which is defined as the ratio of the actual contact area of ​​the tire pattern to the tire footprint area. i and the length L of the tire-road contact line segment of the i-th tire iBy comparing the image frame creation time and the piezoelectric signal peak occurrence time, the visual information and piezoelectric signal from the same tire are matched.

[0119] b. Taking the uncertainty of tire-road footprint shape as the main error body, a priori uncertainty quantification model for contact tire load identification is established:

[0120]

[0121] in, represents the priori result of the hysteresis load identification of the i-th tire, v - is the uncertainty distribution of the a priori piezoelectric measurement, represents the posterior result of piezoelectric load recognition, which needs to be obtained through subsequent fusion and update; μ G and are the mean and variance of the normal distribution obeyed by the imprint shape coefficient.

[0122] Step 4: Through the contact-non-contact multi-source information decision fusion and adaptive update algorithm, the uncertainty results of non-contact load identification and contact load identification are optimized and fused to obtain the a posteriori results with minimized uncertainty, and the initial error distribution is adaptively and dynamically updated according to the identification results.

[0123] Step 4 specifically includes the following steps:

[0124] S4.1: Calculate the fusion gain coefficient K G,i :

[0125]

[0126] S4.2: Calculate a posteriori results for contact-non-contact combined loads

[0127]

[0128] in, are the posterior results of the i-th tire respectively. The mean and variance of the normal distribution that follows. Reach the minimum value, that is, the a posteriori load identification result Get the best estimate.

[0129] S4.3: Use the tire pressure balance model to reversely estimate the actual tire inflation pressure and adaptively update the uncertainty quantification model for non-contact load identification.

[0130] Actual tire inflation pressure p inIt is difficult to obtain under non-laboratory conditions, but pressure measurement data can be obtained through the contact tire load identification system. According to the assumption of the tire pressure balance model, theoretically, the actual tire inflation pressure and the tire-road contact pressure should be in direct proportion. Therefore, the actual tire inflation pressure p can be estimated through pressure measurement data. in It is feasible to perform reverse estimation, and the reverse estimation value The pressure ratio distribution described in step three can be recalibrated, thereby updating the non-contact a priori uncertainty quantification model associated therewith.

[0131] S4.3 specifically includes the following sub-steps:

[0132] S4.3.1: Using the a posteriori result obtained in step S4.2, the actual tire inflation pressure is inversely solved through the tire pressure balance model described in step 3 to obtain the inverse estimate of the actual tire inflation pressure:

[0133]

[0134] S4.3.2: For tires that have completed data collection and calculation, distinguish the type of vehicle to which the tires belong based on the tire specification marks obtained in step 1. When the number of collected tires for cars, light trucks, and heavy trucks exceeds the respective set thresholds M c 、M l 、M h After that, the pressure ratio The calculation results are divided into three groups: Passenger car tire pressure ratio series Light truck tire pressure ratio series Heavy-duty truck tire pressure ratio series Solve the means and standard deviations of the above three sequences respectively, and perform normal distribution tests to obtain the updated tire pressure ratio distributions of the three types of vehicles:

[0135]

[0136] Compared with the uniform pressure ratio initial distribution It is more reasonable to use different pressure ratio distributions to establish the uncertainty quantification model of the corresponding vehicle. In fact, heavy-duty trucks tend to use air pressure values ​​slightly higher than the standard pressure to improve their carrying capacity, while family cars tend to use actual inflation pressures close to the standard pressure for safety reasons. Therefore, it is necessary to distinguish the pressure ratio distribution by vehicle type.

[0137] S4.3.3: Replace the pressure ratio p in the prior uncertainty quantification model for non-contact tire load identification using three types of updated pressure ratio distributions in / p st Normal distribution Then the uncertainty quantification model of non-contact tire load identification is classified and updated. Subsequent tire load identification will determine the vehicle type based on tire specification information and select the corresponding non-contact load identification system uncertainty quantification model for decision fusion.

[0138] The inverse estimation, data accumulation, and distribution update methods described in this step can be executed repeatedly, that is, after the system completes the data collection and accumulation of kM tires, it should also perform k rounds of updates on the initial distribution, so that the uncertainty quantification model of the non-contact load identification system gradually approaches the actual situation and reaches a stable state.

[0139] Those skilled in the art can understand that the above are only preferred examples of the invention and are not intended to limit the invention. Although the invention is described in detail with reference to the above examples, those skilled in the art can still modify the technical solutions recorded in the above examples or replace some of the technical features therein with equivalents. Any modification, equivalent replacement, etc. made within the spirit and principle of the invention shall be included in the protection scope of the invention.

Claims

1. A road vehicle load contact-non-contact multi-source decision fusion identification system, characterized in that: It includes contact sensor array, data acquisition equipment, video acquisition equipment and data processing equipment; The contact sensor array includes a plurality of contact sensor units, each of which contains a piezoelectric sheet and a temperature sensor; the temperature sensor includes a thermistor and a measurement module thereof; the contact sensor array is embedded in the target lane and is used to collect the piezoelectric signal generated by the passing of the vehicle and the internal temperature of each contact sensor unit in the sensor array; The camera of the video acquisition device is installed on one side of the target lane, and the image plane of the camera is perpendicular to the ground and parallel to the forward direction of the lane, and the optical axis of the image sensor is at the same position in the longitudinal direction of the road as the contact sensing unit embedded in the target lane; the video acquisition device is used to capture complete tire images of various vehicles; The data acquisition device is connected to each contact sensor unit to perform signal conditioning and analog-to-digital conversion on the electrical signals collected by the contact sensor units; The data processing device is electrically connected to the data acquisition device and the video acquisition device, and calculates the pressure measurement value acting on each contact sensing unit according to the electrical signal collected by each contact sensing unit; the video acquisition device is used to capture the complete tire images of various vehicles, obtain tire specification information and deformation characteristics, and then calculate the non-contact tire load identification priori value and establish a non-contact priori uncertainty quantification model; according to the contact pressure obtained by each contact sensing unit and the above-mentioned tire deformation characteristics, the contact load identification priori value is calculated and a contact priori uncertainty quantification model is established; and the contact-non-contact multi-source information decision fusion and adaptive update algorithm are used to perform uncertainty optimization decision fusion on two single-source load identification results containing uncertainty, and the initial uncertainty is adaptively updated through parameter inverse estimation, so as to obtain a posteriori identification result with minimized uncertainty that takes into account vehicle type differences.

2. The highway vehicle load contact-non-contact multi-source decision fusion identification system according to claim 1 is characterized in that: The data processing device calculates the internal temperature of each contact sensing unit based on the electrical signal generated by the measurement module of the temperature sensor of each contact sensing unit; calculates the stress change rate acting on each unit based on the electrical signal generated by the piezoelectric sheet of each contact sensing unit, and corrects the stress change rate calculation result by the actual measured internal temperature of the unit; finally, the modified stress change rate is integrated with respect to time to obtain the pressure measurement value acting on each contact sensing unit.

3. A contact-non-contact multi-source decision fusion identification method for highway vehicle loads, characterized in that: The method is implemented based on the highway vehicle load contact-non-contact multi-source decision fusion identification system according to claim 1 or 2, and specifically comprises the following steps: Step 1: using machine vision methods to identify tire-road contact line segments, rim outer contour lines, tire specification marks and cross-sectional parameters from tire images; at the same time, converting electrical signals generated by contact sensing units into stress rate signals and temperature signals, calculating the relationship between stress change rate and voltage according to the material constitutive structure, structural dimensions, boundary conditions and temperature sensitivity characteristics of the contact sensing units, and obtaining the contact pressure change law of each contact sensing unit through integration; Step 2: Calculate the wheel spatial position based on the rim outer contour line and rim outer diameter parameters obtained in step 1, and then obtain the length L of the tire-road contact line segment; Step 3: Calculate the non-contact tire load identification prior value through the elliptical footprint assumption and tire pressure balance model, and estimate the uncertainty with the uncertainty of the actual tire inflation pressure as the main source of error; Use a contact sensor array and data acquisition equipment embedded in the road surface to obtain the piezoelectric signal generated by the passing of the vehicle and the internal temperature of each contact sensor unit in the sensor array; calculate the stress change rate acting on each contact sensor unit based on the electrical signal generated by the piezoelectric sheet of each contact sensor unit, and correct the stress change rate calculation result by measuring the internal temperature of each contact sensor unit; The stress change rate is integrated with respect to time to obtain the pressure measurement value acting on each contact sensing unit; based on the pressure obtained by the contact tire load identification system and the contact area obtained by the non-contact tire load identification system, the contact tire load identification prior value is calculated and the uncertainty is estimated; Step 4: Through the contact-non-contact multi-source information decision fusion and adaptive update algorithm, the uncertainty results of non-contact load identification and contact load identification are optimized and fused to obtain a posteriori results with minimized uncertainty. The initial error distribution is adaptively and dynamically updated based on the identification results, and finally a posteriori identification results with minimized uncertainty that take into account vehicle type differences are obtained.

4. The contact-non-contact multi-source decision fusion identification method for highway vehicle loads according to claim 3 is characterized in that: In the step 1, a machine vision method is used to identify the tire-road contact line segment, the rim outer contour line, the tire specification mark and the cross-sectional parameters from the tire image, which specifically includes: All wheels and rims in the collected tire video are input into the pre-trained convolutional neural network, and each detected object is recorded as an initial detection box and a category probability; The initial detection frame size, category probability and wheel-rim detection frame position relationship in the image are used as the judgment criteria to clean the initial detection frame data, and the wheel detection frame and rim detection frame in the cleaned effective image are linearly geometrically transformed to obtain the tire-road contact ROI, rim ROI and tire sidewall text ROI; Finally, the tire-ground contact ROI is input into the tire deformation visual recognition framework to identify the tire-road contact area and contact line segment; the rim ROI is input into the rim outer contour visual recognition process to obtain the rim outer contour line; and the various tire parameters are obtained from the tire sidewall text ROI using the method of optical character recognition and standard database query linkage.

5. The contact-non-contact multi-source decision fusion identification method for highway vehicle loads according to claim 4 is characterized in that: In the step 3, for non-contact load identification, the following steps are included: (a) Calculate the load standard value of each tire for non-contact load identification using the tire pressure balance model: F vst,i =p st,i ·x G B i L i Among them, F vst,i represents the load standard value of the ith tire in non-contact load identification, p st,i represents the standard inflation pressure of the ith tire, B i is the tread width of the ith tire, L i is the length of the tire-road contact line segment of the ith tire, ξ G is the footprint shape factor, defined as the footprint area and its circumscribed rectangular area B i L i The ratio of ξ G =(4+π) / 8; (b) Taking the uncertainty of the actual tire inflation pressure as the main error, a priori uncertainty quantification model for non-contact tire load identification is established: in, represents the prior visual load recognition result of the i-th tire, w - is the uncertainty distribution of the prior visual measurement of the i-th tire; μ in and They are the ratio of the actual inflation pressure of road vehicles to the standard inflation pressure p in / p st The mean and variance of the normal distribution it follows; represents the posterior visual load recognition result of the i-th tire, which is obtained by fusion and updating in subsequent steps; For contact load identification, the following steps are involved: (a) Calculate the standard value of contact wheel load identification for each tire: F pst,i =p mp,i ·4ξ C,i B i THE i Among them, F pst,i represents the contact wheel load identification standard value of the i-th tire, p mp,i is the average pressure measurement value of each contact sensor unit on the i-th tire, m is the number of contact sensor units with obvious piezoelectric response when the i-th tire passes over the contact sensor array; p i is the pressure measurement value of each contact sensor unit with obvious piezoelectric response; ξ C is the area reduction rate, which is defined as the ratio of the actual contact area of ​​the tire tread to the tire footprint area; (b) Taking the uncertainty of tire-road footprint shape as the main error body, a priori uncertainty quantification model for contact tire load identification is established: in, represents the priori result of the hysteresis load identification of the i-th tire, v - is the uncertainty distribution of the a priori piezoelectric measurement, represents the posterior result of piezoelectric load recognition, which needs to be obtained through subsequent fusion and update; μ G and are the mean and variance of the normal distribution obeyed by the imprint shape coefficient.

6. The contact-non-contact multi-source decision fusion identification method for highway vehicle loads according to claim 5 is characterized in that: The step 4 specifically includes the following sub-steps: S4.1: Calculate the fusion gain coefficient K G,i : S4.2: Calculate a posteriori results for contact-non-contact combined loads in, are the posterior results of the i-th tire respectively. The mean and variance of the normal distribution it follows; S4.3: Use the tire pressure balance model to reversely estimate the actual tire inflation pressure and adaptively update the uncertainty quantification model for non-contact load identification.

7. The contact-non-contact multi-source decision fusion identification method for highway vehicle loads according to claim 6 is characterized in that: The S4.3 comprises the following sub-steps: S4.3.1: Using the a posteriori result obtained in step S4.2, the actual tire inflation pressure is inversely solved through the tire pressure balance model described in step 3 to obtain the inverse estimate of the actual tire inflation pressure: S4.3.2: For tires that have completed data collection and calculation, distinguish the type of vehicle to which the tires belong based on the tire specification marks obtained in step 1. When the number of collected tires for cars, light trucks, and heavy trucks exceeds the respective set thresholds M c 、M l 、M h After that, the pressure ratio / p st,i The calculation results are divided into three groups: Passenger car tire pressure ratio series Light truck tire pressure ratio series Heavy-duty truck tire pressure ratio series Solve the means and standard deviations of the above three sequences respectively, and perform normal distribution tests to obtain the updated tire pressure ratio distributions of the three types of vehicles: The pressure ratio p in the prior uncertainty quantification model for non-contact tire load identification is replaced by three types of updated pressure ratio distributions. in / p st The normal distribution obeyed by it is used to complete the classification and update of the uncertainty quantification model of non-contact tire load identification.

Images