Vehicle load identification method, electronic device, and computer program product

Abstract

The present disclosure provides a vehicle load identification method, an electronic device and a computer program product, and relates to the technical field of computers. The vehicle load identification method comprises: acquiring a to-be-identified image of a vehicle driving on a target structure, performing target detection on the vehicle in the to-be-identified image, and determining the pixel coordinates of the vehicle; converting the pixel coordinates of the vehicle into physical coordinates of the vehicle on the target structure; mapping the physical coordinates of the vehicle to a degree-of-freedom index set, constructing a prior load information vector based on the degree-of-freedom index set; constructing an objective function based on a response observation equation of the target structure and the prior load information vector; and iteratively solving the objective function to determine the load of the vehicle.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to a vehicle load identification method, electronic device, and computer program product. Background Technology

[0002] Currently, bridge vehicle load identification is mainly divided into direct weighing methods and indirect methods. Direct methods require vehicles to pass slowly and may disrupt traffic, making them unsuitable for real-time monitoring of in-service bridges. The most commonly used indirect method is bridge dynamic weighing technology based on influence line theory, which requires only a small number of sensors to estimate vehicle speed, axle load, and wheelbase. However, because the number of sensors is far less than the structural degrees of freedom, the influence line matrix is ​​often singular, making load identification a typical ill-posed problem. Traditional regularization methods can provide smooth solutions, but they lack real physical spatial constraints, leading to identified loads potentially leaking into areas without vehicles or causing severe overfitting. Load identification is essentially an underdetermined inverse problem. Existing methods rely solely on structural response data, ignoring the real physical constraints of load spatial distribution, making it difficult to accurately define the load application area. Summary of the Invention

[0003] This disclosure provides a vehicle load identification method, electronic device, and computer program product.

[0004] According to one aspect of this disclosure, a vehicle load identification method is provided, comprising: acquiring an image to be identified of a vehicle traveling on a target structure; performing target detection on the vehicle in the image to be identified to determine the pixel coordinates of the vehicle; converting the pixel coordinates of the vehicle into physical coordinates of the vehicle on the target structure; mapping the physical coordinates of the vehicle to a degree-of-freedom index set; constructing a priori load information vector based on the degree-of-freedom index set, wherein the degree-of-freedom index set represents the set of degree-of-freedom numbers of the vehicle in the finite element model of the target structure, and the priori load information vector is used to represent the area of ​​action of the vehicle's load on the target structure; constructing an objective function based on the response observation equation of the target structure and the priori load information vector, wherein the response observation equation is determined based on the structural response of the target structure under a unit load and the actual measured structural response; and iteratively solving the objective function to determine the load of the vehicle.

[0005] According to the vehicle load identification method disclosed herein, by fusing prior information on vehicle position obtained from computer vision with structural response data, the spatial distribution constraint of vehicle load is explicitly introduced into the objective function, transforming the originally ill-posed inverse problem of load identification into a well-posed problem. This effectively suppresses the phenomena of load leakage to vehicle-free areas and overfitting in traditional pure mathematical inversion methods, significantly improving the accuracy and stability of load identification under complex working conditions such as multi-vehicle parallel operation, distributed load, and variable position, and realizing non-contact real-time estimation of bridge vehicle load.

[0006] According to at least one embodiment of the vehicle load identification method of this disclosure, the objective function is iteratively solved to determine the vehicle load, including: solving the objective function to determine the vehicle load estimate; based on the load estimate, replacing the vector elements in the prior load information vector corresponding to the degree of freedom index set with the values ​​at the corresponding positions in the load estimate, until the loss function value is less than the target convergence threshold, to determine the vehicle load.

[0007] According to at least one embodiment of the vehicle load identification method of this disclosure, the objective function includes a regularization parameter, which is determined by a double logarithmic curve constructed from the data residual norm and the regularized denorm under different regularization parameters. The data residual norm represents the deviation between the structural response generated by the load estimation and the actual measured structural response, and the regularized denorm represents the deviation between the load estimation and the prior load information vector.

[0008] According to at least one embodiment of the vehicle load recognition method of this disclosure, target detection is performed on the vehicle in the image to be recognized to determine the pixel coordinates of the vehicle, including: dividing the image to be recognized into grid cells; performing target detection on each grid cell to determine the candidate bounding box of each grid cell and the confidence level of the presence of the vehicle in each candidate bounding box; filtering the candidate bounding boxes of each grid cell based on the confidence level to determine the target bounding box of each grid cell; and determining the center point coordinates or target corner point coordinates of the target bounding box as the pixel coordinates of the vehicle.

[0009] According to at least one embodiment of the vehicle load identification method of this disclosure, converting the pixel coordinates of the vehicle into the physical coordinates of the vehicle on the target structure includes: calibrating the parameters of the spatial plane equation of the target structure using calibration points with known physical coordinates on the target structure; and determining the physical coordinates of the vehicle on the target structure by solving the spatial plane equation of the target structure using a camera pinhole imaging model for the vehicle pixel coordinates, wherein the camera pinhole imaging model represents the projection relationship between the pixel coordinates and the physical coordinates of the vehicle.

[0010] According to at least one embodiment of the vehicle load identification method of this disclosure, the physical coordinates of the vehicle are mapped to a set of degrees of freedom indexes, and a priori load information vector is constructed based on the set of degrees of freedom indexes. The method includes: establishing a finite element model of the target structure, mapping the physical coordinates of the vehicle to the finite element model, and determining the set of degrees of freedom indexes; constructing a priori load information vector based on the set of degrees of freedom indexes, wherein the number of vector elements of the priori load information vector is the same as the total number of degrees of freedom of the finite element model.

[0011] According to at least one embodiment of the vehicle load identification method of this disclosure, before constructing the objective function based on the response observation equation of the target structure and the prior load information vector, the method includes: setting the vector elements in the prior load information vector corresponding to the degree of freedom index set to a preset small amplitude value, i.e., a first value, and setting the remaining vector elements to zero, i.e., a second value.

[0012] According to at least one embodiment of the vehicle load identification method of this disclosure, a target function is constructed based on the response observation equation of the target structure and the prior load information vector, including: constructing a structural response observation equation of the target structure based on the structural response of the target structure under unit load and the actual measured structural response; and constructing a target function based on the prior load information vector and the structural response observation equation.

[0013] According to another aspect of this disclosure, an electronic device is provided, comprising: a memory storing execution instructions; and a processor executing the execution instructions stored in the memory, causing the processor to perform a vehicle load identification method according to any embodiment of this disclosure.

[0014] According to another aspect of this disclosure, a readable storage medium is provided, wherein execution instructions are stored therein, which, when executed by a processor, are used to implement the vehicle load identification method of any embodiment of this disclosure.

[0015] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements a vehicle load identification method according to any embodiment of this disclosure.

[0016] The beneficial effects of the vehicle load identification method disclosed herein are at least as follows: by fusing prior information on vehicle position obtained from computer vision with structural response data, the spatial distribution constraints of vehicle load are explicitly introduced into the objective function, transforming the originally ill-posed inverse problem of load identification into a well-posed problem. This effectively suppresses the phenomena of load leakage to vehicle-free areas and overfitting in traditional pure mathematical inversion methods, significantly improves the accuracy and stability of load identification under complex working conditions such as multi-vehicle parallel operation, distributed load, and variable position, and realizes non-contact real-time estimation of bridge vehicle load. Attached Figure Description

[0017] The accompanying drawings illustrate exemplary embodiments of the present disclosure and, together with the description thereof, serve to explain the principles of the present disclosure. These drawings are included to provide a further understanding of the present disclosure and are incorporated in and constitute a part of this specification.

[0018] Figure 1 This is a schematic diagram of the overall process of a vehicle load identification method according to one embodiment of the present disclosure.

[0019] Figure 2 This is a schematic flowchart illustrating the process of determining the pixel coordinates of a vehicle in a vehicle load identification method according to one embodiment of the present disclosure.

[0020] Figure 3 This is a schematic diagram of the coordinate transformation process in a vehicle load identification method according to one embodiment of the present disclosure.

[0021] Figure 4 This is a schematic diagram of the process for constructing a priori load information vector in a vehicle load identification method according to one embodiment of the present disclosure.

[0022] Figure 5 This is a flowchart illustrating the construction of an objective function in a vehicle load identification method according to one embodiment of the present disclosure.

[0023] Figure 6 This is a schematic diagram of the process for determining the load of a vehicle in a vehicle load identification method according to one embodiment of the present disclosure.

[0024] Figure 7 This is a flowchart of the YOLO network in a vehicle load identification method according to one embodiment of the present disclosure.

[0025] Figure 8 This is a schematic diagram of camera imaging and coordinate transformation in a vehicle load recognition method according to one embodiment of the present disclosure.

[0026] Figure 9 This is a schematic structural block diagram of a vehicle load identification device according to one embodiment of the present disclosure.

[0027] Figure 10This is a schematic structural block diagram of an electronic device according to one embodiment of the present disclosure. Detailed Implementation

[0028] The present disclosure will now be described in further detail with reference to the accompanying drawings and examples. It should be understood that the specific examples described herein are for illustrative purposes only and are not intended to limit the scope of the disclosure. Furthermore, it should be noted that, for ease of description, only the parts relevant to the present disclosure are shown in the accompanying drawings.

[0029] It should be noted that, where there is no conflict, the embodiments and features described in this disclosure can be combined with each other. The technical solutions of this disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0030] When multiple vehicles of different models cross a bridge in parallel or sequentially, it is difficult to distinguish the specific location, axle load, and speed of each vehicle based solely on signals from a few strain sensors on the bridge. This leads to load leakage (i.e., the identified load appears in areas without vehicles) and unstable estimation results in traditional methods. At the same time, the bridge's own vibration and environmental noise can severely interfere with the weak load response signal, making purely data-driven inversion algorithms prone to overfitting.

[0031] To address this issue, this disclosure proposes a vehicle load identification method. By mapping the physical coordinates of vehicles obtained through computer vision to a set of degrees of freedom indexes in a finite element model, and constructing a priori load information vector based on this, the method explicitly introduces spatial position constraints into the objective function of mathematical inversion, allowing the load to exist only in areas with vehicles. This enables accurate identification of the operational area of ​​each vehicle even with only a few sensor signals, effectively preventing the identified load from leaking into areas without vehicles. This solves the problem of load position confusion caused by signal aliasing when multiple vehicles pass in parallel or sequentially, ensuring the spatial rationality of the load identification results.

[0032] The vehicle load identification method disclosed herein can be deployed on a server-side, receiving video streams and structural response data transmitted from multiple bridge sites to centrally perform vehicle target detection, coordinate transformation, and load identification, enabling remote real-time monitoring and statistical analysis of traffic loads across a cross-regional bridge group. Alternatively, it can be deployed on edge computing terminals or embedded smart cameras at the bridge site to perform image acquisition, vehicle location identification, and load inversion locally, reducing network transmission latency and bandwidth consumption, and meeting real-time early warning requirements.

[0033] Figure 1 A schematic diagram illustrating the overall flow of a vehicle load identification method according to one embodiment of this disclosure is shown. Figure 1 The method M100 shown includes steps S110 to S140. This method can be executed by electronic devices such as servers, mobile terminals, edge computing terminals, smart cameras, or industrial control computers.

[0034] In step S110, an image of the vehicle driving on the target structure is acquired, and the vehicle in the image is detected to determine the pixel coordinates of the vehicle.

[0035] By mounting cameras on a target structure (such as a bridge), a sequence of video frames containing moving vehicles is continuously captured. A deep learning-based object detection algorithm performs object detection on each image to be identified in the video frame sequence, determining the pixel coordinates of each vehicle in the image to be identified.

[0036] Preferably, a YOLO convolutional neural network is used to detect vehicles in the image to be identified.

[0037] Alternatively, vehicle detection in the image to be identified can be performed using a Faster R-CNN neural network or an SSD neural network.

[0038] Preferably, the origin is the upper left corner or center of the image to be recognized, the horizontal direction is the horizontal axis and the vertical direction is the vertical axis, and the pixel coordinates are used to describe the specific position of the center point or corner point of the vehicle bounding box in the plane of the image to be recognized.

[0039] In step S120, the pixel coordinates of the vehicle are converted into the physical coordinates of the vehicle on the target structure.

[0040] By establishing a geometric mapping relationship between the image pixel space and the physical space of the target structure, the two-dimensional image position information (i.e. pixel coordinates) obtained by visual detection is transformed into three-dimensional physical position information (i.e. physical coordinates) that can be used for structural analysis.

[0041] Preferably, by utilizing pre-calibrated camera imaging parameters and spatial geometric constraints of the target structure surface, the pixel coordinates of the vehicle are back-projected onto the three-dimensional physical coordinate system of the target structure based on the principle of perspective projection, thereby determining the vehicle's true spatial position coordinates relative to the target structure, i.e., physical coordinates.

[0042] In step S130, the physical coordinates of the vehicle are mapped to the degree-of-freedom index set, and a priori load information vector is constructed based on the degree-of-freedom index set. The degree-of-freedom index set represents the set of degree-of-freedom numbers of the vehicle in the finite element model of the target structure, and the priori load information vector is used to represent the area of ​​action of the vehicle's load on the target structure.

[0043] A finite element model of the target structure is established, and all degrees of freedom (DOF) numbers are determined. The vehicle's physical coordinates are mapped to the corresponding set of DDF numbers, forming a DDF index set. A priori load information vector with the same dimension as the total number of DDFs is constructed. By marking the possible DDF positions of the vehicle loads in the priori load information vector, a mapping relationship between visual position information and the structural mechanics analysis model is established.

[0044] Preferably, before constructing the objective function based on the response observation equation of the target structure and the prior load information vector, the following steps are taken: setting the vector elements in the prior load information vector corresponding to the degree-of-freedom index set to a preset small amplitude value, i.e., the first value, and setting the remaining vector elements to zero, i.e., the second value. The first value is used to mark the location of the corresponding vector element where a vehicle load exists, and the second value is used to indicate the location of the corresponding vector element where a vehicle load does not exist.

[0045] Specifically, the prior load information vector is initialized to a zero vector. Then, all vector elements of the prior load information vector are traversed. For vector elements whose vector indices belong to the degree-of-freedom index set, their values ​​are set to a preset small positive number, i.e., the first value. For all vector elements whose vector indices do not belong to the degree-of-freedom index set, their values ​​remain zero, i.e., the second value.

[0046] Specifically, the first value is 0.01.

[0047] In step S140, an objective function is constructed based on the response observation equation of the target structure and the prior load information vector. The response observation equation is determined based on the structural response of the target structure under unit load and the actual measured structural response.

[0048] The response observation equation of the target structure is established based on influence line theory. The influence line matrix is ​​determined by the structural response of the target structure under a unit load, and the measured response vector is determined by actual sensor measurements. The prior load information vector is embedded into the objective function as a regularization term, so that the estimated load converges to the load action region marked by the visual prior, and the regularization parameter is used as the weight.

[0049] Specifically, the response observation equation is: Y=W·F Y represents the actual measured structural response, W represents the influence line matrix, and F represents the nodal load estimation vector to be solved.

[0050] Specifically, the objective function is: F represents the estimated vector of nodal loads to be solved, Y represents the actual measured structural response, and W represents the influence line matrix. This represents the prior load information vector, where P represents the norm order. This represents the regularization parameter.

[0051] In step S150, the objective function is solved iteratively to determine the vehicle load.

[0052] By iteratively updating the regularized objective function of the fused visual prior, the load estimation results are solved one by one, so that the load estimation results gradually converge to the true load distribution under the constraints of physical space, and the vehicle load is determined.

[0053] Preferably, the objective function is analytically solved based on the prior load information vector to obtain the load estimate. Then, the load estimate is used to update the values ​​at the corresponding degrees of freedom index positions in the prior load information vector, while the values ​​at other positions remain zero. Through multiple iterations, the prior information gradually approximates the actual load magnitude from the initial coarse markings until the convergence condition is met.

[0054] Therefore, the vehicle load identification method disclosed herein transforms the vehicle position information acquired by computer vision into a priori load information vector at the degree of freedom level of the structural finite element model, and explicitly embeds it into a generalized regularized objective function based on influence line theory. This transforms the originally highly ill-posed inverse load identification problem into a well-posed problem under physical space constraints. It effectively suppresses the phenomena of load leakage to vehicle-free areas, sensitivity of estimation results to noise, and overfitting under complex working conditions found in traditional pure mathematical inversion methods. It achieves stable and accurate identification of the load application area and axle load of each vehicle under conditions of a limited number of sensors and only known vehicle position indices.

[0055] Regarding step S110, target detection is performed on the vehicle in the image to be recognized to determine the vehicle's pixel coordinates. In some embodiments of this disclosure, this may include, for example... Figure 2 Steps S1101 to S1104 are shown.

[0056] In step S1101, the image to be recognized is divided to determine the grid cells of the image to be recognized.

[0057] The input image to be recognized is uniformly divided into a preset S×S grid, forming S² non-overlapping grid cells in S rows and S columns. Each grid cell has a fixed pixel size and covers a local area of ​​the image.

[0058] In step S1102, target detection is performed on each grid cell to determine the candidate bounding box of each grid cell and the confidence level of the presence of a vehicle in each candidate bounding box.

[0059] For each grid cell, features are extracted using a convolutional neural network, and the positional parameters (including center offset, width, and height) of multiple candidate bounding boxes are directly regressed and predicted, along with a confidence score for each candidate bounding box indicating the presence of a vehicle. The confidence score comprehensively reflects both the probability of a vehicle being present in a candidate bounding box and the accuracy of the candidate bounding box's location.

[0060] In step S1103, candidate bounding boxes for each grid cell are filtered based on confidence level to determine the target bounding box for each grid cell.

[0061] Candidate bounding boxes are initially filtered based on a preset confidence threshold, eliminating low-quality candidate bounding boxes with confidence scores below the threshold. A non-maximum suppression algorithm is then used to further filter the remaining candidate bounding boxes whose overlap exceeds a preset intersection-union ratio (IU) threshold, retaining only the target bounding box with the highest confidence score to eliminate duplicate detections.

[0062] In step S1104, the center point coordinates or corner point coordinates of the target bounding box are determined as the pixel coordinates of the vehicle.

[0063] The position of the target bounding box is defined by the center point coordinates and width and height or corner point coordinates. The center point coordinates or specific target corner point coordinates of the bounding box are selected as the pixel coordinates of the vehicle in the image plane to be identified according to the vehicle positioning requirements.

[0064] Therefore, by dividing the image into grid cells and performing distributed target detection, combined with confidence threshold filtering and non-maximum suppression screening, the rapid localization and accurate pixel coordinate extraction of vehicles in the image to be identified are achieved.

[0065] Regarding step S120, the pixel coordinates of the vehicle are converted into the physical coordinates of the vehicle on the target structure. In some embodiments of this disclosure, this may include, for example... Figure 3 Steps S1201 to S1202 are shown.

[0066] In step S1201, the parameters of the spatial plane equation of the target structure are calibrated using calibration points with known physical coordinates on the target structure.

[0067] By using multiple calibration points with known physical coordinates on the target structure, and combining the spatial geometric constraints of the camera pinhole imaging model and the target structure surface, the planar parameters of the target structure's spatial plane equation and the similarity coefficients of the camera imaging model are calculated, thereby establishing a geometric mapping relationship between the image pixel space and the target structure's physical space.

[0068] In step S1202, for the vehicle pixel coordinates, the physical coordinates of the vehicle on the target structure are determined by solving the spatial plane equation of the target structure using the camera pinhole imaging model. The camera pinhole imaging model represents the projection relationship between the vehicle's pixel coordinates and physical coordinates.

[0069] The projection relationship between pixel coordinates and physical coordinates is established based on a camera pinhole imaging model, where the physical coordinates are expressed as a function of pixel coordinates, focal length, and similarity coefficient. For the vehicle's pixel coordinates, the camera pinhole imaging model is substituted to obtain a physical coordinate expression containing unknown similarity coefficients. Then, the unknown similarity coefficients are solved by simultaneously solving the spatial plane equations of the pre-calibrated target structure, and finally, the vehicle's physical coordinates on the target structure are determined by substituting these equations back into the model.

[0070] Therefore, by establishing a geometric mapping relationship between the image pixel space and the physical space of the target structure through calibration points, and by simultaneously solving the camera pinhole imaging model and the spatial plane equations, the vehicle pixel coordinates can be accurately converted into physical coordinates, and the spatial positioning of the vehicle load area can be obtained without additional auxiliary equipment.

[0071] Regarding step S130, the vehicle's physical coordinates are mapped to a set of degrees of freedom indexes, and a priori load information vector is constructed based on the set of degrees of freedom indexes. In some embodiments of this disclosure, this may include, for example... Figure 4 Steps S1301 to S1302 are shown.

[0072] In step S1301, a finite element model of the target structure is established, the physical coordinates of the vehicle are mapped to the finite element model, and the set of degrees of freedom indexes is determined.

[0073] Based on the actual geometric dimensions, material properties, and boundary conditions of the target structure, a finite element model is established, and the finite element model is meshed. All nodes in the finite element model used to simulate the bridge deck's load-bearing capacity are identified, and the degrees of freedom (DOF) of these nodes in the vertical direction are extracted to form a complete set of DDF indices.

[0074] Using the vehicle's physical coordinates as input, a search is performed within the geometric space of the finite element model to find one or more finite element nodes that are closest to or covered by those physical coordinates. The vertical degree of freedom numbers corresponding to these nodes are extracted to form a degree of freedom index set, which precisely identifies the location of the vehicle load in the finite element model.

[0075] In step S1302, a priori load information vector is constructed based on the degree-of-freedom index set. The number of vector elements in the priori load information vector is the same as the total number of degrees of freedom of the finite element model.

[0076] Construct a priori load information vector with the same dimension as the total number of degrees of freedom of the finite element model, and mark the elements in the priori load information vector corresponding to the degree of freedom index set as the possible action area of ​​the vehicle load.

[0077] Therefore, by establishing a finite element model of the target structure and mapping the vehicle's physical coordinates to the corresponding degree-of-freedom index set, a priori load information vector with the same dimension as the total number of degrees of freedom in the finite element model was constructed, thus realizing an accurate mapping from visual space to the structural mechanics analysis model.

[0078] Regarding step S140, an objective function is constructed based on the response observation equation of the target structure and the prior load information vector. In some embodiments of this disclosure, it may include, for example... Figure 5 Steps S1401 to S1402 are shown.

[0079] In step S1401, the structural response observation equation of the target structure is constructed based on the structural response of the target structure under unit load and the actual measured structural response.

[0080] The structural response acquired by sensors at various locations on the target structure under a unit load is calculated using a finite element model, forming an influence line matrix. The actual measured structural response is also obtained. Based on the influence line matrix and the measured response vector, an observation equation for the structural response is established.

[0081] Preferably, the structural response is the structural response of data collected from multiple sensors after normalization.

[0082] In step S1402, an objective function is constructed based on the prior load information vector and the structural response observation equation.

[0083] Therefore, by establishing a structural response observation equation based on the influence line matrix, a mathematical model of the relationship between the structural response and the load to be determined is achieved. Prior load information vectors are embedded into the objective function as regularization terms, explicitly incorporating the load position constraints acquired visually into the inversion process. This solves the ill-posed problem caused by the singularity of the influence line matrix, effectively suppresses load leakage into vehicle-free areas and overfitting, and improves the accuracy and stability of load identification.

[0084] Regarding step S150, the objective function is iteratively solved to determine the vehicle load. In some embodiments of this disclosure, this may include, for example... Figure 6 Steps S1501 to S1502 are shown.

[0085] In step S1501, the objective function is solved to determine the vehicle load estimate.

[0086] In step S1502, based on the load estimation, the vector elements in the prior load information vector corresponding to the degree of freedom index set are replaced with the values ​​at the corresponding positions in the load estimation until the loss function value is less than the target convergence threshold, thereby determining the vehicle load.

[0087] Based on the load estimate determined in the current iteration, the elements in the prior load information vector corresponding to the degree of freedom index positions are replaced with the values ​​of the load estimate at the corresponding positions, while the remaining positions remain zero. The loss function value, including the data residual term and the prior constraint term, is calculated. If the loss function value is less than the preset target convergence threshold, the iteration stops and the final load estimate result is output; otherwise, the prior load information vector is updated and the above solution process is repeated until the convergence condition is met.

[0088] Preferably, the objective function includes a regularization parameter, which is determined by constructing a double logarithmic curve using the data residual norm and the regularized denorm under different regularization parameters. The data residual norm represents the deviation between the structural response generated by the load estimation and the actual measured structural response, while the regularized denorm represents the deviation between the load estimation and the prior load information vector.

[0089] Specifically, the regularization parameters are determined using the L-curve method.

[0090] Specifically, the loss function value is: This represents the value of the loss function. Represents the influence line matrix. This represents the load estimation vector for the i-th iteration. This represents the load estimation vector for the (i-1)th iteration. This represents the actual measured structural response vector. This represents the dimensional balance factor.

[0091] Therefore, by iteratively updating the prior load information vector step by step and solving the objective function analytically, the prior constraints that initially only contain position markers and no numerical information gradually approach the actual load distribution. At the same time, the loss function value is used as the convergence criterion to ensure the stability and reliability of the iterative process, thereby improving the accuracy of vehicle load identification under conditions of limited number of sensors and environmental noise interference.

[0092] The technical solution of this disclosure will be further explained below with a specific implementation example.

[0093] Video images (i.e., images to be identified) are acquired using cameras mounted on the bridge (i.e., the target structure). A YOLO convolutional neural network is used to detect vehicles in the video images. The video images are divided into an S×S grid, and B candidate bounding boxes and confidence scores are predicted for each grid. The pixel coordinates (x′, y′) of the vehicle in the video image are output.

[0094] The workflow of the YOLO network is as follows: Figure 7As shown, the input video image is divided into S×S grid cells as input. Each grid cell is responsible for predicting multiple candidate bounding boxes and their confidence scores, while generating a class probability map to identify the area where the vehicle is located. Through confidence score filtering and non-maximum suppression processing, the final output is the target bounding box and class label (such as "truck 100%)" for vehicle detection results.

[0095] Based on the camera pinhole imaging model: (x, y, z) = (x′·t, y′·t, f·t), where f is the focal length and t is the similarity coefficient. The spatial plane equation of the bridge surface is pre-calibrated: Ax + By + Cz + D = 0. The corresponding similarity coefficients (t1, t2, t3, and t4) are calculated using four calibration points of known length on the bridge surface, thereby determining the planar parameters A, B, C, and D. For the detected vehicle pixel coordinates, the physical coordinates (x, y, z) of the vehicle on the bridge surface are solved by simultaneously solving the camera pinhole imaging model and the spatial plane equation of the bridge surface.

[0096] The principles of camera imaging and coordinate transformation are as follows: Figure 8 As shown, the left-hand 3D schematic represents a point P(x,y,z) in the physical space of the bridge surface projected onto a point P'(x',y') on the pixel plane through the camera's optical center. The projection relationship is determined by the focal length f and the similarity coefficient. The right-hand 2D schematic represents a point P' on the pixel plane and a point P in 3D space located on the same straight line passing through the camera's focal point. By combining this projection relationship with the pre-calibrated spatial plane equations of the bridge surface, the physical coordinates (x,y,z) of the vehicle on the bridge surface can be calculated from the pixel coordinates (x',y').

[0097] A finite element model of the bridge is established, and the vertical degrees of freedom of the bridge deck are numbered. Based on the vehicle's physical coordinates, these are mapped to the corresponding degree-of-freedom index set Ω. The prior load information vector F_ is initialized. prior 0 is an n×1 zero vector. The vector element corresponding to the degree of freedom index is set to a small positive number ε (e.g., 0.01). This vector indicates that there may be vehicle load at this location, but the specific value is unknown. The remaining vector elements are zero.

[0098] The response at each sensor location under a unit load is calculated using a finite element model, forming an influence line matrix W (e.g., strain, displacement, rotation angle, etc.). The actual measured response vector Y (which may contain data from multiple sensors, after normalization) is obtained. Then, Y = W·F, where Y represents the actual measured structural response, W represents the influence line matrix, and F represents the estimated nodal load vector to be solved.

[0099] Construct the objective function: F represents the estimated vector of nodal loads to be solved, Y represents the actual measured structural response, and W represents the influence line matrix. This represents the prior load information vector, where P represents the norm order. This represents the regularization parameter. Taking the derivative of the objective function and setting it to zero yields the analytical solution: This represents the transpose of the influence line matrix. Let represent the identity matrix. This analytical solution introduces prior information and, by adding the λI term, makes the originally singular WᵀW invertible, transforming the ill-posed problem into a well-posed one.

[0100] Since the initial prior load information vector only contains position information and the value is very small, it is necessary to iteratively approximate the true value.

[0101] Specifically, the i-th iteration is: by Calculate the current load estimate. This represents the prior load information vector for the (i-1)th iteration. This represents the load estimate obtained in the i-th solution.

[0102] Update the prior vector: Replace the vector elements of the prior load information vector of the i-th iteration in the degree-of-freedom index set with the corresponding values ​​of the load estimates of the i-th solution in the degree-of-freedom index set, and keep the positions of other vector elements at zero; Calculate the loss function value , This represents the value of the loss function. Represents the influence line matrix. This represents the load estimation vector for the i-th iteration. This represents the load estimation vector for the (i-1)th iteration. This represents the actual measured structural response vector. Indicates the dimensional balance factor; If the loss function value Δ < δ (δ is a preset small amount, such as 0.01 μF1ᵀF1, where F1 represents the load estimation vector obtained in the first iteration and F1ᵀ represents the transpose of the load estimation vector obtained in the first iteration), then stop the iteration and output the load estimate of the vehicle.

[0103] Preferably, the L-curve method is used to determine the regularization parameters.

[0104] Specifically, plotting under different regularization parameters ( , , Represents the norm of the data residuals. The double logarithmic curve (representing the prior constraint norm) is L-shaped, and the regularization parameter corresponding to its corner is the optimal value.

[0105] Alternatively, the objective function can be regularized using ℓ1 sparse regularization (suitable for sparse loads) or Bayesian regularization (suitable for high noise).

[0106] Optionally, the regularization parameter can be determined using generalized cross-validation (GCV).

[0107] Optionally, for concentrated load conditions, since the vehicle load actually acts on a local area of ​​the bridge deck, the estimated forces on multiple adjacent degrees of freedom should physically belong to the same concentrated load. Therefore, during the iterative update process, the estimated forces on multiple degrees of freedom near the loading area can be summed and merged into an equivalent concentrated force. This concentrated force is then updated to the center degree of freedom position of the prior load information vector, while the positions of the remaining adjacent degrees of freedom are set to zero. This force aggregation strategy reduces the number of iteration variables, accelerates the convergence speed, and better reflects the physical reality of vehicle axle load as a concentrated force.

[0108] Based on any of the above embodiments, this disclosure also provides a vehicle load identification device.

[0109] Figure 9 This is a schematic block diagram of the structure of a vehicle load identification device according to one embodiment of the present disclosure.

[0110] like Figure 9 As shown, the vehicle load identification device includes: The target detection module 9002 acquires an image of a vehicle traveling on a target structure, performs target detection on the vehicle in the image, and determines the pixel coordinates of the vehicle. The coordinate transformation module 9004 converts the vehicle's pixel coordinates into the vehicle's physical coordinates on the target structure. The prior vector construction module 9006 maps the vehicle's physical coordinates to the degree-of-freedom index set, and constructs a prior load information vector based on the degree-of-freedom index set. The degree-of-freedom index set represents the set of degree-of-freedom numbers of the vehicle in the finite element model of the target structure, and the prior load information vector is used to represent the area of ​​action of the vehicle's load on the target structure. The objective function construction module 9008 constructs the objective function based on the response observation equation of the target structure and the prior load information vector. The response observation equation is determined based on the structural response of the target structure under unit load and the actual measured structural response. The load estimation module 9010 iteratively solves the objective function to determine the vehicle's load.

[0111] The aforementioned vehicle load identification device can be in the form of computer software, and each module of the aforementioned vehicle load identification device can be implemented through computer software modules.

[0112] The specific implementation process of the functions and roles of each module in the above-mentioned vehicle load identification device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0113] The vehicle load identification method in the specific embodiments of this disclosure can be implemented by electronic devices such as servers, mobile terminals, edge computing terminals, smart cameras, or industrial control computers.

[0114] Therefore, based on any of the above embodiments, this disclosure also provides an electronic device that can execute the vehicle load identification method of any of the embodiments described above.

[0115] Figure 10 This is a schematic block diagram of an electronic device 1000 according to one embodiment of the present disclosure.

[0116] The hardware architecture of the electronic device 1000 can be implemented using a bus architecture. The bus architecture can include any number of interconnect buses and bridges, depending on the specific application of the hardware and overall design constraints. Bus 1100 connects various circuits, including one or more processors 1200, memory 1300, and / or hardware modules. Bus 1100 can also connect various other circuits 1400, such as peripheral devices, voltage regulators, power management circuits, external antennas, etc.

[0117] Bus 1100 can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, this diagram uses only one connection line, but this does not imply that there is only one bus or one type of bus.

[0118] This disclosure also provides a readable storage medium storing a computer program that, when executed by a processor, is used to implement the methods described above. A "readable storage medium" can be any means capable of containing, storing, communicating, propagating, or transmitting a program for use by or in conjunction with an instruction execution system, apparatus, or device. More specific examples of a readable storage medium include: an electrical connection with one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), fiber optic devices, and portable read-only memory (CDROM), etc.

[0119] This disclosure also provides a computer program product, the methods of which can be implemented wholly or partially through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented wholly or partially as a computer program product. The computer program product includes one or more computer programs or instructions. When the computer program or instructions are loaded and executed, all or part of the processes or functions of this disclosure are performed.

[0120] Computer programs or instructions can be stored in a readable storage medium or transferred from one readable storage medium to another. For example, the computer program or instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The readable storage medium can be any available medium capable of access, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium, such as a floppy disk, hard disk, or magnetic tape; an optical medium, such as a digital video optical disc; or a semiconductor medium, such as a solid-state drive. The computer-readable storage medium can be a volatile or non-volatile storage medium, or it can include both volatile and non-volatile types of storage media.

[0121] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, systems, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0122] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0123] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0124] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0125] In the description of this specification, the references to terms such as "one embodiment / mode," "some embodiments / modes," "example," "specific example," or "some examples," etc., refer to specific features, structures, or characteristics described in connection with that embodiment / mode or example, which are included in at least one embodiment / mode or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment / mode or example. Moreover, the specific features, structures, or characteristics described may be combined in any suitable manner in one or more embodiments / modes or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments / modes or examples described in this specification, as well as the features of different embodiments / modes or examples.

[0126] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this disclosure, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0127] Those skilled in the art should understand that the above embodiments are merely for illustrating the present disclosure and are not intended to limit the scope of the disclosure. Those skilled in the art can make other changes or modifications based on the above disclosure, and these changes or modifications still fall within the scope of the present disclosure.

Claims

1. A method for identifying vehicle load, characterized in that, include: Acquire an image of a vehicle traveling on a target structure, perform target detection on the vehicle in the image, and determine the pixel coordinates of the vehicle; Convert the pixel coordinates of the vehicle into the physical coordinates of the vehicle on the target structure; The physical coordinates of the vehicle are mapped to a set of degrees of freedom indexes, and a priori load information vector is constructed based on the set of degrees of freedom indexes. The set of degrees of freedom indexes represents the set of degrees of freedom numbers of the vehicle in the finite element model of the target structure, and the priori load information vector is used to represent the area of ​​action of the vehicle's load on the target structure. Based on the response observation equation of the target structure and the prior load information vector, an objective function is constructed. The response observation equation is determined based on the structural response of the target structure under unit load and the actual measured structural response. as well as The objective function is solved iteratively to determine the load on the vehicle.

2. The vehicle load identification method as described in claim 1, characterized in that, The objective function is iteratively solved to determine the vehicle's load, including: Solve the objective function to determine the load estimate of the vehicle; Based on the load estimation, the vector elements in the prior load information vector corresponding to the degree of freedom index set are replaced with the values ​​at the corresponding positions in the load estimation until the loss function value is less than the target convergence threshold, thereby determining the load of the vehicle.

3. The vehicle load identification method as described in claim 2, characterized in that, The objective function includes a regularization parameter, which is determined by constructing a double logarithmic curve using the data residual norm and the regularized denorm under different regularization parameters. The data residual norm represents the deviation between the structural response generated by the load estimation and the actual measured structural response, and the regularized denorm represents the deviation between the load estimation and the prior load information vector.

4. The vehicle load identification method as described in claim 1, characterized in that, Target detection is performed on the vehicle in the image to be identified, and the pixel coordinates of the vehicle are determined, including: The image to be identified is divided into grid cells to determine the grid cells of the image to be identified; Target detection is performed on each grid cell to determine the candidate bounding box of each grid cell and the confidence level of the presence of the vehicle in each candidate bounding box; Based on the confidence level, candidate bounding boxes for each grid cell are filtered to determine the target bounding box for each grid cell. The center point coordinates or corner point coordinates of the target bounding box are determined as the pixel coordinates of the vehicle.

5. The vehicle load identification method as described in claim 1, characterized in that, Converting the pixel coordinates of the vehicle to the physical coordinates of the vehicle on the target structure includes: The parameters of the spatial plane equation of the target structure are calibrated using calibration points with known physical coordinates on the target structure. For the vehicle's pixel coordinates, the physical coordinates of the vehicle on the target structure are determined by solving the spatial plane equation of the target structure using a camera pinhole imaging model. The camera pinhole imaging model represents the projection relationship between the vehicle's pixel coordinates and physical coordinates.

6. The vehicle load identification method as described in claim 1, characterized in that, Mapping the vehicle's physical coordinates to a set of degrees of freedom indexes, and constructing a priori load information vector based on the set of degrees of freedom indexes, including: Establish a finite element model of the target structure, map the physical coordinates of the vehicle to the finite element model, and determine the degree of freedom index set; A priori load information vector is constructed based on the aforementioned degree-of-freedom index set, wherein the number of vector elements in the priori load information vector is the same as the total number of degrees of freedom of the finite element model.

7. The vehicle load identification method as described in claim 1, characterized in that, Before constructing the objective function based on the response observation equation of the target structure and the prior load information vector, the following steps are taken: setting the vector elements in the prior load information vector corresponding to the degree of freedom index set to a preset small amplitude value, i.e., the first value, and setting the remaining vector elements to zero, i.e., the second value.

8. The vehicle load identification method as described in claim 1, characterized in that, Based on the response observation equation of the target structure and the prior load information vector, an objective function is constructed, including: Based on the structural response of the target structure under unit load and the actual measured structural response, the structural response observation equation of the target structure is constructed. Based on the prior load information vector and the structural response observation equation, an objective function is constructed.

9. An electronic device, characterized in that, include: The memory stores execution instructions; A processor that executes the execution instructions stored in the memory, causing the processor to perform the vehicle load identification method according to any one of claims 1 to 8.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the vehicle load identification method according to any one of claims 1 to 8.

Images