Road surface deformation quantity test method for road test detection

Abstract

The invention relates to the technical field of road engineering test detection, and discloses a road surface deformation quantity test method for road test detection, which comprises the following steps of: firstly, establishing a spatial domain discrete trigger model, converting continuous motion into a discrete sampling instruction by monitoring a driving distance, realizing synchronization of a collected action and a vehicle position, responding to the trigger instruction, and determining the deformation quantity of a road surface. Reference state image data and loading state image data for the same road surface area are sequentially collected and reconstructed into a three-dimensional point cloud model, a rigid reference area far away from a load center is defined in a measurement point cloud, the area features are matched with the reference point cloud, rigid body transformation parameters of a mobile carrier platform are calculated, and the rigid body transformation parameters are calculated. And carrying out reverse attitude correction on the measurement point cloud by using the parameter, and generating a pavement deformation field through differential operation. The invention relates to light source modulation based on vehicle speed so as to ensure imaging quality, a non-stress area is used as a static reference, vehicle body vibration interference and pavement tiny deformation are effectively decoupled, and high-precision dynamic deflection detection in a driving state is realized.

Description

Technical Field

[0001] This invention relates to the field of highway engineering testing and inspection technology, specifically to a method for testing pavement deformation in highway testing and inspection. Background Technology

[0002] With rapid economic development, highway transportation infrastructure construction has made leaps and bounds. Throughout the service life of highways, the continuous increase in traffic flow and vehicle axle load poses a severe challenge to the load-bearing capacity and durability of highway pavement structures. Because there is often a difference between the design life of a highway and its actual deformation, and this difference is difficult to verify effectively in the short term, maintenance departments cannot promptly obtain the evolutionary patterns of pavement structural performance, making it difficult to extend road life through early preventative maintenance. Therefore, employing scientific and effective testing methods to accurately detect the deformation of highway pavements, especially dynamic deflection, is of great significance for assessing pavement structural strength and making maintenance decisions.

[0003] However, existing highway pavement deformation detection technologies still have many limitations in practical applications. Traditional detection methods mostly rely on contact measurements or somewhat destructive invasive inspections. These methods not only easily cause irreparable physical damage to the pavement surface and compromise the integrity of the pavement structure, but also have significant bottlenecks in terms of ease of operation and measurement accuracy. They are also highly susceptible to human error and environmental factors, leading to a decrease in the confidence level of the detection data.

[0004] Furthermore, existing detection technologies generally suffer from low operational efficiency and significant traffic interference. Conventional methods typically require closed lanes for fixed-point measurements, making continuous and rapid detection under driving conditions difficult. This makes high-frequency, full-coverage, repeated detection over long road sections extremely challenging. This limitation not only significantly increases the time cost and operational risks for testing personnel, leading to high detection costs, but also, due to sparse sample data, fails to accurately capture the dynamic characteristics of road surface deformation, greatly reducing the utilization efficiency and data value of the detection methods. When attempting dynamic detection, existing technologies often struggle to effectively eliminate interference from the vibrations of the testing vehicle itself, failing to obtain pure road surface deformation information in complex driving environments, thus limiting the further development and application of highway pavement detection technology. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a road surface deformation testing method for highway testing, which solves the problems of low accuracy, poor efficiency, and high testing costs in non-contact road surface micro-deformation measurement due to the difficulty of effectively eliminating vehicle movement interference in dynamic driving environments.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The first aspect of the present invention provides a method for testing pavement deformation in highway testing, the method mainly comprising the following steps: First, a spatial domain discrete triggering model is established. By monitoring the travel distance of the mobile platform, the continuous travel motion in the time domain is transformed into discrete sampling trigger commands in the spatial domain, establishing a rigorous synchronization mechanism between the data acquisition action and the vehicle's travel position. This step aims to eliminate the impact of vehicle speed changes on the consistency of sampling positions and ensure that data acquisition follows a spatial position benchmark.

[0007] Secondly, dual-state road surface data acquisition is performed. In response to the discrete sampling trigger command, the system sequentially acquires reference-state image data and loaded-state image data for the same target road surface area. The reference-state data corresponds to the initial morphology of the road surface before tire load, while the loaded-state data corresponds to the stress morphology of the road surface under tire load.

[0008] Subsequently, 3D point cloud reconstruction is performed. The acquired reference state image data and loaded state image data are processed for 3D reconstruction to generate high-precision reference 3D point cloud models and measurement 3D point cloud models, thereby realizing the digital restoration of the road surface morphology.

[0009] Next, rigid body attitude decoupling and correction are performed. This is the core processing step of this method. The system determines a rigid reference region in the measured 3D point cloud model. This region is located outside the influence range of the tire load and is physically considered to be approximately free of vertical deformation. By spatially matching the feature data within the rigid reference region with the reference 3D point cloud model, the rigid body transformation parameters of the mobile platform caused by changes in vehicle attitude (such as pitch, roll, and heave) between two acquisition times are calculated.

[0010] Finally, the dense deformation field is calculated. The calculated rigid body transformation parameters are used to perform inverse attitude correction on the measured 3D point cloud model, eliminating the displacement components introduced by vehicle body motion. Based on this, the deformation field data purely generated by road surface forces is extracted through differential calculations between the corrected measured 3D point cloud model and the reference 3D point cloud model.

[0011] In a preferred embodiment, to ensure consistent imaging quality at different vehicle speeds, this method includes a vehicle speed-based light source modulation step. The system acquires the instantaneous speed of the moving carrier platform and determines the flicker frequency of the projected light source by the ratio of the instantaneous speed to a preset spatial sampling step size, ensuring a constant spatial sampling density. Simultaneously, the exposure time is controlled based on the instantaneous speed and optical magnification to meet motion blur constraints, ensuring that the optical pattern projected onto the road surface remains spatially phase-frozen at the moment of exposure, preventing a decrease in 3D reconstruction accuracy due to motion blur.

[0012] Regarding the specific implementation of spatial domain triggering, this invention accurately calculates the real-time driving distance using the wheel rolling radius and the pulse count fed back by the encoder. When the incremental driving distance reaches an integer multiple of the preset spatial sampling step size, reference-state acquisition is triggered and the current position is recorded. Subsequently, by calculating the difference between the current driving distance and the recorded position in real time, loading-state acquisition is triggered when this difference equals the physical baseline distance between the front and rear acquisition devices. This logic ensures that the two acquisitions accurately cover the same physical road surface area, achieving distance-based triggering rather than time-based triggering.

[0013] At the algorithmic level of rigid body attitude decoupling, to accurately distinguish between deformable and non-deformable regions, this method employs distance-based region partitioning logic. The system determines the tire loading center and a preset critical influence radius, and calculates the horizontal Euclidean distance from each point in the point cloud to the loading center. Regions with distances greater than the critical influence radius are marked as rigid reference regions, which are assumed to have zero vertical displacement and contain only rigid body motion components; regions with distances less than or equal to the critical influence radius are marked as deformation observation regions.

[0014] Based on the aforementioned region division, this method constructs a target loss function, which characterizes the Euclidean distance error between feature points within the rigid reference region and their corresponding points on the reference model after rigid body transformation. By minimizing this target loss function, the optimal rigid body transformation parameters, including rotation matrices and translation vectors, are solved. To improve the robustness of the solution, the algorithm employs a random sample consensus algorithm to eliminate mismatched points, uses singular value decomposition to obtain the initial solution, and uses an iterative nearest-point algorithm for fine-tuning iterations until the error converges.

[0015] To enhance the matching accuracy of poorly textured road surfaces, the image data includes pseudo-random speckle patterns or phase-shifted fringe patterns projected onto the road surface. Based on these artificial textures, feature descriptors containing three-dimensional geometric coordinates, reflection intensity, and local normal vector information are constructed, thereby establishing a highly reliable spatial matching relationship within a rigid reference region.

[0016] In the final deformation field generation stage, attitude correction is achieved by subtracting the translation vector from the coordinates of the measured 3D point cloud model and multiplying by the inverse of the rotation matrix. Subsequently, a 2D virtual mesh is constructed under a unified coordinate reference, and the elevation matrix is ​​obtained by interpolation and resampling of the two sets of point clouds. The difference matrix is ​​obtained by subtracting the elevation matrix, which is the dynamic deflection distribution field of the road surface. Based on this, the maximum central deflection value and the deflection basin profile curve are extracted.

[0017] This invention provides a method for testing pavement deformation in highway testing. It has the following beneficial effects: 1. This invention defines a rigid reference region in the measurement three-dimensional point cloud model, and uses the relative static characteristics of this non-stressed region to establish spatial matching with the reference three-dimensional point cloud, accurately calculating the rigid body transformation parameters of the mobile carrier platform. This method can effectively decouple the rigid body motion components such as pitch, roll, and heave caused by vehicle vibration from the actual stress deformation of the road surface, eliminating the interference of vehicle dynamic driving on the measurement of small deflection, and realizing high signal-to-noise ratio extraction of the road surface deformation field under driving conditions.

[0018] 2. This invention establishes a spatial domain discrete triggering model coupled with vehicle speed, which transforms continuous motion in the time domain into discrete sampling commands based on driving distance. Combined with the asynchronous acquisition of dual-state data controlled by physical baseline distance, this mechanism ensures that the reference state and loaded state image data accurately cover the same physical road surface area, avoids misalignment of front and rear sampling positions caused by vehicle speed fluctuations, guarantees the consistency of spatial reference for differential operations, and improves the positioning accuracy and data reproducibility of dynamic detection.

[0019] 3. This invention adopts a light source modulation strategy based on vehicle speed. It dynamically adjusts the flicker frequency of the projected light source and the exposure time of image acquisition according to the instantaneous speed. By maintaining a constant spatial sampling density and limiting the maximum blurred pixel size, it effectively solves the problems of data sparsity and motion blur caused by the speed change of the moving carrier platform. It ensures that the optical pattern projected on the road surface still maintains clear spatial phase characteristics under high-speed movement, thereby ensuring the integrity and accuracy of the 3D point cloud reconstruction. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram of the algorithm for optimizing rigid body motion parameters according to the present invention. Detailed Implementation

[0021] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] This invention provides a road surface deformation test method for highway testing, which is based on a specific vehicle-mounted testing system.

[0023] The vehicle-mounted inspection system includes a mobile carrier platform, a first vision acquisition module, a second vision acquisition module, a displacement sensing unit, and a central processing unit. The mobile carrier platform is an inspection vehicle with a standard axle load, and its chassis structure includes a front axle area and a rear axle area. The tires in the rear axle area serve as the loading source for applying a standard load to the road surface, causing instantaneous elastic deformation of the road surface.

[0024] The first visual acquisition module is rigidly mounted at the front of the chassis of the mobile platform, located between the front and rear axles, and in front of the rear axle tires in the direction of travel. The field of view of the first visual acquisition module is set to cover the road surface area before the vehicle passes, that is, the original state area of ​​the road surface unaffected by the rear axle load. The first visual acquisition module includes a first high-speed industrial camera and a first coded structured light projector. The first high-speed industrial camera and the first coded structured light projector maintain a fixed relative position.

[0025] The second vision acquisition module is rigidly mounted on the rear axle wheel clearance or on the outer rear wheel bracket of the mobile platform. The field of view of the second vision acquisition module is set to include the road surface area surrounding the rear axle tire contact point. This field of view is divided into two logical regions: a deformation observation area and a rigid reference area. The deformation observation area corresponds to the tire contact surface and the area affected by load-induced deflection; the rigid reference area corresponds to the area at the edge of the field of view far from the tire contact surface and where the load effect is negligible. The second vision acquisition module includes a second high-speed industrial camera and a second coded structured light projector. The second high-speed industrial camera and the second coded structured light projector maintain a fixed relative position.

[0026] Both the first and second coded structured light projectors are used to project optical patterns with specific coded features onto the road surface. These optical patterns are selected from phase-shifted fringe patterns, Gray code patterns, or pseudo-random speckle patterns. Both the first and second high-speed industrial cameras are equipped with external trigger interfaces to receive synchronous exposure signals from the central processing unit.

[0027] The displacement sensing unit is mounted at the center of the wheel hub of the mobile platform to measure the vehicle's travel distance and instantaneous speed in real time. The displacement sensing unit uses a high-resolution rotary encoder to output pulse signals as the wheels rotate. These pulse signals are transmitted to the central processing unit via signal transmission cables.

[0028] The central processing unit (CPU) is electrically connected to the first vision acquisition module, the second vision acquisition module, and the displacement sensing unit. The CPU receives pulse signals from the displacement sensing unit, calculates the trigger time based on the pulse signals, and sends trigger commands to the first and second vision acquisition modules. The CPU also receives image data acquired by the first and second high-speed industrial cameras and executes 3D reconstruction and differential calculation algorithms.

[0029] To describe the spatial location of each component and road surface deformation data, the system defines the following spatial coordinate system: World Coordinate System Reference camera coordinate system and measuring camera coordinate system World coordinate system Defined on a stationary plane of the road surface, with the Z-axis perpendicular to the road surface and pointing upwards. Reference camera coordinate system. The origin is located at the optical center of the first high-speed industrial camera. The axis points along the camera's optical axis towards the road surface. (Measurement camera coordinate system) The origin is located at the optical center of the second high-speed industrial camera, and the Z-axis points along the camera's optical axis toward the road surface.

[0030] Under the condition that the system is stationary and the road surface is flat and deformation-free, the reference camera coordinate system With the measuring camera coordinate system The relative pose relationship between them is determined in advance through calibration. This relative pose relationship is represented by a transformation matrix. It includes rotation matrix components and translation vector components.

[0031] The imaging models of the first and second high-speed industrial cameras follow a combination of the pinhole camera model and a distortion correction model. For any pixel on the image plane... Its corresponding three-dimensional space point The following mapping relationship is satisfied: ; in, Indicates the scale factor; and Represents the x and y coordinates of a pixel in the image coordinate system; This represents the camera's intrinsic parameter matrix, which includes focal length, principal point coordinates, and tilt factor. This represents the rotation matrix of the camera coordinate system relative to the world coordinate system; This represents the translation vector of the camera coordinate system relative to the world coordinate system; , , This represents the coordinates of a point in three-dimensional space in the world coordinate system.

[0032] The central processing unit stores a pre-calibrated intrinsic parameter matrix of the first high-speed industrial camera. Second high-speed industrial camera intrinsic parameter matrix The relative structural parameters of the first coded structured light projector and the first high-speed industrial camera, and the relative structural parameters of the second coded structured light projector and the second high-speed industrial camera are used to support subsequent 3D point cloud reconstruction calculations.

[0033] The physical installation distance between the first vision acquisition module and the second vision acquisition module in the vehicle's driving direction is defined as the baseline distance. The baseline distance refers to the straight-line distance between the vertical projection points of the optical centers of the first high-speed industrial camera and the second high-speed industrial camera on the road surface. The central processing unit utilizes this baseline distance. By combining the data from the displacement sensing unit, the first vision acquisition module and the second vision acquisition module are controlled to asynchronously acquire data from the same physical road surface area.

[0034] See attached document Figure 1 This invention provides a road surface deformation test method for highway testing. This method establishes a spatiotemporal synchronization mechanism between vehicle dynamics and optical acquisition system, and uses the texture features of non-deformed areas of the road surface for self-reference pose correction, thereby achieving accurate measurement of the vertical deformation of the road surface under dynamic load.

[0035] The method includes steps S100 to S500.

[0036] In step S100, a spatial domain discrete triggering model coupled with vehicle speed is established. The central processing unit monitors the driving status of the mobile carrier platform in real time through the displacement sensing unit, converting the continuous motion in the time domain into discrete sampling points in the spatial domain. This step involves generating a synchronous trigger signal according to a preset spatial sampling step size, and modulating the stroboscopic frequency of the first and second coded structured light projectors in real time according to the instantaneous driving speed to ensure that the optical pattern projected onto the road surface maintains spatial phase freeze relative to the road surface texture during the exposure cycle.

[0037] In step S200, asynchronous pipeline acquisition of dual-state road surface data is performed. This step utilizes the movement of the mobile platform to acquire reference-state image data of a specific road surface area at a first moment using a first vision acquisition module. This reference-state image data represents the initial morphology of the road surface without rear axle load. Subsequently, loaded-state image data of the same physical road surface area is acquired at a second moment using a second vision acquisition module. This loaded-state image data represents the stress morphology of the road surface under rear axle tire load. The central processing unit controls the time interval between the first and second moments to precisely correspond to the time required for the vehicle to pass the installation baseline distance between the first and second vision acquisition modules.

[0038] In step S300, a 3D point cloud is reconstructed based on the principle of structured light triangulation. The central processing unit calls the pre-stored camera intrinsic parameter matrix and structured light system calibration parameters to process the reference state image data and the loaded state image data acquired in step S200. This processing includes decoding the coded grating features in the image, extracting the sub-pixel coordinates of the feature points, and using a triangulation algorithm to map the 2D image coordinates to 3D spatial coordinates, thereby generating a reference 3D point cloud model representing the unloaded state and a measurement 3D point cloud model representing the loaded state, respectively.

[0039] In step S400, rigid body attitude decoupling and correction are performed based on the features of the non-deformation zone. This step defines logical partitions for the measured 3D point cloud model, dividing it into a deformation observation zone containing the tire's action area and a rigid reference zone far from the tire's action area. The central processing unit extracts road texture features and structured light features within the rigid reference zone and performs spatial matching with the corresponding regions in the reference 3D point cloud model. By minimizing the spatial distance error between feature point pairs within the rigid reference zone, the six-degree-of-freedom rigid body transformation parameters of the mobile platform caused by vehicle vibration, pitch, or roll between the first and second time moments are calculated, including the rotation matrix and translation vector.

[0040] In step S500, the dense deformation field is calculated and the deflection detection results are output. This step utilizes the rigid body transformation parameters calculated in step S400 to perform an inverse rigid body transformation on the measured 3D point cloud model, unifying it to the coordinate system reference of the reference 3D point cloud model, thus eliminating the pose error introduced by vehicle motion. Subsequently, under the unified coordinate reference, the coordinate difference between corresponding spatial points in the deformation observation area in the direction perpendicular to the road surface is calculated, generating the dynamic deflection distribution field of the road surface. This distribution field is output in the form of a 3D surface or a data matrix, used to characterize the mechanical response characteristics of the highway pavement under standard axle load. Step S100 specifically involves converting the vehicle's continuous mechanical motion into discrete optical acquisition commands and modulating lighting parameters through real-time speed feedback. The central processing unit (CPU) first initializes a first-in-first-out (FIFO) position queue for storing trigger position information. When the mobile platform begins to move, the CPU receives pulse signals from the displacement sensing unit in real time via an interrupt service routine. The displacement sensing unit uses an incremental rotary encoder coaxially connected to the wheel hub, generating A / B phase quadrature pulses as the wheel rotates. The CPU counts the pulse signals and converts the accumulated pulse count into the real-time travel distance of the mobile platform relative to the starting point. Real-time driving distance The calculation follows the following relationship: ; in, express Real-time driving distance at any given moment; This indicates the rolling radius of the wheel on which the displacement sensing unit is installed; express The total number of valid pulses read at any given time; This indicates the resolution of the number of pulses output per revolution of the rotary encoder.

[0041] Based on the real-time driving distance, the central processing unit does not generate trigger signals according to fixed time intervals, but rather according to a preset spatial sampling step size. Spatial domain discretization decision is performed. The spatial sampling step size is... Based on the minimum resolution requirement of the road surface texture to be measured and the projection coverage of the first coded structured light projector, a predetermined overlap rate is ensured between adjacent acquisition frames. When the real-time driving distance is monitored... The increment reached When the value is an integer multiple of the specified value, the central processing unit immediately generates the first trigger signal.

[0042] The first trigger signal is synchronously sent via a hardware trigger circuit to the first high-speed industrial camera and the first coded structured light projector in the first vision acquisition module. Within a microsecond delay of receiving the signal, the first high-speed industrial camera opens its electronic shutter, and the first coded structured light projector is simultaneously illuminated and projects a coded pattern. At the same time, the central processing unit calculates the absolute travel distance corresponding to the current trigger moment. Record and push the location into the position queue.

[0043] For the trigger control of the second vision acquisition module, the central processing unit continuously polls the position queue in a parallel processing thread. This thread calculates the current driving distance in real time. Distance from the head of the position queue The difference. When this difference is equal to the physical baseline distance between the first vision acquisition module and the second vision acquisition module. When, that is, satisfied When the conditions are met, the central processing unit determines that the second vision acquisition module has reached the location of the first vision acquisition module. At the same physical location at that moment, the central processing unit generates a second trigger signal, sends it to the second vision acquisition module, and simultaneously... Popped from the position queue. This logic ensures precise spatial alignment of the front and rear cameras, eliminating timeline alignment errors caused by vehicle speed fluctuations.

[0044] To address the blurring issue caused by the relative motion between the structured light pattern and the road surface at high speeds, the central processing unit (CPU) also implements a vehicle speed-coupled strobe control strategy. The CPU calculates the vehicle's instantaneous speed by differentiating the position pulses. Based on this instantaneous speed, the central processing unit dynamically adjusts the flicker drive frequency of the coded structured light projector. and the camera's exposure time Strobe drive frequency With instantaneous speed Maintain the following linear coupling relationship: ; in, Indicates the trigger frequency of the projector and camera; Indicates the instantaneous speed of the vehicle; This indicates the aforementioned spatial sampling step size. Through this coupling control, the system ensures that the sampling point density on the road surface remains constant regardless of changes in vehicle speed.

[0045] At the same time, in order to ensure image sharpness, the camera's exposure time... Limited by instantaneous speed constraints, the central processing unit sets the exposure time to not exceed the maximum motion blur allowed by the system. The threshold value is determined. This constraint is expressed as: ; in, Indicates the exposure time of a single frame of an image; This represents the maximum allowable blurred pixel size on the image sensor plane, and is typically less than 1 pixel width. The instantaneous speed of the vehicle; This refers to the imaging magnification of the optical system. The central processing unit adjusts the camera's electronic shutter speed to always satisfy this inequality condition, thereby freezing the structured light texture projected onto the road surface with an extremely short exposure time when the vehicle is traveling at high speed, preventing grating phase slip or contrast reduction caused by relative motion.

[0046] Steps S200 and S300 specifically involve using front-and-back distributed visual modules, driven by a spatiotemporal synchronization mechanism, to acquire raw geometric data characterizing the mechanical response characteristics of the road surface, and to transform the two-dimensional image information into a three-dimensional point cloud model with physical metrics.

[0047] These two steps are divided into an asynchronous data acquisition phase and a three-dimensional spatial reconstruction phase.

[0048] During the asynchronous data acquisition phase, the system executes a location-triggered pipeline operation. When the mobile platform reaches a specific spatial sampling location, the central processing unit first activates the first vision acquisition module located at the front of the vehicle. A first coded structured light projector projects a frame of coded optical pattern with high-contrast features onto the road surface within the field of view. This coded optical pattern preferably employs pseudo-random speckle or phase-shifting fringes to construct artificial feature textures on asphalt or cement roads with uniform optical texture. A first high-speed industrial camera simultaneously exposes and acquires an image containing this coded feature, denoted as the reference image. At this point, the area of ​​the road surface being photographed has not yet entered the stress influence range of the rear axle tires, and the road surface is in a naturally relaxed state with zero deformation.

[0049] As the mobile platform continues to move, when the central processing unit determines that the rear axle tires have accurately moved to the same physical road surface location captured by the first vision acquisition module, it triggers the second vision acquisition module located at the rear axle. A second coded structured light projector projects an optical pattern onto the same road surface area that is spatially phase-consistent with or has a known mapping relationship to the aforementioned coded features. A second high-speed industrial camera simultaneously exposes and acquires an image of that location, denoted as the loaded image. At this moment, the photographed road surface area is under the vertical force of the standard axle load of the rear axle tires, causing the road structure to generate an elastic deflection bowl. It should be noted that this is a loaded image. The recorded geometric shape is the result of the superposition of the road surface elastic deformation and the rigid body posture change of the vehicle body caused by driving vibration.

[0050] During the 3D spatial reconstruction stage, the central processing unit utilizes the active triangulation principle of structured light to analyze the reference image. and loaded state image The processing is performed in parallel. This process aims to map the pixel coordinates on the image plane back to the three-dimensional world coordinate system.

[0051] First, the central processing unit decodes the coded optical patterns in the image. For pseudo-random speckle patterns, the disparity of feature points is determined using a digital image correlation (DIC) subset matching algorithm; for phase-shifted fringe patterns, the absolute phase value of each pixel is calculated using a phase unwrapping algorithm. This decoding process establishes the pixel values ​​on the camera image plane. A unique correspondence between the light stripes or spots and the light on the structured light projection plane. Subsequently, based on pre-calibrated system parameters, including the camera intrinsic matrix... Lens distortion coefficient and rotation / translation matrix between the camera optical center and the projector optical center. The three-dimensional spatial coordinates of each valid pixel are calculated using the ray triangulation method. For any pixel successfully decoded in the image, its three-dimensional coordinate vector... The solution follows the following inverse projection transformation model: ; in, This represents the three-dimensional coordinate vector of a point on the road surface in the camera coordinate system. ; This represents the depth scaling factor, which is uniquely determined by the intersection of the equation of the projector's projection plane and the camera's line-of-sight vector. The inverse matrix representing the camera intrinsic parameter matrix; Represents the normalized homogeneous pixel coordinates after distortion correction. .

[0052] Through the above calculations, the central processing unit will process the reference state image. Convert to reference 3D point cloud model Loading state image Transform into a measurement 3D point cloud model .

[0053] Reference 3D point cloud model It is a dense point cloud collection that accurately describes the microscopic texture and macroscopic undulations of the road surface at an unloaded time. Measurement of the 3D point cloud model. Similarly, this is a dense point cloud set describing the surface morphology of the road surface at the moment of loading. To facilitate subsequent differential calculations, the system preprocesses the generated point cloud, including outlier removal and downsampling, to remove noise data caused by road surface reflections or shadows, and to unify the data density of the two sets of point clouds. These two 3D point cloud models form the data foundation for subsequent separation of vehicle body vibration errors and the true deformation of the road surface.

[0054] In step S400, in order to achieve high-precision rigid body attitude decoupling, it is necessary to measure the three-dimensional point cloud model. The process clearly distinguishes between reference data used to calculate pose and observation data used to calculate deformation. This step involves spatial region logical segmentation techniques based on pavement mechanical response characteristics.

[0055] The central processing unit measures the 3D point cloud model. The execution logic partition is based on the application of Saint-Venant's principle in road engineering, which states that the vertical deformation effect of wheel load on the road surface is limited to a certain radius centered on the loading point. The road surface area outside this range can be regarded as a rigid body at the moment of loading, and its vertical displacement relative to the geodetic coordinate system is zero.

[0056] First, the system needs to determine the coordinates of the loading center. During the system calibration phase, the projected coordinates of the rear axle tire contact point in the second vision acquisition module coordinate system are determined through physical measurements or with the aid of a calibration board, and this coordinate is denoted as the loading center. The loading center is the geometric pole of the deflection basin and also the point where the pavement deformation gradient reaches its maximum value.

[0057] Next, the central processing unit defines a critical influence radius. The value of this critical influence radius is determined based on the pavement structure type of the highway under test (such as rigid or flexible pavement) and the design axle load standard, and is usually set as the theoretical maximum edge radius of the deflection basin. In the design of the field of view, the width and length of the field of view of the second vision acquisition module must be greater than the coverage area of ​​this critical influence radius to ensure that the edge of the field of view contains a sufficient number of non-deformation points.

[0058] Based on Load Center and critical influence radius The central processing unit generates a two-dimensional logic mask. This mask is a matrix corresponding to the image pixel resolution, used to mark the functional region to which each spatial point belongs. For measuring 3D point cloud models... any point in Calculate the Euclidean distance from its projection point on the horizontal plane to the loading center. ; in, Indicates matching point With the center point Geometric distance between them Indicates the first The coordinates of the matching points (usually the points to be observed or pixels), The coordinates of the center point.

[0059] rigid reference region subset According to distance With critical influence radius The comparison results divide the point cloud data into two mutually exclusive subsets: When the condition is met At that time, determine the point Located outside the area affected by road surface loads. Point cloud data within this region is labeled as rigid reference data. Physically, the road surface in this region is relative to the reference state at the time of acquisition ( At any given moment, no elastoplastic deformation occurred; its coordinate changes in the measuring camera coordinate system were entirely caused by the rigid body motion of the detected vehicle body (such as body swaying or bumping). Therefore, It is the only data source for subsequent calculations of the rigid body transformation matrix.

[0060] Deformation observation region subset When the condition is met At that time, determine the point Located within the load-bearing area of ​​the road surface, specifically in the deflection basin region, the point cloud data within this area is labeled as deformation observation data. This region contains the actual mechanical response information of the road surface under tire pressure. In subsequent pose correction steps, the data in this region is strictly masked and not included in the regression calculation of rigid body motion parameters to prevent road surface deformation from interfering with the accuracy of vehicle body attitude calculation.

[0061] Furthermore, to further improve the robustness of the rigid reference region characteristics, the central processing unit can also introduce a confidence filtering mechanism. In the partitioning... During this process, outliers in the point cloud with excessively large abrupt changes in normal vectors (corresponding to road surface cracks or foreign objects) or excessively low reflection intensity (corresponding to water accumulation or oil stains) are removed, retaining only points with smooth surfaces and clear texture features as the final attitude calculation anchor points. Through the above logical division, the system mathematically decouples the complex field data, which mixes vehicle motion and road surface deformation, into a background field for self-calibration and a target field for numerical measurement.

[0062] Following the aforementioned field-of-view segmentation steps, the central processing unit operates within the defined rigid reference region. The internal feature extraction and matching operations are performed to establish a reference 3D point cloud model. With measurement of 3D point cloud models The precise point-to-point correspondence between them. Due to the differences in time and perspective between the two acquisitions, and the fact that the road surface texture is usually relatively simple, the matching process relies heavily on the artificial texture features constructed by the coded structured light patterns projected in the preceding steps.

[0063] The central processing unit first performs feature descriptor calculations on the point cloud data within the rigid reference region. For and For each spatial point, the system utilizes not only its three-dimensional geometric coordinates but also its corresponding reflection intensity and local normal vector to generate a high-dimensional feature descriptor vector. This feature descriptor preferably employs the Fast Point Feature Histogram (FPFH) or Signature Histogram (SHOT) algorithm to ensure the invariance of features under rotation and translation transformations. Due to the projection of speckle or stripe patterns with spatial phase freezing, each point on the road surface is endowed with a unique optical fingerprint, resulting in highly distinctive feature descriptions.

[0064] Subsequently, the central processing unit performs a nearest neighbor search in the feature space to analyze the measured point cloud model. Feature points and reference point cloud model within a rigid reference region The system performs feature point matching across the entire field. To eliminate the interference of mismatches on pose calculation, the system introduces the Random Sample Consensus (RANSAC) algorithm for geometric constraint filtering. The system randomly selects a subset of point pairs to calculate the transformation hypothesis and verifies the interior point rate of the remaining point pairs, iteratively eliminating abnormal matching pairs that, although feature-similar, do not satisfy rigid body consistency in geometric space. After filtering, the system outputs a set of high-confidence corresponding point sets. , which includes For each matching point, each pair is represented as , respectively, represent the three-dimensional position of the same physical point on the road surface in the reference coordinate system and the measurement coordinate system.

[0065] See attached document Figure 2 Based on the set of corresponding points with the same name This step aims to solve the problem of mobile carrier platforms in... Time to The relative positional changes caused by vibrations during driving at different times.

[0066] The central processing unit constructs a least-squares optimization model with the physical objective of finding an optimal rigid body transformation matrix such that the rigid reference region in the measured point cloud, after transformation, perfectly coincides spatially with the corresponding region in the reference point cloud. This rigid body transformation matrix consists of a rotation matrix. Translation vector Composition. Among them, It characterizes the vehicle's pitch, roll, and yaw attitude changes. This characterizes the vehicle's vertical heave, lateral offset, and longitudinal displacement. A target loss function is defined. The sum of squared Euclidean distances between all matching pairs: ; in, For the first A column vector of coordinates of each matching point in the reference camera coordinate system; For the total number of matching point pairs, The square of the L2 norm.

[0067] To find the minimum of the objective function, the central processing unit employs either Singular Value Decomposition (SVD) or the Levenberg-Marquardt nonlinear iterative algorithm. First, the centroids of the two point sets are calculated. The influence of translation components is eliminated through centroid decomposition. Then, the optimal rotation matrix is ​​directly obtained analytically from the covariance matrix using SVD decomposition. Then, substituting back into the centroid formula, the optimal translation vector is solved. .

[0068] Furthermore, to further improve the solution accuracy, the system executes the Iterative Closest Point (ICP) fine registration algorithm after obtaining the analytical solution. (The analytical solution is then used for further processing.) As initial values, point-to-plane error minimization iterations are performed within a rigid reference region until the error converges to a preset threshold range. The final calculated value... This precisely describes the rigid body motion components of the vehicle within the interval between two shots. This parameter will serve as a key correction factor in subsequent steps to eliminate systematic errors and restore the true deformation of the road surface. Through this self-referenced solution method based on the characteristics of the non-deformation zone of the road surface, the system effectively decouples the complex dynamic motion of the vehicle itself from the minute mechanical deformation of the road surface mathematically.

[0069] Step S500 specifically performs deformation field calculation based on a unified spatial reference. The core of this step lies in accurately extracting the microscopic elastic displacement of the road surface in the vertical direction from complex geometric measurement data after eliminating systematic interference from vehicle motion attitude. The central processing unit first performs inverse spatial correction of the measurement data. It then reads the optimal rigid body transformation parameters, i.e., the rotation matrix, calculated in step S400. Translation vector These two parameters quantitatively describe the six-degree-of-freedom attitude deviation of the vehicle at the moment the loaded state image was acquired, relative to the moment the reference state image was acquired. To measure the three-dimensional point cloud model of the loaded state... With reference 3D point cloud model When compared within the same stationary reference frame, the central processing unit... All three-dimensional spatial points are subjected to inverse rigid body transformation. The corrected coordinates of the measured points are then obtained. Follow the following transformation logic: ; in, The coordinates of any point in the original measured point cloud; It is the inverse of the rotation matrix (for an orthogonal matrix, its inverse is equal to its transpose); This is a translation vector. After this inverse transformation operation, the measurement point cloud belongs to the rigid reference region ( The portion of it spatially overlaps with the reference point cloud and belongs to the deformation observation area. The portion that retains the relative displacement information with respect to the reference plane.

[0070] After spatial alignment, given the spatial discreteness and unstructured nature of point cloud data, direct point-to-point subtraction cannot obtain a continuous deformation field. Therefore, the central processing unit performs resampling and interpolation processing based on a virtual mesh. The system operates in the reference camera coordinate system... A regular two-dimensional grid with a preset resolution is constructed on a plane, covering the entire deformation observation area. For each node in the grid... The central processing unit respectively references the 3D point cloud model and the corrected measurement 3D point cloud model The system searches for neighboring points and calculates the vertical elevation at each node using either inverse distance weighted interpolation (IDW) or kriging interpolation. This generates two digital elevation matrices corresponding to the same grid coordinate system: a reference elevation matrix and a reference elevation matrix. and loading elevation matrix .

[0071] Subsequently, the central processing unit calculates the dense deformation field matrix. This calculation process involves more than just simple numerical subtraction; it also includes smoothing to account for road surface texture noise. For any grid point within the deformation observation area, its dynamic deflection value... The calculation is as follows: ; Calculated The matrix is ​​a three-dimensional digital representation of the dynamic deflection basin of the road surface. In this matrix, negative values ​​indicate that the road surface has undergone subsidence displacement, and positive values ​​indicate that the road surface has undergone heave displacement (usually occurring in the heave area at the edge of the deflection basin).

[0072] In order to output test indicators that meet traffic engineering standards, the central processing unit... The system extracts feature parameters from the matrix. First, it iterates through the matrix to find the point with the largest absolute negative displacement, defining it as the maximum central deflection value. The system records the planar coordinates of the load center. Then, using this center as the origin, deflection profile curves are extracted along the vehicle's travel direction (longitudinal) and perpendicular to the travel direction (lateral). Based on these profile curves, the system further calculates the radius of curvature, radius of diffusion, and steepness of the deflection basin. Finally, the central processing unit packages the calculated three-dimensional deflection field data with the corresponding vehicle position, speed, and ambient temperature information to generate a complete test record for a single measurement, which is then uploaded to a database or display terminal via the vehicle communication interface. Through this entire process, the present invention achieves non-contact, full-field measurement of micron-level elastic deformation of the road surface during dynamic vehicle movement.

Claims

1. A method for testing pavement deformation in highway testing, characterized in that, Includes the following steps: Step S100: By monitoring the travel distance of the mobile carrier platform, the continuous travel motion in the time domain is converted into discrete sampling trigger commands in the spatial domain, and a synchronization mechanism between the acquisition action and the vehicle's travel position is established. Step S200: In response to the discrete sampling trigger command, reference state image data and loaded state image data for the same target road surface area are acquired sequentially; Step S300: By performing three-dimensional reconstruction processing on the reference state image data and the loaded state image data, a reference three-dimensional point cloud model representing the unstressed state of the road surface and a measurement three-dimensional point cloud model representing the stressed state of the road surface are generated. Step S400: Determine a rigid reference region in the measured three-dimensional point cloud model, and calculate the rigid body transformation parameters of the mobile carrier platform during the acquisition process by spatial matching between the feature data in the rigid reference region and the reference three-dimensional point cloud model. Step S500: The attitude of the measured three-dimensional point cloud model is corrected using the rigid body transformation parameters, and the road surface deformation field is calculated by the difference operation between the corrected measured three-dimensional point cloud model and the reference three-dimensional point cloud model.

2. The method for testing pavement deformation according to claim 1, characterized in that, The road surface deformation test method further includes a vehicle speed-based light source modulation step, which includes: Obtain the instantaneous speed of the mobile carrier platform; The flicker frequency of the projected light source is calculated by dividing the instantaneous velocity by a preset spatial sampling step size. Simultaneously, the exposure time for image acquisition is controlled so that the exposure time is less than or equal to the maximum allowable blurred pixel size divided by the product of the instantaneous speed and the optical imaging magnification.

3. The method for testing pavement deformation according to claim 1, characterized in that, In step S100, the specific process of converting continuous driving motion in the time domain into discrete sampling trigger commands in the spatial domain is as follows: The real-time travel distance of the mobile carrier platform is calculated using the wheel rolling radius and the pulse count fed back by the encoder. When the increment of the real-time driving distance is detected to be an integer multiple of the preset spatial sampling step size, a first trigger signal is generated to control the acquisition action of the reference state image data, and the distance value corresponding to the current trigger position is recorded in the position queue.

4. The method for testing pavement deformation according to claim 3, characterized in that, In step S200, the logic for sequentially acquiring reference state image data and loaded state image data for the same target road surface area is as follows: The difference between the current travel distance of the mobile carrier platform and the trigger position recorded in the location queue is calculated. When the difference is equal to the preset physical baseline distance, a second trigger signal is generated to control the acquisition of loaded image data, so that the loaded image data and the reference image data cover the same physical road surface area.

5. The method for testing pavement deformation according to claim 1, characterized in that, In step S400, determining the rigid reference region in the measured three-dimensional point cloud model includes: Determine the load center coordinates of the tire and the preset critical influence radius; The horizontal Euclidean distance from the arbitrary spatial point to the loading center is calculated by geometric operations on the coordinates of the arbitrary spatial point in the measured three-dimensional point cloud model and the coordinates of the loading center. When the horizontal Euclidean distance is greater than the critical influence radius, the spatial point is marked as belonging to the rigid reference region, which is assumed to have zero vertical displacement and contains only the rigid motion components of the moving carrier platform.

6. The method for testing pavement deformation according to claim 5, characterized in that, In step S400, the calculation of rigid body transformation parameters of the mobile carrier platform during the acquisition process includes: Construct a target loss function, which is defined as the sum of squares of the Euclidean distances between the feature points in the rigid reference region and the corresponding feature points in the reference 3D point cloud model after the rigid body transformation to be solved. The optimal rigid body transformation parameters, including rotation matrix and translation vector, are obtained by minimizing the target loss function.

7. The method for testing pavement deformation according to claim 6, characterized in that, The rigid body transformation parameters of the mobile carrier platform during the data acquisition process also include: The random sampling consensus algorithm is used to remove mismatched feature point pairs between the rigid reference region and the reference 3D point cloud model. The initial solution of the rotation matrix and translation vector is calculated using the singular value decomposition method; The initial solution is refined by using the iterative nearest point algorithm until the spatial matching error converges to a preset threshold.

8. The method for testing pavement deformation according to claim 5, characterized in that, In step S300, the reference state image data and the loaded state image data contain pseudo-random speckle patterns or phase-shifted stripe patterns projected onto the road surface; In step S400, the spatial matching between the feature data within the rigid reference area and the reference three-dimensional point cloud model is performed based on the feature descriptor generated from the artificial texture features constructed by the pattern on the road surface. The feature descriptor includes three-dimensional geometric coordinates, reflection intensity values, and local normal vector information.

9. The method for testing pavement deformation according to claim 6, characterized in that, In step S500, the step of performing attitude correction on the measured three-dimensional point cloud model using the rigid body transformation parameters includes: The corrected 3D point cloud model is calculated by subtracting the translation vector from the coordinates in the measured 3D point cloud model and then multiplying it by the inverse of the rotation matrix. After this correction, the corrected measurement 3D point cloud model is unified to the coordinate system reference of the reference 3D point cloud model.

10. The method for testing pavement deformation according to claim 9, characterized in that, In step S500, the calculation of the road surface deformation field through the difference operation between the calibrated measured three-dimensional point cloud model and the reference three-dimensional point cloud model includes: A two-dimensional virtual mesh is constructed under a unified spatial coordinate reference; interpolation and resampling are performed on the reference three-dimensional point cloud model and the corrected measurement three-dimensional point cloud model respectively to obtain the reference elevation matrix and the loading elevation matrix. The difference matrix is ​​calculated by subtracting the loaded elevation matrix from the reference elevation matrix, and the difference matrix is ​​the road surface deformation field. The maximum central deflection value and the deflection basin profile curve are extracted based on the difference matrix.

Images