A method and device for repairing missing three-dimensional texture point cloud data of asphalt pavement

Abstract

The present invention discloses a method and device for repairing missing three-dimensional texture point cloud data of an asphalt pavement, which relates to the field of road traffic safety technology and solves the problem of insufficient point cloud repair accuracy. The method comprises: obtaining original three-dimensional texture point cloud data of an asphalt pavement through a three-dimensional laser sensor; performing tilt suppression processing on the original three-dimensional texture point cloud data to obtain corrected point cloud data; performing denoising processing on the corrected point cloud data to obtain denoised point cloud data; and performing repair processing on the denoised point cloud data using a repair model to obtain repaired point cloud data; wherein the repair model comprises: a discriminator, a generator, a discriminator loss function and a generator loss function; and achieving an improvement in the ability of a network output to fit a pavement texture by defining a loss function of the repair model, thereby significantly improving the quality and availability of pavement texture data and providing a good data basis for subsequent three-dimensional texture characterization and accurate evaluation of pavement anti-skid performance.

Description

Technical Field

[0001] The present invention relates to the field of road traffic safety technology, and in particular to a method and device for repairing missing three-dimensional texture point cloud data of an asphalt pavement. Background Art

[0002] As the main part that directly contacts the tire and provides friction, road surface texture significantly affects the anti-skid performance of the road surface. Therefore, the acquisition and processing of road surface texture data is crucial. Road surface texture data collected by precision optical equipment has data problems such as non-measurement points and noise. The main reasons include two aspects: first, the presence of water, markings, polished aggregates, etc. on the road surface causes changes in surface reflectivity; second, the road surface morphology is irregularly distributed, and the concave parts of the surface cannot effectively feedback depth information, resulting in missed detection points in the observation data. If the problematic data is simply filtered out, a large amount of attributes and information will be lost, which will have a great impact on the extraction of relevant texture features and the construction of anti-skid performance evaluation models. Therefore, it is necessary to accurately diagnose anomalies, repair them, and improve the data quality of surface texture data obtained by optical sensor equipment. Summary of the Invention

[0003] The present invention solves the problem of insufficient point cloud repair accuracy by providing a method and device for repairing missing three-dimensional texture point cloud data of asphalt pavement. By defining the loss function of the repair model, the network output's ability to fit the pavement texture is improved, and the quality and availability of pavement texture data are significantly improved, providing a good data foundation for subsequent three-dimensional texture characterization and accurate evaluation of pavement anti-skid performance.

[0004] In a first aspect, the present invention provides a method for repairing missing three-dimensional texture point cloud data of an asphalt pavement, the method comprising:

[0005] Obtain the original 3D texture point cloud data of the asphalt pavement through a 3D laser sensor;

[0006] Performing tilt suppression processing on the original three-dimensional texture point cloud data to obtain corrected point cloud data;

[0007] Performing denoising on the corrected point cloud data to obtain denoised point cloud data;

[0008] The denoised point cloud data is repaired using a repair model to obtain repaired point cloud data; wherein the repair model includes: a discriminator, a generator, a discriminator loss function, and a generator loss function.

[0009] In conjunction with the first aspect, in one possible implementation, performing tilt suppression processing on the original three-dimensional texture point cloud data to obtain corrected point cloud data includes:

[0010] Setting an equation of a fitting surface, and calculating the fitting surface equation using the original three-dimensional texture point cloud data and the least squares principle;

[0011] Calculating the normal vector of the fitting surface and the direction vector of the z-axis using the fitting surface equation, and then obtaining the spatial rotation angle according to the normal vector and the direction vector of the z-axis;

[0012] The spatial rotation angle is used to obtain a spatial variation matrix and a displacement matrix, and the original three-dimensional texture point cloud data is processed using the spatial variation matrix and the displacement matrix to obtain the corrected point cloud data.

[0013] In conjunction with the first aspect, in one possible implementation, performing denoising on the corrected point cloud data to obtain denoised point cloud data includes:

[0014] Calculating the arithmetic mean, standard deviation and median of the corrected point cloud data within the filter window;

[0015] The noise of the corrected point cloud data is determined by using the arithmetic mean, the standard deviation and the median to obtain denoised point cloud data.

[0016] With reference to the first aspect, in one possible implementation, performing noise judgment on the corrected point cloud data using the arithmetic mean, the standard deviation, and the median to obtain denoised point cloud data specifically includes:

[0017] Converting the corrected point cloud data within the filtering area within the filtering window into a three-dimensional point cloud data matrix, indexing the three-dimensional point cloud data matrix by row, and sequentially traversing the data of each row in the three-dimensional point cloud data matrix;

[0018] Each data in each row is judged using a discriminant formula. If the discriminant formula is satisfied, the data is replaced with the arithmetic mean; otherwise, no replacement is performed.

[0019] In conjunction with the first aspect, in a possible implementation, the discriminant formula is specifically expressed as:

[0020]

[0021] Among them, z ij Represents a data in the 3D point cloud data matrix; represents the standard deviation; S ij represents the median; k represents the filter coefficient.

[0022] In conjunction with the first aspect, in a possible implementation, the using a restoration model to restore the denoised point cloud data to obtain restored point cloud data includes:

[0023] Performing data processing on the denoised point cloud data to obtain a denoised point cloud data matrix;

[0024] Marking the data in the denoised point cloud data matrix to obtain a marking matrix;

[0025] Filling the missing positions in the denoised point cloud data matrix using a priori distribution to obtain a noise matrix;

[0026] Inputting the denoised point cloud data matrix, the label matrix and the noise matrix into the generator to obtain a generator output matrix;

[0027] Input the generator output matrix and the observation point cloud data into the discriminator, judge the authenticity of the generator output matrix, and adjust the generator parameters according to the judgment result until the number of cycles is reached to obtain a trained generator;

[0028] The trained generator is used to supplement the missing parts of the denoised point cloud data to obtain the repaired point cloud data.

[0029] In combination with the first aspect, in a possible implementation, the repair model further includes a prompt mechanism, which provides missing data prompts to the discriminator in the repair model.

[0030] In combination with the first aspect, in a possible implementation, the discriminator loss function is expressed as:

[0031]

[0032] Where a represents the generated data of the generator; b represents the real data; D(x) represents the value measured by the discriminator on the real data x; Represents the real data distribution p data (x) hope; Represents the input noise zp to the generator z (z) hope; G(z) represents the false data generated by the generator based on the noise z; pdata ( x ) table has known data distribution.

[0033] In combination with the first aspect, in one possible implementation, the generator loss function is expressed as:

[0034]

[0035] in, Represents the real data distribution p data (x) hope; Represents the input noise zp to the generatorz (z) hope; G(z) represents the generator's false data based on the noise z; c represents a constant; avg represents the mean of the known data; D(x) represents the discriminator's measurement of the real data x; β represents the second scaling factor; α represents the first scaling factor.

[0036] In a second aspect, the present invention provides a device for repairing missing three-dimensional texture point cloud data of an asphalt pavement, the device comprising:

[0037] A data acquisition module is used to obtain original three-dimensional texture point cloud data of the asphalt pavement through a three-dimensional laser sensor;

[0038] a tilt suppression module, configured to perform tilt suppression processing on the original three-dimensional texture point cloud data to obtain corrected point cloud data;

[0039] A denoising module, configured to perform denoising on the corrected point cloud data to obtain denoised point cloud data;

[0040] A restoration module is used to perform restoration processing on the denoised point cloud data using a restoration model to obtain restored point cloud data; wherein the restoration model includes: a discriminator, a generator, a discriminator loss function, and a generator loss function.

[0041] One or more technical solutions provided in the present invention have at least the following technical effects or advantages:

[0042] (1) The present invention adopts a holistic perspective of three-dimensional point cloud data. Compared with the traditional single-dimensional missing value repair algorithm, the completion method provided by the present invention can more accurately repair the missing values ​​in the point cloud data.

[0043] (2) For the case where the original point cloud data contains a large number of missing values, the missing value repair method proposed in the present invention effectively avoids problems such as poor point cloud completion effect caused by the lack of locally available information, and effectively improves the quality and usability of the three-dimensional point cloud.

[0044] (3) Combined with the actual application scenarios, considering the consistency and continuity of asphalt pavement point cloud data, the least squares principle was introduced to improve the stability and accuracy of the missing value repair model. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 A flowchart of the steps of a method for repairing missing 3D texture point cloud data of an asphalt pavement provided by an embodiment of the present invention;

[0046] Figure 2 A scene diagram for obtaining original three-dimensional texture point cloud data of an asphalt pavement provided in an embodiment of the present invention;

[0047] Figure 3An example diagram showing a comparison of tilt suppression of asphalt pavement point clouds provided by an embodiment of the present invention;

[0048] Figure 4 This is a comparative example diagram of asphalt pavement point cloud denoising provided by an embodiment of the present invention;

[0049] Figure 5 Flowchart of the repair model (PR-Net algorithm) provided in an embodiment of the present invention;

[0050] Figure 6 This is a diagram of the input data processing of the generator in the repair model (PR-Net algorithm) provided in an embodiment of the present invention;

[0051] Figure 7 This is a design diagram of the generator structure in the repair model (PR-Net algorithm) provided in an embodiment of the present invention;

[0052] Figure 8 This is a diagram of the input data processing of the discriminator in the repair model (PR-Net algorithm) provided in an embodiment of the present invention;

[0053] Figure 9 This is a diagram of the discriminator structure design in the repair model (PR-Net algorithm) provided in an embodiment of the present invention;

[0054] Figure 10 A reconstructed image of the original asphalt pavement 3D point cloud data provided by an embodiment of the present invention;

[0055] Figure 11 This is a graph of missing value repair results provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0056] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present invention.

[0057] The present invention provides a method for repairing missing 3D texture point cloud data of asphalt pavement. Figure 1 , is a flowchart of a method for repairing missing three-dimensional texture point cloud data of an asphalt pavement, and the method includes the following steps S101 to S104.

[0058] S101, obtaining original three-dimensional texture point cloud data of the asphalt pavement through a three-dimensional laser sensor.

[0059] Here, see Figure 2This is a diagram of a scenario for acquiring raw 3D texture point cloud data of an asphalt pavement, provided by an embodiment of the present invention. This embodiment uses a 3D laser sensor to collect raw 3D texture point cloud data of a specimen's surface texture. Due to factors such as the angle between the sensor and the specimen surface and the complex surface texture of the specimen itself, raw point cloud data often suffers from reference surface tilt, noise, and missing values.

[0060] S102: performing tilt suppression processing on the original three-dimensional texture point cloud data to obtain corrected point cloud data.

[0061] Specifically, in step S102, the original three-dimensional texture point cloud data is subjected to tilt suppression processing to obtain corrected point cloud data, which includes the following steps S1021 to S1023.

[0062] S1021, setting the equation of the fitting surface, and using the original three-dimensional texture point cloud data and the least squares principle to calculate the fitting surface equation.

[0063] Here, let the equation of the fitting surface be z=Ax+By+C, and the original 3D texture point cloud data is represented as: {(x i ,y i ,z i )|i≤n}, where n represents the size of the original 3D texture point cloud data.

[0064] Using the least squares principle, we can get formula (1), and calculate the values ​​of A, B, and C in the set fitting surface equation according to formula (1), and then we can get the fitting surface equation. Specifically, formula (1) is expressed as:

[0065]

[0066] Among them, A represents the slope of the plane in the x-axis direction; B represents the slope of the plane in the y-axis direction; C represents the intercept of the plane in the z-axis direction; x i Represents the x-axis coordinate of a three-dimensional data point; y i Represents the y-axis coordinate of the three-dimensional data point; z i Indicates the z-axis coordinate of a 3D data point.

[0067] S1022: Calculate the normal vector of the fitting surface and the direction vector of the z-axis using the fitting surface equation, and then obtain the spatial rotation angle based on the normal vector and the direction vector of the z-axis.

[0068] Here, after calculating the fitting surface equation, the spatial rotation angle can be calculated by calculating the normal vector of the fitting surface and the direction vector of the z-axis. Specifically, the normal vector m of the fitting surface is designed to be (x m ,y m ,z m), the direction vector n of the z-axis is (0,0,1), and the spatial rotation angle is expressed as:

[0069]

[0070] S1023, using the spatial rotation angle, obtains a spatial variation matrix and a displacement matrix, and uses the spatial variation matrix and the displacement matrix to process the original 3D texture point cloud data to obtain corrected point cloud data. Here, the spatial transformation matrix is ​​specifically expressed as:

[0071]

[0072] Here, the displacement matrix is ​​specifically expressed as:

[0073]

[0074] Among them, t x Indicates the displacement distance in the x-axis direction; t y Indicates the displacement distance in the y-axis direction.

[0075] The tilt can be suppressed by using the spatial transformation matrix and the displacement matrix.

[0076] For example, see Figure 3 , Figure 3 This is an example of the comparison of tilt suppression of asphalt pavement point clouds.

[0077] S103: De-noising the corrected point cloud data to obtain de-noised point cloud data.

[0078] Specifically, in step S103 , the corrected point cloud data is subjected to denoising processing to obtain denoised point cloud data, which includes the following steps S1031 to S1032 .

[0079] S1031, calculate the arithmetic mean, standard deviation and median of the corrected point cloud data within the filter window. Here, the corrected point cloud data is recorded as Z m×n , record the filter matrix size l (l=5 is selected for illustration in this invention), where m and n represent the number of rows and columns of the 3D point cloud data matrix respectively. The corrected point cloud data is read from top to bottom according to the row index, and the mean of the filter template area is calculated. Standard deviation S ij and median ξ ij .

[0080]

[0081]

[0082] ξ ij =median(z st)(s=i-2,...,i+2; t=j-2,...,j+2) (7)

[0083] Among them, z st Represents the data in the sth row and tth column of the matrix; median(z st ) represents the median value of the data within the filtering window.

[0084] S1032: Using the arithmetic mean, standard deviation, and median to perform noise determination on the corrected point cloud data to obtain denoised point cloud data. Specifically, in step S1032, using the arithmetic mean, standard deviation, and median to perform noise determination on the corrected point cloud data to obtain denoised point cloud data includes the following steps.

[0085] (1) Convert the corrected point cloud data within the filter area within the filter window into a three-dimensional point cloud data matrix, index the three-dimensional point cloud data matrix by row, and traverse the data of each row in the three-dimensional point cloud data matrix in turn.

[0086] (2) Use the discriminant formula to judge each data in each row. If the discriminant formula is satisfied, the data is replaced with the arithmetic mean. If not, no replacement is performed. The discriminant formula here is expressed as:

[0087]

[0088] Among them, z ij Represents a data in the 3D point cloud data matrix; represents the standard deviation; S ij represents the median; k represents the filter coefficient, and k in the present invention is 1.6.

[0089] For example, each data in the row is traversed in turn, and each data in the row is judged using formula (8). If data z ij If the conditions expressed by formula (8) are satisfied, then the data z ij is the noise and the arithmetic mean R is used ij Replace the original value z of the point data ij , if the data z ij If the conditions are not met, no processing will be performed, where k represents the filter coefficient, which is generally 1.0 to 3.

[0090] In this embodiment, see Figure 4 , Figure 4 This is an example of the comparison of asphalt pavement point cloud denoising. The corrected point cloud data is denoised to obtain the denoised point cloud data.

[0091] S104, using a restoration model to perform restoration processing on the denoised point cloud data to obtain restored point cloud data; wherein the restoration model includes: a discriminator, a generator, a discriminator loss function, and a generator loss function.

[0092] Specifically, refer to Figure 5 , Figure 5 This is a flow chart of the repair model (PR-Net algorithm) provided by an embodiment of the present invention. The core module of the network for repairing missing values ​​in the repair model (PR-Net algorithm) of this embodiment is composed of a generator and a discriminator. The function of the generator is to accurately fill in the missing values ​​in the denoised point cloud data. The discriminator is responsible for receiving the generator output matrix and the observed point cloud data, and determining whether the received data is filled or real. The generator will continue to enhance the ability to generate missing data, so that the generated missing values ​​are closer and closer to the observed point cloud data, that is, it becomes increasingly difficult for the discriminator to distinguish the authenticity of the generated samples. By repeatedly iterating the above process, until the discriminator finds it difficult to distinguish whether the received sample is the observed point cloud data or the generated missing data. In order to improve the accuracy of missing value filling, a prompt mechanism is added to the constructed repair model (PR-Net algorithm). In order to ensure that the discriminator forces the generator to learn the desired data distribution, additional information is provided to the discriminator in the form of a prompt vector. This hint vector reveals information about missing values ​​in the original three-dimensional point cloud data to the discriminator. These hints ensure that the generator learns to generate missing components of the point cloud according to the distribution of the real three-dimensional point cloud data, thereby making the generated point cloud data closer and closer to the real observed point cloud data.

[0093] In step S104 , the denoised point cloud data is repaired using the repair model to obtain repaired point cloud data, which includes the following steps S1041 to S1046 .

[0094] S1041, performing data processing on the denoised point cloud data to obtain a denoised point cloud data matrix. For example, referring to Figure 6 , Figure 6 The input data processing diagram of the generator in the repair model (PR-Net algorithm) is shown in the figure. The repair model (PR-Net algorithm) of the present invention uses a two-dimensional matrix with a shape of 1000*500 as input X. It is known that the data obeys a certain data distribution defined as p data (x), the data distribution learned by the generator is defined as p z There are some missing values ​​in the input X. Here, we define a Mask matrix (M) with the same dimension as the input X. The values ​​in M ​​range from {0, 1} to indicate whether the data at this position is missing.

[0095] S1042: Mark the data in the denoised point cloud data matrix to obtain a marking matrix. If data exists at the corresponding position in the matrix, it is marked as 1, and if not, it is marked as 0.

[0096] S1043, fill the missing positions in the denoised point cloud data matrix using the prior distribution to obtain a noise matrix. For example, a noise matrix (Noise matrix N) with the same dimension as the input X is defined, and the value of N is as follows:

[0097]

[0098] Where noi∈[0,0.1]. The matrix N is a sampled Gaussian random distribution, which serves as a prior distribution. After the input data enters the model, after several rounds of iterations, the distribution p learned by the generator is z (z) will get closer and closer to the known data distribution p data (x). The output of the generator is The output of the discriminator is

[0099] S1044, input the denoised point cloud data matrix, the label matrix and the noise matrix into the generator to obtain the generator output matrix.

[0100] For example, refer to Figure 7 , Figure 7 This is a diagram of the generator structure design in the repair model (PR-Net algorithm). The input layer of the generator includes X, M, and N, which respectively represent the data matrix with missing values, the label matrix for the missing positions, and the noise filled in the missing positions. These three matrices are combined as the final input of the generator; the feature extraction layer includes N connected weighted neurons and N Relu activation layers, and the output layer includes 1 neuron and 1 Sigmoid activation layer. Generator output The dimension is the same as the input X, No longer contains missing data.

[0101] It should be pointed out that the values ​​involved in the neural network calculation need to be normalized. Therefore, when reconstructing the three-dimensional texture point cloud, the values ​​need to be denormalized.

[0102] S1045, input the generator output matrix and the observation point cloud data into the discriminator, judge the authenticity of the generator output matrix, and adjust the generator parameters according to the judgment result until the number of cycles is reached to obtain a trained generator.

[0103] For example, refer to Figure 8 , Figure 8 This is the input data processing diagram of the discriminator in the repair model (PR-Net algorithm). The discriminator input is and H, respectively, represent the complete matrix generated by the generator and the prompt matrix that randomly prompts the discriminator for a certain position, and the output The dimension is the same as X, and the value range is {0,1}, which represents the judgment matrix for determining whether each position is generated data.

[0104] refer to Figure 9 , Figure 9 This is a diagram of the discriminator structure design in the repair model (PR-Net algorithm). The discriminator network sequentially connects the input layer, feature extraction layer, and output layer. The input layer includes the point cloud matrix and prompt matrix output by the generator. The feature extraction layer includes N connected weighted neurons and N Leaky Relu activation layers. The output layer includes 1 neuron and 1 Sigmoid activation layer. The output layer of the discriminator indicates that the discriminator will judge the value of each position in the point cloud matrix output by the generator. If the discriminator believes that this value is a point that is not missing in the original point cloud, it will be judged as True. Otherwise, if the discriminator believes that this value is a point that was repaired by the generator, that is, a point that was missing in the original point cloud, it will be judged as False.

[0105] S1046: Use the trained generator to supplement the missing parts of the denoised point cloud data to obtain repaired point cloud data.

[0106] In step S104, considering the continuity of the pavement structure, the design of the objective function should ensure that the generated data is close enough to the existing real data and that the generated data fits the real pavement as closely as possible.

[0107] Specifically, this embodiment optimizes the objective function of the missing value filling network, introduces the least squares loss function to replace the cross entropy loss function used in the traditional network, and makes the distribution obtained by the generator's learning approach the distribution of the true value by minimizing the square error objective function. Choosing to minimize the square error to update the neural network parameters has two advantages: (1) The objective function more strictly punishes "outliers" far away from the known data, so that the data generated by the generator is more consistent with the distribution of the known data. (2) It solves the problem that the training process of the original GAN ​​architecture network is difficult to fit. Combined with the actual application scenario, it is considered that there is a certain position continuity between the point clouds from the asphalt pavement, that is, the height difference between two adjacent points will not be large, and the adjacent continuous areas have the same undulating trend. Therefore, the minimization of the square error objective function is introduced to construct the missing value repair model, which ensures the consistency and coherence of the surface structure of the asphalt pavement after the point cloud is completed.

[0108] The objective function of the proposed model is designed as follows, and the discriminator loss function is expressed as:

[0109]

[0110] Where a represents the generated data of the generator; b represents the real data; D(x) represents the predicted value of the discriminator for the real data x; Represents the real data distribution p data (x) expectations; Denotes the distribution p of the generator input noise z z (z); G(z) represents the false data generated by the generator based on the noise z; p data (x) represents the known data distribution.

[0111] In step S104, the generator loss function is expressed as:

[0112]

[0113] in, Represents the real data distribution p data (x) expectations; Denotes the distribution p of the generator input noise z z (z); G(z) represents the fake data generated by the generator based on the noise z; c represents a constant; avg represents the mean of the known data; D(x) represents the predicted value of the discriminator for the real data x; β represents the second scaling factor; α represents the first scaling factor.

[0114] The generator (G) and the discriminator (D) interact and update their respective weights in a competitive manner. The task of the discriminator D is to distinguish whether the input data comes from the real data set and clearly distinguish between real data and generated data. The task of the generator G is to generate fake data that is as close to the real data as possible to deceive the discriminator D. V(G) generally encourages the data generated by the generator to be misclassified as real data by the discriminator. During the training process, the generator G attempts to minimize V(G). This training process is repeated until a certain convergence condition or a predetermined number of training rounds is reached. In this way, the generator G and the discriminator D evolve together in competition, the quality of the data generated by the generator G gradually improves, and the discriminator D's ability to distinguish is also enhanced.

[0115] The repair model also includes a prompting mechanism, which prompts the discriminator in the repair model about missing data.

[0116] Exemplarily, the repair model takes denoised point cloud data containing missing values ​​as input, the size of which is 1000*500, and the generator output matrix is ​​a matrix of size 1000*500. It updates its own weights according to the generator output matrix, and finally outputs a complete matrix to obtain the repaired point cloud data.

[0117] In a specific embodiment provided by the present invention, the missing value of the asphalt mixture surface texture point cloud data is repaired, and the original three-dimensional texture point cloud data is reconstructed. Figure 10 , the missing value repair results are shown in Figure 11 .

[0118] In this example, in order to prove the accuracy of the model restoration in the present invention, the proposed model is compared and evaluated with the KNN point cloud completion method, the Linear point cloud completion method, and the Three-Spine point cloud completion method. The obtained accuracy and error evaluation indicators are shown in Tables 2 to 4:

[0119] Table 2 Performance of different completion algorithms on AC13 point cloud

[0120]

[0121] Table 3 Performance of different completion algorithms on OGFC13 point cloud

[0122]

[0123] Table 4 Performance of different completion algorithms on SMA13 point cloud

[0124]

[0125] In this embodiment, in order to prove that the method proposed in the present invention can more accurately evaluate the anti-skid performance of asphalt pavement, the following experiments are designed: First, the missing point cloud dataset P0 is processed using the PR-Net point cloud completion method, the KNN point cloud completion method, the Linear point cloud completion method, and the Three-Spine point cloud completion method, respectively, to obtain the complete point cloud dataset P0. 1_1 、P 1_2 、P 1_3 、P 1_4 Then, from the complete point cloud dataset P 1_1 、P 1_2 、P 1_3 、P 1_4 Extract three-dimensional texture features R a 、R q ,MTD,S m , S, R s 、R sk 、R ku , and obtain feature sets F1, F2, F3, and F4 respectively. Finally, the feature sets F1, F2, F3, and F4 are used to train the anti-slip performance evaluation model (denoted as Estimator).

[0126] The present invention provides a device for repairing missing three-dimensional texture point cloud data of an asphalt pavement. The device comprises a data acquisition module, a tilt suppression module, a denoising module and a repair module.

[0127] The data acquisition module is used to obtain the original three-dimensional texture point cloud data of the asphalt pavement through a three-dimensional laser sensor.

[0128] The tilt suppression module is used to perform tilt suppression processing on the original three-dimensional texture point cloud data to obtain corrected point cloud data.

[0129] The denoising module is used to denoise the corrected point cloud data to obtain denoised point cloud data.

[0130] The restoration module is used to use the restoration model to restore the denoised point cloud data to obtain restored point cloud data; wherein the restoration model includes: a discriminator, a generator, a discriminator loss function and a generator loss function.

[0131] The devices or modules described in the above embodiments can be implemented as computer chips or physical devices, or as products with certain functions. For ease of description, the above devices are described separately by function, with each module described separately. When implementing the present invention, the functions of each module can be implemented in the same or multiple software and / or hardware components. Of course, a module that implements a certain function can also be implemented by combining multiple submodules or subunits.

[0132] The methods, devices, or modules described in the present invention can be implemented in the form of computer-readable program code. The controller can be implemented in any appropriate manner. For example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (such as software or firmware) that can be executed by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also know that in addition to implementing the controller in the form of pure computer-readable program code, it is entirely possible to implement the same function of the controller in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, such a controller can be considered a hardware component, and the devices included therein for implementing various functions can also be considered as structures within the hardware component. Or even, the means for implementing various functions may be considered to be both a software module for implementing the method and a structure within a hardware component.

[0133] Some modules within the apparatus described herein may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, classes, and the like that perform specific tasks or implement specific abstract data types. The present invention may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communications network. In a distributed computing environment, program modules may reside in local and remote computer storage media, including storage devices.

[0134] Through the description of the above implementation methods, it can be seen that those skilled in the art can clearly understand that the present invention can be implemented by means of software plus necessary hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, or can be embodied through the implementation process of data migration. The computer software product can be stored in a storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a mobile terminal, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present invention or certain parts of the embodiments.

[0135] The various embodiments in this specification are described in a progressive manner. References to the same or similar parts between the various embodiments are sufficient. Each embodiment focuses on the differences from other embodiments. All or part of the present invention can be used in a variety of general or specialized computer system environments or configurations. For example, personal computers, server computers, handheld or portable devices, tablet devices, mobile communication terminals, multiprocessor systems, microprocessor-based systems, programmable electronic devices, network PCs, minicomputers, mainframe computers, and distributed computing environments that include any of the above systems or devices.

[0136] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it should be understood by those skilled in the art that the technical solutions described in the aforementioned embodiments may still be modified, or some or all of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the present invention.

Claims

1. A method for repairing missing 3D texture point cloud data of asphalt pavement, characterized in that: include: Obtain the original 3D texture point cloud data of the asphalt pavement through a 3D laser sensor; Performing tilt suppression processing on the original three-dimensional texture point cloud data to obtain corrected point cloud data; Performing denoising on the corrected point cloud data to obtain denoised point cloud data; Performing repair processing on the denoised point cloud data using a repair model to obtain repaired point cloud data; wherein the repair model includes: a discriminator, a generator, a discriminator loss function, and a generator loss function; The discriminator loss function is expressed as: The generator loss function is expressed as: in, Represents the real data distribution p data (x) expectations; Denotes the distribution p of the generator input noise z z (z); G(z) represents the fake data generated by the generator based on the noise z; c represents a constant; avg represents the mean of the known data; D(x) represents the predicted value of the discriminator for the real data x; β represents the second scaling factor; α represents the first scaling factor; a represents the generated data of the generator; b represents the real data; p data (x) represents the known data distribution.

2. The method for repairing missing asphalt pavement 3D texture point cloud data according to claim 1, characterized in that: The step of performing tilt suppression processing on the original three-dimensional texture point cloud data to obtain corrected point cloud data includes: Setting an equation of a fitting surface, and calculating the fitting surface equation using the original three-dimensional texture point cloud data and the least squares principle; Calculating the normal vector of the fitting surface and the direction vector of the z-axis using the fitting surface equation, and then obtaining the spatial rotation angle according to the normal vector and the direction vector of the z-axis; The spatial rotation angle is used to obtain a spatial variation matrix and a displacement matrix, and the original three-dimensional texture point cloud data is processed using the spatial variation matrix and the displacement matrix to obtain the corrected point cloud data.

3. The method for repairing missing asphalt pavement 3D texture point cloud data according to claim 1, characterized in that: The step of performing denoising on the corrected point cloud data to obtain denoised point cloud data includes: Calculating the arithmetic mean, standard deviation and median of the corrected point cloud data within the filter window; The noise of the corrected point cloud data is determined by using the arithmetic mean, the standard deviation and the median to obtain denoised point cloud data.

4. The method for repairing missing asphalt pavement 3D texture point cloud data according to claim 3 is characterized in that: The performing noise judgment on the corrected point cloud data by using the arithmetic mean, the standard deviation, and the median to obtain denoised point cloud data specifically includes: Converting the corrected point cloud data within the filtering area within the filtering window into a three-dimensional point cloud data matrix, indexing the three-dimensional point cloud data matrix by row, and sequentially traversing the data of each row in the three-dimensional point cloud data matrix; Each data in each row is judged using a discriminant formula. If the discriminant formula is satisfied, the data is replaced with the arithmetic mean; otherwise, no replacement is performed.

5. The method for repairing missing asphalt pavement 3D texture point cloud data according to claim 4 is characterized in that: The discriminant formula is specifically expressed as: Among them, z ij Represents a data in the 3D point cloud data matrix; represents the standard deviation; S ij represents the median; k represents the filter coefficient.

6. The method for repairing missing asphalt pavement 3D texture point cloud data according to claim 1, characterized in that: The repairing of the denoised point cloud data using the repair model to obtain repaired point cloud data includes: Performing data processing on the denoised point cloud data to obtain a denoised point cloud data matrix; Marking the data in the denoised point cloud data matrix to obtain a marking matrix; Filling the missing positions in the denoised point cloud data matrix using a priori distribution to obtain a noise matrix; Inputting the denoised point cloud data matrix, the label matrix and the noise matrix into the generator to obtain a generator output matrix; Input the generator output matrix and the observation point cloud data into the discriminator, judge the authenticity of the generator output matrix, and adjust the generator parameters according to the judgment result until the number of cycles is reached to obtain a trained generator; The trained generator is used to supplement the missing parts of the denoised point cloud data to obtain the repaired point cloud data.

7. The method for repairing missing asphalt pavement 3D texture point cloud data according to claim 1, characterized in that: The repair model further includes a prompting mechanism, which provides missing data prompts to the discriminator in the repair model.

8. A device for repairing missing 3D texture point cloud data of asphalt pavement, characterized in that: include: A data acquisition module is used to obtain original three-dimensional texture point cloud data of the asphalt pavement through a three-dimensional laser sensor; a tilt suppression module, configured to perform tilt suppression processing on the original three-dimensional texture point cloud data to obtain corrected point cloud data; A denoising module, configured to perform denoising on the corrected point cloud data to obtain denoised point cloud data; A restoration module is configured to perform restoration processing on the denoised point cloud data using a restoration model to obtain restored point cloud data; wherein the restoration model includes: a discriminator, a generator, a discriminator loss function, and a generator loss function; the discriminator loss function is expressed as: The generator loss function is expressed as: in, Represents the real data distribution p data (x) expectations; Denotes the distribution p of the generator input noise z z (z); G(z) represents the fake data generated by the generator based on the noise z; c represents a constant; avg represents the mean of the known data; D(x) represents the predicted value of the discriminator for the real data x; β represents the second scaling factor; α represents the first scaling factor; a represents the generated data of the generator; b represents the real data; p data (x) represents the known data distribution.

Images