Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In order to better explain the working process of the present invention, the following explains the principle of implementing the present invention.
The existing pavement single index detection method based on three-dimensional pavement data detects a single index by using a certain disease of a pavement or the elevation characteristic of the index, can acquire pavement track information by using data with lower precision, and can detect pavement crack information by using data with higher precision (the transverse resolution is 1mm, and the elevation resolution is less than or equal to 0.5 mm).
However, when the precision of the three-dimensional data is high enough (the transverse resolution is 1mm, and the elevation resolution is less than or equal to 0.5mm), the three-dimensional pavement elevation data contains relatively complex pavement scene information, which not only contains macroscopic pavement deformation, pavement curvature and other information, but also contains microscopic cracks, marked lines, repair and other information, and even the pavement structure depth can be reflected in the three-dimensional pavement data with the precision; in high-precision data, different types of diseases have different degrees of mutual influence relationship on different pavements. For example, the pavement texture fluctuation in a pavement with a large structural depth is similar to the depth characteristic of a crack, and the robustness and the practicability of the crack extraction method can be influenced by only utilizing the crack depth characteristic without considering the influence of the pavement structural depth. Therefore, the high-precision three-dimensional road surface data contains complex scene information and various indexes have mutual influence, and if one index is analyzed only through simple elevation information without considering the influence of other indexes, the robustness and the universality of the method are influenced.
Therefore, in order to find the mutual influence relationship among various road surface information, the data characteristics of cracks, marked lines, pit grooves, textures, ruts and the like on the road surface are analyzed, fig. 1(a) is a schematic diagram of high-precision line scanning three-dimensional road surface data, fig. 1(b) is a schematic diagram of rut data, marked line data, crack data, pit groove data and texture data contained in the three-dimensional road surface data, and fig. 1(c) is an abstract representation diagram of components of the three-dimensional road surface data, as can be seen from fig. 1, the cracks show a sharp downward-prick-like characteristic in the cross section of the three-dimensional road surface, the marked lines show a regular elevation step convex characteristic, edges of the pit grooves also usually have a sharp drop characteristic, and from the perspective of the three components all have high-frequency characteristics and a sparse characteristic, so that the cracks can be divided into frequency domains, The marked lines and the pits are summarized into sparse characteristic data.
The slow deformation of road ruts and the like and the standard contour of the road and the like usually do not contain high-frequency components, and belong to low-frequency components in three-dimensional cross section data.
The road surface texture shows rapid fluctuation characteristics in a specific range in the cross section of the three-dimensional road surface, and is also high-frequency characteristics, and compared with road surface cracks, marked lines and the like, the fluctuation of the road surface texture does not have sparse characteristics.
In addition, in the three-dimensional road surface cross section data, from the perspective of a spatial domain, various components are usually mixed with each other, for example, the influence of a road surface texture background is usually considered in three-dimensional crack extraction; the influence of cracks mixed in the texture of the pavement needs to be removed when evaluating the depth of the pavement structure.
By combining the above analysis, for any three-dimensional road surface cross section data y, it can be divided into high frequency component data and low frequency component data, and the high frequency component data includes sparse characteristic data and vibration characteristic data, that is, the three-dimensional road surface cross section data can be expressed as follows:
f=f+x+t,
wherein y represents input three-dimensional pavement cross section elevation data, and the length of the three-dimensional pavement cross section elevation data is N; f represents low-frequency component data and can represent information such as tracks, standard contours, slowly-changing deformation diseases and the like; x is road surface sparse characteristic data, and can represent information which has mutation characteristics and occupies a small proportion in the cross section of the road surface, such as cracks, pit and groove diseases or edges of artificial marked lines; and t is road surface vibration characteristic data which can represent fluctuation information of road surface textures.
f is a low frequency component, which is very low compared to x and t, and can be first obtained by a low pass filter. Modeling t as a signal satisfying statistical white Gaussian noise characteristics, and setting the variance of the signal to be sigma2. And x has a sparse characteristic and can be expressed by combining a difference equation.
According to the analysis process, any three-dimensional pavement cross section data can be expressed as low-frequency component data and high-frequency component data, and the high-frequency component data comprises sparse characteristic data and vibration characteristic data. Based on this conclusion, a specific implementation scheme of the embodiment of the present invention is described below, and fig. 2 is a flowchart of a line scanning three-dimensional road surface data component analysis method based on frequency domain filtering and a total variation model according to the embodiment of the present invention, as shown in fig. 2, the method includes: designing a low-pass filter suitable for three-dimensional pavement data through pavement component characteristic analysis, performing frequency domain low-pass filtering on the three-dimensional pavement cross section data to be detected, respectively acquiring low-frequency component data and high-frequency component data corresponding to the three-dimensional pavement cross section data to be detected, wherein, the low-frequency component data comprises slow deformation information on the cross section of the road surface to be detected, the high-frequency component data comprises sparse characteristic data and vibration characteristic data, the sparse characteristic data comprises violent change information on the cross section of the road surface to be detected, the vibration characteristic data comprises texture characteristic information on the cross section of the road surface to be detected, the slow deformation information comprises ruts, the violent change information comprises one or more of cracks, marked lines and pits, and the cut-off frequency of the low-pass filter is obtained according to the influence width of the deformation of the road surface to be detected and the elevation resolution of the three-dimensional data;
further dividing the high-frequency component data into the sparse characteristic data and the vibration characteristic data through a full variation model;
according to the low-frequency component data, the sparse characteristic data and the vibration characteristic data, one or more of ruts, cracks, marked lines, pits and textures in the cross section of the pavement to be detected can be analyzed and identified.
Before this, the following steps are also required: the method comprises the steps of firstly, acquiring cross section data of a three-dimensional road surface to be detected, wherein the cross section data of the three-dimensional road surface to be detected is obtained by measuring the road surface to be detected through a line scanning three-dimensional measuring sensor, the three-dimensional measuring sensor obtains the relative elevation condition of the surface of a measured object based on the triangulation principle, and the acquired cross section data of the three-dimensional road surface to be detected can reflect the elevation information of the surface of the measured object.
The line scanning three-dimensional measuring sensor can realize the synchronous measurement of the profile of the section at the same attitude and the same moment, and the acquisition mode comprises two modes: the three-dimensional measuring sensor is arranged on a fixed support, a measured object passes through a measuring area at a certain speed within the measuring range of the three-dimensional measuring sensor, and the three-dimensional data of the outline of the measured object is acquired in the moving process of the measured object; and secondly, the three-dimensional measuring sensor is arranged on a moving carrier, and in the moving process of the measuring carrier, three-dimensional data of the outline of the measured object is acquired.
In the data acquisition, a three-dimensional measuring sensor is arranged on a moving carrier, and in the moving process of the measuring carrier, the data acquisition is carried out on the three-dimensional outline of a measured object.
Due to the interference of the measurement environment, for example, the water stain and oil stain on the road surface or the foreign matter in the measured area, a small amount of abnormal noise (zero point) may exist in the acquired data, and therefore, the acquired data of the cross section of the three-dimensional road surface to be detected needs to be preprocessed, and the specific preprocessing steps are as follows: and carrying out abnormal value elimination and data calibration on the three-dimensional pavement cross section data to be detected.
Because the line scanning three-dimensional measurement sensor is formed by combining the area-array camera and the line laser, the distortion at the center of the camera is minimum, and the acquired three-dimensional data of the cross section of the road surface is most stable near the center point of the section.
In a pavement three-dimensional measuring system consisting of an area-array camera and a high-power line laser, system errors such as sensor installation angle, laser line collimation degree, laser light intensity distribution unevenness and the like exist. These systematic errors weaken the characteristics of the target of interest on the road surface, and therefore, the data collected by the three-dimensional measurement sensor needs to be corrected by a calibration file, and the image data is converted into object data y.
Just in the frequency domain, the three-dimensional pavement cross section data to be detected can be divided into high-frequency component data and low-frequency component data, so that the collected three-dimensional pavement cross section data to be detected is converted into the frequency domain through Fourier transform, and then the low-pass filter is used for filtering the frequency domain data to distinguish the low-frequency component data from the high-frequency component data in the three-dimensional pavement cross section data to be detected.
The most important issue with low-pass filters is how to set the cut-off frequency of the low-pass filter,
the system related to the embodiment of the invention acquires the cross section data of the three-dimensional pavement to be detected, and the fluctuation of the data components is related to the data resolution besides the fluctuation of the components. Considering that the resolution of the three-dimensional road surface cross section data of the road surface adopted by the data of the embodiment of the invention is known (the section data length N is 2048, the transverse resolution R _ x is 1mm, and the elevation resolution R _ z is 0.1mm), the link mainly combines the fluctuation of the data component under the fixed resolution to obtain the cut-off frequency f for distinguishing the low-frequency component and the high-frequency component of the three-dimensional road surface cross section datac。
In general, in consideration of factors such as road drainage, the asphalt pavement has a certain radian; on the other hand, asphalt pavements usually have slowly changing but large transverse range, tracks with depth greater than 10mm affect the driving safety and need to be detected, and the width W _ r affected by the tracks on one side is generally 0.5m-1 m.
T=W_r/R_x;
T_m=min(T);
fc=1/T_m;
When the influence width of the rut is 0.5m to 1m, the minimum period T _ m in the data of R _ x 1mm is considered to be 500. So that the cut-off frequency f for the low frequency components allowed in the sectionc=1/500=0.002。
For section data with the data length of N, combining with cut-off frequency, performing low-pass filtering on the data by utilizing Fourier transform, frequency domain ideal low-pass filtering and inverse Fourier transform, and separating low-frequency component data f and high-frequency component data of the three-dimensional pavement cross section data y to be detected, wherein the high-frequency component data comprises sparse characteristic data x and vibration characteristic data w.
Therefore, the three-dimensional pavement cross section data to be detected are filtered through a low-pass filter, and low-frequency component data and high-frequency component data are obtained according to the filtered frequency domain data, and the method specifically comprises the following steps:
(1) carrying out Fourier transform on the three-dimensional pavement cross section data y to be detected: and performing Fourier transform on the three-dimensional pavement cross section data Y to be detected discretely by using a fast algorithm of discrete Fourier transform, wherein the length of Y is N, the obtained Fourier transform sequence is recorded as Y, the length of Y is also N (the frequency band corresponding to [ N/2,1] is [0,1]), and frequency distribution information of the three-dimensional pavement cross section data Y to be detected is represented. The knowledge of the related theory of Fourier transform shows that Y is centrosymmetric, the middle corresponds to the low-frequency component of the signal Y, and the two sides correspond to the high-frequency component of the signal Y.
(2) Ideal low-pass filtering in frequency domain: for the sequence Y obtained in the above steps, the cut-off frequency f of the low-frequency component obtained by analyzing the data characteristics and the road surface low-frequency component is combined
cThe sequence is low-pass filtered in the frequency domain. Then for a sequence Y of length N, f left and right of the Y midpoint are retained
cFrequency of N points (i.e. middle)
Frequency amplitude of Y is kept) and the frequency amplitudes of the other parts of Y are all set to 0, the frequency domain low-pass filtered sequence is marked as Y _ LF.
(3) Performing inverse Fourier transform on the frequency sequence Y _ LF: and performing inverse Fourier transform on the frequency sequence Y _ LF, taking a real number part of the frequency sequence Y _ LF, obtaining low-frequency component data corresponding to the frequency sequence Y _ LF, and marking the obtained low-frequency component data as Y _ LF, wherein the Y _ LF is the low-frequency component data f of the three-dimensional pavement cross section data Y to be detected.
And (y-f) after the low-frequency component data f of the three-dimensional pavement cross section data y to be detected is obtained, the low-frequency component data f is the high-frequency component data h of the rest part, namely the sum of the sparse characteristic data x and the vibration characteristic data w in the model.
After the low-frequency component data and the high-frequency component data are solved, the high-frequency component data are further divided into sparse characteristic data and vibration characteristic data through a full variation model.
For the data noise reduction problem containing sparse characteristics and sparse derivative (e.g. piecewise constant) characteristics, Total Variation Denoising (TVD) has the well-known characteristics of retaining signal details and effectively removing noise. For a sparse signal containing noise, a classic TVD acquires a condition extreme value by constructing a sparse optimization objective function and utilizing a Lagrange multiplier method, converts a sparse solving problem into a signal model energy functional minimization problem, and has the following analysis process:
for a discrete time series, the first order differential matrix D (N-1) × N is defined as:
x is a sparse derivative (sparse derivative) signal, and w is a signal satisfying a variance σ2When h is x + w, there is a gaussian white noise signal according to the classical total variation model
arg min||Dx||1L1Norm regularization, sparsity is satisfied.
Constraint conditions are as follows:
two norms, the linear distance of two vectors in space.
The above minimization problem can be converted to the following objective function by the lagrange multiplier method:
by using the formula, the sparse characteristic data x and the vibration characteristic data w in the high-frequency component data h are solved, wherein the parameter lambda is a regularization parameter and is used for adjusting the weight occupied by two parts forming the objective function in optimization, the value of the parameter lambda is more than 0, the value setting of the parameter lambda is generally proportional to the standard deviation of the vibration component in the signal, and the value lambda is more suitable for 1.2 in the application (the range of the macroscopic structure depth of the road surface is 1-2 mm).
And finally, identifying one or more of ruts, cracks, marked lines, pits and textures in the cross section of the pavement to be detected according to the low-frequency component data, the sparse characteristic data and the vibration characteristic data.
According to the embodiment of the invention, the three-dimensional pavement cross section data is decomposed into the low-frequency component, the sparse component and the vibration component by combining the model, and the low-frequency component, the sparse characteristic data and the vibration characteristic data of the three-dimensional pavement can be respectively obtained after all the components are spliced, so that the method is used for accurately extracting various indexes and disease information of the pavement.
By combining practical application requirements of pavement disease detection, maintenance and the like, the sparse characteristic data of model decomposition is verified mainly by using cracks and marked lines as example indexes; three-dimensional pavement cross section data with different construction depths are used as example indexes to verify vibration characteristic data decomposed by the model; the accuracy of the low frequency component is verified by the envelope method disclosed in the reference 201710861318.2.
For crack information in the sparse component, it is typically lower than normal road surfaces. According to the characteristic, the mean value of the sparse characteristic data is x _ aver, the fluctuation amplitude t _ a of the texture in the acquired high-frequency component data h can be counted, and the region points which are lower than the mean value and have fluctuation amplitude larger than t _ a in the sparse characteristic data, namely the point set of which the elevation is lower than x _ aver-t _ a in the sparse characteristic data, are used as the suspected crack region. And the crack information can be obtained after splicing each adjacent section. And similarly, the edge of the marked line is also taken, the marked line is usually higher than a normal road surface, and the area points which are higher than the mean value and have fluctuation amplitude larger than t _ a in the sparse characteristic data, namely the point set with the height higher than x _ aver + t _ a in the sparse characteristic data is taken as the suspected area of the marked line.
For the vibration characteristic data obtained by the model, only the area mean and variance of the vibration characteristic data are used for representing the magnitude of the vibration component in the embodiment of the invention.
In order to verify the effectiveness and reliability of the scheme, the embodiment of the invention takes the three-dimensional data of the asphalt pavement containing the cracks and the marked lines and the three-dimensional data of the asphalt pavement containing different structural depths as examples, and describes the asphalt pavement data component analysis method based on line scanning three-dimensional measurement.
Due to the interference of a measuring environment (water stain and oil stain on a road surface or foreign matters in a measured area), part of abnormal noise (zero value points) may exist in the acquired data, and the abnormal noise points are replaced by non-abnormal sampling points close to the central area of the section; and correcting system errors caused by sensor installation, laser line radian and uneven light intensity distribution in the object section profile measured by the three-dimensional measuring sensor by using the calibration file, and converting image data into object data. And simultaneously splicing a series of pretreated sections along the driving direction to obtain the cross section data of the asphalt three-dimensional pavement.
For common pavement index detection applications, three-dimensional pavement cross section data y is modeled into three types of components:
y=f+x+t,
wherein y represents input three-dimensional pavement cross section data, and the length of the data is N; f is low-frequency component data of the pavement, and can represent slow deformation information such as pavement ruts and the like; x is road surface sparse characteristic data, and can represent information which has mutation characteristics and occupies a small proportion in the cross section of the road surface, such as cracks, pit and groove diseases or edges of artificial marked lines; and t is road surface vibration characteristic data. f. The length of x and t is N.
For the acquired three-dimensional pavement cross section data y, the data length is N, the transverse resolution of the data is set to be R _ x, the elevation resolution is set to be R _ z, and the slow deformation influence width of the pavement is set to be W _ R. The road surface evaluation length is T, and the road surface radian cutoff frequency is fc。
T=W_r/R_x;
T_m=min(T),
fc=1/T_m,
In combination with practical situations, W _ r is typically 0.5m-1 m. Using the above formula analysis, the cut-off frequency f is determined for data with a resolution of 1mmc=0.002。
For three-dimensional pavement cross section data y, the cutoff frequency f is combinedcThe method comprises the steps of carrying out low-pass filtering on data by utilizing Fourier transform, frequency domain ideal low-pass filtering and inverse Fourier transform, and separating low-frequency component data f and high-frequency component data of three-dimensional pavement cross section data y, wherein the high-frequency component data comprises sparse characteristic data x and vibration characteristic data w.
An embodiment of a process for separating low-frequency and high-frequency components of three-dimensional pavement cross-sectional data based on frequency-domain low-pass filtering is shown in the figure, and fig. 3 is a schematic diagram of pre-processed three-dimensional pavement cross-sectional data, which is represented by y; FIG. 4 is a schematic representation of three-dimensional pavement cross-sectional data in the frequency domain, the three-dimensional pavement cross-sectional data in the frequency domain being represented by Y; FIG. 5 is a diagram of a low pass filter utilizing fcDesigning a frequency domain ideal low-pass filter L with a cut-off frequency fc(ii) a Fig. 6 is a schematic diagram of three-dimensional pavement cross-sectional data after frequency-domain filtering, where L is used to filter Y to obtain frequency distribution Y _ LF; fig. 7 is a schematic diagram of low-frequency component data of three-dimensional pavement cross-sectional data, and as shown in fig. 7, inverse fourier transform is performed on a frequency sequence Y _ LF, and a real number part of the frequency sequence Y _ LF is taken to obtain a corresponding low-frequency component, that is, low-frequency component data f of a signal Y; FIG. 8 is a three-dimensional cross-section of a pavementAs shown in fig. 8, after the low-frequency component data f is removed from the y of the three-dimensional road surface cross section data, the remaining high-frequency component data h includes vibration characteristic data t and sparse characteristic data x.
Solving the sparse characteristic data x and the vibration characteristic data t in h by using an all-variation solving model, wherein the embodiment is shown in fig. 9 and 10, and fig. 9 is a sparse characteristic data schematic diagram of three-dimensional pavement cross section data; fig. 10 is a schematic view of vibration characteristic data of three-dimensional road surface cross-sectional data.
And obtaining a component splicing diagram after splicing each component by using the obtained low-frequency component data f, the sparse characteristic data x and the vibration characteristic data t. Correspondingly, for the acquired sparse characteristic data, the link is verified by utilizing crack information and marking information in the sparse characteristic data. For the vibration characteristic data obtained by the model, only the area mean and variance of the vibration characteristic data are used for representing the size of the vibration characteristic data in the patent. Fig. 11 is a graph for testing effectiveness of sparse characteristic data by using three-dimensional pavement cross section data, fig. 11(a) is a depth grayscale conversion graph of the three-dimensional pavement cross section data when a pavement cross section to be detected contains cracks, fig. 11(b) is a schematic diagram of sparse characteristic data containing cracks in the pavement cross section to be detected, and fig. 11(c) is a schematic diagram of cracks in the pavement cross section to be detected.
Fig. 12(a) is a depth-to-grayscale map of three-dimensional pavement cross section data when the pavement cross section to be detected includes marked lines, fig. 12(b) is a schematic diagram of sparse characteristic data including marked lines in the pavement cross section to be detected, and fig. 12(c) is a schematic diagram of marked lines in the pavement cross section to be detected.
As can be seen from fig. 11 and 12, the sparse feature data solved by the model includes more complete crack information and reticle information.
In addition, in order to verify the validity of the low-frequency component data, the three-dimensional cross section data containing the ruts is used for verification, fig. 13(a) is a depth grayscale map of the three-dimensional cross section data of the cross section of the road surface to be detected containing the ruts, fig. 13(b) is a schematic diagram of vibration characteristic data, fig. 13(c) is a schematic diagram of low-frequency component data, and fig. 13(d) is a schematic diagram of extracted rut information. It can be seen from fig. 13 that the extracted low-frequency component data can be used for extracting slow deformation information such as ruts.
To sum up, the embodiment of the invention processes part abnormal zero-value noise points of the road surface section profile measured by the three-dimensional measuring sensor, which are caused by the interference of the measuring environment, through preprocessing, and obtains the image space section profile; the calibration file is utilized to effectively correct system errors caused by sensor installation, laser line radian and uneven light intensity in the road surface section profile measured by the three-dimensional measuring sensor, and the image direction and object space conversion is carried out to obtain the real object space section profile information of the measured road surface, so that good data input is provided for subsequent marking detection and information extraction.
According to the embodiment of the invention, the characteristics of various components contained in a three-dimensional road scene and the characteristics of sparsity, vibratility, low frequency and the like of some components of the road are respectively represented, so that local rapidly-changed information such as cracks, marking edges and pit and groove edges of the road scene are abstracted into sparse components through the sparse characteristics; the method comprises the following steps of abstracting the road surface texture fluctuation characteristic into vibration components through the vibration characteristic, abstracting the road surface gradual change characteristic into low-frequency components through the low-frequency characteristic, wherein the three-dimensional road surface typical component types after abstract expression comprise: low frequency information components, sparse components, and vibration components.
According to the embodiment of the invention, the adopted three-dimensional data resolution characteristic and length characteristic are utilized, the range characteristic of slow deformation of the road surface is combined, the cut-off frequency for distinguishing the low-frequency component from the high-frequency component in the three-dimensional road surface data is determined, the frequency domain ideal low-pass filter is constructed, and the three-dimensional road surface cross section data is decomposed into the low-frequency component and the high-frequency component.
The embodiment of the invention combines the full variational model to have good denoising capability for keeping the signal detail characteristic for solving the noise-containing signal with the sparse characteristic, obtains the high-frequency component simultaneously containing the vibration component and the sparse component obtained in the steps and obtains the sparse component and the vibration component through the full variational decomposition model.
According to the embodiment of the invention, a decomposition model is combined, sparse components, low-frequency components and vibration components are obtained by decomposing three-dimensional data of a road surface, the sparse components are obtained by using data containing cracks and marked lines through the model, then crack and marked line regions are obtained by using a threshold method and are compared with marked data, and the effectiveness of the method is explained.
The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.