Pit segmentation method based on adaptive statistical analysis and point cloud geometric features

Abstract

The application discloses a kind of based on adaptive statistical analysis and point cloud geometric feature's road surface pit groove segmentation method, this method includes: converting road surface three-dimensional point cloud data into depth map;Local variance of depth map is calculated, and according to local variance, contrast factor is obtained, CLAHE processing is carried out to depth map, and according to contrast factor, bilinear interpolation is carried out to the depth map after CLAHE processing, and contrast enhancement map is obtained;Binarization processing is carried out to contrast enhancement map, and binary result is mapped back to road surface three-dimensional point cloud, and target area point cloud is segmented out;Target area point cloud is carried out smoothing processing, and the point cloud after smoothing processing is carried out multi-feature combination's preselected point screening, and preselected point set is obtained;Plane fitting is carried out to preselected point set, the distance of target area point cloud to fitting plane is calculated, and based on distance, road surface pit groove point cloud is extracted.The application can quickly and effectively extract out the pit groove of road surface, especially suitable for the road condition of existing slope.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional road detection technology, specifically to a method for road surface pothole segmentation based on adaptive statistical analysis and point cloud geometric features. Background Technology

[0002] Road defect detection is a challenging aspect of road maintenance. Traditional methods rely on images, employing algorithms such as image segmentation and edge detection. However, images are susceptible to factors like weather and lighting conditions. Point clouds, on the other hand, are datasets composed of three-dimensional coordinate points that directly describe the spatial morphology of a target object or environment. In the field of road detection, point clouds offer richer details and texture information compared to images. Summary of the Invention

[0003] The main objective of this invention is to provide a method for road surface pothole segmentation based on adaptive statistical analysis and point cloud geometric features, which can quickly and effectively extract potholes on the road surface.

[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: In a first aspect, the present invention provides a method for road surface pothole segmentation based on adaptive statistical analysis and point cloud geometric features, the method comprising: Convert road surface 3D point cloud data into depth maps; Calculate the local variance of the depth map, obtain the contrast factor based on the local variance, perform CLAHE processing on the depth map, and perform bilinear interpolation on the CLAHE-processed depth map based on the contrast factor to obtain a contrast-enhanced map. The contrast enhancement image is binarized, and the binarization result is mapped to the 3D point cloud of the loop surface to segment the point cloud of the target region. The point cloud of the target region is smoothed, and the smoothed point cloud is filtered by combining multiple features to obtain a pre-selected point set; Plane fitting is performed on the pre-selected point set, the distance from the point cloud of the target area to the fitted plane is calculated, and the point cloud of road potholes is extracted based on the distance.

[0005] Following the above technical solution, the local variance of the depth map is calculated, and the contrast factor is obtained based on the local variance, including: Calculate the local mean of each pixel location in the depth map, and calculate the local variance of each pixel location based on the local mean; The local variance is normalized to obtain the normalized intensity at each pixel location. ,in x and y Indicates pixel position; Based on the normalized intensity, the contrast factor with contrast limitation is obtained.c_desired ,as follows: ; In the formula, This represents the lower limit of the contrast factor. This represents the upper limit of the contrast factor. It is a non-linear adjustment index.

[0006] Following the above technical solution, the depth map is subjected to CLAHE processing, and bilinear interpolation is performed on the CLAHE-processed depth map according to the contrast factor to obtain a contrast-enhanced map, including: Set tile size set T ={ t 1, t 2, ..., t m , ..., t M} and contrast limit threshold set C ={ c 1, c 2, ..., c k , ..., c K};in, t m For the first m Each tile size, M For the number of tile sizes, c k For the first k One contrast limit threshold, K Limit the number of contrast thresholds, and c 1<…< c k <…< c K ; Combining different tile sizes with different contrast limit thresholds to refine the depth map In each ( c k , t m Combined with contrast-limited adaptive histogram equalization: ; In the formula, CLAHE represents contrast-limited adaptive histogram equalization. Indicates after CLAHE processing The pixel value at the pixel location; Based on the contrast factor, the same size after CLAHE processing... t m Perform bilinear interpolation on the depth map below to obtaint m Interpolated image at a given size; By combining the normalized intensity, the interpolated images at different sizes are weighted and fused to obtain a contrast-enhanced image.

[0007] Following the above technical solution, based on the contrast factor, the same size after CLAHE processing... t m Perform bilinear interpolation on the depth map below to obtain t m Interpolated images at various sizes include: When contrast limiting factor c_desired Meet the conditions c j ≤ c_desired ≤ c j+1 At that time, apply the following formula to each pixel: ; In the formula, α =( c_desired - c j ) / ( c j+1 - c j ), c j and c j+1 The contrast limit threshold is set for two adjacent elements. express t m Interpolated image at a given size; By combining the normalized intensity, interpolated images of different sizes are weighted and fused to obtain a contrast-enhanced image, including: For different tile sizes, calculate the Gaussian weighting factor based on normalized intensity: ; In the formula, for t m Size Gaussian weighting factor at pixel location β Sensitivity weighting factor; u m Is related to tile size t m The corresponding eigenvalues ​​are uniformly distributed in the interval [0, 1]. The Gaussian weight factors at the same pixel position under different tile sizes are normalized to obtain the final weight. ; Based on the final weight The interpolated images at different sizes are weighted and fused to obtain a contrast-enhanced image: ; In the formula, This is a contrast-enhanced image.

[0008] Following the above technical solution, the contrast enhancement image is binarized, and the binarization result is mapped to the 3D point cloud of the loop surface to segment the target region point cloud, including: Perform global pixel histogram statistics on the contrast enhancement map to obtain the pixel mean. and standard deviation ; Define dynamic deviation factor : ; Combined with custom thresholds Perform binarization judgment: ; Where tanh is the hyperbolic tangent function, and k is an adjustable parameter; For each pixel, based on a custom threshold Categorize: ; In the formula, This represents a binary image, with values ​​ranging from {0, 1}; where, =1 indicates The pixel location could be a pothole in the road. =0 indicates The pixel location represents a normal road surface. The binarized result is mapped to the 3D point cloud of the loop surface, and the target area point cloud that may be a road pothole is segmented out.

[0009] Following the above technical solution, the point cloud of the target region is smoothed, including: The point cloud of the target region is smoothed by three B-spline processes along the rows, and then Savitzky-Golay filtering is applied along the columns.

[0010] Following the above technical solution, the smoothed point cloud is subjected to multi-feature combination pre-selection point screening to obtain a pre-selected point set, including: Calculate the neighborhood uniformity feature for each point Calculation point p i Neighborhood point-to-point p i The distance, and the standard deviation of the distance are calculated based on the distance. σ i Get points pi neighborhood uniformity characteristics ; Calculate the normal vector consistency feature for each point Calculation point p i normal vector and points p i neighborhood points normal vector Get points p i Normal vector consistency feature ;in, Point p i The neighborhood point set, For point p i The number of neighborhood points; Calculate the local smoothness features at each point Calculation point p i eigenvalues ​​of the covariance matrix Get points p i Local smoothness features ; Neighborhood uniformity features Normal vector consistency feature and local smoothness features Normalization is performed to obtain the corresponding normalization index. , and ; Each normalized index is divided into an equal number of continuous intervals (bins), and the probability of each normalized index falling into each interval is calculated. : ; In the formula, The total number of normalized index i; Calculate the entropy of each normalized index. ;in, It is an interval number; Normalizing the entropy yields adaptive weights: ; In the formula, This represents the adaptive weight of the normalization index i; , and These are the normalized indices of entropy for neighborhood uniformity, normal vector consistency, and local smoothness, respectively. Based on adaptive weights, the neighborhood uniformity feature of each point is... Normal vector consistency feature and local smoothness features After weighted fusion, we get: ; In the formula, For the weighted fusion result, , and These are the adaptive weights for neighborhood uniformity features, normal vector consistency features, and local smoothness features, respectively. Set threshold coefficient ,like Then point p i We obtain a set of pre-selected points.

[0011] Following the above technical solution, the extraction of road surface pothole point clouds based on distance includes: Set a distance threshold, filter out points whose distance exceeds the threshold, and perform connectivity analysis to finally obtain the point cloud of road potholes.

[0012] In a second aspect, the present invention provides a computer device / apparatus / system, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method described in the first aspect.

[0013] Thirdly, the present invention provides a computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implement the steps of the method described in the first aspect.

[0014] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: This invention combines adaptive statistical analysis with point cloud geometric features to achieve point cloud segmentation of road potholes. Through adaptive contrast factor adjustment based on local variance, combined with CLAHE processing using multi-size tiles and multiple contrast thresholds, bilinear interpolation, and Gaussian weighted fusion, precise enhancement of depth map contrast is achieved. The contrast-enhanced image is adaptively binarized using a custom threshold based on a dynamic deviation factor, and the result is mapped to a loop surface 3D point cloud to segment the target region point cloud. Then, adaptive weighted fusion of multiple features such as neighborhood uniformity, normal vector consistency, and local smoothness is used to filter pre-selected points. Combined with MLESAC plane fitting and point-to-surface distance and connectivity analysis, pothole point clouds are extracted, significantly improving segmentation accuracy and robustness. This method is particularly suitable for road scenarios with slopes, enabling rapid and effective extraction of pothole point clouds from the road surface. It provides reliable technical support for defect detection in road maintenance, significantly improving the accuracy and scene adaptability of road pothole detection. Attached Figure Description

[0015] Figure 1 This is a flowchart of a road surface pothole segmentation method based on adaptive statistical analysis and point cloud geometric features according to an embodiment of the present invention; Figure 2 This is a depth map of a three-dimensional point cloud of a road surface according to an embodiment of the present invention; Figure 3 This is a contrast enhancement diagram according to an embodiment of the present invention; Figure 4 This is a map showing the point cloud extraction result of road surface potholes according to an embodiment of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. All other embodiments obtained by those skilled in the art based on the embodiments provided by this invention without inventive effort are within the scope of protection of this invention.

[0017] Obviously, the accompanying drawings described below are merely some examples or embodiments of the present invention. Those skilled in the art can apply the present invention to other similar scenarios based on these drawings without any inventive effort. Furthermore, it is understood that although the efforts made in this development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this invention, modifications to design, manufacturing, or production based on the technical content disclosed in this invention are merely conventional technical means and should not be construed as insufficient disclosure of the present invention.

[0018] In this invention, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this invention may be combined with other embodiments without conflict.

[0019] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "a," "an," "an," "the," and similar words used in this invention do not indicate quantity limitation and may indicate singular or plural. The terms "comprising," "including," "having," and any variations thereof used in this invention are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to these processes, methods, products, or devices. The terms "connected," "linked," "coupled," and similar words used in this invention are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "A plurality" used in this invention refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships may exist; for example, "A and / or B" can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following objects have an "or" relationship. The terms "first," "second," and "third" used in this invention are merely to distinguish similar objects and do not represent a specific ordering of the objects.

[0020] This invention provides a method for road surface pothole segmentation based on adaptive statistical analysis and point cloud geometric features. The method includes: obtaining a depth map from point cloud data through elevation projection. I 0; Calculate the image I The local variance of 0 is used to obtain the contrast factor. Then, CLAHE processing is performed on the image, and bilinear interpolation is performed on the CLAHE-processed data based on the contrast factor to finally obtain the contrast-enhanced image. I C ; for contrast-enhanced images I CBinarization is performed, and the binarized result is mapped back to the original 3D point cloud of the road surface as the Region of Interest (ROI). Smaller point clouds are then segmented around the ROIs. Cubic B-spline smoothing and Savitzky-Golay filtering are applied to these smaller point clouds. A pre-selected point set with multiple features is then used to filter the smoothed and filtered point clouds. MLESAC plane fitting is performed on the selected point set. The point set after MLESAC plane fitting is extracted using the points' distances to the fitted plane. Connectivity analysis is performed on the extracted results to obtain the final extraction result. This invention can quickly and effectively extract potholes from 3D road surfaces, and is particularly suitable for roads with slopes.

[0021] like Figure 1 As shown, the road surface pothole segmentation method based on adaptive statistical analysis and point cloud geometric features of the present invention specifically includes the following steps: S1. Obtain a depth map from the point cloud data through elevation projection. I 0; S2, Calculate the image I The local variance of 0 is used to obtain the contrast factor. Then, CLAHE processing is performed on the image, and bilinear interpolation is performed on the CLAHE-processed data based on the contrast factor to finally obtain the contrast-enhanced image. I C ; S3, Enhance contrast image I C Binarization is performed, and the binarization result is mapped back to the original 3D point cloud of the road surface as the ROI region. The point cloud of small regions is then segmented with the ROI region as the center. S4. Perform cubic B-spline smoothing and Savitzky-Golay filtering on the small region point cloud to smooth the point cloud. Then, perform multi-feature combination pre-selection point set screening on the smoothed and filtered point cloud, and perform MLESAC plane fitting on the screened point set. S5. Extract the point set after fitting the MLESAC plane using the points to the fitted plane and the distance, perform connectivity analysis on the extracted results, and obtain the final extraction result.

[0022] Specifically, in step S1, the 3D point cloud data of the road surface is projected onto an elevation map to obtain a depth map, as follows: First, the original point cloud is divided into several blocks. m × n Calculate the average height of the small blocks. : ; in, N This represents the number of points in a small point cloud block. Z iThis represents the height value of the center point of the small dot cloud.

[0023] It should be noted that the size of the small point cloud in this embodiment is 5×5.

[0024] like Figure 2 As shown, a depth map of the point cloud is generated based on the average height. A small patch of the point cloud is represented as a pixel in the depth map: The average height is normalized, and then the normalized data is processed to generate a depth map. The normalization formula is: ; in, It is the minimum value of the average height of the point cloud blocks. It is the maximum value of the average height of the point cloud blocks. It is the average height of the current small point cloud.

[0025] Generate depth map: ; in, Represents the depth map The pixel value at the pixel location.

[0026] In step S2, the contrast factor is obtained from the image based on the local variance, including: For input grayscale image I 0. First, calculate the local mean using window w: ; in, It is the number of pixels within window w.

[0027] Then calculate the local variance of the window: ; Meanwhile, to avoid the influence of extreme values, the local variance is linearly normalized to obtain the normalization intensity at each pixel location: ; in, It is the minimum value of the calculated local variance. It is the maximum value of the calculated local variance.

[0028] Then, by using the normalized intensity, the contrast factor with contrast limitation is obtained. c_desired Its mathematical representation is: ; In the formula, This is the lower limit of the contrast factor, representing the minimum intensity of contrast enhancement; This is the upper limit of the contrast factor, representing the maximum intensity of contrast enhancement; η It is a non-linear adjustment index. In areas with complex textures, Approaching 1, at this point Approaching 0, c_desired = c min In smooth / low contrast areas, Approaching 0, at this point Approaching 1, c_desired = c max By setting maximum and minimum clipping thresholds, contrast can be enhanced in areas with small pixel variations, while the upper limit of variation in areas with strong textures is lowered, preventing excessively large peak values ​​in areas with strong textures.

[0029] In step S2, the CLAHE result is bilinearly interpolated based on the contrast factor to obtain a contrast-enhanced image. The specific steps are as follows: Different tile sizes T ={ t 1, t 2, ..., t m , ..., t M} and different clip_levels Contrast Limit Threshold C ={ c 1, c 2, ..., c k , ..., c K Combine the images I 0 in each ( c k , t m Combined with contrast-limited adaptive histogram equalization: ; In the formula, CLAHE represents contrast-limited adaptive histogram equalization. Indicates after CLAHE processing The pixel value at the pixel location.

[0030] To avoid boundary effects and pixel value jumps, for the same size t m The results are then bilinearly interpolated. This is done when the contrast limiting factor meets the condition. c j ≤ c_desired ≤ c j+1 At that time, the following formula is applied to each pixel of the image: ; in, α =( c_desired - c j ) / ( c j+1 - c j ), c j This indicates the current contrast limit threshold. c j+1 Indicates and c j Adjacent contrast thresholds are used. This operation yields the interpolated image result. I t .

[0031] Then, for different tile sizes, normalized strength is calculated. The Gaussian weighted fusion is performed as follows: First, calculate the Gaussian weighting factor: ; in, for t m Size Gaussian weighting factor at pixel position β Control the sensitivity weighting factor. u m Is related to tile size t m The corresponding feature values ​​are evenly distributed in the interval [0, 1]. That is, the interval [0, 1] is divided into the same number of sub-intervals according to the number of tile sizes, and the tile sizes correspond to the interval endpoint values ​​in order from small to large.

[0032] Will Weight normalization is performed to obtain w m , w m satisfy: .

[0033] like Figure 3 As shown, the final interpolated images of different sizes... Weighted fusion yields an enhanced graph I C : ; In the formula, This is a contrast-enhanced image.

[0034] It should be noted that in this embodiment, the window size is 3×3 when calculating the local variance, and the tile size set is... T ={8, 16, 32}; clip_levels The set is C ={2.0, 3.5, 6.0}. Of course, these three parameters can also be set according to the actual situation.

[0035] In step S3, the image after CLAHE enhancement is... I C Perform linear normalization, and then adaptively binarize the normalization result. The specific steps are as follows: Perform global pixel histogram statistics on the contrast enhancement map to obtain the pixel mean. and standard deviation ; Define a dynamic deviation factor : ; Combined with custom thresholds Perform binarization judgment: ; Where tanh is the hyperbolic tangent function. k It is an adjustable parameter.

[0036] For each pixel Based on a custom threshold Categorize: ; At this time, it was obtained This is a binary image with values ​​ranging from {0, 1}. =1 indicates that the area may be a road pothole. =0 indicates that the area is a normal road surface.

[0037] It should be noted that, in this embodiment, the adjustable parameters k The value is 0.25.

[0038] In step S3, the binarized result is mapped back to the original 3D point cloud of the road surface as the Region of Interest (ROI). This ROI is essentially a portion of the pothole, primarily the bottom of the pothole. The point cloud is then segmented into smaller regions centered on the ROI. Specifically, each ROI boundary is used as the center, and a circle with a radius of 5 is extended outwards. The elevation difference between points on the extended circle and the boundary points is determined. If the elevation difference is less than a threshold, the desired location has been reached. If the elevation difference is greater than the threshold, the circle is used as the new center, and the expansion continues until a suitable location is found. This location is then used as the boundary to cut the point cloud into smaller regions. After finding a suitable location, the maximum and minimum X and Y values ​​of all new boundaries of this smaller region's point cloud are used as cutting lines to further divide the point cloud into smaller regions. The final smaller region's point cloud is a rectangle.

[0039] In step S4, cubic B-spline smoothing is performed on the segmented small region point cloud. The specific steps are as follows: First, perform cubic B-spline smoothing on the point cloud along the Y direction (each row). For each fixed point cloud... x = x i Smoothing is achieved by adjusting the Z value of each point: ; At the same time, when processing the boundary, the points need to be copied, that is: .

[0040] After the cubic B-spline smoothing is completed, Savitzky-Golay filtering is then performed along the X direction (for each column): For each column j, obtain the height sequence: ; At each scan line position i At this point, take a length of 2. m +1 local window: ; in, m It is half the width of the window.

[0041] Within this window, a least-squares fit is performed on the height sequence. p order polynomial: ; in, τ ( τ = m , m +1,…, m() represents the displacement relative to the center of the window; a 0, a 1, ..., a p These are the polynomial coefficients.

[0042] The principle of polynomial fitting is: select a set of coefficients. a The goal is to make the polynomial curve approximate the true height sequence within the window. This is transformed into a least squares problem: ; in, w r ≥ 0 represents a weight, used to suppress the influence of outliers or missing points on the fit.

[0043] The above least squares problem can be expressed in matrix form: 1) Construct the Vandermonde matrix: ; Each row corresponds to the offset of a point within the window, and each column corresponds to a different order of the polynomial.

[0044] 2) Construct the weight matrix W: ; Among them, diagonal elements w r Indicates the weight of the corresponding window point; w r = 0 indicates that the point is an outlier or a missing point and will not be included in the fitting; w r = 1 indicates that the point is valid and fully participates in the fitting; off-diagonal elements are zero, and the weight matrix is ​​a diagonal matrix used for weighted least squares.

[0045] 3) Matrix-form weighted least squares problem: ; Corresponding normal equation: ; This equation shows the polynomial coefficients. a It uses matrix V (window offset information), weight W (outlier control), and window height. v j (i) The solution was obtained by combining the two methods.

[0046] Seeking system a =[ a 0, a 1, ..., a p ] T Then, the polynomial is in the center of the window. τ The predicted value for = 0 is: ; This value is used to replace the original height. v i,j .

[0047] For each column j All scan line positions i =1.. N x Repeat the above process to merge all columns and obtain the smoothed point cloud result.

[0048] It should be noted that in this embodiment, the window width m = 6 and the polynomial order p = 3. The combined processing of cubic B-spline interpolation and Savitzky-Golay filtering results in a smoother point cloud, facilitating subsequent processing.

[0049] In step S4, the smoothed and filtered point cloud undergoes pre-screening of planar point sets by combining multiple features. The specific steps are as follows: Calculate the neighborhood uniformity feature for each point p. i Calculate the distance from the neighboring points to the center point: ; Calculate the average distance of the neighborhood and standard deviation σ i : ; ; Define neighborhood uniformity features U i : ; Among them, point p j Point p i The nearest neighbor, k Point p i The number of nearest neighbors, N k Point p i The set of nearest neighbor numbers.

[0050] Normalize the neighborhood uniformity to obtain U .

[0051] Calculate the normal vector consistency feature for each point p. i Calculate the consistency of the normal vectors of its neighboring points: ; Normalize the consistency of the normal vectors of the neighboring points to obtain C .

[0052] Local smoothness features of calculation points S i : ; in, λ i The eigenvalues ​​of the covariance matrix of the points satisfy the following conditions: λ 0≤ λ 1≤ λ 2.

[0053] Normalize the smoothness feature to obtain S .

[0054] Each normalization index is divided into N bin Each of the continuous intervals bin The scope is: ; in, b 0=min( X ), b Nbin =max(X), and the other dividing points are evenly divided.

[0055] No. j indivual bin probability p j i for: ; in, j = 1, 2, ..., N bin .

[0056] Calculate the entropy index: ; Entropy normalization to adaptive weights: ; In the formula, This represents the adaptive weight of the normalization index i; , and These are the normalized index entropy for neighborhood uniformity features, normal vector consistency features, and local smoothness features, respectively.

[0057] By weighting and fusing the three features of uniformity, normal vector consistency, and smoothness, we obtain: ; In the formula, For the weighted fusion result, , and These are the adaptive weights for neighborhood uniformity features, normal vector consistency features, and local smoothness features, respectively. Based on the threshold coefficient The weighted fusion results are filtered to obtain pre-selected points for plane fitting: ; in, T =1 indicates that the corresponding point is a pre-selected point.

[0058] It should be noted that, in this embodiment, the interval N bin The size is set to 10, the number of neighboring points is set to 30, and the threshold coefficient is set to... = 0.7.

[0059] In step S5, the selected candidate point set is fitted to a plane using the MLESAC method, and the distance from each point to the fitted plane is used to segment the pits. The fitted plane obtained using the MLESAC method is Ax + By + Cz + D = 0. The normal vector of the plane is... n = ( A , B , C ), calculate non-planar points p i = ( x i , y i , z i The distance to the fitted plane is d i = Ax i + By i + Cz i + D Points meeting certain criteria are selected based on a distance threshold, and connectivity analysis is performed on these points. Specifically, using the point cloud mapped back from the previous binarization results as a baseline, points exceeding the distance threshold are selected, and connectivity analysis is performed to obtain the final extraction results. Finally, the pothole extraction results can be processed, and all extracted road surface potholes can be returned to the original road surface point cloud according to their coordinates for display, allowing for a more intuitive view of the road surface potholes, such as... Figure 4 As shown.

[0060] Furthermore, the present invention also provides a computer device / apparatus / system, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above-described method.

[0061] The present invention also provides a computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0062] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, systems, and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in one or more blocks of the flowchart illustrations and / or block diagrams.

[0063] These computer program instructions may also be stored in a computer-readable storage medium capable of directing a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more flowcharts and / or one or more blocks in a block diagram.

[0064] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable data processing apparatus to produce a computer-implemented process, thereby providing steps for implementing the functions specified in one or more flowcharts and / or one or more blocks in a block diagram.

[0065] It should be noted that, depending on the implementation needs, the various steps / components described in this invention can be broken down into more steps / components, or two or more steps / components or parts of the operation of steps / components can be combined into new steps / components to achieve the purpose of this invention.

[0066] Those skilled in the art will readily understand that the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for road surface pothole segmentation based on adaptive statistical analysis and point cloud geometric features, characterized in that, The method includes: Convert road surface 3D point cloud data into depth maps; Calculate the local variance of the depth map, obtain the contrast factor based on the local variance, perform CLAHE processing on the depth map, and perform bilinear interpolation on the CLAHE-processed depth map based on the contrast factor to obtain a contrast-enhanced map. The contrast enhancement image is binarized, and the binarization result is mapped to the 3D point cloud of the loop surface to segment the point cloud of the target region. The point cloud of the target region is smoothed, and the smoothed point cloud is filtered by combining multiple features to obtain a pre-selected point set; Perform plane fitting on the pre-selected point set, calculate the distance from the point cloud of the target area to the fitted plane, and extract the point cloud of road potholes based on the distance; The calculation of the local variance of the depth map, and the determination of the contrast factor based on the local variance, includes: Calculate the local mean of each pixel location in the depth map, and calculate the local variance of each pixel location based on the local mean; The local variance is normalized to obtain the normalized intensity at each pixel location. ,in x and y Indicates pixel position; Based on the normalized intensity, the contrast factor with contrast limitation is obtained. c_desired ,as follows: ； In the formula, This represents the lower limit of the contrast factor. This represents the upper limit of the contrast factor. It is a non-linear adjustment index.

2. The method for road surface pothole segmentation based on adaptive statistical analysis and point cloud geometric features according to claim 1, characterized in that, The depth map is subjected to CLAHE processing, and bilinear interpolation is performed on the CLAHE-processed depth map according to the contrast factor to obtain a contrast-enhanced map, including: Set tile size set T ={ t 1, t 2, ..., t m , ..., t M } and contrast limit threshold set C ={ c 1, c 2, ..., c k , ..., c K };in, t m For the first m Each tile size, M For the number of tile sizes, c k For the first k One contrast limit threshold, K Limit the number of contrast thresholds, and c 1<…< c k <…< c K ; Combining different tile sizes with different contrast limit thresholds to refine the depth map In each ( c k , t m Combined with contrast-limited adaptive histogram equalization: ； In the formula, CLAHE represents contrast-limited adaptive histogram equalization. Indicates after CLAHE processing The pixel value at the pixel location; Based on the contrast factor, the same size after CLAHE processing... t m Perform bilinear interpolation on the depth map below to obtain t m Interpolated image at a given size; By combining the normalized intensity, the interpolated images at different sizes are weighted and fused to obtain a contrast-enhanced image.

3. The method for road surface pothole segmentation based on adaptive statistical analysis and point cloud geometric features according to claim 2, characterized in that, Based on the contrast factor, the same size after CLAHE processing... t m Perform bilinear interpolation on the depth map below to obtain t m Interpolated images at various sizes include: When contrast limiting factor c_desired Meet the conditions c j ≤ c_desired ≤ c j+1 At that time, apply the following formula to each pixel: ； In the formula, α =( c_desired - c j ) / ( c j+1 - c j ), c j and c j+1 The contrast limit threshold is set for two adjacent elements. express t m Interpolated image at a given size; By combining the normalized intensity, interpolated images of different sizes are weighted and fused to obtain a contrast-enhanced image, including: For different tile sizes, calculate the Gaussian weighting factor based on normalized intensity: ； In the formula, for t m Size Gaussian weighting factor at pixel location β Sensitivity weighting factor; u m Is related to tile size t m The corresponding eigenvalues ​​are uniformly distributed in the interval [0, 1]. The Gaussian weight factors at the same pixel position under different tile sizes are normalized to obtain the final weight. ; Based on the final weight The interpolated images at different sizes are weighted and fused to obtain a contrast-enhanced image: ； In the formula, To enhance contrast in the image The pixel value at the pixel location.

4. The method for road surface pothole segmentation based on adaptive statistical analysis and point cloud geometric features according to claim 1, characterized in that, The contrast enhancement image is binarized, and the binarization result is mapped to the 3D point cloud of the loop surface to segment the target region point cloud, including: Perform global pixel histogram statistics on the contrast enhancement map to obtain the pixel mean. and standard deviation ; Define dynamic deviation factor : ； Combined with custom thresholds Perform binarization judgment: ； Where tanh is the hyperbolic tangent function. It is an adjustable parameter; For each pixel Based on a custom threshold Categorize: ； In the formula, This represents a binary image, with values ​​ranging from {0, 1}; where, =1 indicates The pixel location could be a pothole in the road. =0 indicates The pixel location represents a normal road surface. The binarized result is mapped to the 3D point cloud of the loop surface, and the target area point cloud that may be a road pothole is segmented out.

5. The method for road surface pothole segmentation based on adaptive statistical analysis and point cloud geometric features according to claim 1, characterized in that, Smoothing of the point cloud in the target region includes: The point cloud of the target region is smoothed by three B-spline processes along the rows, and then Savitzky-Golay filtering is applied along the columns.

6. The method for road surface pothole segmentation based on adaptive statistical analysis and point cloud geometric features according to claim 1, characterized in that, The smoothed point cloud is subjected to multi-feature combination pre-selection point filtering to obtain a pre-selected point set, including: Calculate the neighborhood uniformity feature for each point Calculation point p i Neighborhood point-to-point p i The distance, and the standard deviation of the distance are calculated based on the distance. σ i Get points p i neighborhood uniformity characteristics ; Calculate the normal vector consistency feature for each point Calculation point p i normal vector and points p i neighborhood points normal vector Get points p i Normal vector consistency feature ;in, Point p i The neighborhood point set, For point p i The number of neighborhood points; Calculate the local smoothness features at each point Calculation point p i eigenvalues ​​of the covariance matrix Get points p i Local smoothness features ; Neighborhood uniformity features Normal vector consistency feature and local smoothness features Normalization is performed to obtain the corresponding normalization index. , and ; Each normalized index is divided into an equal number of continuous intervals (bins), and the probability of each normalized index falling into each interval is calculated. : ； In the formula, Normalized index The total quantity; Calculate the entropy of each normalized index. ;in, It is an interval number; Normalizing the entropy yields adaptive weights: ； In the formula, Indicates the normalization index Adaptive weights; , and These are the normalized indices of entropy for neighborhood uniformity, normal vector consistency, and local smoothness, respectively. Based on adaptive weights, the neighborhood uniformity feature of each point is... Normal vector consistency feature and local smoothness features After weighted fusion, we get: ； In the formula, For the weighted fusion result, , and These are the adaptive weights for neighborhood uniformity features, normal vector consistency features, and local smoothness features, respectively. Set threshold coefficient ,like Then point p i We obtain a set of pre-selected points.

7. The method for road surface pothole segmentation based on adaptive statistical analysis and point cloud geometric features according to claim 1, characterized in that, Distance-based extraction of road surface pothole point clouds includes: Set a distance threshold, filter out points whose distance exceeds the threshold, and perform connectivity analysis to finally obtain the point cloud of road potholes.

8. A computer device or apparatus or system, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 7.

Images