CN115810153B - Highway point cloud extraction method based on relative entropy and semantic constraint - Google Patents
Highway point cloud extraction method based on relative entropy and semantic constraint Download PDFInfo
- Publication number
- CN115810153B CN115810153B CN202310077847.9A CN202310077847A CN115810153B CN 115810153 B CN115810153 B CN 115810153B CN 202310077847 A CN202310077847 A CN 202310077847A CN 115810153 B CN115810153 B CN 115810153B
- Authority
- CN
- China
- Prior art keywords
- voxels
- point cloud
- voxel
- point
- monomer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
The invention relates to the technical field of expressway point cloud extraction, in particular to an expressway point cloud extraction method based on relative entropy and semantic constraint, which comprises the following steps of: acquiring road original point cloud data and RGB information thereof, and obtaining a ground point cloud set T and a non-ground point cloud F through self-adaptive filtering; initializing a ground point cloud set T; performing similarity constraint of points, clustering point clouds into voxels at the same time, and establishing a voxel adjacency list; further expanding and contracting the voxels by using a relative entropy function, and superposing the voxels by combining the structural similarity index with a relative entropy probability value; and carrying out semantic constraint on the single body, and carrying out flatness judgment to finally obtain a final road point cloud result. The invention solves the problem of unstable or invalid extraction result in complex road environment by setting point similarity constraint, relative entropy function and structural similarity index; by setting semantic constraint optimization classification, a more accurate road extraction result is obtained.
Description
Technical Field
The invention relates to the technical field of expressway point cloud extraction, in particular to an expressway point cloud extraction method based on relative entropy and semantic constraint.
Background
With the rapid development of laser radar technology and the popularization of multi-platform laser scanners, a large amount of obtained point cloud data provides effective and reliable data support for three-dimensional reconstruction of the earth surface. The vehicle-mounted laser is suitable for road scenes with high precision requirements, makes up the limitation of acquiring accurate ground objects on board, has the advantages of high point density, stability and the like, and how to quickly convert complex mass point cloud data into identifiable and determinable effective data in the traffic industry after the vehicle-mounted laser scans the point cloud data, so that the vehicle-mounted laser becomes a focus of attention, wherein the extraction of the vehicle-mounted road point cloud is an important link of the vehicle-mounted point cloud for road modeling, vehicle-road coordination and other applications.
In the production process, the highway pavement extraction is mainly divided into two modes, one is that by manual plotting, the working efficiency is low, the cost is high, and the road information updating period is longer; the other is to extract the road by using road characteristics, empirical threshold and other construction algorithms; the vehicle-mounted road point cloud extraction algorithm is mainly divided into the following three types: 1. the road point cloud is extracted by utilizing the height difference and the gradient difference, and the method has strict requirements on the precision of the original data, is only suitable for extracting the road surface information with single environmental background and obvious high Cheng Tidu difference, and cannot be applied to the expressway road surface extraction work in mountain areas with complex natural topography; 2. generating scanning lines by utilizing the point cloud, and obtaining the road point cloud by combining with filtering and segmentation such as a moving window or low-pass filtering, wherein the method does not consider the interrelation between ground objects and is easy to misjudge; 3. the spatial clustering algorithm considers various characteristic parameters such as geometric lines, elevations, textures and the like of the pavement, but the algorithm consumes a long time, a large number of threshold judgment conditions are required to be set in the clustering process, and the phenomenon of excessive segmentation is easy to occur when the characteristic parameters are not fully considered.
Disclosure of Invention
The invention aims to overcome the defects in the prior art, and discloses a highway point cloud extraction method based on relative entropy and semantic constraint.
In order to achieve the above object, the present invention provides the following technical solutions:
a highway point cloud extraction method based on relative entropy and semantic constraint comprises the following steps:
s1, acquiring road original point cloud data and RGB information thereof, and filtering ground points and non-ground points on the original point cloud data and the RGB information thereof through self-adaptive filtering to obtain a ground point cloud set T and a non-ground point cloud F;
s2, initializing a ground point cloud set T in the S1, wherein the ground point cloud set T comprises discrete points with similar spatial distribution, RGB information and intensity value information, obtaining point clouds with clear adjacency relations through similarity constraint of set points, clustering the point clouds into voxels at the same time, and establishing a voxel adjacency list;
s3, establishing a monomer and a monomer adjacency list, and further expanding and contracting voxels in the S2 by using a relative entropy function, wherein the relative entropy function further optimizes the probability that the ground point cloud set T falls on the corresponding voxels; superposing the structural similarity indexes and the relative entropy probability values to obtain more accurate similarity measurement, distinguishing and dividing voxels into monomers, and updating the monomers and the monomer adjacency list;
S4, carrying out semantic constraint on the monomers in the S3, selecting the monomers to judge whether the class labels are 'roads' or 'non-roads', carrying out flatness judgment, and traversing and judging all voxels and the monomers to obtain a final road point cloud result.
Further, the step S1 includes:
s11, acquiring road origin point cloud data with intensity information, and simultaneously acquiring a road panoramic image;
s12, fusing the original point cloud data and the panoramic image to obtain point cloud data with RGB information and intensity information;
s13, dividing the point cloud data in the S12 into ground points and non-ground points through an adaptive filtering algorithm.
Further, the specific steps of the adaptive filtering algorithm for dividing the point cloud data into ground points and non-ground points are as follows:
when the self-adaptive filtering is used, a grid window is constructed, the points to be classified are filtered gradually from large to small, the filtering threshold value is self-adaptively adjusted through an energy formula and the size of the grid window, other points adhered to the ground point cloud are removed, and iteration is carried out until the threshold value or the iteration times are set; repeating the steps until the iteration times are larger than the set maximum iteration times, and obtaining a ground point cloud set T and a non-ground point cloud F after the self-adaptive filtering process is finished.
Further, the step S2 includes:
s21, constructing an OCTree (OCTree), wherein the OCTree constructing comprises:
traversing the ground point cloud set T, and finding the minimum size of a cube bounding box (Axis Aligned Bounding Box, AABB), wherein the coordinates are (Xmin, xmax, ymin, ymax, zmin, zmax); iteratively processing all the ground point clouds, for each point, respectively comparing the three-dimensional coordinates (X, Y, Z) of the point with the current AABB range, if the point falls outside the AABB range, updating the AABB range, calculating the maximum value R=MAX (Xmax-Xmin, ymax-Ymin, zmax-Zmin), updating the AABB range, and finally, obtaining the side length R= (Xmax-Xmin) = (Ymax-Ymin) = (Zmax-Zmin);
three-dimensional space where ground point cloud set T is locatedIs regarded as by
A cubic composition with a side length r, wherein m, n, k are positive integers, wherein +.>
Each minimum cube unit containing three-dimensional points is regarded as one leaf node of the octree, so that an octree structure of the three-dimensional point cloud can be constructed, and the depth of the octree is set as follows;
equal division of octree space is performed, wherein q in formula (1) is octal, from q 0 To q n-1 The path of each leaf node in the octree to the root is shown in its entirety;
Decomposing the bounding box into 8 subcubes Q according to equation (1) 8 Numbering the cubes, if the maximum traversing depth is not reached and the node does not have sub-nodes, dividing the cubes into 8 equally divided sub-cubes until the traversing depth is set depth, stopping traversing, and constructing an octree structure of a three-dimensional point cloud with the resolution of r, wherein each cube, namely one voxel, in the octree has a definite 26 neighborhood, each point represents the center of the voxel, and 26 adjacent voxels are arranged around the center voxel;
if the sub-node (cube) has no three-dimensional point, marking the point as an empty node, and stopping segmentation to obtain an octree structure;
s22, initializing a ground point cloud set T by using the octree after obtaining the octree structure of the voxels, regarding the three-dimensional point set contained in each non-empty leaf node in the octree as an initialized point cloud voxel, and recording coordinate information, RGB information and intensity information of each point cloud voxel by using a list; the coordinate information, RGB information and intensity information of the point cloud voxels are respectively statistical average values of the coordinate information, RGB information and intensity information of three-dimensional points in the voxels; as shown in the formula (2),
Wherein SV represents an average value, x, y and z represent three-dimensional coordinates of points in the voxel, N represents a natural number and represents the number of all points in the point cloud voxel, I represents an ith point, I i The intensity value of the I-th point, SV (x, y, z), SV (R, G, B), SV (I), respectively, represents coordinate information, RGB information, and intensity information;
s23, further processing voxels, establishing a voxel adjacency list, using a KNN neighborhood search algorithm, finding a nearest 26 neighborhood point cloud voxel according to the initialized voxels, searching points in the 26 neighborhood region to determine the furthest distance of the search range, inquiring the voxels in the furthest distance, searching to judge the furthest distance of the update, using the KNN algorithm to write the octree search result into two storage vectors, respectively storing the search result and the square distance between the corresponding search point and the adjacent point by the two storage vectors, and establishing the voxel adjacency list; judging whether the node area is overlapped with the voxels with the farthest distance as the radius, and searching all the voxels, so that the algorithm time is shortened and the efficiency is improved; setting a K value, wherein the K value is the number of voxels, and obtaining an adjacent list of each voxel;
s24, further adjusting points in the voxels, judging the voxels of each three-dimensional point according to the point similarity measure in an iterative mode, and updating the information of the voxels; record the set of all voxels as 
Wherein k is the number of all non-empty leaf nodes in the octree; selecting a point a of any single point belonging to the point cloud set T, calculating the similarity between the point a and any voxel, as shown in a formula (3),
wherein point cloud build similarity relationship considerationsThree factors of coordinate space, color space and intensity space are taken into account, in equation (3)
Representing the i-th voxel in the set of voxels SV, < >>
Euclidean distance of two-point space coordinates, euclidean distance of RGB component, two-point intensity distance, respectively, for the +.>
Normalization processing is performed, as shown in formula (4),
r in the formula (4) is the maximum size resolution of a single voxel, and m is the maximum gray value of each RGB component;
representing the absolute value of the difference in intensity values of the two points;
the final point similarity function is calculated by the following steps:
wherein w in formula (5) 1 ，w 2 ，w 3 Three coefficients, three coefficients w 1 ，w 2 ，w 3 Sizing according to specific experiments, wherein
The smaller the value obtained, the stronger the similarity between the two points;
s25, under the point similarity function, carrying out the voxel formation of the road point cloud, and utilizing the point similarity function
Calculating the similarity of any point a and the three-dimensional center point +.>
And the similarity of any point a and the three-dimensional central point neighborhood voxels +. >
If said->
The value of (2) is greater than said +.>
The point a is given to the voxel sv j In the following, the judgment of sv j If the traversal is finished, if so, updating the voxels, and if not, re-entering the calculation loop step S24;
s26, after the voxel updating is completed, a new voxel set SV= { SV is generated 1 ,sv 2 ....,sv k }。
Further, at the voxel set sv= { SV1, SV2 k Further clustering and label discrimination are carried out on the basis of the }, so that a road monomer is obtained, and the value range of the label of each voxel class is = { road, non-road }, wherein the step S3 comprises:
s31, setting a monomer adjacent list for the monomer, wherein the monomer adjacent list records index numbers of all voxels of corresponding categories adjacent to the monomer in the voxel adjacent list, and the corresponding adjacent list with the same category number as the monomer is empty and is also recorded as the monomer, wherein the category number is a value of a road or a non-road;
s32, further expanding and contracting the voxels by using a relative entropy function, wherein the relative entropy function transforms the target ground point cloud so that the probability of each ground point cloud falling on the corresponding voxel is increased; carrying out mathematical statistics on the intensity values of the point clouds in the voxels, mapping the intensity values to a 0-255 interval, and obtaining the intensity values of the point clouds in the voxels according to the following formula,
Wherein the method comprises the steps of
Representing the intensity value of a certain point, I x The intensity values mapped to the interval from 0 to 255 are represented, N represents the number of points in the voxel, the frequency and probability of each intensity value are defined as p (x), and therefore the intensity entropy of the three-dimensional point cloud is set as follows:
q (x) represents the true distribution of the point cloud voxel intensity values, the relative entropy function simulates the true distribution q (x) by using P (x), and the relative entropy between P (x) and q (x) is as follows:
s33, acquiring gray values from RGB information of the road point cloud by using a structural similarity index, so as to obtain brightness, contrast and structure, wherein the structural similarity index is used for measuring similarity between two images, establishing a brightness, contrast and structural relationship, calculating structural similarity probability, and combining probability values of the relative entropy functions to carry out superposition to obtain similarity measurement, so as to carry out discrimination segmentation from road voxels to monomers; the structural similarity index is commonly used for measuring the similarity between two images, imitates human perception by establishing brightness, contrast and structural relation, focuses on edge and texture similarity, and accords with visual perception of human eyes;
s34, automatically selecting voxels of normal vectors or vectors represented by straight lines forming an included angle of 85-90 degrees with a plane as judging starting points, wherein neighbor voxels of the voxels and neighbor voxels of the neighbor voxels are normal vectors or vectors represented by straight lines forming an included angle of 85-90 degrees with the plane, and the eligible voxels are assigned to road labels, specifically, calculating neighbor points of voxel points of each normal vector or vectors represented by straight lines forming an included angle of 85-90 degrees with the plane, wherein the number of the neighbor points is equal to or less than 26; recording a vector label normal_flag of each voxel point, marking normal_flag=1 if normal is more than or equal to 85 degrees, marking normal_flag=0 if normal is less than 85 degrees, and adding the marking bits of all 26 neighborhood of the voxel point to be used as the final flag value of the voxel point; the marking process is circularly calculated until the voxel with the flag more than or equal to 8 is obtained, the conforming voxel is selected as a starting point voxel y judged in the step S35, and the rest 26 neighborhood voxels are put into a to-be-judged list Q;
S35, overlapping the similarity index obtained by using the structural similarity index and the relative entropy result, and by overlapping two probability values, starting from three aspects of intensity probability distribution, structure and RGB, ensuring the reliability of the voxel classification result, being more suitable for complex environments, as shown in a formula (9),
W=α*（1-SSIM(x,y)）+（1-α）I(p||q) （9）
wherein, alpha is more than 0 and less than 1, and when W is more than 0.1, the voxel is judged to be a road;
s36, matching the voxels of the judgment starting point y selected in the step S34 with the voxels of the list Q to be judged, judging one by one, and counting the voxels with W less than 0.1, wherein the voxels with the minimum value are regarded as roads and classified as roads;
s37, judging one by one from the adjacent list of the road voxels matched at last in the step S36, and counting voxels with W less than 0.1, wherein the voxels with the minimum value are regarded as roads and classified as the roads;
s38, circularly traversing until the voxels in each voxel adjacent list do not meet the threshold value, returning to the step S36, traversing again, searching the smallest voxel which is removed from the voxels and meets the threshold value, and carrying out matching judgment to obtain the final monomer and the monomer adjacent list thereof.
Further, the step S33 includes the following steps:
using the structural similarity index, for point cloud voxels, first convert from RGB space to one-dimensional gray scale, the formula is as follows:
Gray = R * 0.299 + G * 0.587 + B * 0.114 （10）
Taking the Gray value Gray as a calculated parameter x, taking the average Gray value Hx of the voxels as an estimation of brightness space measurement, and giving an 'x' of one voxel for comparison and a 'y' of a road voxel;
the luminance estimation function C (x, y) is shown below, C 1 Is a constant that is not equal to zero,
the contrast estimation function g (x, y) is shown below as C 2 Is a constant that is not equal to zero,
the structure contrast function s (x, y) is shown below,
the structural similarity index formula is as follows, and consists of a brightness estimation function c (x, y), a contrast estimation function g (x, y), and a structural comparison function s (x, y):
selecting simplified command
The method comprises the following steps of:
further, adding knowledge in the semantic level as constraint, and judging and optimizing the category of the monomer obtained in the step S3 according to a category knowledge rule; the step S4 includes:
s41, selecting a single body, judging whether the category label is a road, if so, performing step S42, and if not, performing step S43;
s42, if the number of voxels recorded in the monomer adjacent list is greater than the number of voxels contained in the monomer and the voxel categories recorded in the monomer are all 'roads', reclassifying the monomer as 'roads';
S43, if the number of voxels recorded in the monomer adjacent list is greater than the number of voxels contained in the monomer and the voxel categories recorded in the voxel adjacent list are all non-roads, reclassifying the monomer as non-roads;
further, the step S4 further includes:
s44, carrying out flatness judgment on the monomer optimization processing obtained in the step S42 and the step S43, and classifying the monomer as a road if the value of a flatness function of the voxel is smaller than a threshold value of 0.5 and more than 50% of the categories of the voxels recorded in the monomer adjacent list are non-roads; the flatness function is used for calculating the average value of the integral elevation standard deviation of all K individual elements
As shown in the following formula,
wherein n represents the number of points in the voxel, z i The elevation value of the i-th point is indicated,
representing an elevation value of the kth voxel;
and (3) automatically traversing and judging all the point cloud voxels and monomers in the ground point cloud data by utilizing the semantic constraint of the steps S42 to S44 to obtain a final road point cloud result.
As a preferred embodiment of the present invention,
an electronic device comprising at least one processor, and a memory communicatively coupled to the at least one processor; the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of the preceding claims.
Compared with the prior art, the invention has the beneficial effects that:
1. the invention solves the problem of unstable or invalid extraction result in complex road environment by setting point similarity constraint, relative entropy function and structural similarity index;
2. according to the invention, by setting semantic constraint optimization classification, a more accurate road extraction result is obtained.
Drawings
FIG. 1 is a schematic workflow diagram of the present invention;
FIG. 2 is a schematic diagram of a data structure according to the present invention;
FIG. 3 is a schematic diagram of an octree structure according to the present invention;
fig. 4 is a schematic diagram of actual data results of the present invention.
Detailed Description
The present invention will be described in further detail with reference to test examples and specific embodiments. It should not be construed that the scope of the above subject matter of the present invention is limited to the following embodiments, and all techniques realized based on the present invention are within the scope of the present invention.
Example 1
Referring to fig. 1 to 4, the present embodiment includes the following steps:
s1, acquiring road original point cloud data and RGB information thereof, and filtering ground points and non-ground points on the original point cloud data and the RGB information thereof through self-adaptive filtering to obtain a ground point cloud set T and a non-ground point cloud F;
S2, initializing a ground point cloud set T in the S1, wherein the ground point cloud set T comprises discrete points with similar spatial distribution, RGB information and intensity value information, obtaining point clouds with clear adjacency relations through similarity constraint of set points, clustering the point clouds into voxels at the same time, and establishing a voxel adjacency list;
s3, establishing a monomer and a monomer adjacency list, and further expanding and contracting voxels in the S2 by using a relative entropy function, wherein the relative entropy function further optimizes the probability that the ground point cloud set T falls on the corresponding voxels; superposing the structural similarity indexes and the relative entropy probability values to obtain more accurate similarity measurement, distinguishing and dividing voxels into monomers, and updating the monomers and the monomer adjacency list;
s4, carrying out semantic constraint on the monomers in the S3, selecting the monomers to judge whether the class labels are 'roads' or 'non-roads', carrying out flatness judgment, and traversing and judging all voxels and the monomers to obtain a final road point cloud result.
Example 2
Referring to fig. 1 to 4, the present embodiment includes the following steps:
s1, acquiring road original point cloud data and RGB information thereof, and filtering ground points and non-ground points on the original point cloud data and the RGB information thereof through self-adaptive filtering to obtain a ground point cloud set T and a non-ground point cloud F;
S2, initializing a ground point cloud set T in the S1, wherein the ground point cloud set T comprises discrete points with similar spatial distribution, RGB information and intensity value information, obtaining point clouds with clear adjacency relations through similarity constraint of set points, clustering the point clouds into voxels at the same time, and establishing a voxel adjacency list;
s3, establishing a monomer and a monomer adjacency list, and further expanding and contracting voxels in the S2 by using a relative entropy function, wherein the relative entropy function further optimizes the probability that the ground point cloud set T falls on the corresponding voxels; superposing the structural similarity indexes and the relative entropy probability values to obtain more accurate similarity measurement, distinguishing and dividing voxels into monomers, and updating the monomers and the monomer adjacency list;
s4, carrying out semantic constraint on the monomers in the S3, selecting the monomers to judge whether the class labels are 'roads' or 'non-roads', carrying out flatness judgment, and traversing and judging all voxels and the monomers to obtain a final road point cloud result.
The step S1 includes:
s11, acquiring road origin point cloud data with intensity information, and simultaneously acquiring a road panoramic image;
s12, fusing the original point cloud data and the panoramic image to obtain point cloud data with RGB information and intensity information;
S13, dividing the point cloud data in the S12 into ground points and non-ground points through an adaptive filtering algorithm.
The self-adaptive filtering algorithm divides the point cloud data into ground points and non-ground points, and the specific steps are as follows: when the self-adaptive filtering is used, a grid window is constructed, the points to be classified are filtered gradually from large to small, the filtering threshold value is self-adaptively adjusted through an energy formula and the size of the grid window, other points adhered to the ground point cloud are removed, and iteration is carried out until the threshold value or the iteration times are set; repeating the steps until the iteration times are greater than the set maximum iteration times, wherein in the embodiment, the maximum iteration times are set to be 500, and after the self-adaptive filtering process is finished, the ground point cloud set T and the non-ground point cloud F are obtained.
The step S2 includes:
s21, constructing an OCTree (OCTree), wherein the OCTree constructing comprises:
traversing the ground point cloud set T, and finding the minimum size of a cube bounding box (Axis Aligned Bounding Box, AABB), wherein the coordinates are (Xmin, xmax, ymin, ymax, zmin, zmax); iteratively processing all the ground point clouds, for each point, respectively comparing the three-dimensional coordinates (X, Y, Z) of the point with the current AABB range, if the point falls outside the AABB range, updating the AABB range, calculating the maximum value R=MAX (Xmax-Xmin, ymax-Ymin, zmax-Zmin), updating the AABB range, and finally, obtaining the side length R= (Xmax-Xmin) = (Ymax-Ymin) = (Zmax-Zmin);
Consider the three-dimensional space in which the ground point cloud set T is located as a virtual space
A cubic composition with a side length r, wherein m, n, k are positive integers, wherein +.>
Each minimum cube unit containing three-dimensional points is regarded as one leaf node of the octree, then the octree structure of the three-dimensional point cloud can be constructed, and the depth of the octree is set as follows
；
Equal division of octree space is performed, wherein q in formula (1) is octal, from q 0 To q n-1 The path of each leaf node in the octree to the root is shown in its entirety;
decomposing the bounding box into 8 subcubes Q according to equation (1) 8 Numbering the cubes, if the maximum traversing depth is not reached and the node does not have sub-nodes, dividing the cubes into 8 equally divided sub-cubes until the traversing depth is set depth, stopping traversing, and constructing an octree structure of a three-dimensional point cloud with the resolution of r, wherein each cube, namely one voxel, in the octree has a definite 26 neighborhood, each point represents the center of the voxel, and 26 adjacent voxels are arranged around the center voxel;
if the sub-node (cube) has no three-dimensional point, marking the point as an empty node, and stopping segmentation to obtain an octree structure;
S22, initializing a ground point cloud set T by using the octree after obtaining the octree structure of the voxels, regarding the three-dimensional point set contained in each non-empty leaf node in the octree as an initialized point cloud voxel, and recording coordinate information, RGB information and intensity information of each point cloud voxel by using a list; the coordinate information, RGB information and intensity information of the point cloud voxels are respectively statistical average values of the coordinate information, RGB information and intensity information of three-dimensional points in the voxels; as shown in the formula (2),
wherein SV represents an average value, x, y and z represent three-dimensional coordinates of points in the voxel, N represents a natural number and represents the number of all points in the point cloud voxel, I represents an ith point, I i The intensity value of the I-th point, SV (x, y, z), SV (R, G, B), SV (I), respectively, represents coordinate information, RGB information, and intensity information;
s23, further processing voxels, establishing a voxel adjacency list, using a KNN neighborhood search algorithm, finding a nearest 26 neighborhood point cloud voxel according to the initialized voxels, searching points in the 26 neighborhood region to determine the furthest distance of the search range, inquiring the voxels in the furthest distance, searching to judge the furthest distance of the update, using the KNN algorithm to write the octree search result into two storage vectors, respectively storing the search result and the square distance between the corresponding search point and the adjacent point by the two storage vectors, and establishing the voxel adjacency list; judging whether the node area is overlapped with the voxels with the farthest distance as the radius, and searching all the voxels, so that the algorithm time is shortened and the efficiency is improved; setting a K value, wherein the K value is the number of voxels, and in this embodiment, setting k=10, so as to finally obtain a neighbor list of each voxel;
S24, further adjusting points in the voxels, judging the voxels of each three-dimensional point according to the point similarity measure in an iterative mode, and updating the information of the voxels; record the set of all voxels as
Wherein k is the number of all non-empty leaf nodes in the octree; selecting a point a of any single point belonging to the point cloud set T, calculating the similarity between the point a and any voxel, as shown in a formula (3),
in equation 3
Representing the i-th voxel in the set of voxels SV, < >>
Euclidean distance of two-point space coordinates, euclidean distance of RGB component, two-point intensity distance, respectively, for the +.>
Normalization processing is performed, as shown in formula (4),
r in the formula (4) is the maximum size resolution of a single voxel, and m is the maximum gray value of each RGB component;
representing the absolute value of the difference in intensity values of the two points;
the final point similarity function is calculated by the following steps:
wherein w in formula (5) 1 ，w 2 ，w 3 Three coefficients, three coefficients w 1 ，w 2 ，w 3 According to the specific experimental setting size, in this embodiment, w is set 1 ，w 2 ，w 3 0.5, 0.2 and 0.3; wherein the method comprises the steps of
The smaller the value obtained, the stronger the similarity between the two points;
s25, under the point similarity function, carrying out the voxel formation of the road point cloud, and utilizing the point similarity function 
Calculating the similarity of any point a and the three-dimensional center point +.>
And the similarity of any point a and the three-dimensional central point neighborhood voxels +.>
If said->
The value of (2) is greater than said +.>
The point a is given to the voxel sv j In the following, the judgment of sv j If the traversal is finished, if so, updating the voxels, and if not, re-entering the calculation loop step S24;
s26, after the voxel updating is completed, a new voxel set SV= { SV is generated 1 ,sv 2 ....,sv k }。
At the voxel set sv= { SV1, SV2 k Further clustering and label discrimination are carried out on the basis of the }, so that a road monomer is obtained, and the value range of the label of each voxel class is = { road, non-road }, wherein the step S3 comprises:
s31, setting a monomer adjacent list for the monomer, wherein the monomer adjacent list records index numbers of all voxels of corresponding categories adjacent to the monomer in the voxel adjacent list, and the corresponding adjacent list with the same category number as the monomer is empty and is also recorded as the monomer, wherein the category number is a value of a road or a non-road;
s32, further expanding and contracting the voxels by using a relative entropy function, wherein the relative entropy function transforms the target ground point cloud so that the probability of each ground point cloud falling on the corresponding voxel is increased; carrying out mathematical statistics on the intensity values of the point clouds in the voxels, mapping the intensity values to a 0-255 interval, and obtaining the intensity values of the point clouds in the voxels according to the following formula,
Wherein the method comprises the steps of
Representing the intensity value of a certain point, I x The intensity values mapped to the interval from 0 to 255 are represented, N represents the number of points in the voxel, the frequency and probability of each intensity value are defined as p (x), and therefore the intensity entropy of the three-dimensional point cloud is set as follows:
q (x) represents the true distribution of the point cloud voxel intensity values, the relative entropy function simulates the true distribution q (x) by using P (x), and the relative entropy between P (x) and q (x) is as follows:
s33, acquiring gray values from RGB information of the road point cloud by using a structural similarity index, so as to obtain brightness, contrast and structure, wherein the structural similarity index is used for measuring similarity between two images, establishing a brightness, contrast and structural relationship, calculating structural similarity probability, and combining probability values of the relative entropy functions to carry out superposition to obtain similarity measurement, so as to carry out discrimination segmentation from road voxels to monomers; the structural similarity index is commonly used for measuring the similarity between two images, imitates human perception by establishing brightness, contrast and structural relation, focuses on edge and texture similarity, and accords with visual perception of human eyes;
s34, automatically selecting voxels of normal vectors or vectors represented by straight lines forming an included angle of 85-90 degrees with a plane as judging starting points, wherein neighbor voxels of the voxels and neighbor voxels of the neighbor voxels are normal vectors or vectors represented by straight lines forming an included angle of 85-90 degrees with the plane, and the eligible voxels are assigned to road labels, specifically, calculating neighbor points of voxel points of each normal vector or vectors represented by straight lines forming an included angle of 85-90 degrees with the plane, wherein the number of the neighbor points is equal to or less than 26; recording a vector label normal_flag of each voxel point, marking normal_flag=1 if normal is more than or equal to 85 degrees, marking normal_flag=0 if normal is less than 85 degrees, and adding the marking bits of all 26 neighborhood of the voxel point to be used as the final flag value of the voxel point; the marking process is circularly calculated until the voxel with the flag more than or equal to 8 is obtained, the conforming voxel is selected as a starting point voxel y judged in the step S35, and the rest 26 neighborhood voxels are put into a to-be-judged list Q;
S35, overlapping the similarity index obtained by using the structural similarity index and the relative entropy result, and by overlapping two probability values, starting from three aspects of intensity probability distribution, structure and RGB, ensuring the reliability of the voxel classification result, being more suitable for complex environments, as shown in the following formula,
W=α*（1-SSIM(x,y)）+（1-α）I(p||q) （9）
wherein 0 < α < 1, in this example, let α=0.6; when W is less than 0.1, judging that the voxel is a road;
s36, matching the voxels of the judgment starting point y selected in the step S34 with the voxels of the list Q to be judged, judging one by one, and counting the voxels with W less than 0.1, wherein the voxels with the minimum value are regarded as roads and classified as roads;
s37, judging one by one from the adjacent list of the road voxels matched at last in the step S36, and counting voxels with W less than 0.1, wherein the voxels with the minimum value are regarded as roads and classified as the roads;
s38, circularly traversing until the voxels in each voxel adjacent list do not meet the threshold value, returning to the step S36, traversing again, searching the smallest voxel which is removed from the voxels and meets the threshold value, and carrying out matching judgment to obtain the final monomer and the monomer adjacent list thereof.
The step S33 comprises the following steps:
using the structural similarity index, for point cloud voxels, first convert from RGB space to one-dimensional gray scale, the formula is as follows:
Gray = R * 0.299 + G * 0.587 + B * 0.114 （10）
Taking the Gray value Gray as a calculated parameter x, taking the average Gray value Hx of the voxels as an estimation of brightness space measurement, and giving an 'x' of one voxel for comparison and a 'y' of a road voxel;
the luminance estimation function C (x, y) is shown below, C 1 For a constant different from zero, let C in this embodiment 1 =0.1；
The contrast estimation function g (x, y) is shown below as C 2 For a constant different from zero, let C in this embodiment 2 =0.1；
The structure contrast function s (x, y) is shown below,
the structural similarity index formula is as follows, and consists of a brightness estimation function c (x, y), a contrast estimation function g (x, y), and a structural comparison function s (x, y):
selecting simplified command
The method comprises the following steps of:
adding semantic knowledge as constraint, and judging and optimizing the category of the monomer obtained in the step S3 according to category knowledge rules; the step S4 includes:
s41, selecting a single body, judging whether the category label is a road, if so, performing step S42, and if not, performing step S43;
s42, if the number of voxels recorded in the monomer adjacent list is greater than the number of voxels contained in the monomer and the voxel categories recorded in the monomer are all 'roads', reclassifying the monomer as 'roads';
S43, if the number of voxels recorded in the monomer adjacent list is greater than the number of voxels contained in the monomer and the voxel categories recorded in the voxel adjacent list are all non-roads, reclassifying the monomer as non-roads;
the step S4 further includes:
s44, carrying out flatness judgment on the monomer optimization processing obtained in the step S42 and the step S43, and classifying the monomer as a road if the value of a flatness function of the voxel is smaller than a threshold value of 0.5 and more than 50% of the categories of the voxels recorded in the monomer adjacent list are non-roads; the flatness function is used for calculating the average value of the integral elevation standard deviation of all K individual elements
As shown in the following formula,
wherein n represents the number of points in the voxel, z i The elevation value of the i-th point is indicated,
representing an elevation value of the kth voxel;
and (3) automatically traversing and judging all the point cloud voxels and monomers in the ground point cloud data by utilizing the semantic constraint of the steps S42 to S44 to obtain a final road point cloud result.
Example 3
Referring to fig. 1 to 3, an electronic device includes at least one processor and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor to enable the at least one processor to perform a method as described in the previous embodiments. The input/output interface can comprise a display, a keyboard, a mouse and a USB interface, and is used for inputting and outputting data; the power supply is used for providing power for the electronic device.
Those skilled in the art will appreciate that: all or part of the steps for implementing the above method embodiments may be implemented by hardware related to program instructions, and the foregoing program may be stored in a computer readable storage medium, where the program, when executed, performs steps including the above method embodiments; and the aforementioned storage medium includes: a mobile storage device, a Read Only Memory (ROM), a magnetic disk or an optical disk, or the like, which can store program codes.
The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, and alternatives falling within the spirit and principles of the invention.
Claims (9)
1. The expressway point cloud extraction method based on relative entropy and semantic constraint is characterized by comprising the following steps of:
s1, acquiring road original point cloud data and RGB information thereof, and filtering ground points and non-ground points on the original point cloud data and the RGB information thereof through self-adaptive filtering to obtain a ground point cloud set T and a non-ground point cloud F;
s2, initializing a ground point cloud set T in the S1, wherein the ground point cloud set T comprises discrete points with similar spatial distribution, RGB information and intensity value information, obtaining point clouds with clear adjacency relations through similarity constraint of set points, clustering the point clouds into voxels at the same time, and establishing a voxel adjacency list;
S3, establishing a monomer and a monomer adjacency list, and further expanding and contracting voxels in the S2 by using a relative entropy function, wherein the relative entropy function further optimizes the probability that the ground point cloud set T falls on the corresponding voxels; superposing the structural similarity indexes and the relative entropy probability values to obtain more accurate similarity measurement, distinguishing and dividing voxels into monomers, and updating the monomers and the monomer adjacency list;
s4, carrying out semantic constraint on the monomers in the S3, selecting the monomers to judge whether the class labels are 'roads' or 'non-roads', carrying out flatness judgment, traversing and judging all voxels and the monomers to obtain a final road point cloud result,
wherein the step S3 includes:
s31, setting a monomer adjacent list for the monomer, wherein the monomer adjacent list records index numbers of all voxels of corresponding categories adjacent to the monomer in the voxel adjacent list, and the corresponding adjacent list with the same category number as the monomer is empty and is also recorded as the monomer, wherein the category number is a value of a road or a non-road;
s32, performing further expansion and contraction on the voxels by using a relative entropy function; carrying out mathematical statistics on the intensity values of the point clouds in the voxels, mapping the intensity values to a 0-255 interval, and obtaining the intensity values of the point clouds in the voxels according to the following formula,
Wherein the method comprises the steps of
Representing the intensity value of a certain point, I x The intensity values mapped to the interval from 0 to 255 are represented, N represents the number of points in the voxel, the frequency and probability of each intensity value are defined as p (x), and therefore the intensity entropy of the three-dimensional point cloud is set as follows:
q (x) represents the true distribution of the point cloud voxel intensity values, the relative entropy function simulates the true distribution q (x) by using P (x), and the relative entropy between P (x) and q (x) is as follows:
s33, acquiring gray values from RGB information of the road point cloud by using a structural similarity index, so as to obtain brightness, contrast and structure, wherein the structural similarity index is used for measuring similarity between two images, establishing a brightness, contrast and structural relationship, calculating structural similarity probability, and combining probability values of the relative entropy functions to carry out superposition to obtain similarity measurement, so as to carry out discrimination segmentation from road voxels to monomers;
s34, automatically selecting voxels of normal vectors or vectors represented by straight lines forming an included angle of 85-90 degrees with a plane as judging starting points, wherein neighbor voxels of the voxels and neighbor voxels of the neighbor voxels are also normal vectors or vectors represented by straight lines forming an included angle of 85-90 degrees with the plane, and the eligible voxels are assigned to road labels, specifically, calculating neighbor points of voxel points of each normal vector or vectors represented by straight lines forming an included angle of 85-90 degrees with the plane, wherein the number of the neighbor points is equal to or less than 26; recording a vector label normal_flag of each voxel point, marking normal_flag=1 if normal is more than or equal to 85 degrees, marking normal_flag=0 if normal is less than 85 degrees, and adding the marking bits of all 26 neighborhood of the voxel point to be used as the final flag value of the voxel point; the marking process is circularly calculated until the voxel with the flag more than or equal to 8 is obtained, the conforming voxel is selected as a starting point voxel y judged in the step S35, and the rest 26 neighborhood voxels are put into a to-be-judged list Q;
S35, superposing a similarity index obtained by using the structural similarity index and a relative entropy result, and superposing two probability values, wherein the three aspects of intensity probability distribution, structure and RGB are considered as shown in a formula (9),
W=α*（1-SSIM(x,y)）+（1-α）I(p||q) （9）
wherein, alpha is more than 0 and less than 1, and when W is more than 0.1, the voxel is judged to be a road;
s36, matching the voxels of the judgment starting point y selected in the step S34 with the voxels of the list Q to be judged, judging one by one, and counting the voxels with W less than 0.1, wherein the voxels with the minimum value are regarded as roads and classified as roads;
s37, judging one by one from the adjacent list of the road voxels matched at last in the step S36, and counting voxels with W less than 0.1, wherein the voxels with the minimum value are regarded as roads and classified as the roads;
s38, circularly traversing until the voxels in each voxel adjacent list do not meet the threshold value, returning to the step S36, traversing again, searching the smallest voxel which is removed from the voxels and meets the threshold value, and carrying out matching judgment to obtain the final monomer and the monomer adjacent list thereof.
2. The method for extracting the point cloud of the highway based on the relative entropy and the semantic constraint according to claim 1, wherein the step S1 comprises:
s11, acquiring road origin point cloud data with intensity information, and simultaneously acquiring a road panoramic image;
S12, fusing the original point cloud data and the panoramic image to obtain point cloud data with RGB information and intensity information;
s13, dividing the point cloud data in the S12 into ground points and non-ground points through an adaptive filtering algorithm.
3. The method for extracting the point cloud of the expressway based on the relative entropy and the semantic constraint according to claim 2, wherein the specific steps of dividing the point cloud data into the ground points and the non-ground points by the adaptive filtering algorithm are as follows:
when the self-adaptive filtering is used, a grid window is constructed, the points to be classified are filtered gradually from large to small, the filtering threshold value is self-adaptively adjusted through an energy formula and the size of the grid window, other points adhered to the ground point cloud are removed, and iteration is carried out until the threshold value or the iteration times are set; repeating the steps until the iteration times are larger than the set maximum iteration times, and obtaining a ground point cloud set T and a non-ground point cloud F after the self-adaptive filtering process is finished.
4. The method for extracting the point cloud of the highway based on the relative entropy and the semantic constraint according to claim 1, wherein the step S2 comprises:
S21, constructing an octree, wherein constructing the octree comprises:
traversing the ground point cloud set T, and finding the minimum cube bounding box size, wherein the coordinates are (Xmin, xmax, ymin, ymax, zmin, zmax); iteratively processing all the ground point clouds, for each point, respectively comparing the three-dimensional coordinates (X, Y, Z) of the point with the current AABB range, if the point falls outside the AABB range, updating the AABB range, calculating the maximum value R=MAX (Xmax-Xmin, ymax-Ymin, zmax-Zmin), updating the AABB range, and finally, obtaining the side length R= (Xmax-Xmin) = (Ymax-Ymin) = (Zmax-Zmin); wherein, AABB is a cubic bounding box;
consider the three-dimensional space in which the ground point cloud set T is located as a virtual space
Each side is of lengthr, wherein m, n, k are positive integers, wherein +.>
Each minimum cube unit containing three-dimensional points is regarded as one leaf node of the octree, then the octree structure of the three-dimensional point cloud can be constructed, and the depth of the octree is set as follows
；
Equal division of octree space is performed, wherein q in formula (1) is octal, from q 0 To q nt-1 The path of each leaf node in the octree to the root is shown in its entirety;
decomposing the bounding box into 8 subcubes Q according to equation (1) 8 Numbering the cubes, if the maximum traversing depth is not reached and the node does not have sub-nodes, dividing the cubes into 8 equally divided sub-cubes until the traversing depth is set depth, stopping traversing, and constructing an octree structure of a three-dimensional point cloud with the resolution of r, wherein each cube, namely one voxel, in the octree has a definite 26 neighborhood, each point represents the center of the voxel, and 26 adjacent voxels are arranged around the center voxel;
if the sub-node has no three-dimensional point, marking the point as an empty node, and stopping segmentation to obtain an octree structure;
s22, initializing a ground point cloud set T by using the octree, and taking a three-dimensional point set contained in each non-empty leaf node in the octree as an initialized point cloud voxel, wherein the coordinate information, RGB information and intensity information of each point cloud voxel are recorded by a list; the coordinate information, RGB information and intensity information of the point cloud voxels are respectively statistical average values of the coordinate information, RGB information and intensity information of three-dimensional points in the voxels; as shown in the formula (2),
wherein SV represents an average value, x, y and z represent three-dimensional coordinates of points in the voxel, N represents a natural number and represents the number of all points in the point cloud voxel, I represents an ith point, I i The intensity value of the I-th point, SV (x, y, z), SV (R, G, B), SV (I), respectively, represents coordinate information, RGB information, and intensity information;
s23, further processing voxels, establishing a voxel adjacency list, using a KNN neighborhood search algorithm, finding a nearest 26 neighborhood point cloud voxel according to the initialized voxels, searching points in the 26 neighborhood region to determine the furthest distance of the search range, inquiring the voxels in the furthest distance, searching to judge the furthest distance of the update, using the KNN algorithm to write the octree search result into two storage vectors, respectively storing the search result and the square distance between the corresponding search point and the adjacent point by the two storage vectors, and establishing the voxel adjacency list; setting a K value, wherein the K value is the number of voxels, and obtaining an adjacent list of each voxel;
s24, judging voxels to which each three-dimensional point belongs according to the point similarity measure in an iterative mode, and updating the voxel information; record the set of all voxels as
Wherein k is the number of all non-empty leaf nodes in the octree; selecting a point a of any single point belonging to the point cloud set T, calculating the similarity between the point a and any voxel, as shown in a formula (3),
In equation 3
Representing the i-th voxel in the set of voxels SV, < >>
Euclidean distance of two-point space coordinates, euclidean distance of RGB component, two-point intensity distance, respectively, for the +.>
Normalization is performed, as shown in formula (4),/->
R in the formula (4) is the maximum size resolution of a single voxel, and m is the maximum gray value of each RGB component;
representing the absolute value of the difference in intensity values of the two points;
the final point similarity function is calculated by the following steps:
wherein w in formula (5) 1， w 2 ，w 3 Three coefficients;
s25, under the point similarity function, carrying out the voxel formation of the road point cloud, and utilizing the point similarity function
Calculating the similarity of any point a and the three-dimensional center point +.>
And the similarity of any point a and the three-dimensional central point neighborhood voxels +.>
If said->
The value of (2) is greater than said +.>
The point a is given to the voxel sv j In the following, the judgment of sv j If the traversal is finished, if so, updating the voxels, and if not, re-entering the calculation loop step S24;
s26, after the voxel updating is completed, a new voxel set SV= { SV is generated 1 ,sv 2 ....,sv k }。
5. The method for extracting point cloud from highway according to claim 4, wherein said step S3 is performed on said voxel set sv= { SV1, SV2 k And carrying out further clustering and label discrimination on the basis of the }, thereby obtaining a road monomer, wherein the value range of each voxel class label is = { road, non-road }.
6. The method for extracting the point cloud of the highway based on the relative entropy and the semantic constraint according to claim 5, wherein the step S33 comprises the following steps:
using the structural similarity index, for point cloud voxels, first convert from RGB space to one-dimensional gray scale, the formula is as follows:
Gray = R * 0.299 + G * 0.587 + B * 0.114 （10）
taking the Gray value Gray as a calculated parameter x, taking the average Gray value Hx of the voxels as an estimation of brightness space measurement, and giving an 'x' of one voxel for comparison and a 'y' of a road voxel;
the luminance estimation function C (x, y) is shown below, C 1 Is a constant that is not equal to zero,
the contrast estimation function g (x, y) is shown below as C 2 Is a constant that is not equal to zero,
the structure contrast function s (x, y) is shown below,
the structural similarity index formula is as follows, and consists of a brightness estimation function c (x, y), a contrast estimation function g (x, y), and a structural comparison function s (x, y):
selecting simplified command
The method comprises the following steps of:
7. the expressway point cloud extraction method based on relative entropy and semantic constraint of claim 6, wherein semantic knowledge is added as constraint, and the category of the monomer obtained in the step S3 is judged and optimized according to category knowledge rules; the step S4 includes:
S41, selecting a single body, judging whether the category label is a road, if so, performing step S42, and if not, performing step S43;
s42, if the number of voxels recorded in the monomer adjacent list is greater than the number of voxels contained in the monomer and the voxel categories recorded in the monomer are all 'roads', reclassifying the monomer as 'roads';
s43, if the number of voxels recorded in the monomer adjacent list is greater than the number of voxels contained in the monomer and the voxel categories recorded in the voxel adjacent list are all non-roads, the monomer is reclassifying as non-roads.
8. The method for extracting the point cloud of the highway based on the relative entropy and the semantic constraint of claim 7, wherein the step S4 further comprises:
s44, carrying out flatness judgment on the monomer optimization processing obtained in the step S42 and the step S43, and classifying the monomer as a road if the value of a flatness function of the voxel is smaller than a threshold value of 0.5 and more than 50% of the categories of the voxels recorded in the monomer adjacent list are non-roads; the flatness function is used for calculating the average value of the integral elevation standard deviation of all K individual elements 
As shown in the following formula,
wherein n represents the number of points in the voxel, z i Indicating the ith pointIs used for the height value of the (c),
representing an elevation value of the kth voxel;
and (3) automatically traversing and judging all the point cloud voxels and monomers in the ground point cloud data by utilizing the semantic constraint of the steps S42 to S44 to obtain a final road point cloud result.
9. An electronic device comprising at least one processor, and a memory communicatively coupled to the at least one processor; the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310077847.9A CN115810153B (en) | 2023-02-08 | 2023-02-08 | Highway point cloud extraction method based on relative entropy and semantic constraint |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310077847.9A CN115810153B (en) | 2023-02-08 | 2023-02-08 | Highway point cloud extraction method based on relative entropy and semantic constraint |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115810153A CN115810153A (en) | 2023-03-17 |
| CN115810153B true CN115810153B (en) | 2023-05-02 |
Family
ID=85487594
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310077847.9A Active CN115810153B (en) | 2023-02-08 | 2023-02-08 | Highway point cloud extraction method based on relative entropy and semantic constraint |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115810153B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116758238B (en) * | 2023-08-17 | 2024-01-23 | 山东高速工程检测有限公司 | Road guardrail automatic modeling method based on vehicle-mounted laser point cloud |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111382496A (en) * | 2018-12-29 | 2020-07-07 | 达索系统公司 | Learning neural networks for inferring editable feature trees |
| CN112132840A (en) * | 2020-09-01 | 2020-12-25 | 济南市房产测绘研究院(济南市房屋安全检测鉴定中心) | Vehicle-mounted street tree point cloud classification and characteristic information extraction method |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110390706B (en) * | 2018-04-13 | 2023-08-08 | 北京京东尚科信息技术有限公司 | Object detection method and device |
| CN110570428B (en) * | 2019-08-09 | 2023-07-07 | 浙江合信地理信息技术有限公司 | Method and system for dividing building roof sheet from large-scale image dense matching point cloud |
| CN111274976B (en) * | 2020-01-22 | 2020-09-18 | 清华大学 | Lane detection method and system based on multi-level fusion of vision and laser radar |
| US11393158B2 (en) * | 2020-04-02 | 2022-07-19 | Adobe Inc. | Utilizing voxel feature transformations for deep novel view synthesis |
| CN111861946B (en) * | 2020-08-03 | 2023-03-28 | 中国科学院空天信息创新研究院 | Adaptive multi-scale vehicle-mounted laser radar dense point cloud data filtering method |
-
2023
- 2023-02-08 CN CN202310077847.9A patent/CN115810153B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111382496A (en) * | 2018-12-29 | 2020-07-07 | 达索系统公司 | Learning neural networks for inferring editable feature trees |
| CN112132840A (en) * | 2020-09-01 | 2020-12-25 | 济南市房产测绘研究院(济南市房屋安全检测鉴定中心) | Vehicle-mounted street tree point cloud classification and characteristic information extraction method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115810153A (en) | 2023-03-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110120097B (en) | Semantic modeling method for airborne point cloud of large scene | |
| EP0526881B1 (en) | Three-dimensional model processing method, and apparatus therefor | |
| Bernardini et al. | Automatic reconstruction of 3D CAD models from digital scans | |
| US20070052706A1 (en) | System and Method for Performing Domain Decomposition for Multiresolution Surface Analysis | |
| CN110502973B (en) | Automatic extraction and identification method for road marking based on vehicle-mounted laser point cloud | |
| CN111899328B (en) | Point cloud three-dimensional reconstruction method based on RGB data and generation countermeasure network | |
| CN106548520A (en) | A kind of method and system of cloud data denoising | |
| CN116310192A (en) | Urban building three-dimensional model monomer reconstruction method based on point cloud | |
| CN115810153B (en) | Highway point cloud extraction method based on relative entropy and semantic constraint | |
| CN115372989A (en) | Laser radar-based long-distance real-time positioning system and method for cross-country automatic trolley | |
| CN109118588B (en) | Automatic color LOD model generation method based on block decomposition | |
| CN112215958B (en) | Laser radar point cloud data projection method based on distributed computation | |
| Yogeswaran et al. | 3d surface analysis for automated detection of deformations on automotive body panels | |
| CN111784840A (en) | LOD level three-dimensional data unitization method and system based on vector data automatic segmentation | |
| CN116452852A (en) | Automatic generation method of high-precision vector map | |
| Zhu et al. | 3D reconstruction of plant leaves for high-throughput phenotyping | |
| Thiemann et al. | 3D-symbolization using adaptive templates | |
| Omidalizarandi et al. | Segmentation and classification of point clouds from dense aerial image matching | |
| CN111861946A (en) | Adaptive multi-scale vehicle-mounted laser radar dense point cloud data filtering method | |
| CN117115012A (en) | Road surface point cloud marking segmentation denoising method | |
| CN115186347B (en) | Building CityGML modeling method combining house type plan view and inclination model | |
| Becker et al. | Lidar inpainting from a single image | |
| CN111612869B (en) | Analysis method for geological mapping based on raster data | |
| Assfalg et al. | Spin images for retrieval of 3D objects by local and global similarity | |
| Smith | 3D data representation, storage and processing |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |