CN110969624B - Laser radar three-dimensional point cloud segmentation method - Google Patents
Laser radar three-dimensional point cloud segmentation method Download PDFInfo
- Publication number
- CN110969624B CN110969624B CN201911082345.5A CN201911082345A CN110969624B CN 110969624 B CN110969624 B CN 110969624B CN 201911082345 A CN201911082345 A CN 201911082345A CN 110969624 B CN110969624 B CN 110969624B
- Authority
- CN
- China
- Prior art keywords
- point cloud
- dimensional
- segmentation
- point
- neighborhood
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
- G06T17/005—Tree description, e.g. octree, quadtree
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/66—Analysis of geometric attributes of image moments or centre of gravity
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10028—Range image; Depth image; 3D point clouds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
- G06T2207/10044—Radar image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20092—Interactive image processing based on input by user
- G06T2207/20101—Interactive definition of point of interest, landmark or seed
-
- 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
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Geometry (AREA)
- Computer Graphics (AREA)
- Software Systems (AREA)
- Optical Radar Systems And Details Thereof (AREA)
Abstract
The invention discloses a laser radar three-dimensional point cloud segmentation method, which comprises the steps of firstly extracting original three-dimensional point cloud data acquired by a laser radar, preprocessing the original point cloud, including denoising, simplifying and coordinate transformation of the original point cloud data, constructing basic point cloud data under a three-dimensional Cartesian coordinate system, storing the three-dimensional data in a two-dimensional array mode, adopting a variable neighborhood dispersion search strategy, dynamically adjusting the resolution of a region growing method around the neighborhood range of seeds and a search matching range, developing the point cloud preliminary segmentation work, designing a point cloud segmentation envelope diffusion strategy on the basis, further searching around a point cloud segmentation set to realize the fusion of multiple sets, further obtaining a point cloud segmentation set, and finally designing a visual function of a point cloud segmentation result for checking the point cloud segmentation effect. The invention effectively improves the segmentation rate, effectively inhibits the over-segmentation condition, maintains the integrity of each target, and is convenient for observing the scanning result of the segmented targets.
Description
Technical Field
The invention relates to the technical field of computer vision, in particular to a method for dividing laser radar 3D point cloud, which can be used for rapidly dividing three-dimensional point cloud data into independent point cloud sets.
Background
With the development of scientific technology, the acquisition of surrounding information of a ship in the running process gradually expands from a two-dimensional image mainly based on optical imaging to a three-dimensional image mainly based on laser scanning, the ship laser radar is widely used as a sensor for acquiring three-dimensional point cloud data by virtue of the characteristics of high precision, interference resistance, small volume, light weight and the like, is gradually popularized and installed at the high position of a ship mast, but the original three-dimensional point cloud data acquired by the laser radar cannot be directly reflected to surrounding ships, the point cloud segmentation work needs to be carried out, the subsequent further use of point cloud is directly influenced by the point cloud segmentation result, and therefore, the three-dimensional point cloud segmentation of the ship laser radar is carried out, and the method has important theoretical research and practical application significance.
The conventional three-dimensional point cloud segmentation method is mostly based on a traditional region growing method, on the basis of keeping good boundary, the method can segment the connected region with the same characteristics, for example, huang Zhen provides a 3D point cloud segmentation method in document 'fast 3D point cloud segmentation combining super voxel geometry and color information', and a new measurement variable for capturing scanned target geometry and color information is utilized to obtain a hierarchical structure of the segmented region, and the method has good segmentation efficiency; for example Wang, in the literature 'an improved region growing color 3D point cloud segmentation algorithm', a clustering idea is fused in a region growing method, and a clustering threshold is set by referring to the normal angle and the color difference in a local neighborhood, so that a three-dimensional point cloud segmentation task is realized, and the method can improve the point cloud segmentation speed and stability. In the three-dimensional point cloud segmentation process, most of the documents process the point cloud segmentation speed or stability in a targeted manner, but the segmentation accuracy of the point cloud is not ideal, especially in the point cloud segmentation of the ship laser radar, the problems of under-segmentation of the point cloud caused by mutual shielding of multiple ships and over-segmentation of the point cloud caused by the fact that a large ship is close to the laser radar are not effectively processed, and therefore a method for rapidly and accurately segmenting the three-dimensional point cloud of the ship laser radar is needed.
Disclosure of Invention
Aiming at the prior art, the technical problem to be solved by the invention is to provide the laser radar three-dimensional point cloud segmentation method capable of effectively improving segmentation accuracy and rapidity.
In order to solve the technical problems, the method for dividing the laser radar three-dimensional point cloud comprises the following steps:
step 1: acquiring laser radar three-dimensional point cloud data:
acquiring original three-dimensional point cloud data by a laser radar, wherein the point cloud data is represented by three elements (r, theta, phi), r represents the distance of a scanning point, theta represents a horizontal direction angle, and phi represents a vertical direction angle;
step 2: preprocessing an original three-dimensional point cloud, which specifically comprises the following steps:
step 2.1: denoising the original three-dimensional point cloud: calculating average distance R from each point to all points in the neighborhood ave Assuming that the resulting distribution is a gaussian distribution having a mean and standard deviation, an outlier decision boundary radius R is set set R is taken as ave And R is set Comparing when R ave ≤R set Reserving the judged point; when R is ave >R set The determined points are outliers, and the outliers are deleted from the original point cloud data set;
step 2.2: original three-dimensional point cloud downsampling is simplified:
downsampling the point cloud by voxel grid filtering, specifically: dividing the point cloud into cubic voxels by using octree, calculating the gravity centers of all points in the cubic voxels, and replacing all point clouds in the voxels by using the gravity center points;
step 2.3: two-dimensional representation of a three-dimensional point cloud:
converting the three-dimensional vector coordinates into three-dimensional rectangular coordinates, wherein the conversion formula is as follows:
wherein (x, y, z) in the above formula is the coordinates of each point in the point cloud data under a three-dimensional rectangular coordinate system;
two-dimensional array coordinate (x, y) location and storage distance data r for that location xy The relationship with the corresponding original three-dimensional point cloud vector coordinates (r, θ, ψ) can be expressed by the following formula:
(x,y,r xy ) For the preprocessed point cloud data, the horizontal direction angle resolution of the laser radar is as followsThe angular resolution in the vertical direction is +.>
Step 3: point cloud preliminary segmentation based on variable neighborhood dispersion search strategy:
the variable neighborhood distributed search strategy specifically comprises the following steps:
the eight neighborhood is further expanded outwards to form a larger neighborhood, the number of turns of the neighborhood expanded outwards is set to be dynamic, the outward detection distance of the laser radar in the horizontal direction is omega, the outward detection distance of the laser radar in the vertical direction is h, and the number of expansion cells in the horizontal and vertical directions is N x_exd And N y_exd The method meets the following conditions:
simplifying the traversing range, namely properly selecting the non-searching range in the variable neighborhood range
When expanding from eight adjacent areas of the selected seed, the width W of the horizontal and vertical non-search range after changing the adjacent areas Unser And height H Unser The method meets the following conditions:
through the above, the point cloud searching work is not performed in the area between the eight neighborhood areas of the selected seeds and the outer layer;
performing preliminary point cloud segmentation according to the variable neighborhood dispersion searching strategy;
step 4: and (3) performing point cloud re-segmentation based on the preliminary segmentation envelope diffusion strategy:
obtaining a point cloud set according to the region growing method in the step 3, extracting a segmentation envelope boundary of the set, expanding sigma distance outwards along the normal direction, calculating the gravity center O of the set, optionally selecting one boundary point N of the set, and calculating the distance d between two points ON Then in the horizontal and vertical directionsExtended distance sigma xec Sum sigma yec The method meets the following conditions:
comparing the extended range with other sets if there is an intersection, the two sets can be fused into one set, if there is no intersection, they are still represented as two different sets.
The invention also includes:
1. the point cloud segmentation result is visualized, specifically: and displaying the point cloud segmentation result in the three-dimensional scene by adopting an OpenGL three-dimensional scene display technology.
2. The preliminary point cloud segmentation in the step 3 specifically comprises the following steps:
step 3.1: sequentially scanning the point cloud data, finding out the 1 st data point which is not attributed yet as a seed, and recording the point as S (x 0 ,y 0 );
Step 3.2: by S (x) 0 ,y 0 ) The method comprises the steps of centering, determining a neighborhood and a non-searching range of a seed point according to the coordinate position of the seed point, and determining a point S in an effective searching range new (x * ,y * ) S is calculated according to the following new (x * ,y * ) And S (x) 0 ,y 0 ) Is the spatial distance d of (2) S :
In the above formula (x) * ,y * ,z * ) Setting the judgment distance as d meters for the space coordinates of the points, when d S When d is smaller, the region growing criterion is satisfied, representing S new (x * ,y * ) And S (x) 0 ,y 0 ) Belonging to the same set, simultaneously S new (x * ,y * ) Pushing the stack;
step 3.3: taking a pixel out of the stack, taking the pixel as a new seed point, repeating the step 3.2, and continuously searching points belonging to the same set;
step 3.4: returning to the step 3.1 when the stack is empty;
step 3.5: and (3) repeating the steps 3.1 to 3.4 until each point in the three-dimensional point cloud has a belonging set, and ending the region growth.
The invention has the beneficial effects that: aiming at the problems of segmentation accuracy and rapidity in the laser radar three-dimensional point cloud segmentation process, the invention dynamically changes the neighborhood scanning range according to the difference of the distance between the laser radar and the neighborhood scanning range by improving the traditional region growing method, and simultaneously provides a scattered search strategy for avoiding repeated judgment of part of point clouds and solving the over-segmentation problem caused by different arc lengths between adjacent points of the point clouds with different distances. And on the basis of preliminary segmentation, an envelope diffusion strategy is provided, further matching association work of the point cloud set is carried out, the problem of under-segmentation of the three-dimensional point cloud is solved, and finally the task of improving the accuracy and the speed of point cloud segmentation and guaranteeing the integrity of a target is realized.
According to the invention, on the basis of the original point cloud, three-dimensional point cloud data denoising and simplification can be realized by adopting a filtering and downsampling mode, and the three-dimensional point cloud is stored as a two-dimensional plane mode, so that the subsequent point cloud processing is facilitated; the variable neighborhood dispersion searching strategy is adopted, so that the invariance of the geometric characteristics of the point cloud in the three-dimensional space is kept, meanwhile, the segmentation rate is improved, and the over-segmentation condition is effectively restrained; adopting a preliminary segmentation envelope diffusion strategy to perform re-diffusion searching operation on the set after preliminary segmentation, so that the under-segmentation condition is effectively restrained, and the integrity of each target is maintained; and the visual mode is adopted to display the algorithm segmentation effect, so that the scanning result of the segmentation target ship can be conveniently observed. In conclusion, the method and the device realize the task of quickly and accurately dividing the laser radar three-dimensional point cloud.
Drawings
FIG. 1 is a flow chart of a three-dimensional point cloud segmentation process of the present invention;
FIG. 2 is a schematic view of a lidar scan;
FIG. 3a is a schematic diagram of the denoising case of the point A cloud according to the present invention;
FIG. 3B is a schematic diagram of the B-point cloud denoising process according to the present invention;
FIG. 4 is a schematic diagram of a three-dimensional point cloud downsampling principle;
FIG. 5 is a schematic view of laser radar point cloud data expansion;
FIG. 6 is a schematic diagram of a variant neighborhood range;
FIG. 7 is a simplified traversal scope diagram;
FIG. 8 is a schematic diagram of a preliminary segmentation envelope expansion strategy;
FIG. 9 shows the effect of the point cloud segmentation results;
FIG. 10 is a graph of the effect of conventional region growing over-segmentation;
FIG. 11 is a graph of the present invention suppressing over-segmentation effects;
FIG. 12 is a diagram of the undersegmentation effect of the conventional region growing method;
fig. 13 is a view showing the effect of suppressing the undersegmentation in the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings.
Referring to fig. 1, a three-dimensional point cloud segmentation flow of the laser radar of the present invention is shown in the following chart. Firstly extracting original three-dimensional point cloud data acquired by a 360-degree laser radar, then carrying out preprocessing work on the original point cloud, wherein the preprocessing process comprises denoising, simplifying and coordinate transformation on the original point cloud data, constructing basic point cloud data under a three-dimensional Cartesian coordinate system, storing the three-dimensional data in a two-dimensional array form, adopting a variable neighborhood dispersion search strategy, dynamically adjusting the resolution of a region growing method around the neighborhood range of seeds and a search matching range, carrying out point cloud preliminary segmentation work, designing a point cloud segmentation envelope diffusion strategy on the basis, further searching around a point cloud segmentation set to realize fusion of multiple sets, further obtaining a point cloud segmentation set, and finally designing a visual function of a point cloud segmentation result, wherein the visual function is specially used for checking the point cloud segmentation effect.
1. Acquisition of laser radar three-dimensional point cloud data
The three-dimensional point cloud data come from the ship laser radar, the laser radar is installed at the high position of the ship deck mast, other ships around the ship can be scanned by rotating 360 degrees in the horizontal direction, and the laser radar scanning schematic diagram is shown in fig. 2.
The original point cloud data acquired by the laser radar is represented by three elements (r, θ, ψ), where r represents the distance of the scan points, θ represents the horizontal direction angle, and ψ represents the vertical direction angle.
2. Preprocessing of original three-dimensional point cloud
(1) Original three-dimensional point cloud denoising
According to the invention, the measurement error of the ship laser radar hardware is considered, so that noise interference often exists in the original point cloud to generate a small amount of outliers, and denoising work is carried out on the original point cloud data before the point cloud is segmented: according to the invention, based on the distributed calculation of the distance from the point to the neighborhood, the average distance from each point to all points in the neighborhood is calculated by carrying out statistical analysis on the neighborhood of each point, and if the obtained distribution is Gaussian distribution with mean value and standard deviation, whether the laser radar scanning point is reserved or not is judged by comparing the average distance with the set spherical distance mean value and standard deviation;
in connection with fig. 3a and 3B, if decision points a and B, i.e. the center point in the figure, are reserved, an outlier decision boundary radius R is set set For 0.4 m, respectively judging the average distance R between the point A and the point B and all points in the respective neighborhood (namely pointed by arrows) ave Namely the radius of the average distance indicating circle between each point in the neighborhood and R in the upper graph set Comparing, determining whether to retain the determined point is performed by:
according to the above formula, the point A can be obtained to be reserved, the distance between the point B and other points is too large, and the point B is regarded as an outlier, and the point A is deleted from the original point cloud data set.
(2) Original three-dimensional point cloud downsampling reduction
Considering that the original point cloud is large in quantity and high in density, the point cloud segmentation rate and the practicability of a segmentation method are seriously affected, the 3D point cloud is downsampled by utilizing voxel grid filtering, and a schematic diagram is shown in fig. 4;
referring to fig. 4, the point cloud is divided into cubic voxels by using octree, wherein the downsampled point is a solid point, the gravity center of the voxel is calculated to obtain a hollow point, and all the point clouds in the voxel are respectively replaced by the gravity centers by using the method.
In addition, by considering the characteristic that the ship which can be scanned by the ship laser radar is high, the point cloud data which is within 0.5 meter higher than the horizontal plane are filtered, the number of the point clouds can be reduced on the premise of keeping the geometric characteristics unchanged by the mode, and a data foundation is laid for improving the point cloud segmentation speed.
(3) Two-dimensional representation of a three-dimensional point cloud
In order to facilitate subsequent calculation, the three-dimensional vector coordinates are converted into three-dimensional rectangular coordinates, and the conversion formula is as follows:
in the above formula, (x, y, z) is the coordinates of each point in the point cloud data in a three-dimensional rectangular coordinate system.
In order to improve the traditional region growing method, the three-dimensional point cloud data is stored in a two-dimensional array form, so that the subsequent data access and extraction are convenient, and the stored two-dimensional array form is as follows: the x-axis coordinate is the number of the laser beam in the horizontal direction, the y-axis coordinate is the number of the laser beam in the vertical direction, and r xy The distance to the target is scanned for the laser beam.
Because the data volume of the point cloud is large, the original data acquired by the laser radar is stored by the two-dimensional array, the subsequent data access and extraction are convenient, and the point cloud is unfolded into a two-dimensional array form as shown in fig. 5.
In FIG. 5, the x-axis coordinates are the number of the laser beam in the horizontal direction, the y-axis coordinates are the number of the laser beam in the vertical direction, and r xy The laser beam is scanned to the distance of the object encountered. Let the horizontal angular resolution of the laser radar beThe angular resolution in the vertical direction is +.>Two-dimensional array coordinate (x, y) location and storage distance data r for that location xy The relationship with the corresponding original three-dimensional point cloud vector coordinates (r, θ, ψ) can be expressed by the following formula:
the invention will (x, y, r) xy ) As the preprocessed point cloud data, and develop the introduction of the subsequent processing method on the basis.
3. Point cloud preliminary segmentation based on variable neighborhood distributed search strategy
The invention provides a variable neighborhood dispersion search strategy based on the eight neighborhood of the traditional region growing method, and the specific implementation process is shown in figure 6: in the traditional region growing method, the searching neighborhood is eight neighborhood, such as grid region 1, and the invention provides a variable neighborhood, such as grid region 2, namely the traditional eight neighborhood is further expanded outwards for one circle to form a larger neighborhood, so that the A part and the C part are connected and divided into one class, and the over-division condition is restrained to a certain extent.
In the above manner, for a target ship point cloud with a shorter distance, because the point cloud density is larger in this case, if the number of expansion turns is smaller, the over-segmentation situation may still not be avoided; for a target ship point cloud at a longer distance, if the number of expansion turns is large, two adjacent ships may be split into one ship, resulting in a point cloud under-split condition. The invention provides a three-dimensional point cloud with both short-distance and long-distance, which sets the number of outward expansion neighborhood turns as dynamic, sets the outward detection distance of a laser radar in the horizontal direction as omega, the outward detection distance of the laser radar in the vertical direction as h, and the number of expansion grids in the horizontal and vertical directions as N x_exd And N y_exd Calculated using the following formula:
according to the method, the neighborhood expansion can increase the point cloud segmentation time, particularly for ships closer to the laser radar, the corresponding three-dimensional point cloud is large in number and high in density, the neighborhood range to be traversed is larger, and in order to determine whether the point cloud array slightly far from the selected seed belongs to the same set with the seed, the traversing range is simplified, namely in the variable neighborhood range, the non-searching range is properly selected, and the principle diagram is shown in fig. 7.
Referring to FIG. 7, after expansion from eight neighborhoods of a selected seed, the width W of the horizontal and vertical convenient non-search range after the neighborhood is changed according to the present invention Unser And height H Unser Calculated from the following formula:
through the method, the point cloud searching work is not performed in the area between the eight neighborhood areas of the selected seeds and the outer layer, compared with the traditional area growing method, the method can accelerate the point cloud segmentation speed, the eight neighborhood areas ensure the growing continuity, each circle of the outer layer can check whether the point cloud of the array with a longer distance belongs to the same class as the seeds, and the problem of over-dividing caused by over-density of the point cloud is solved.
Based on the above variable neighborhood dispersion searching strategy, the preliminary point cloud segmentation steps are as follows:
step 1: sequentially scanning the point cloud data, finding out the 1 st data point which is not attributed yet as a seed, and recording the point as S (x 0 ,y 0 );
Step 2: by S (x) 0 ,y 0 ) The method comprises the steps of centering, determining a neighborhood and a non-searching range of a seed point according to the coordinate position of the seed point, and determining a point S in an effective searching range new (x * ,y * ) S is calculated according to the following new (x * ,y * ) And S (x) 0 ,y 0 ) Is the spatial distance d of (2) S :
In the above formula (x) * ,y * ,z * ) Is the spatial coordinates of the point. The invention sets the judging distance to be 5 meters, when d S When the number of the regions is less than 5, the region growing criterion is satisfied, representing S new (x * ,y * ) And S (x) 0 ,y 0 ) Belonging to the same set, simultaneously S new (x * ,y * ) Pushing the stack;
step 3: taking a pixel out of the stack, taking the pixel as a new seed point, repeating Step 2, and continuously searching points belonging to the same set;
step 4: returning to Step 1 when the stack is empty;
step 5: repeating steps Step 1 to Step 4 until each point in the three-dimensional point cloud has a belonging set, and ending the region growth.
4. Point cloud re-segmentation based on preliminary segmentation envelope diffusion strategy
After the three-dimensional point cloud primary segmentation result is obtained, the invention further considers the under segmentation situation which still possibly occurs in the primary segmentation result, namely when two vessels are shielded, one vessel is easily segmented into two parts by the other vessel. The invention provides a preliminary segmentation envelope expansion strategy, and the principle is shown in figure 8.
Referring to fig. 8, if A, B, C three point cloud sets are obtained, taking the set a as an example, extracting the dividing envelope boundary of the set a, expanding the sigma distance outwards along the normal direction, calculating the center of gravity O of the set a, optionally one boundary point N of the set a, and calculating the distance d between two points ON Then the distance sigma is extended in the horizontal and vertical directions xec Sum sigma yec Calculated from the following formula:
comparing the extended range with other sets if there is an intersection, the two sets can be fused into one set, if there is no intersection, they are still represented as two different sets. Through the step, the under-segmentation condition of the point cloud can be further effectively restrained. In fig. 8, the relationship between the envelope extension boundary of the set a and the set C exists in the graph, and the envelope extension boundary of the set a and the set B do not exist in the graph, so that the set a and the set C are represented as the same ship, and the set B is separately represented as one ship.
5. Point cloud segmentation result visualization
According to the invention, an OpenGL three-dimensional scene display technology is adopted, a point cloud segmentation result is displayed in a three-dimensional scene, a display result of the three-dimensional scene is in a white or black background, and each part set after the point cloud segmentation is represented by different colors. The visualization effect of the point cloud segmentation result is as shown in fig. 9, and different part sets are represented as different gray scales and represent different colors.
6. The method of the invention is compared with the test results of the traditional method
In order to verify the effectiveness of the point cloud segmentation method, the comparison work is carried out with the traditional region growing method, and the comparison result is as follows:
(1) Comparison of processing results for the case of over-segmentation
When the simulation working condition is that one large ship is close to the laser radar, the point cloud segmentation effect of the conventional region growing method is shown in fig. 10, it can be seen that after the three-dimensional point cloud of the ship is segmented, the three-dimensional point cloud is composed of colors 1 to 4, and different gray scales are displayed in fig. 10, so that the conventional region growing method is used for segmenting one ship into a plurality of ships, and obvious over-segmentation is caused.
By using the method of the invention to divide the three-dimensional point cloud data of the ship, the effect graph is shown in fig. 11, and it can be found that although the ship still consists of a plurality of small blocks and gaps are reserved among the small blocks, the small blocks are red, and are shown as the same gray scale in fig. 11, namely, the three-dimensional point cloud data are divided into one class by the dividing algorithm, and no over-dividing phenomenon occurs.
(2) Comparison of processing results for under-segmentation cases
When the simulation working condition is that two ships are shielded, the point cloud segmentation effect of the conventional area growth method is shown in fig. 12, it can be seen that colors 1 to 3 exist after the two ship point clouds are segmented, 3 different colors are all present, and three different gray scales are shown in fig. 12, namely, the two ships are segmented into three ships by the conventional method, which is incorrect.
By using the method of the invention, the two ship point clouds are split under the working conditions, the effect is shown in fig. 13, and it can be seen that the split result is only two colors of color 1 and color 2, and is shown as two different gray scales in fig. 13, so that the point cloud split is accurately completed, and the under-split condition is effectively restrained.
(3) Comparison of the results of the segmentation time
Selecting samples containing 40 ten thousand points in the three-dimensional point cloud data, wherein the time for completing the segmentation calculation is about 10 seconds by testing the running time in the traditional method; by adopting the point cloud segmentation method, the time for completing the point cloud segmentation is reduced to about 1.0 second, and the time is shortened to 1/10 of the segmentation time of the traditional method, so that the method obviously shortens the point cloud segmentation time and has higher speed.
Claims (2)
1. The laser radar three-dimensional point cloud segmentation method is characterized by comprising the following steps of:
step 1: acquiring laser radar three-dimensional point cloud data:
acquiring original three-dimensional point cloud data by a laser radar, wherein the point cloud data is represented by three elements (r, theta, phi), r represents the distance of a scanning point, theta represents a horizontal direction angle, and phi represents a vertical direction angle;
step 2: preprocessing an original three-dimensional point cloud, which specifically comprises the following steps:
step 2.1: denoising the original three-dimensional point cloud: calculating average distance R from each point to all points in the neighborhood ave Assuming that the resulting distribution is a gaussian distribution having a mean and standard deviation, an outlier decision boundary radius R is set set R is taken as ave And R is set Comparing when R ave ≤R set Reserving the judged point; when R is ave >R set Then is covered byDetermining that the point is an outlier, and deleting the outlier from the original point cloud data set;
step 2.2: original three-dimensional point cloud downsampling is simplified:
downsampling the point cloud by voxel grid filtering, specifically: dividing the point cloud into cubic voxels by using octree, calculating the gravity centers of all points in the cubic voxels, and replacing all point clouds in the voxels by using the gravity center points;
step 2.3: two-dimensional representation of a three-dimensional point cloud:
converting the three-dimensional vector coordinates into three-dimensional rectangular coordinates, wherein the conversion formula is as follows:
wherein (x, y, z) in the above formula is the coordinates of each point in the point cloud data under a three-dimensional rectangular coordinate system;
two-dimensional array coordinate (x, y) location and storage distance data r for that location xy The relationship with the corresponding original three-dimensional point cloud vector coordinates (r, θ, ψ) can be expressed by the following formula:
(x,y,r xy ) For the preprocessed point cloud data, the horizontal direction angle resolution of the laser radar is as followsThe angular resolution in the vertical direction is +.>
Step 3: point cloud preliminary segmentation based on variable neighborhood dispersion search strategy:
the variable neighborhood distributed search strategy specifically comprises the following steps:
further expanding the eight neighborhood outwards to form a larger neighborhood, and setting the number of the neighborhood turns expanded outwards as dynamicSetting the outward detection distance of the laser radar in the horizontal direction as omega, the outward detection distance in the vertical direction as h, and the number of expansion lattices N in the horizontal and vertical directions x_exd And N y_exd The method meets the following conditions:
simplifying the traversing range, namely properly selecting the non-searching range in the variable neighborhood range
When expanding from eight adjacent areas of the selected seed, the width W of the horizontal and vertical non-search range after changing the adjacent areas Unser And height H Unser The method meets the following conditions:
through the above, the point cloud searching work is not performed in the area between the eight neighborhood areas of the selected seeds and the outer layer;
performing preliminary point cloud segmentation according to the variable neighborhood dispersion searching strategy;
step 4: and (3) performing point cloud re-segmentation based on the preliminary segmentation envelope diffusion strategy:
obtaining a point cloud set, extracting a segmentation envelope boundary of the set, expanding sigma distance outwards along a normal direction, calculating the gravity center O of the set, optionally selecting one boundary point N of the set, and calculating the distance d between two points ON Then the distance sigma is extended in the horizontal and vertical directions xec Sum sigma yec The method meets the following conditions:
comparing whether the expanded range has an intersection with other sets, if so, representing that the two sets can be fused into one set, and if not, representing that the two sets are still two different sets, wherein the obtained point cloud set is specifically:
step a: sequentially scanning the point cloud data, finding out the 1 st data point which is not attributed yet as a seed, and recording the point as S (x 0 ,y 0 );
Step b: by S (x) 0 ,y 0 ) The method comprises the steps of centering, determining a neighborhood and a non-searching range of a seed point according to the coordinate position of the seed point, and determining a point S in an effective searching range new (x * ,y * ) S is calculated according to the following new (x * ,y * ) And S (x) 0 ,y 0 ) Is the spatial distance d of (2) S :
In the above formula (x) * ,y * ,z * ) Setting the judgment distance as d meters for the space coordinates of the points, when d S When d is smaller, the region growing criterion is satisfied, representing S new (x * ,y * ) And S (x) 0 ,y 0 ) Belonging to the same set, simultaneously S new (x * ,y * ) Pushing the stack;
step c: taking a pixel out of the stack, taking the pixel as a new seed point, repeating the step b, and continuously searching points belonging to the same set;
step d: when the stack is empty, returning to the step a;
step e: and d, repeating the steps a to d until each point in the three-dimensional point cloud has a belonging set, and ending the region growth.
2. The laser radar three-dimensional point cloud segmentation method according to claim 1, wherein the method comprises the following steps of: the point cloud segmentation result is visualized, specifically: and displaying the point cloud segmentation result in the three-dimensional scene by adopting an OpenGL three-dimensional scene display technology.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911082345.5A CN110969624B (en) | 2019-11-07 | 2019-11-07 | Laser radar three-dimensional point cloud segmentation method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911082345.5A CN110969624B (en) | 2019-11-07 | 2019-11-07 | Laser radar three-dimensional point cloud segmentation method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110969624A CN110969624A (en) | 2020-04-07 |
CN110969624B true CN110969624B (en) | 2023-08-01 |
Family
ID=70030357
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911082345.5A Active CN110969624B (en) | 2019-11-07 | 2019-11-07 | Laser radar three-dimensional point cloud segmentation method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110969624B (en) |
Families Citing this family (22)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111507919B (en) * | 2020-04-16 | 2023-07-14 | 北京深测科技有限公司 | Denoising processing method for three-dimensional point cloud data |
CN111859612B (en) * | 2020-06-08 | 2024-06-25 | 北京经纬恒润科技股份有限公司 | Laser radar simulation method and device |
CN111724323B (en) * | 2020-06-19 | 2024-01-26 | 武汉海达数云技术有限公司 | Method and device for removing sunlight noise of laser radar point cloud |
CN111722201A (en) * | 2020-06-23 | 2020-09-29 | 常州市贝叶斯智能科技有限公司 | Data denoising method for indoor robot laser radar |
CN111951197B (en) * | 2020-08-14 | 2023-10-20 | 中国科学院自动化研究所苏州研究院 | Point cloud segmentation method based on structured light |
CN111929657B (en) * | 2020-08-26 | 2023-09-19 | 北京布科思科技有限公司 | Noise filtering method, device and equipment for laser radar |
CN112199991B (en) * | 2020-08-27 | 2024-04-30 | 广州中国科学院软件应用技术研究所 | Simulation point cloud filtering method and system applied to vehicle-road cooperation road side perception |
WO2022061604A1 (en) * | 2020-09-23 | 2022-03-31 | 株式会社Ntt都科摩 | Electronic device and signal envelop obtaining method |
CN112184867A (en) * | 2020-09-23 | 2021-01-05 | 中国第一汽车股份有限公司 | Point cloud feature extraction method, device, equipment and storage medium |
CN112465829B (en) * | 2020-10-26 | 2022-09-27 | 南京理工大学 | Interactive point cloud segmentation method based on feedback control |
CN112560747A (en) * | 2020-12-23 | 2021-03-26 | 苏州工业园区测绘地理信息有限公司 | Vehicle-mounted point cloud data-based lane boundary interactive extraction method |
CN112991435B (en) * | 2021-02-09 | 2023-09-15 | 中国农业大学 | Orchard end-of-line and head-of-line identification method based on 3D LiDAR |
CN113011113B (en) * | 2021-03-23 | 2022-04-05 | 中国空气动力研究与发展中心计算空气动力研究所 | Method for rapidly searching discrete point cloud repetition points |
CN113034682A (en) * | 2021-04-13 | 2021-06-25 | 阿波罗智联(北京)科技有限公司 | Point cloud data processing method, device, equipment and storage medium |
CN113177477A (en) * | 2021-04-29 | 2021-07-27 | 湖南大学 | Target detection and identification method based on three-dimensional point cloud analysis |
CN113866743B (en) * | 2021-12-06 | 2022-03-15 | 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) | Roadside laser point cloud simplification method and system for cooperative vehicle and road sensing |
CN114111628A (en) * | 2021-12-07 | 2022-03-01 | 西安理工大学 | Three-dimensional reconstruction algorithm for underwater target laser point cloud data |
CN114820986B (en) * | 2022-05-13 | 2024-04-09 | 广西微车检智能科技有限公司 | Laser radar-based trailer outline parameter measurement method |
CN116012613B (en) * | 2023-01-04 | 2024-01-16 | 北京数字绿土科技股份有限公司 | Method and system for measuring and calculating earthwork variation of strip mine based on laser point cloud |
CN116168386A (en) * | 2023-03-06 | 2023-05-26 | 东南大学 | Bridge construction progress identification method based on laser radar scanning |
CN116523414B (en) * | 2023-06-29 | 2023-09-05 | 深圳市鑫冠亚科技有限公司 | Production management method and system for composite nickel-copper heat dissipation bottom plate |
CN116797704B (en) * | 2023-08-24 | 2024-01-23 | 山东云海国创云计算装备产业创新中心有限公司 | Point cloud data processing method, system, device, electronic equipment and storage medium |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106056614A (en) * | 2016-06-03 | 2016-10-26 | 武汉大学 | Building segmentation and contour line extraction method of ground laser point cloud data |
CN108010092B (en) * | 2017-12-26 | 2018-11-13 | 哈尔滨工业大学 | A kind of city high density area Solar use potential evaluation method based on low altitude photogrammetry |
CN109961440B (en) * | 2019-03-11 | 2021-06-18 | 重庆邮电大学 | Three-dimensional laser radar point cloud target segmentation method based on depth map |
-
2019
- 2019-11-07 CN CN201911082345.5A patent/CN110969624B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN110969624A (en) | 2020-04-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110969624B (en) | Laser radar three-dimensional point cloud segmentation method | |
Cheng et al. | 3D building model reconstruction from multi-view aerial imagery and lidar data | |
CN111784840B (en) | LOD (line-of-sight) level three-dimensional data singulation method and system based on vector data automatic segmentation | |
CN112099046B (en) | Airborne LIDAR three-dimensional plane detection method based on multi-value voxel model | |
CN108074232B (en) | Voxel segmentation-based airborne LIDAR building detection method | |
CN111325138A (en) | Road boundary real-time detection method based on point cloud local concave-convex characteristics | |
US20230098480A1 (en) | Apparatus and method for detecting intersection edges | |
CN114926739B (en) | Unmanned collaborative acquisition processing method for geographical space information on water and under water of inland waterway | |
CN112033385A (en) | Pier pose measuring method based on mass point cloud data | |
Tian et al. | Lidar super-resolution based on segmentation and geometric analysis | |
CN107993242B (en) | Method for extracting boundary of missing area based on airborne LiDAR point cloud data | |
Gujski et al. | Machine learning clustering for point clouds optimisation via feature analysis in Cultural Heritage | |
Elkhrachy | Feature extraction of laser scan data based on geometric properties | |
Ni et al. | Applications of 3d-edge detection for als point cloud | |
CN112581511A (en) | Three-dimensional reconstruction method and system based on approximate vertical scanning point cloud rapid registration | |
CN115861566A (en) | LiDAR point cloud building integrity enhancement method based on intensity voxel model | |
CN111881964A (en) | Linear building mode identification method and system based on Delaunay triangulation network | |
Liu et al. | Displacement field reconstruction in landslide physical modeling by using a terrain laser scanner–Part 1: Methodology, error analysis and validation | |
CN109118565B (en) | Electric power corridor three-dimensional model texture mapping method considering shielding of pole tower power line | |
Guinard et al. | Planar polygons detection in lidar scans based on sensor topology enhanced RANSAC | |
CN111598941A (en) | Method, device and equipment for measuring gradient of tower and storage medium | |
CN106709473B (en) | Voxel-based airborne LIDAR road extraction method | |
CN117253205A (en) | Road surface point cloud rapid extraction method based on mobile measurement system | |
CN114879180A (en) | Seamless situation perception method for real-time fusion of unmanned ship-borne multi-element multi-scale radar | |
CN111145201B (en) | Steady and fast unmanned aerial vehicle photogrammetry mark detection and positioning method |
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 |