CN109186608A - A kind of rarefaction three-dimensional point cloud towards reorientation ground drawing generating method - Google Patents
A kind of rarefaction three-dimensional point cloud towards reorientation ground drawing generating method Download PDFInfo
- Publication number
- CN109186608A CN109186608A CN201811131452.8A CN201811131452A CN109186608A CN 109186608 A CN109186608 A CN 109186608A CN 201811131452 A CN201811131452 A CN 201811131452A CN 109186608 A CN109186608 A CN 109186608A
- Authority
- CN
- China
- Prior art keywords
- point
- tracing
- track
- curvature
- point cloud
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
- G01C21/20—Instruments for performing navigational calculations
Landscapes
- Engineering & Computer Science (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Automation & Control Theory (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Manipulator (AREA)
- Control Of Position, Course, Altitude, Or Attitude Of Moving Bodies (AREA)
Abstract
A kind of present invention with providing the rarefaction three-dimensional point cloud towards reorientation drawing generating method, belongs to three-dimensional point cloud map structuring and field of locating technology.This method firstly generates robot three-dimensional point cloud map and corresponding track, then carries out distance restraint to track point data and angle restriction pre-processes；Carry out the screening of track point data based on curvature limitation and distance interval constraint simultaneously later；Finally construct the three-dimensional point cloud map of rarefaction.The method of the invention curvature limitation based on tracing point and distance interval constraint simultaneously, can accurately select the redundant points to be deleted, can satisfy the application demand of robot rapid relocation in three-dimensional space.
Description
Technical field
The invention belongs to three-dimensional point cloud map structuring and field of locating technology, and in particular to one kind is based on time and space about
The curvature estimation algorithm of beam with carrying out a three-dimensional point cloud for cloud LS-SVM sparseness drawing generating method.
Background technique
With the continuous development of robot technology and enriching constantly for application scenarios, laser range sensor is by its measurement
The advantage for the such environmental effects such as precision is high, is not illuminated by the light is widely used in a variety of applications.Positioning based on laser point cloud is exactly base
The current pose of robot is determined in the registration of cloud scene.Currently used method is ICP (iterative closest
Point) registration and NDT (normal distribution transform) registration, wherein ICP algorithm is based between point set
Matching relationship is registrated, and NDT algorithm is registrated based on point cloud statistical model.But in real-time application, these sides
Method will receive a restriction for cloud map midpoint cloud data volume.It is per second by taking Velodyne VLP-16 three-dimensional laser sensor as an example
About 300,000 laser ranging points are acquired, the huge dense point cloud of data volume can be generated in this way in working continuously.If directly deposited
Dense point cloud is stored up, then with the continuous passage of time, the continuous accumulation of point cloud data can bring heavy burden to subsequent calculating.
Reorientation based on laser point cloud seeks to determine pose of the robot in existing map, to improve machine
Task efficiency.Since robot does not know that it is presently in position when reorientation, it is therefore desirable to be carried out in existing map
Global search.If existing map storage is dense point cloud, the calculation amount of global search is very big and is easy to appear mistake
Matching；If storage is grating map, since Raster Data Model can make the geometric center of each grid midpoint cloud or center of gravity
It is saved for new point cloud position, and this is not the truthful data that sensor obtains, therefore will lead to the decline of matching precision；Such as
Fruit storage is topological map, then due to topological side and not comprising any observation information that can be used for positioning, can only have
Robot localization is realized on the topological node of limit.
Pass through the analysis to robot obtained point cloud data in practical applications, it is known that when robot is run, if machine
Device people will appear a large amount of redundant data along linear running.This is because laser ranging range is big, it can needle in linear motion
A large amount of overlapped datas are generated to same local scene, to cause data redundancy.If carried out to this partial data reasonable
LS-SVM sparseness, does not interfere with the precision and accuracy of reorientation not only, but also will be greatly reduced the calculating of global search
Burden, it is therefore desirable to which reasonable LS-SVM sparseness is carried out to dense three-dimensional point cloud map.
Document (Zhang J, Singh S.LOAM:Lidar Odometry and Mapping in Real-time
[C] //Robotics:Science and Systems Conference.2014.) in propose it is a kind of in real time for six from
By the method for the three-dimensional laser scanner low drifting odometer and map structuring of degree movement.The algorithm is based on improved ICP algorithm
Obtain high-precision robot real time execution track and three-dimensional laser point cloud.The advantages of this method be can generate in real time it is high-precision
Dense three-dimensional laser point cloud map, while can also provide and put the track that cloud has time synchronizing relation.The deficiency of this method is
The data volume of generated dense point cloud is excessive, not can guarantee the real-time of the point cloud registering in robot reorientation.
Octree is a kind of hierarchical data structure that can divide three-dimensional space.It, will table by establishing tree
The three-dimensional space stated is divided into the identical sub- 3 d grid of eight sizes, is then also split to every sub- grid, so passs
Return until the size of sub- 3 d grid is identical as intended size.Document (Hornung A, Wurm K M, Bennewitz M,
et al.OctoMap:An efficient probabilistic 3D mapping framework based on
Octrees [J] .Autonomous Robots, 2013,34 (3): 189-206.) in propose a kind of ground based on Octree
Figure, this method reach the mesh of rarefaction three-dimensional point cloud by storing the centre coordinate for the grid for having a cloud to be distributed in grid
's.The deficiency of this method is if raster resolution will lead to greatly very much the decline of registration accuracy, and resolution ratio is too small, will increase
Computation burden causes the real-time of point cloud registering to decline.Furthermore Octree map will limit the range in entire space at the very start,
The scalability that this will lead to map is insufficient.
In order to guarantee the scalability of constructed map, document (Dub é R, Dugas D, Stumm E, et
al.SegMatch:Segment based place recognition in 3D point clouds[C]//IEEE
International Conference on Robotics and Automation.IEEE, 2017:5266-5272.) in mention
Go out a kind of dynamic grid map based on the segmentation of partial 3 d space, passes through point cloud center of gravity in a certain grid of dynamic memory
Coordinate stores three-dimensional point cloud.Its advantage is that the space that grid indicates allows dynamic to update, therefore the efficiency of global search is higher.
The disadvantages of the method are as follows the calculation amount of map rejuvenation is bigger, and registration accuracy ratio since its minimum grid is small-sized
Dense point cloud decreases.
Point cloud algorithms library PCL (Point Cloud Libarary, http://pointclouds.org/) is a large size
The cross-platform open source projects for being directed to two dimension, 3-D image and points cloud processing.It is provided in the algorithms library largely relevant to point cloud
General-purpose algorithm and efficient data structure.VoxelGrid filter in the algorithms library carries out a cloud by gridding method
Down-sampled, principle is with the center of gravity of grid midpoint cloud come these points of approximate representation, therefore can be carried out to three-dimensional point cloud sparse
Change.The shortcomings that algorithm, is that input point cloud is directly integrally carried out rarefaction by it, thus than stored in Octree grid center and
The speed that point cloud center of gravity is updated in dynamic grid map is slow, and point cloud registering precision is also below dense point cloud.
In conclusion a kind of rarefaction method for dense point cloud how is found, so that the point cloud map generated can
Adequately characterization three-dimensional environment, and the data storage capacity of three-dimensional point cloud map can be effectively reduced, while also needing the essence for guaranteeing to position
Degree, this carries out reorientation based on three-dimensional point cloud map to robot and is very important.
Summary of the invention
For various problems existing for the prior art and method, the invention proposes a kind of based on time and space constraint
Drawing generating method, this method have fully considered robot fortune to curvature estimation algorithm with carrying out a three-dimensional point cloud for cloud LS-SVM sparseness
Will lead to when row along straight way operation has mass of redundancy data in cloud map, and can be before guaranteeing positioning accuracy
It puts, LS-SVM sparseness is carried out to point cloud data.Robot can also synchronize preservation while building three-dimensional point cloud map in real time
The running track data of robot.Although robot running track itself is that continuously, robot is actually with discrete
The form of point saves track data, therefore the discrete point of this statement robot running track is referred to as tracing point, and by rail
The data that mark point is accumulated as are defined as track point data.Because there is time synchronization passes between track data and point cloud data
System, that is to say, that each of track location point is all corresponding in the point three-dimensional point cloud collected with robot, therefore logical
Crossing can achieve the purpose that carry out targeted rarefaction to cloud to the deletion of track redundant data.
Pretreated track point data can be obtained based on position constraint and angle restriction, and then by calculating tracing point
Curvature also just deletes and tracing point phase simultaneously in this way come redundancy tracing point generated when rejecting robot along linear running
Corresponding three dimensional point cloud, to achieve the purpose that three-dimensional point cloud rarefaction.Since robot actual motion track is usual
It will not be absolute straight line, therefore when robot is run along the track that tracing point curvature is less than certain threshold value, so that it may recognize
It is along approximate straight line motion at this time for robot.
How accurately to select the redundant points to be deleted is the core of this algorithm.It is this selection need based on two aspects because
Element, one be tracing point curvature, the other is distance interval constrain.It, can basis after completing the rejecting of redundancy track point data
The time synchronizing relation between tracing point and three-dimensional point cloud retained after rejecting constructs rarefaction three-dimensional point cloud map.The present invention
The point cloud map rarefaction method of proposition overcome conventional method reorientation when point cloud registering real-time it is poor, it is computationally intensive, dynamically
The limitation of performance difference, while the high advantage of dense point cloud registration accuracy is remained, it can satisfy robot in three-dimensional space
The application demand of rapid relocation.
Technical solution of the present invention:
A kind of rarefaction three-dimensional point cloud towards reorientation ground drawing generating method, steps are as follows:
(1) generation of robot three-dimensional point cloud map and corresponding track
By the real-time low drifting robot SLAM algorithm based on three-dimensional laser point cloud feature extraction and continuous point cloud registering,
Obtain robot three-dimensional point cloud map and corresponding track；
(2) distance restraint is carried out to track point data and angle restriction pre-processes
2.1) distance restraint pre-processes
If distance threshold is δd, the space length d between two neighboring tracing point is calculated, when d is greater than distance threshold δdWhen,
It then enters step 2.2), otherwise does not deal with.
2.2) angle restriction pre-processes
If angle threshold is δθ, calculate the yaw angle θ of two neighboring tracing pointyaw, roll angle θrollAnd pitching angle thetapitchIt
Between absolute value of the difference Δ θyaw、ΔθrollWith Δ θpitch, and by it and angle threshold δθIt is compared, as Δ θyaw、ΔθrollAnd Δ
θpitchRespectively less than angle threshold δθWhen, then previous tracing point in the two tracing points is deleted, otherwise is not dealt with.
(3) screening of track point data is carried out based on curvature limitation and distance interval constraint simultaneously
3.1) based on the screening of curvature limitation
A curvature threshold is set) as δthreshold, track point data includes s tracing point altogether, and each tracing point is by its three-dimensional
Each tracing point is numbered according to the chronological order of generation, obtains track point sequence P for coordinate representation1,P2,P3,…,
Pi,…,Ps。
B) the selection of track point set
In the track point sequence that step A) is obtained, in addition to n point of n point and the sequence ends part that sequence starts,
Remaining tracing point calculates curvature；As i-th (i ∈ [n+1, s-n]) a tracing point P of calculatingiCurvature when, then with PiCentered on,
Take tracing point PiAnd its continuous n point in front and back (total 2n+1 tracing point) constitutes set Si；
C track point set S) is calculatediCovariance matrix
In set SiIn, tracing point is numbered again, obtains sequence P1′,P2′,P3′,…,Pn′；Tracing point Pi′(i
∈ [1, n]) coordinate be pi'=[xi1,xi2,…,xim]T, then set SiIt is expressed as matrix
Wherein, n is set SiThe number of middle tracing point, m indicate the dimension of track point data, m=3.
Calculating matrix XiThe mean value of each columnThen matrix column vector subtracts mean value and obtainsAccording to [X1,...,Xl,...,Xm]T[X1,...,Xl,...,Xm] obtain track point set
Close SiCovariance matrix:
D singular value decomposition) is carried out to covariance matrix
When m=3, by carrying out singular value decomposition to covariance matrix, three eigenvalue λs are obtained0、λ1And λ2, wherein λ0≤
λ1≤λ2.Three eigenvalue λs0、λ1And λ2Corresponding feature vector respectively indicates three directions of data distribution, then i-th of track
The curvature δ of pointiAre as follows:
E step B) is repeated) arrive step D), the curvature of all tracing points is calculated.
F) by the curvature of the obtained each tracing point of step E) and curvature threshold δthresholdIt is compared, when the tracing point
Curvature be greater than the curvature and then retain the tracing point, it is on the contrary then delete.
3.2) screening based on distance interval constraint
It while considering curvature limitation, is constrained according to distance interval, is retained in the point data of track every fixed intervals
One tracing point avoids Robot linear running tracing point generated from being deleted excessive.
(4) the three-dimensional point cloud map of rarefaction is constructed
According to step (3) screening after required track point data, according to time synchronizing relation scene dense three-dimensional point
Corresponding three-dimensional point cloud is extracted in cloud, constitutes new three-dimensional point cloud map.
It can be used in relocating to guarantee to construct map, need that reasonable threshold value is arranged in each step.Step (2) by
Three rotation angles vary less when Robot linear running, while in view of the speed of service of robot is to track point data
The influence of number, distance threshold δdIt can be 0.1-0.3 meters, angle threshold δθValue range can be 5 ° -30 °, it is specific visual
The robot speed of service and application scenarios adjust.In step (3), the curvature of tracing point when due to Robot linear motion
Meeting very little, and in turning, curvature is larger, therefore can be by curvature threshold δthresholdIt is set as 0.001-0.005.Meanwhile in order to
Guarantee tracing point be unlikely to be removed it is excessive, while also to guarantee will not distance interval constraint will not excessively interfere curvature limitation,
The value range of fixed intervals is 5-10 meters, and the ranging range of specific visual three-dimensional laser sensor adjusts.
Beneficial effects of the present invention: the present invention generates the three-dimensional point cloud map of rarefaction, and which obviate be based on dense point
Cloud Orientation on map real-time is poor, insufficient based on grating map positioning accuracy, and asking based on topological map positioning continuity difference
Topic.The time that the present invention is taken full advantage of when constructing rarefaction three-dimensional point cloud map between track point data and three-dimensional point cloud is same
Step relationship, to ensure that mobile robot can still realize reliable reorientation based on the point cloud after rarefaction.The map is abundant
While stating three-dimensional environment, the point cloud data amount in three-dimensional point cloud map is reduced, ensure that subsequent robot's reorientation
Real-time and reliability, and since include in map is initial data, the point of sparse cloud map when relocating
The precision and dense point cloud map of cloud registration are essentially identical.
Detailed description of the invention
Fig. 1 is the flow chart of the method for the present invention.
Fig. 2 is dense point cloud map.
Fig. 3 is the corresponding tracing point of dense point cloud map.
Fig. 4 is the three-dimensional scenic schematic diagram comprising various structures.
Fig. 5 is tracing point data prediction schematic diagram.
Fig. 6 is the tracing point schematic diagram data generated according to curvature and distance interval constraint.
Fig. 7 is the three-dimensional point cloud map of rarefaction.
Specific embodiment
Technical solution of the present invention is further detailed below in conjunction with specific embodiments and the drawings, skill of the present invention
The flow chart of art scheme is as shown in Figure 1.
It is to generate rarefaction three-dimensional point based on 16 line three-dimensional laser sensors used in specific implementation process of the present invention
Cloud map, the three-dimensional lasers sensors such as 32 lines, 64 lines are equally applicable to this method.
The 16 line three-dimensional laser sensors that the present invention uses, the frequency for generating three-dimensional point cloud is 10Hz, per second to export
300000 three-dimensional laser points, measurement range is up to 100 meters.
A kind of rarefaction three-dimensional point cloud towards reorientation ground drawing generating method, steps are as follows:
(1) generation of robot three-dimensional point cloud map and corresponding track
Utilize aforementioned documents (Zhang J, Singh S.LOAM:Lidar Odometry and Mapping in Real-
Time [C] //Robotics:Science and Systems Conference.2014.) in mentioned method by by SLAM
(simutaneous localization and mapping) PROBLEM DECOMPOSITION is at two problems, firstly, LOAM algorithm is to each
The laser point of three-dimensional laser sensor scanning carries out feature extraction and obtains set of characteristic points.Then, based on scanning twice in succession
To set of characteristic points be registrated, obtain the odometer of a high frequency low distortion to estimate the speed of laser scanner.It connects down
Come, the characteristic point accumulation data based on Current Scan set of characteristic points and all before scanning after previous step registration are matched
Standard obtains the odometer and rotation translation relation of high-precision low-drift.Finally, dense three-dimensional is generated based on the rotation translation relation
Point cloud map and the track point data for having good time synchronizing relation therewith.It is illustrated in figure 2 the dense point cloud of a certain scene, is schemed
3 show track point data corresponding to Fig. 2.
The initial data for choosing a full scene includes the track point data and three-dimensional point cloud number under global coordinate system
According to two kinds of data retain the time that scanning every time obtains data when storing.It, should in order to more intuitively verify effectiveness of the invention
Scene may include the structures such as any number of straight way, bend, such as Fig. 4.
(2) distance restraint is carried out to track point data and angle restriction pre-processes
Tracing point data prediction is to reduce the calculation amount that subsequent track point data is rejected, and it includes two aspects:
Distance restraint and angle restriction.
2.1) distance restraint pre-processes
Since track point data generates, frequency is very high, and the distance between adjacent track point is smaller.If distance threshold
δdIt is 0.2 meter, calculates the space length d between two neighboring tracing point, when d is greater than distance threshold δdWhen, then it enters step
2.2) it, otherwise does not deal with.
2.2) angle restriction pre-processes
Since track point data generates, frequency is very high, posture of Robot when moving along a straight line between adjacent tracing point
Too big mutation is not had.If angle threshold δθIt is 30 °, calculates the yaw angle θ of two neighboring tracing pointyaw, roll angle θrollWith
Pitching angle thetapitchBetween absolute value of the difference Δ θyaw、ΔθrollWith Δ θpitch, and by it and angle threshold δθIt is compared, works as Δ
θyaw、ΔθrollWith Δ θpitchRespectively less than angle threshold δθWhen, then previous tracing point in the two tracing points is deleted, otherwise not
It deals with.Processing result such as Fig. 5.
(3) screening of track point data is carried out based on curvature limitation and distance interval constraint simultaneously
3.1) based on the screening of curvature limitation
The discrete curvature at the point can be calculated according to tracing point coordinate, and the discrete curvature can reflect out tracing point institute
Locate the dispersion degree of track, therefore track point data can further be rejected based on curvature limitation.Because of turning rail
The curvature of tracing point differs greatly in the curvature and straight way of mark point, therefore suitable curvature threshold is arranged can be by two kinds of tracks
Point distinguishes.
A curvature threshold δ) is setthresholdIt is 0.005, track point data includes s tracing point altogether, and each tracing point is by it
Three-dimensional coordinate indicates, is numbered to each tracing point according to the chronological order of generation, obtains track point sequence P1,P2,P3,…,
Pi,…,Ps。
B) the selection of track point set
In the track point sequence that step A) is obtained, in addition to 5 points of 5 points and the sequence ends part that sequence starts,
Remaining tracing point calculates curvature, and the size and calculating of curvature are influenced whether due to being included in the number of tracing point of calculating
Real-time, as i-th (i ∈ [n+1, s-n]) a tracing point P of calculatingiCurvature when, then with PiCentered on, take tracing point PiAnd its it is preceding
Continuous 5 points (totally 11 tracing points) constitute set S afterwardsi；
C track point set S) is calculatediCovariance matrix
In set SiIn, tracing point is numbered again, obtains sequence P1′,P2′,P3′,…,Pn′；Tracing point Pi′(i
∈ [1, n]) coordinate be pi'=[xi1,xi2,…,xim]T, then set SiIt is expressed as matrix
Wherein, n is set SiThe number of middle tracing point, m indicate the dimension of track point data, m=3.
Calculating matrix XiThe mean value of each columnThen matrix column vector subtracts mean value and obtainsAccording to [X1,...,Xl,...,Xm]T[X1,...,Xl,...,Xm] obtain track point set
Close SiCovariance matrix:
D singular value decomposition) is carried out to covariance matrix
It is real symmetric matrix by the covariance matrix that step C) is obtained.When m=3, by carrying out singular value to covariance matrix
It decomposes, obtains three eigenvalue λs0、λ1And λ2, wherein λ0≤λ1≤λ2.Three eigenvalue λs0、λ1And λ2Corresponding feature vector point
Not Biao Shi data distribution three directions, then the curvature δ of i-th of tracing pointiAre as follows:
E step B) is repeated) arrive step D), the curvature of all tracing points is calculated.
F) by the curvature of the obtained each tracing point of step E) and curvature threshold δthresholdIt is compared, when the tracing point
Curvature be greater than the curvature and then retain the tracing point, it is on the contrary then delete.
3.2) screening based on distance interval constraint
Since the curvature of Robot linear running tracing point generated is less than along curve motion tracing point generated
Curvature the lesser tracing point of curvature can be rejected therefore in order to realize rarefaction.However, when Robot linear running,
It is based only upon curvature and reject that will lead to retained tracing point number very few, this also means that three-dimensional point corresponding with tracing point
Cloud data can be excessively sparse, this will be unable to guarantee that robot is reliably relocated.In order to overcome the problem, curvature is being applied
Distance interval constraint is needed while be arranged when constraint, every 5 meters of reservations, one tracing point in the point data of track, avoids machine
Device people is deleted excessive along linear running tracing point generated.Such as Fig. 6.
(4) the three-dimensional point cloud map of rarefaction is constructed
According to step (3) screening after required track point data, according to time synchronizing relation scene dense three-dimensional point
Corresponding three-dimensional point cloud is extracted in cloud, constitutes new three-dimensional point cloud map.So far, the rarefaction three-dimensional point towards reorientation problem
The generation of cloud map terminates.The three-dimensional point cloud map ultimately generated is as shown in Figure 7.
Claims (2)
1. a kind of rarefaction three-dimensional point cloud towards reorientation ground drawing generating method, which is characterized in that steps are as follows:
(1) generation of robot three-dimensional point cloud map and corresponding track
By the real-time low drifting robot SLAM algorithm based on three-dimensional laser point cloud feature extraction and continuous point cloud registering, obtain
Robot three-dimensional point cloud map and corresponding track；
(2) distance restraint is carried out to track point data and angle restriction pre-processes
2.1) distance restraint pre-processes
If distance threshold is δd, the space length d between two neighboring tracing point is calculated, when d is greater than distance threshold δdWhen, then into
Enter step 2.2), otherwise does not deal with；
2.2) angle restriction pre-processes
If angle threshold is δθ, calculate the yaw angle θ of two neighboring tracing pointyaw, roll angle θrollAnd pitching angle thetapitchBetween it is poor
Absolute value delta θyaw、ΔθrollWith Δ θpitch, and by it and angle threshold δθIt is compared, as Δ θyaw、ΔθrollAnd Δ
θpitchRespectively less than angle threshold δθWhen, then previous tracing point in the two tracing points is deleted, otherwise is not dealt with；
(3) screening of track point data is carried out based on curvature limitation and distance interval constraint simultaneously
3.1) based on the screening of curvature limitation
A curvature threshold is set) as δthreshold, track point data includes s tracing point altogether, and each tracing point is by its three-dimensional coordinate
It indicates, each tracing point is numbered according to the chronological order of generation, obtains track point sequence P1,P2,P3,…,Pi,…,Ps；
B) the selection of track point set
It is remaining in addition to n point of n point and the sequence ends part that sequence starts in the track point sequence that step A) is obtained
Tracing point calculate curvature；As i-th (i ∈ [n+1, s-n]) a tracing point P of calculatingiCurvature when, then with PiCentered on, take rail
Mark point PiAnd its continuous n point in front and back (total 2n+1 tracing point) constitutes set Si；
C track point set S) is calculatediCovariance matrix
In set SiIn, tracing point is numbered again, obtains sequence P1′,P2′,P3′,…,Pn′；Tracing point Pi′(i∈[1,
N]) coordinate be pi'=[xi1,xi2,…,xim]T, then set SiIt is expressed as matrix
Wherein, n is set SiThe number of middle tracing point, m indicate the dimension of track point data, m=3；
Calculating matrix XiThe mean value of each columnThen matrix column vector subtracts mean value and obtainsAccording to [X1,...,Xl,...,Xm]T[X1,...,Xl,...,Xm] obtain track point set
Close SiCovariance matrix:
D singular value decomposition) is carried out to covariance matrix
When m=3, by carrying out singular value decomposition to covariance matrix, three eigenvalue λs are obtained0、λ1And λ2, wherein λ0≤λ1≤
λ2；Three eigenvalue λs0、λ1And λ2Corresponding feature vector respectively indicates three directions of data distribution, then i-th tracing point
Curvature δiAre as follows:
E step B) is repeated) arrive step D), the curvature of all tracing points is calculated；
F) by the curvature of the obtained each tracing point of step E) and curvature threshold δthresholdIt is compared, when the song of the tracing point
Rate is greater than the curvature and then retains the tracing point, on the contrary then delete；
3.2) screening based on distance interval constraint
It while considering curvature limitation, is constrained according to distance interval, retains one every fixed intervals in the point data of track
Tracing point avoids Robot linear running tracing point generated from being deleted excessive；
(4) the three-dimensional point cloud map of rarefaction is constructed
According to the required track point data after step (3) screening, according to time synchronizing relation in the dense three-dimensional point cloud of scene
Corresponding three-dimensional point cloud is extracted, new three-dimensional point cloud map is constituted.
2. a kind of rarefaction three-dimensional point cloud towards reorientation according to claim 1 drawing generating method, feature exist
In the distance threshold δdIt is 0.1-0.3 meters, angle threshold δθIt is 5 ° -30 °, curvature threshold δthresholdFor 0.001-0.005,
The value of fixed intervals is 5-10 meters.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811131452.8A CN109186608B (en) | 2018-09-27 | 2018-09-27 | Repositioning-oriented sparse three-dimensional point cloud map generation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811131452.8A CN109186608B (en) | 2018-09-27 | 2018-09-27 | Repositioning-oriented sparse three-dimensional point cloud map generation method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109186608A true CN109186608A (en) | 2019-01-11 |
| CN109186608B CN109186608B (en) | 2021-10-15 |
Family
ID=64907395
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811131452.8A Active CN109186608B (en) | 2018-09-27 | 2018-09-27 | Repositioning-oriented sparse three-dimensional point cloud map generation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109186608B (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109848996A (en) * | 2019-03-19 | 2019-06-07 | 西安交通大学 | Extensive three-dimensional environment map creating method based on figure optimum theory |
| CN110415281A (en) * | 2019-07-30 | 2019-11-05 | 西安交通大学深圳研究院 | A kind of point set rigid registration method based on Loam curvature weighting |
| CN110515054A (en) * | 2019-08-23 | 2019-11-29 | 斯坦德机器人（深圳）有限公司 | Filtering method and device, electronic equipment, computer storage medium |
| CN110728288A (en) * | 2019-10-12 | 2020-01-24 | 上海高仙自动化科技发展有限公司 | Corner feature extraction method based on three-dimensional laser point cloud and application thereof |
| CN110928312A (en) * | 2019-12-16 | 2020-03-27 | 深圳市银星智能科技股份有限公司 | Robot position determination method, non-volatile computer-readable storage medium, and robot |
| CN112269386A (en) * | 2020-10-28 | 2021-01-26 | 深圳拓邦股份有限公司 | Method and device for repositioning symmetric environment and robot |
| CN112446952A (en) * | 2020-11-06 | 2021-03-05 | 杭州易现先进科技有限公司 | Three-dimensional point cloud normal vector generation method and device, electronic equipment and storage medium |
| CN112767456A (en) * | 2021-01-18 | 2021-05-07 | 南京理工大学 | Three-dimensional laser point cloud rapid relocation method |
| CN113447949A (en) * | 2021-06-11 | 2021-09-28 | 天津大学 | Real-time positioning system and method based on laser radar and prior map |
| CN113776530A (en) * | 2020-09-11 | 2021-12-10 | 北京京东乾石科技有限公司 | Point cloud map construction method and device, electronic equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120089295A1 (en) * | 2010-10-07 | 2012-04-12 | Samsung Electronics Co., Ltd. | Moving robot and method to build map for the same |
| US20120316784A1 (en) * | 2011-06-09 | 2012-12-13 | Microsoft Corporation | Hybrid-approach for localizaton of an agent |
| CN105069840A (en) * | 2015-09-14 | 2015-11-18 | 南开大学 | Three-dimensional normal distribution transformation point cloud registration method based on curvature feature |
| CN106599108A (en) * | 2016-11-30 | 2017-04-26 | 浙江大学 | Method for constructing multi-mode environmental map in three-dimensional environment |
| CN108205566A (en) * | 2016-12-19 | 2018-06-26 | 北京四维图新科技股份有限公司 | A kind of method and device being managed based on track to cloud, navigation equipment |
-
2018
- 2018-09-27 CN CN201811131452.8A patent/CN109186608B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120089295A1 (en) * | 2010-10-07 | 2012-04-12 | Samsung Electronics Co., Ltd. | Moving robot and method to build map for the same |
| US20120316784A1 (en) * | 2011-06-09 | 2012-12-13 | Microsoft Corporation | Hybrid-approach for localizaton of an agent |
| CN105069840A (en) * | 2015-09-14 | 2015-11-18 | 南开大学 | Three-dimensional normal distribution transformation point cloud registration method based on curvature feature |
| CN106599108A (en) * | 2016-11-30 | 2017-04-26 | 浙江大学 | Method for constructing multi-mode environmental map in three-dimensional environment |
| CN108205566A (en) * | 2016-12-19 | 2018-06-26 | 北京四维图新科技股份有限公司 | A kind of method and device being managed based on track to cloud, navigation equipment |
Non-Patent Citations (2)
| Title |
|---|
| 王丽英等: "多回波激光雷达点云冗余数据辨识及消除", 《测绘科学》 * |
| 闫飞等: "基于拓扑高程模型的室外三维环境模型与路径规划", 《自动化学报》 * |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109848996A (en) * | 2019-03-19 | 2019-06-07 | 西安交通大学 | Extensive three-dimensional environment map creating method based on figure optimum theory |
| CN110415281B (en) * | 2019-07-30 | 2022-04-22 | 西安交通大学深圳研究院 | Loam curvature weighting-based point set rigid registration method |
| CN110415281A (en) * | 2019-07-30 | 2019-11-05 | 西安交通大学深圳研究院 | A kind of point set rigid registration method based on Loam curvature weighting |
| CN110515054A (en) * | 2019-08-23 | 2019-11-29 | 斯坦德机器人（深圳）有限公司 | Filtering method and device, electronic equipment, computer storage medium |
| CN110728288A (en) * | 2019-10-12 | 2020-01-24 | 上海高仙自动化科技发展有限公司 | Corner feature extraction method based on three-dimensional laser point cloud and application thereof |
| CN110728288B (en) * | 2019-10-12 | 2022-06-28 | 上海高仙自动化科技发展有限公司 | Corner feature extraction method based on three-dimensional laser point cloud and application thereof |
| CN110928312A (en) * | 2019-12-16 | 2020-03-27 | 深圳市银星智能科技股份有限公司 | Robot position determination method, non-volatile computer-readable storage medium, and robot |
| CN110928312B (en) * | 2019-12-16 | 2021-06-29 | 深圳市银星智能科技股份有限公司 | Robot position determination method, non-volatile computer-readable storage medium, and robot |
| CN113776530A (en) * | 2020-09-11 | 2021-12-10 | 北京京东乾石科技有限公司 | Point cloud map construction method and device, electronic equipment and storage medium |
| CN112269386A (en) * | 2020-10-28 | 2021-01-26 | 深圳拓邦股份有限公司 | Method and device for repositioning symmetric environment and robot |
| CN112269386B (en) * | 2020-10-28 | 2024-04-02 | 深圳拓邦股份有限公司 | Symmetrical environment repositioning method, symmetrical environment repositioning device and robot |
| CN112446952A (en) * | 2020-11-06 | 2021-03-05 | 杭州易现先进科技有限公司 | Three-dimensional point cloud normal vector generation method and device, electronic equipment and storage medium |
| CN112446952B (en) * | 2020-11-06 | 2024-01-26 | 杭州易现先进科技有限公司 | Three-dimensional point cloud normal vector generation method and device, electronic equipment and storage medium |
| CN112767456A (en) * | 2021-01-18 | 2021-05-07 | 南京理工大学 | Three-dimensional laser point cloud rapid relocation method |
| CN112767456B (en) * | 2021-01-18 | 2022-10-18 | 南京理工大学 | Three-dimensional laser point cloud rapid relocation method |
| CN113447949A (en) * | 2021-06-11 | 2021-09-28 | 天津大学 | Real-time positioning system and method based on laser radar and prior map |
| CN113447949B (en) * | 2021-06-11 | 2022-12-09 | 天津大学 | Real-time positioning system and method based on laser radar and prior map |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109186608B (en) | 2021-10-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109186608A (en) | A kind of rarefaction three-dimensional point cloud towards reorientation ground drawing generating method | |
| Ding et al. | Automatic registration of aerial imagery with untextured 3d lidar models | |
| CN110717983A (en) | Building facade three-dimensional reconstruction method based on knapsack type three-dimensional laser point cloud data | |
| CN111696210A (en) | Point cloud reconstruction method and system based on three-dimensional point cloud data characteristic lightweight | |
| Li et al. | Leveraging planar regularities for point line visual-inertial odometry | |
| Alidoost et al. | An image-based technique for 3D building reconstruction using multi-view UAV images | |
| Kersten et al. | Automated generation of an historic 4D city model of Hamburg and its visualisation with the GE engine | |
| He et al. | Incremental 3D line segment extraction from semi-dense SLAM | |
| CN115131514A (en) | Method, device and system for simultaneously positioning and establishing image and storage medium | |
| CN115690138A (en) | Road boundary extraction and vectorization method fusing vehicle-mounted image and point cloud | |
| Javanmardi et al. | Autonomous vehicle self-localization based on probabilistic planar surface map and multi-channel LiDAR in urban area | |
| Palma et al. | Detection of geometric temporal changes in point clouds | |
| Withers et al. | Modelling scene change for large-scale long term laser localisation | |
| CN117706577A (en) | Ship size measurement method based on laser radar three-dimensional point cloud algorithm | |
| CN116051980B (en) | Building identification method, system, electronic equipment and medium based on oblique photography | |
| JP6761388B2 (en) | Estimator and program | |
| CN113227713A (en) | Method and system for generating environment model for positioning | |
| Li et al. | Efficient laser-based 3D SLAM in real time for coal mine rescue robots | |
| Goshin et al. | Parallel implementation of the multi-view image segmentation algorithm using the Hough transform | |
| CN114577196A (en) | Lidar positioning using optical flow | |
| Nilosek et al. | Geo-accurate model extraction from three-dimensional image-derived point clouds | |
| Kang et al. | 3D urban reconstruction from wide area aerial surveillance video | |
| Alkhatib et al. | Camera pose estimation based on structure from motion | |
| Amer et al. | Crowd-sourced visual data collection for monitoring indoor construction in 3d | |
| He et al. | Automatic docking recognition and location algorithm of port oil loading arm based on 3D laser point cloud |
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 |