CN114440901A - Global hybrid map creation method - Google Patents

Global hybrid map creation method Download PDF

Info

Publication number
CN114440901A
CN114440901A CN202011236206.6A CN202011236206A CN114440901A CN 114440901 A CN114440901 A CN 114440901A CN 202011236206 A CN202011236206 A CN 202011236206A CN 114440901 A CN114440901 A CN 114440901A
Authority
CN
China
Prior art keywords
map
sub
dimensional probability
unmanned vehicle
global
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.)
Pending
Application number
CN202011236206.6A
Other languages
Chinese (zh)
Inventor
齐建永
高文祥
张哲华
龚建伟
陈慧岩
熊光明
吴绍斌
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beili Huidong Beijing Technology Co ltd
Bit Intelligent Vehicle Technology Co ltd
Beijing Institute of Technology BIT
Original Assignee
Beili Huidong Beijing Technology Co ltd
Bit Intelligent Vehicle Technology Co ltd
Beijing Institute of Technology BIT
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Beili Huidong Beijing Technology Co ltd, Bit Intelligent Vehicle Technology Co ltd, Beijing Institute of Technology BIT filed Critical Beili Huidong Beijing Technology Co ltd
Priority to CN202011236206.6A priority Critical patent/CN114440901A/en
Publication of CN114440901A publication Critical patent/CN114440901A/en
Pending legal-status Critical Current

Links

Images

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
    • G01C21/26—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 specially adapted for navigation in a road network
    • G01C21/28—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 specially adapted for navigation in a road network with correlation of data from several navigational instruments
    • G01C21/30—Map- or contour-matching
    • G01C21/32—Structuring or formatting of map data

Landscapes

  • Engineering & Computer Science (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Remote Sensing (AREA)
  • Automation & Control Theory (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Navigation (AREA)

Abstract

The invention provides a global hybrid map model which can simultaneously serve pose estimation and unmanned vehicle navigation based on the advantages and disadvantages of a grid map, a topological map and a geometric feature map and the advantages of various map models. The global mixed map is mainly composed of sub-map nodes and the communication relation among the sub-map nodes. The map data in the sub-map nodes includes a three-dimensional probability feature sub-map, a three-dimensional probability grid sub-map, and a passable area. The three-dimensional probability characteristic sub-map and the three-dimensional probability grid map adopt a mixed data structure based on a sequential storage structure and a tree structure to carry out data organization so as to improve the real-time performance of data updating of the three-dimensional probability characteristic sub-map and the three-dimensional probability grid map, and the robustness of the sub-map to system noise is improved through probability filtering. And matching the three-dimensional probability characteristic sub-map and the three-dimensional probability grid map with the point cloud together for pose estimation. A passable area for unmanned vehicle navigation is extracted based on the three-dimensional probability grid sub-map.

Description

Global hybrid map creation method
Technical Field
The invention relates to the technical field of unmanned vehicles, in particular to a method capable of simultaneously serving pose estimation and unmanned vehicle navigation.
Background
Currently, the map models commonly used for environment modeling include geometric maps, grid maps, topological maps, point cloud maps, hybrid maps, and the like. The geometric feature map carries out environment modeling by extracting geometric features such as points, lines, surfaces and the like in the environment, is a compact map model, is convenient for pose estimation, has low consumption of computing resources and small occupation of storage space, but requires the environment to have obvious features such as points, lines, surfaces and the like, can better describe environment targets with obvious geometric features such as the ground, trees and the like, but is useless for other complex environment conditions. It is not possible to distinguish, by means of geometric features, where an obstacle area is passable, and therefore it is generally not possible to use it for unmanned vehicle navigation. The point cloud map is used for simply splicing original point clouds of the sensors, almost all information of environmental observation is kept, but the data size of the point cloud map is too large, and the point cloud map is not suitable for constructing an online map. The grid map divides an environment into grids of uniform size, each of which is given an attribute of whether or not to be occupied, and is divided into a two-dimensional grid map and a three-dimensional grid map. The grid map description environment is more detailed than a geometric map and is more sparse than a point cloud map, and the grid map description environment is a compromise map model in terms of data volume and memory space occupation. The grid map can reduce the negative influence of environmental noise and observation noise on the map precision, is easy to create and update, and can be conveniently used for unmanned vehicle path planning and pose estimation. However, as the resolution of the grid map is reduced and the size of the map is increased, the number of grids is increased rapidly, the memory occupation and the updating calculation amount are increased rapidly, and the real-time performance of map updating is reduced. The topological map expresses the environment based on the topological theory, represents the targets in the environment as nodes, and represents the relation between the targets by using the connecting lines between the nodes, so that the topological map has low positioning requirement on the unmanned vehicle, and the accurate positioning of the unmanned vehicle is difficult to realize based on the topological map. The topological map is higher to the abstract degree of environmental data, can be very convenient be used for unmanned vehicle's global planning, to indoor environment, topological structure can be convenient target such as expression office, corridor, can be used for the navigation of indoor autonomous platform by the efficient, but cross-country environment is comparatively complicated, the extraction of node and the description of relation between the node are all more difficult, and topological map generally can not realize the accurate positioning of decimetre level in addition, consequently is difficult to be used for local navigation and location
Disclosure of Invention
In view of the foregoing analysis, the present invention is directed to a method for simultaneously performing pose estimation and unmanned vehicle navigation, so as to solve the problem that pose estimation and unmanned vehicle navigation cannot be performed simultaneously in the prior art.
The purpose of the invention is mainly realized by the following technical scheme:
a method of simultaneously servicing pose estimation and unmanned vehicle navigation, comprising the steps of:
creating and updating a map;
extracting a passable area;
further, the map creation and update includes:
1) map creation initialization, which is divided into two parts of global mixed map creation initialization and sub-map creation initialization;
2) the sub-map data are matched to obtain the relative pose relationship between the unmanned vehicle and the original pose of the current sub-map, so that the relative pose between two adjacent moments of the unmanned vehicle can also be obtained;
3) updating and switching the sub-map;
4) updating a global topological structure;
the invention has the following beneficial effects:
the global mixed map model capable of simultaneously serving pose estimation and unmanned vehicle navigation fully considers road surface jolt in the cross-country environment, various road conditions such as a ramp, a fork road, a bridge, an earth road and a cement road, and various environment appearance scenes such as grasslands, earth roads, bushes and forests, on one hand, the real-time map updating is ensured, on the other hand, a map with small error is created, so that a foundation is provided for unmanned vehicle positioning and safe navigation, and the global mixed map model has a wide application prospect in the field of unmanned driving.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
Drawings
The drawings are only for purposes of illustrating particular embodiments and are not to be construed as limiting the invention, wherein like reference numerals are used to designate like parts throughout.
FIG. 1 is a schematic diagram of a three-dimensional probability feature sub-map in an embodiment of the invention;
FIG. 2 is a schematic diagram illustrating a point cloud wire harness numbering according to an embodiment of the present invention;
Detailed Description
The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form a part hereof, and which together with the embodiments of the invention serve to explain the principles of the invention.
Firstly, in order to ensure the real-time performance of map updating and reduce the memory occupation and updating calculation amount of map data, the global map is divided into a plurality of sub-maps, each sub-map is used as the edge of the topology structure formed by the connection relationship between one node in the global topology structure, so the global mixed map can be written as:
Figure BSA0000223943120000031
wherein s isiTo indicate a sectionPoints, each node being defined as
Figure BSA0000223943120000032
Wherein N isiRepresenting a node siUnique number
Figure BSA0000223943120000041
Representing the pose of the node when it was initially created, called a sub-map siThe original pose of the node, once set, is not changed;
Figure BSA0000223943120000042
representing the pose of the node after global pose optimization, called the optimized pose of the sub-map i, and keeping the attribute consistent with the result after each optimization;
Figure BSA0000223943120000043
IMU pose when built for the sub-map node
Figure BSA0000223943120000044
Stored in the sub-map is the original pose xi of all vehicles contained in the sub-mapVVehicle optimizing position xiVopIMU pose xiSimuPoint cloud numbering
Figure BSA0000223943120000045
And the rotation histogram feature of the corresponding point cloud at each pose
Figure BSA0000223943120000046
The rotating histogram feature will be used for the first layer detection in closed loop detection;
Figure BSA0000223943120000047
the map data of the node, namely the sub map corresponding to the node.
EiIs associated with each SiCorresponding edge constraint information defined as
Figure BSA0000223943120000048
Wherein N isiThe number of the node corresponding to the edge is represented;
Figure BSA0000223943120000049
representation and node SiThe number of geographically adjacent connected nodes, the attribute being continuously updated as the global hybrid map is continuously updated and expanded.
Figure BSA00002239431200000410
The map data updating method comprises the steps that all map data information in one node is contained, a map model is in a three-dimensional grid map in a basic form, in order to improve the real-time performance of map updating, a mixed data structure formed by combining a tree-shaped data structure and a sequential storage structure is used for data organization, the real-time performance of data modification is improved to a greater extent at the cost of smaller data access real-time performance, and the real-time performance of map data updating is improved on the whole.
The sub-map data model adopts a three-layer architecture for data organization. The first layer is a dynamic expansion layer which adopts a sequential storage structure and expands along with the increasing of the detection range of the unmanned vehicle. The subdivision grids in each data grid in the dynamic expansion layer are organized in a tree structure and are divided into an intermediate data layer and a voxel layer from the dynamic expansion layer downwards. Each data grid in the dynamic extension layer contains 23*23*23512 intermediate data layer grids, each of which also contains 23*23*23A grid in a voxel layer is also called a voxel, 512 voxel layer grid. The resolution of the submap refers to the side length of each voxel in the submap, and when the resolution of the submap is 0.2m, each grid in the dynamic expansion layer can represent 12.8 × 12.8m3The environment of the range. Different types of data may be stored in each voxel as desired, and the difference in the type of data stored in the voxel is used hereinThe sub-maps are divided into two types. Stored in the first seed map voxel is the probability of whether the voxel is hit by the point cloud. The sub-map can be regarded as a sub-map formed after point cloud sparsification, and is called as a three-dimensional probability grid sub-map, and the resolution r of the sub-mapgThe sub-map is 0.2m, and the influence of environmental noise, observation noise and pose estimation noise can be well reduced by the probability of the reliability of the voxel state. The second seed map voxel stores the point cloud distribution characteristics in the voxel and the characteristic credibility in the voxel, which is called three-dimensional probability characteristic sub map in the invention, and the resolution r of the sub map isfWhich is 0.5m, is a schematic diagram of a three-dimensional probability feature sub-map as shown in fig. 1. The resolution of the three-dimensional probability feature sub-map is set to be larger so as to count more accurate point cloud distribution features in a larger range, and the point cloud distribution features are divided into line features, surface features and bulk features; in order to distinguish the voxel in the three-dimensional probability grid sub-map in terms of name, the voxel in the three-dimensional probability characteristic sub-map is called a characteristic voxel, the reliability of the characteristic can be represented by counting the point cloud source state in the characteristic voxel, and the reliability of the characteristic of the voxel is higher when the types of sources from different wiring harnesses at different moments are more. The three-dimensional probability grid sub-map and the three-dimensional probability feature sub-map serve pose matching and closed-loop detection together. The passable area is further extracted on the basis of the three-dimensional probability grid sub-map, and the passable area is used as an unmanned vehicle navigation map for a path planning module to carry out local path planning.
Step S1, creating and updating a map;
the creation of the global hybrid map of the unmanned vehicle is divided into four parts, namely map creation initialization, sub-map data matching, sub-map updating and switching and global topological structure updating.
The map creation initialization is divided into two parts, global hybrid map creation initialization and sub-map creation initialization. When the unmanned vehicle system is initially operated, the global hybrid map is initialized, and at this time, the global hybrid map only includes one sub-map node, and there is no side information, which may be expressed as:
G＝{S0}
therein, a sub-map S0N of (A)0Is set to 0;
Figure BSA0000223943120000061
is arranged as
Figure BSA0000223943120000062
Is arranged as
Figure BSA0000223943120000063
Set to the IMU observed pose at that time; the original pose of the unmanned vehicle is initialized to be
Figure BSA0000223943120000064
At this time, because the sub-map does not have data to be matched, the preprocessed point cloud data is directly added into the three-dimensional probability grid sub-map and the three-dimensional probability feature sub-map.
The sub-map data matching is to obtain the original poses of the unmanned vehicle and the current sub-map
Figure BSA0000223943120000065
The relative pose relationship between the unmanned vehicles can be obtained, and the map data matching process is the pose estimation process of the unmanned vehicles. In order to perform data matching, firstly, the point clouds are preprocessed, a number is assigned to each point cloud, and a harness number is assigned to each point in the point clouds, as shown in fig. 2, the harness number corresponding to the point a is 0, the harness number corresponding to the point B is 1, and the harness number corresponding to the point C is 2. For the three-dimensional probability feature sub-map, point cloud distribution in each feature voxel is counted, and the number N of points from different moments or different line bundle numbers in the voxel is recordedfea。NfeaEvery time the numerical value of (2) is increased, the higher the credibility of the feature classification in the feature extraction is represented, the Bayesian probability updating method is adopted to update the first step
Confidence of feature classification in i feature voxels
Figure BSA0000223943120000066
As shown in
Figure BSA0000223943120000067
Wherein
Figure BSA0000223943120000068
Is set to 0.6, represents the current characteristic voxel state confidence level,
Figure BSA0000223943120000069
after the probability is updated, based on the credibility of the state of the characteristic voxel obtained by all the observations in the sub-map,
Figure BSA00002239431200000610
the reliability of the state of the characteristic voxel after probability updating at the last time is shown, when the probability updating is carried out for the first time,
Figure BSA00002239431200000611
is set to 0.55.
The three-dimensional Gaussian distribution is used to statistically characterize the point cloud distribution in the voxels, as
D～N(μ，∑)
Wherein mu is the mean value of point cloud distribution in the characteristic voxel, and sigma is the covariance of point cloud distribution in the characteristic voxel. When the reliability of the characteristic voxel is larger than a certain threshold value Pthresh_fThen, the characteristic voxel is set as a credible characteristic voxel, and the characteristic value and the characteristic vector of the covariance matrix are solved:
∑＝VTΛV
where Λ is a diagonal matrix with diagonal elements λ1、λ2And λ3Three eigenvalues of the covariance matrix are arranged according to the size, and each column of V is an eigenvector corresponding to the eigenvalue in turn. In Principal Component Analysis (PCA), a feature vector represents a point cloudThe magnitude of the eigenvalue is the characterization of the variance of the point cloud distribution in the direction, therefore, when the eigenvalue λ in the covariance matrix is1Much greater than λ2And λ3When the point cloud in the characteristic voxel is in line characteristic distribution; when the characteristic value lambda1Much greater than λ2And λ3Then, the point cloud in the characteristic voxel is distributed in a surface characteristic manner; and when the three eigenvalues in the covariance matrix have no obvious difference, the point cloud in the characteristic voxel is distributed in a cluster-shaped characteristic.
For a three-dimensional probability grid sub-map, when a certain laser point in a frame of point cloud hits a certain voxel, the state of the voxel in the sub-map can be divided into three types, namely hit, pass and irrelevant, and for the voxel that the laser beam passes through, the voxel position can be considered to be without any barrier, namely the voxel is empty; for a hit voxel, it is then generally considered to have a blockage of the obstacle, i.e. the voxel is occupied by the obstacle. The probability value of the voxel exceeds 0.5, which means that the voxel is possibly occupied, and the larger the probability value is, the greater the confidence degree of the occupied voxel is; when the probability value of the voxel is less than 0.5, the voxel is considered to be empty, and the smaller the probability value is, the greater the reliability of the empty voxel is; the probability value of a voxel is initialized to 0. The observation errors are divided into two types, one type is the characteristic observation errors based on a three-dimensional probability characteristic sub-map, and the other type is the voxel observation errors based on a three-dimensional probability grid map. For characteristic observation errors, if the distance TCk，jThe nearest characteristic voxel is a credible characteristic voxel, and then:
Figure BSA0000223943120000081
wherein σ1T is a coordinate transformation matrix from the current vehicle coordinate system to the sub-map coordinate system, which is also the result of map matching, C is a weight coefficientk，jFor the current frame point cloud CkA certain point mu is a distance TCk，jThe point cloud distribution mean value in the nearest characteristic voxel, V is the distance TCk，jPoint cloud distribution eigenvector matrix, Λ, in the nearest eigen voxel-1The inverse matrix of the point cloud distribution eigenvalue matrix in the nearest characteristic voxel is used, and in order to highlight the distribution characteristics of the line characteristic and the surface characteristic, the inverse matrix is used-1For line feature order 1/λ10 for face feature order 1/λ1＝0，1/λ2＝0。
For voxel observation errors are:
Figure BSA0000223943120000082
wherein σ2Is a weight coefficient, p (v | C)k，j) For the distance TC in the current three-dimensional probability grid sub-mapk，jProbability values in the nearest voxels. Finally, the sub-map data matching problem is transformed into a non-linear least squares problem:
Figure BSA0000223943120000083
wherein e (C, S) is the observation error of the current point cloud, SigmaeA covariance matrix of e (C, S), a transformation matrix T satisfying the above formula*The result is a map matching result, and therefore the current original pose of the unmanned vehicle can be obtained:
Figure BSA0000223943120000084
wherein the content of the first and second substances,
Figure BSA0000223943120000085
for the current original pose of the unmanned vehicle to be estimated,
Figure BSA0000223943120000086
and obtaining the original pose of the current sub-map. In order to provide a better iteration heuristic value, the angular velocity and the linear velocity of the IMU are utilized to carry out dead reckoning so as to obtain a better iteration initial value Tinit。
Figure BSA0000223943120000091
Wherein the content of the first and second substances,
Figure BSA0000223943120000092
is the original pose, t, of the unmanned vehicle at the previous moment1To calculate out
Figure BSA0000223943120000093
Time stamp of time, t2As the current time, the time of day,
Figure BSA0000223943120000094
for IMU at t1And t2Angular velocity of the unmanned vehicle observed during the time, rodriguess (·), is a formula of Rodriguess, vlimFor the unmanned vehicle speed estimated from the original pose of the unmanned vehicle in the first two frames, Delta lin is relative to
Figure BSA0000223943120000095
Δ R is relative to
Figure BSA0000223943120000096
Attitude change matrix of Rt1Is t1A matrix of poses of the unmanned vehicle at the time.
The aim of updating the sub-map is to solve the pose transformation matrix according to the data matching of the sub-map
Figure BSA0000223943120000097
The current frame point cloud CkAnd aligning the vehicle coordinate system to the current frame sub-map coordinate system, and updating the corresponding three-dimensional probability feature sub-map, the three-dimensional probability grid sub-map and the node data according to the point cloud data. In order to save computing resources, the sub-map data are updated after the relative movement or interval time of the original pose of the vehicle in the current frame relative to the original pose of the vehicle in the previous frame exceeds a certain threshold. When updating, the current frame point cloud CkA certain point C ink，jThe result of the conversion to the current sub-map coordinate system is as follows:
C′k，j＝Tik·Ck，j＝RikCk，j+tik
point cloud data C 'aligned to the current sub-map coordinate system can be obtained according to formula'k，jAnd updating the data in the sub-map according to the data:
Si，k＝Si，k-1∪C′k
wherein S isi，k-1As a sub-map of the previous moment, Si，kIs the updated sub-map, C'kThe current frame point cloud under the sub-map coordinate system. When the complete point cloud of a frame is updated, the point cloud distribution mean and covariance matrix of each characteristic voxel in the three-dimensional probability characteristic sub-map are calculated again.
After updating the sub-map data, the current point cloud C 'needs to be counted'kTo facilitate rotational histogram matching during closed loop detection. First, the current point cloud C'kDividing the point cloud into multiple pieces of point clouds in the Z-axis direction as the main direction at step length of 0.2m, and then dividing the point in each piece of point cloud into the center C of the piece-shaped point cloud according to the point in each piece of point cloud0The angles of the point cloud are sorted in ascending order of angle, and the height information of the point cloud is ignored. For a certain point p in the sheet-like point cloudjIf point pjThere are some points { p) within a certain Euclidean distance neighborhoodm1, 2 … M, point pjWill become a landmark point, at which point pjIs defined as:
Figure BSA0000223943120000101
point pjTo point pmThe azimuth angles of the formed vectors are called rotation characteristic angles, the rotation characteristic values with the same rotation characteristic angles are added in all the sheet-shaped point clouds, and the rotation characteristic angles are mapped from 0-2 pi to 0-255 to obtain a sub-map siThe rotation histogram feature of the point cloud
Figure BSA0000223943120000102
Finally, the rotating histogram feature is processed
Figure BSA0000223943120000103
Current point cloud number and current vehicle original pose
Figure BSA0000223943120000104
And its corresponding IMU attitude
Figure BSA0000223943120000105
Added to the current sub-map
Figure BSA0000223943120000106
In (1).
Each sub-map containing NsFrame effective point cloud data, NsThe specific numerical value of (A) can be adjusted according to the precision of front-end attitude estimation, the higher the precision is, the slower the increase of the accumulated error is, and N issLarger values can be taken. In order to ensure that certain effective data exists in the map after map switching, the text updates the front map S and the rear map S simultaneouslyi-1、SiWhen S isiThe number of point cloud frames in the point cloud satisfies N is more than or equal to NsAt/2, Si-1In exactly stores NsFrame-effective point cloud data, middle Si-1The map data of the map is deleted from the existing active map and stored in the external memory, and a new sub-map S is opened up+1It is composed of
Figure BSA0000223943120000107
And quilt
Figure BSA0000223943120000108
Setting as the original pose of the unmanned vehicle during sub-map switching
Figure BSA0000223943120000109
And
Figure BSA00002239431200001010
and remains empty at this point.
And (3) global topological structure updating: when a map switch occurs, the old sub-map Si-1Is deleted while adding a new sub-map Si+1At this time, the sub-map node information S needs to be added to the global hybrid mapi+1And its corresponding constraint information Ei+1，Ei+1In (1)
Figure BSA0000223943120000111
Now only including one sub-map number, i.e. SiSub-map number NiDue to the addition of a sub-map node Si+1。SiAlso needs to update its constraint information NiTo it is
Figure BSA0000223943120000117
Adding sub-map Si+1Numbered Ni+1。
Step S2, passable area extraction
The three-dimensional probability grid sub-map and the three-dimensional probability characteristic sub-map do not contain clear obstacles and passable information, so the three-dimensional probability grid sub-map and the three-dimensional probability characteristic sub-map are not suitable for being used as a navigation map of an unmanned vehicle, but the three-dimensional probability grid sub-map can well filter environmental noise, observation noise and pose estimation noise in a probability filtering mode, so the passable area map is extracted on the basis of the three-dimensional probability grid sub-map, and is used as the navigation map of the unmanned vehicle for local path planning of the unmanned vehicle.
When the passable area is extracted, any two-dimensional grid in the passable area is counted along the Z direction in the corresponding three-dimensional grid mapsMaximum height of direction effective voxel center
Figure BSA0000223943120000112
Minimum height
Figure BSA0000223943120000113
And the number of significant voxels niAnd then can calculate to obtain theHeight difference at two-dimensional grid
Figure BSA0000223943120000114
By
Figure BSA0000223943120000115
The theoretical effective number of voxels at the two-dimensional grid can be calculated
Figure BSA0000223943120000116
On the basis of the above, the effective voxel density rho of the two-dimensional grid can be obtainedi＝ni/NiThe state of a two-dimensional grid in the passable area can be determined by the following formula.
Figure BSA0000223943120000121
In the passable area, a safe area and an unknown area can be passed by the unmanned vehicle, the positive barrier area is an area which the unmanned vehicle must avoid, and the suspended area such as a crown area and a suspended billboard is regarded as a suboptimal area which the unmanned vehicle should avoid passing through as much as possible, but if no other area can be selected in the environment, the unmanned vehicle can pass through the suspended area at a low speed.
In conclusion, the rapid terrain identification method based on the vehicle attitude provided by the embodiment of the invention fully considers various working conditions causing the vehicle attitude change under the running working condition of the off-road environment; different terrain working conditions can be identified with more than 80% of accuracy, and the identification speed is high; the intelligent vehicle driving system is independent of a vehicle longitudinal dynamics model, has good universality among different platforms, improves the rapid recognition and reaction adjustment capability of the intelligent vehicle when the driving condition changes suddenly, and has wide application prospect in the field of unmanned driving.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention.

Claims (6)

1. A global hybrid map model that can serve pose estimation and unmanned vehicle navigation simultaneously, comprising the steps of:
creating and updating an unmanned vehicle global hybrid map;
and on the basis of the three-dimensional probability grid sub-map, extracting a passable area map on the basis of filtering out environmental noise, observation noise and pose estimation noise by probability filtering.
2. The global hybrid map model of claim 1, wherein the map creation and update comprises:
1) initializing map creation;
2) matching sub map data;
3) updating and switching the sub-map;
4) and updating the global topological structure.
3. The global hybrid map model of claim 2, wherein the map creation initialization comprises:
establishing and initializing a global mixed map;
child map creation initialization.
4. The global hybrid map model of claim 2, wherein the sub-map data matching is performed to obtain a relative pose relationship between the unmanned vehicle and an original pose of a current sub-map, so that a relative pose between two adjacent moments of the unmanned vehicle can also be obtained.
5. The global hybrid map model of claim 2, wherein the sub-map update and switching purpose is to align the current frame point cloud from the vehicle coordinate system into the current frame sub-map coordinate system according to the pose transformation matrix solved by sub-map data matching, and update the corresponding three-dimensional probability feature sub-map, three-dimensional probability grid sub-map and node data according to the point cloud data.
6. The global hybrid map model of claim 1, wherein the passable region extraction is based on a three-dimensional probability grid sub-map, and the passable region map is extracted and used as a navigation map of the unmanned vehicle for local path planning of the unmanned vehicle on the basis of filtering out environmental noise, observation noise and pose estimation noise through probability filtering.
CN202011236206.6A 2020-11-05 2020-11-05 Global hybrid map creation method Pending CN114440901A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011236206.6A CN114440901A (en) 2020-11-05 2020-11-05 Global hybrid map creation method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011236206.6A CN114440901A (en) 2020-11-05 2020-11-05 Global hybrid map creation method

Publications (1)

Publication Number Publication Date
CN114440901A true CN114440901A (en) 2022-05-06

Family

ID=81361021

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011236206.6A Pending CN114440901A (en) 2020-11-05 2020-11-05 Global hybrid map creation method

Country Status (1)

Country Link
CN (1) CN114440901A (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104764457A (en) * 2015-04-21 2015-07-08 北京理工大学 Urban environment composition method for unmanned vehicles
CN109934868A (en) * 2019-03-18 2019-06-25 北京理工大学 One kind is based on three-dimensional point cloud and the matched vehicle positioning method of satellite mapping
CN111598916A (en) * 2020-05-19 2020-08-28 金华航大北斗应用技术有限公司 Preparation method of indoor occupancy grid map based on RGB-D information
CN111735451A (en) * 2020-04-16 2020-10-02 中国北方车辆研究所 Point cloud matching high-precision positioning method based on multi-source prior information

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104764457A (en) * 2015-04-21 2015-07-08 北京理工大学 Urban environment composition method for unmanned vehicles
CN109934868A (en) * 2019-03-18 2019-06-25 北京理工大学 One kind is based on three-dimensional point cloud and the matched vehicle positioning method of satellite mapping
CN111735451A (en) * 2020-04-16 2020-10-02 中国北方车辆研究所 Point cloud matching high-precision positioning method based on multi-source prior information
CN111598916A (en) * 2020-05-19 2020-08-28 金华航大北斗应用技术有限公司 Preparation method of indoor occupancy grid map based on RGB-D information

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
BEHNAM IRANI: "基于定位能力的移动机器人路径规划研究", 中国优秀硕士学位论文全文数据库, 15 January 2020 (2020-01-15) *
刘忠泽 等: "无人平台越野环境下同步定位与地图创建", 兵工学报, 31 December 2019 (2019-12-31), pages 2399 - 2406 *
李安旭: "移动机器人即时定位与地图构建技术研究", 中国优秀硕士学位论文全文数据库, 15 January 2020 (2020-01-15) *
肖强: "地面无人车辆越野环境多要素合成可通行区域检测", 中国优秀硕士学位论文全文数据库, 15 March 2016 (2016-03-15) *
胡玉文: "城市环境中基于混合地图的智能车辆定位方法研究", 中国博士学位论文全文数据库, 15 April 2015 (2015-04-15) *

Similar Documents

Publication Publication Date Title
US20220026232A1 (en) System and method for precision localization and mapping
US11393216B2 (en) Method of computer vision based localisation and navigation and system for performing the same
CN114419152B (en) Target detection and tracking method and system based on multi-dimensional point cloud characteristics
CN112923933A (en) Laser radar SLAM algorithm and inertial navigation fusion positioning method
Engel et al. Deeplocalization: Landmark-based self-localization with deep neural networks
CN112486197B (en) Fusion positioning tracking control method based on self-adaptive power selection of multi-source image
WO2021021862A1 (en) Mapping and localization system for autonomous vehicles
CN108426582B (en) Indoor three-dimensional map matching method for pedestrians
US11521113B2 (en) System and method for unifying heterogenous datasets using primitives
CN113238251B (en) Target level semantic positioning method based on vehicle-mounted laser radar
CN113110455A (en) Multi-robot collaborative exploration method, device and system for unknown initial state
CN115900708A (en) Robot multi-sensor fusion positioning method based on GPS (global positioning system) guided particle filter
KR102408981B1 (en) Method for Creating ND Map and Updating map Using it
US20230360244A1 (en) Using Deep Learning and Structure-From-Motion Techniques to Generate 3D Point Clouds From 2D Data
CN114440901A (en) Global hybrid map creation method
CN113379915B (en) Driving scene construction method based on point cloud fusion
CN112904388A (en) Fusion positioning tracking control method based on navigator strategy
KR102664699B1 (en) Method for building modeling aerial lidar, and computer program recorded on record-medium for executing method therefor
KR102616435B1 (en) Method for map update, and computer program recorded on record-medium for executing method therefor
CN114926649A (en) Data processing method, device and computer readable storage medium
Du et al. 3D LiDAR-Based Place Recognition Techniques: A Review of the Past 10 Years
WANG et al. Curb Detection and Mapping via Robust Iterative Gaussian Process Regression
CN118091672A (en) Unmanned aerial vehicle indoor scene positioning method and system based on laser radar
CN115560780A (en) Laser radar odometer positioning optimization method based on OSM road network constraint
CN118068840A (en) Mowing robot path planning method, mowing robot path planning equipment, mowing robot path planning medium and mowing robot path planning product

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