CN103575272A - Method for extracting natural landmarks for mobile robot in forest environment - Google Patents
Method for extracting natural landmarks for mobile robot in forest environment Download PDFInfo
- Publication number
- CN103575272A CN103575272A CN201310568954.8A CN201310568954A CN103575272A CN 103575272 A CN103575272 A CN 103575272A CN 201310568954 A CN201310568954 A CN 201310568954A CN 103575272 A CN103575272 A CN 103575272A
- Authority
- CN
- China
- Prior art keywords
- trunk
- segmentation
- environment
- forest environment
- robot
- 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
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
- G01C21/34—Route searching; Route guidance
- G01C21/36—Input/output arrangements for on-board computers
- G01C21/3602—Input other than that of destination using image analysis, e.g. detection of road signs, lanes, buildings, real preceding vehicles using a camera
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/50—Depth or shape recovery
- G06T7/521—Depth or shape recovery from laser ranging, e.g. using interferometry; from the projection of structured light
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Optics & Photonics (AREA)
- Theoretical Computer Science (AREA)
- Automation & Control Theory (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a method for extracting natural landmarks for a mobile robot in a forest environment. Landmarks are used as indications for positioning and navigation on a robot, and in an unknown environment, preset landmarks do not exist and natural features in the unknown environment need to be extracted and are used as landmarks. The method comprises the following steps of carrying out three-dimensional laser scanning on a forest environment, carrying out tree trunk-oriented point cloud segmentation on the scanning result, carrying out geometrical element extraction, and carrying out tree trunk section clustering and integration to obtain a straight trunk section which uniquely represents the tree and is used as the landmark. The method utilizes a tree trunk in the forest environment as the natural landmark, realizes extraction of the tree trunk and its geometric features in the complex forest environment and utilizes the tree trunk and its geometric features to realize positioning and navigation on the mobile robot. The method solves the problem of high difficulty of positioning and navigation on the robot in the forest environment, and realizes extraction of high-quality natural feature landmarks in the non-structural forest environment in which a coded disc, a visual sensor and a GPS system difficultly realize effective robot positioning.
Description
Technical field
The invention belongs to localization for Mobile Robot and field of navigation technology, relate to the extracting method of natural landmark for mobile robot in a kind of forest environment, the intelligent mobile robot or the Autonomous Vehicles that are adapted at operation in the forest environment of field are used.
Background technology
Through development for many years, intelligent independent mobile robot has demonstrated more and more wide application prospect in fields such as industry, agricultural, service sectors.Particularly in unknown or dangerous environment, mobile robot can replace the mankind to carry out the operations such as exploration, detection and operation, and this makes mobile robot have huge potential value and demand in fields such as celestial body detecting, subsurface investigation, ocean development, military anti-terrorism, disaster relief, dangerous material processing, nuclear environment operations.
Regardless of mobile robot's applied environment, in its motion process, to solve all the time self location and navigation problem.Development along with mobile robot's research and application, orientation problem research turns to destructuring outdoor environment from structurized buildings environment gradually, from two dimensional surface environment, turn to three-dimensional environment, from small-sized simple environment, turn to complex environment on a large scale, from the constant environment of static state, turn to the environment of dynamic change.Yet outdoor physical environment self complexity, irregular feature have brought very large challenge to location and navigation, particularly difficult especially for this mixed and disorderly extensive environment of forest.In forest environment, gps signal can block and then lose efficacy due to tree branches in the region of the branches and leaves by dense and vegetation covering, and in this case, by robot self-sensor device, positioning navigation is the unique method addressing this problem.Vision sensor and laser sensor are the most frequently used robot sensors, yet visual sensing data are subject to the impact of the factors such as illumination variation, shade larger, thus therefrom extract object information become difficulty and unreliable.This makes to adopt laser sensor to obtain environmental information becomes the most appropriate solution.In addition, earth's surface that forest landform has is coarse, slope change is large etc., and feature not only makes us cannot predict definite ground level, and causes the six-freedom degree pose of robot all to have larger error, is badly in need of proofreading and correct.The object in scene that makes in a jumble of environment blocks mutually, even the slight variation at robot location and visual angle all can cause the violent difference of visible scenery.Scenery is carried out identification, the understanding of semantic hierarchies in non-structured dense forest environment, and then the extraction that completes high-level geometric element is also a complexity and difficult task.
Summary of the invention
The object of the invention is to solve the orientation problem of mobile robot in forest environment, natural landmark for mobile robot extracting method in a kind of forest environment is provided.
The present invention adopts three-dimensional laser sensor to obtain forest environment information, has invented a kind ofly brand-new from the three-dimensional point cloud of laser measurement data, to extract trunk as the method for natural landmark.This natural landmark extracting method take that scene is understood and trees automatic analysis is basic, and from counting rareness and extracted the geometric model of trunk the serious 3-D scanning of noise, concrete grammar is divided into three steps:
The 1.th pair of forest environment carries out 3 D laser scanning detection, utilizes the circular feature of trunk profile that three-dimensional point cloud is divided into sweep segment, according to space adjacency, sweep segment is organized into trunk segmentation.
The inadequate trunk segmentation of the 2.th geological information obtaining from the 1st step, extract circle model, and according to the center of circle, simulate the axis (this process is an iterative process) of trunk segmentation.
The 3.th by analyzing the discontinuous type in space of trunk segmentation, they are carried out to cluster and integration, and the difference according to trunk segmentation in space distribution, combines the segmentation that belongs to same trunk, remove crotch, and it is dry as road sign finally to obtain unique Naoki that represents trees simultaneously.
Obtain trunk as high-level characteristic after, can use paired scan matching method based on feature to follow the tracks of the six-dimensional pose of robot, realize robot location.
advantage of the present invention and beneficial effect:
Natural landmark extracting method based on heat transfer agent provided by the invention, solved a difficult problem of carrying out robot location in forest environment, be applicable to be blocked at gps signal, code-disc, vision sensor be difficult to the stable forest environment that reliable location is provided, this method is not subject to weather and illumination effect, is a kind of round-the-clock field robot localization method.The fields such as the method robot, Autonomous Vehicles, military affairs, exploration rescue in the wild all have application prospect.
Accompanying drawing explanation
Fig. 1 is that natural landmark is extracted process flow diagram.
Fig. 2 is two-dimensional scan plane and the crossing sweep segment schematic diagram that obtained of trunk.
Fig. 3 is the upper bound and the lower bound of circle model radius, (a) lower bound, (b) upper bound.
Fig. 5 is the discontinuous three kinds of situations of trunk segment space, (a) fracture, (b) bending, (c) bifurcated.
Fig. 6 is the original 3D scanning with sloping scale Forest Scene.
Fig. 7 is that in Fig. 6, the road sign with sloping scale Forest Scene extracts result.
Embodiment
one, towards the three-dimensional point cloud of trunk, cut apart
Forest environment is carried out to 3 D laser scanning detection, and robot scans resulting three-dimensional point cloud with certain pose to special scenes and is called as a width 3-D scanning.One width 3-D scanning operates in by pitching the 2D scan-data that different scanning plane obtains by two-dimensional laser sensor and forms, and therefore the data of obtaining can be called to a line of 3-D scanning in a plane of scanning motion, and arranging of three-dimensional point cloud can be regarded one as accordingly
two-dimensional matrix.Due to the intrinsic sampling mode of pitching type three-dimensional laser sensor,
individual two-dimensional scan plane is not parallel to each other, and spacing is also not quite similar.Use
represent the scanning angle between the current plane of scanning motion and local reference plane, can obtain following matrix description:
3-D scanning
comprise
oK,
, wherein,
represent the
the scan-data that the individual plane of scanning motion obtains is capable,
be
the scanning angle that line scanning is corresponding, and have
.Each provisional capital comprises
individual data point,
,
position represents the
row,
the analyzing spot of column position,
.
In order to obtain environmental information from a cloud, need to cut apart a cloud, object is according to whether hitting same object for according to being divided into segmentation by each two-dimensional scan line.Because the final goal of this method is to extract trunk, so algorithm is cut apart the geometric properties around trunk first targetedly to a cloud.
Utilize two spaces between object known, if there is the laser beam that does not hit any object between two analyzing spots in same two-dimensional scan plane, two analyzing spots are not continuous, and they necessarily belong to different objects.
In addition, for continuous sweep point
with
if from the view of robot, what they hit is two trees that block mutually front and back, also should be separated, and in order to cut apart two-dimensional line segment now, according to the cylindrical character of trunk profile, analyzing spot is carried out to projection.Background context knowledge according to solid geometry can be learnt, any cross section of right circular cylinder, no matter be to tilt or level, ellipse is still just round, along after its axis projection all corresponding to cylindrical positive rounded bottom surface.And trees can carry out modeling with right circular cylinder just.Although hit between the Scanning Section of same trunk and not parallel, their projection can be summed up as same circular arc, so just 3D problem has been changed into 2D problem.If point
with
meet one of following two conditions, they can distinguish two adjacent sectional
with
:
2)
with
be
on continuity point, but Euclidean distance between their projections is greater than threshold value.
When two-dimensional scan plane, cut apart completely, sweep segment screening need to be combined into the trunk segmentation of containing high layer information.The basic ideas that address this problem are to use the barycenter of sweep segment to describe its position, according to each segmentation, at the vertical adjacency in space, they are connected into grouping in turn, and what according to the height of the length of sweep segment and grouping, differentiate which grouping describes is trunk or sturdy crotch.
For weighing the relation of each sweep segment in three dimensions, first need to describe the position of each sweep segment.By 3-D scanning
on row
individual segmentation is denoted as
.Because the sweep segment height in same two-dimensional scan plane is not quite similar, therefore use line index
the position of sign sweep segment in the vertical direction.In addition, in the horizontal direction, will
projection obtains in partial reference plane
.
by width
and barycenter
describe, wherein
be defined as
starting point
and end point
at the Euclidean distance of overall reference plane subpoint,
be
middle a little mean place in partial reference plane.
Trunk segmentation can be regarded as by adjacent sweep segment on one group of space and form.These sweep segment are from continuous two-dimensional scan plane.Therefore, k trunk segmentation is represented as
, and respectively scan between angle and have
, wherein,
be
the index of individual continuous two-dimensional scan plane.
In order to construct trunk segmentation, need to filter out long sweep segment.If
width meet following condition:
Explanation
can be regarded as radius and be no more than the upper limit
right circular cylinder on arc,
may be obtained from trunk.Otherwise,
necessarily take from unbroken ground or dense tree crown.
After filtering for the first time, the segmentation of candidate's trunk will successively be constructed.Construction algorithm is by the trunk set from empty
start, according to
ascending order, be each sweep segment
search for the trunk segmentation of its subordinate.For current
, algorithm inspection
in existing each trunk segmentation
if, do not have trunk segmentation to meet neighbour's condition, explanation up to the present, in space, still not do not form and
contiguous grouping, therefore
start alone a new grouping
, and will
add set
.Otherwise, will
the trunk segmentation that adds its subordinate as last sweep segment.
Form trunk segmentation
each sweep segment by according to the angle of its place two-dimensional scan plane
ascending order is arranged, and has maximum
sweep segment be denoted as
.According to the scanning sequency in vertical direction,
be
in the most approaching
sweep segment.Therefore, if
belong to
, so
should be positioned at
neighborhood in.Two components based on description sweep segment mentioned above
with
, when following two conditions meet simultaneously,
be considered to be positioned at
neighbour:
Condition 1) mean explanation
with
be positioned at adjacent two-dimensional scan plane, and the former
larger.Condition 2) mean
horizontal centroid be positioned at
in the border circular areas determining, the center of circle in this region is exactly
horizontal centroid, radius is
half of width, requirement
with
projection is overlapping, be because in adjacent scanning, the segmentation that belongs to same object has continuity, therefore in the vertical direction not only, and in the horizontal direction can be apart from not too far away, the center of a segmentation is bound to drop in another segmentation limit of adjacent scanning slice.
If scanning
comprise
oK,
, wherein
.
row comprises
individual sweep segment,
,
.Current set
comprise altogether
individual trunk segmentation.Algorithm according to row from
arrive
order to each sweep segment in each row
search for and distribute.Finally, we have obtained trunk segmentation set
, and each
in sweep segment all store from bottom to up,
,
,
it is sweep segment
the z coordinate of barycenter.
two, the extraction of geometric element
The estimation of circle model
After completing the identification of trunk segmentation, extract the potential geometric model of trunk segmentation.The fitting algorithm adopting is an iterative process, comprises two steps:
1) each sweep segment in trunk segmentation is justified to matching.
2) center of circle in trunk segmentation is fitted to trunk axis.
Because original point cloud cannot provide enough stem form information, the information of trunk curvature is particularly described, can cause larger error of fitting, we wish from the analyzing spot of each segmentation, to excavate not by the constant of noise, reduce the real radius of potential round model.
The sweep segment that the pitching scanning of same trees is obtained is not necessarily perpendicular to trunk axis, thus algorithm by sweep segment projection in the projection plane of trunk.Projection plane is crossed laser launching centre
and perpendicular to trunk axis.Owing to can not learning trunk axis direction in advance, so the normal of plane is initialized to the Z axis of robot local coordinate system.
Owing to hitting that the laser of trees counts, be invariant just, therefore obtain after projection scanning segmentation, we will utilize its angular span to estimate the radius of circle model, then according to subpoint, estimate the center of circle.
If projected segmentation
comprise
individual, be denoted as
.If projection radius of a circle is
,
it is variable to be estimated.
meet:
with
be respectively
the upper bound and lower bound.Their value only depends on end points
with
.But the position of these two points is uncertain due to the interference of noise.The situation when radius that Figure 3 shows that projection circle model reaches extreme value.Point
laser emission center,
it is the center of circle of projection circle.On projection plane with
for initial point is set up coordinate system.
,
polar coordinates be respectively
,
.Below they will be used as known quantity statement
with
.
As shown in Fig. 3 (a), when
get lower bound
time,
with
two projection laser beams at place will be tangent with circle model.Due to
,
be two tangent lines of circle, therefore have
With
the length that represents tangent section, but in practice
with
may strictly not equate, therefore by their average, estimate
:
Be different from
,
can not reach the upper bound
.As shown in Fig. 3 (b), when
during appearance, virtual point
with
two projection laser beams and the trunk at place are tangent.
be
previous point,
be
a rear point,
really do not comprise this two points.If their polar coordinates under robot local coordinate system are
with
.The laser range of virtual point can arbitrarily arrange, and establishes
.Due to
with
the two-dimensional scan plane at place is known, therefore scans angle
with
determine.According to the angular resolution of two-dimensional laser sensor,
with
value can be by
with
know by inference.
Especially, when
be in the plane of scanning motion first time
So, according to the relation between polar coordinate system and cartesian coordinate system, can be from
with
local pole coordinate obtain its overall Cartesian coordinates, and then obtain the polar coordinates of subpoint on projection plane
with
.
(6)
If
laser beam that tangent line is adjacent with it angle on projection plane,
the length of tangent line projection.At triangle
,
,
middle utilization straight line and the tangent character of circle, sine, the cosine law, has following system of equations:
Wherein,
,
,
known quantity,
,
,
it is unknown quantity.Three known quantities are denoted as respectively
,
,
, solving equations (7), obtains two solutions, get before radical sign minor for+solution for final, separate.
1) work as use
with
while calculating
If end points
with
in between two extreme value places separately, the probability of any point equates, therefore, gets
expectation value conduct
estimation:
Will
as known quantity, use least square method to projection arc
each point
justify matching, try to achieve the center of circle
coordinate
estimated value.
According to pertinent literature, have:
be nonlinear function, use Levenberg-Marquardt method to minimize objective function, and with point set
the x of barycenter, the conduct of y coordinate
initial value.
Suppose the original sweep segment before projection
corresponding its center of circle of round model is
, radius is
.
be defined as
the z coordinate of barycenter.
Given
after,
can pass through will
from projected coordinate system, be transformed into robot local coordinate system
obtain.Under projected coordinate system, the axis mistake of tree
and perpendicular to projection plane X-Y.Appoint and get 2 points on axis
with
, calculate they
coordinate under coordinate system
with
, can determine thus
with
straight-line equation.Due to
point be also positioned on this straight line and
known,
point exists
coordinate under coordinate system
also can determine.
the centerline fit of trunk segmentation
From sweep segment, extract after circle model trunk segmentation
in effective fitting circle center of circle will be used to matching center line
.Fitting algorithm still adopts least square method.If
the center of circle set of the effective fitting circle comprising is
, radius set is
,
.
Right
carry out SVD decomposition, select maximum singular value characteristic of correspondence vector conduct
estimated value.
After centerline fit completes, the direction vector of tree will become the new normal of projection plane.In order to improve the accuracy of matching, algorithm is carried out circle matching and these two steps of centerline fit repeatedly by iterative, until the direction vector of tree is stable, the iterations generally needing is twice.
three, the cluster of trunk segmentation and integration
The foundation of using trees to locate as robot as natural landmark, must guarantee that every one tree is only used unique and representative trunk segmentation as feature.The road sign so obtaining could be used to carry out associated one to one in different scanning.We utilize the spatial relationship of trunk segmentation axis first trunk segmentation to be carried out to cluster, then to belonging to the segmentation of same one tree, integrate.
the cluster of trunk segmentation
Pass through some steps of algorithm, some trunks, because the uncontinuity in its space has excessively been divided into many segmentations, will divide into groups to trunk segmentation here before, and each segmentation can be assigned to the trees grouping that it should subordinate.
Extraction result based on geometric model, by trunk segmentation
with
between distance definition be axis
with
between distance.
In addition, use line segment distance but not air line distance is to fall within on the extended line of line segment and cause the trunk segmentation on two far trees of space length to be divided into same cluster by mistake for fear of the public affairs point that hangs down.Real determine trunk segmentation in all dimensions in space position be its axis segment rather than central line.
Known line segment
with
the central line at place is
with
, their common vertical line is handed over
with
respectively at point
with
.If
with
drop on respectively line segment
with
upper, axial line distance equals air line distance.
(13)
Here,
mean trunk segmentation
on minimum point not higher than
on minimum point, now claim
below, by the distance definition of axis is
upper extreme point
to straight line
distance and
lower extreme point
to straight line
distance between smaller value.Fig. 4 has shown this situation.
If scanning
comprise trunk segmentation set
, will
be divided into cluster (Cluster), the trunk segmentation in each cluster is subordinated to same one tree.
Trunk segmentation
to cluster
distance definition be
arrive
in from its nearest trunk segmentation
distance.
(14)
In the distance calculating method of trunk segmentation and cluster, we have emphasized the trunk segmentation impact on tolerance result near the end points at discontinuous place.Why doing like this, is because may be divided into multistage with the trunk of one tree, but only have current trunk segmentation and neutralize with one tree the discontinuous place, space existing between its nearest trunk segmentation, is only and causes both sides to be forced to the reason of segmentation.When near discontinuous place, the distance between trunk segmentation can be more and more less.Therefore, dissimilar according to discontinuous place, space, can analyze in same cluster the upper bound of distance between each segmentation.
As shown in Figure 5, discontinuously on the space of trunk point cloud be divided into following three classes, the distance between trunk segmentation also should meet corresponding conditions.
1) fracture, as Fig. 5 (a)), the center section of tree, because analyzing spot lacks, causes the trunk of its above and below to be forced to segmentation.Causing the reason of analyzing spot disappearance may be that trunk is blocked by the object in leaf or the place ahead, may be also that trees are apart from robot is too far away or trunk is meticulous.Now, the axial line distance of trunk segmentation is almost 0.
2) bending, as Fig. 5 (b), the trunk of differing heights has lofty slope difference.The axial line distance of trunk segmentation can not surmount the higher value in both sides' radius.
3) bifurcated, branch and trunk belong to different trunk segmentations.Their junction is the change point of topological structure.Branch type is discontinuous can be divided into again two kinds of situations:
I) as shown in left side situation in Fig. 5 (c), branch appears at trunk top.Both sides are in the vertical direction relation that continues, and the radius of trunk is generally large than branch radius.Now between branch and trunk, slope is different, the situation while being similar to bending, and the axial line distance between them can not be greater than trunk radius.
Ii) as shown in right side situation in Fig. 5 (c), branch grows from a side of trunk, and both sides are certain juxtaposition relationship.Now the distance of axis often branch lower extreme point to the distance between trunk axis, this distance be necessarily less than branch and trunk radius and.
In practice, fracture, bending, the situations such as bifurcated may be in minute intersegmental appearance simultaneously of same one tree.
We may safely draw the conclusion from the above analysis, works as segmentation
be subordinated to cluster
time,
with in cluster from its nearest trunk segmentation
between distance should not be greater than their maximum radius sum.
Due in actual conditions, the radius of trunk segmentation is not to equate everywhere, and therefore, the radius that we are used as threshold value should be two sections of trunks arc radius at close discontinuous place, and should not be the average of all arc radius in trunk segmentation.Yet discontinuous place, these spaces is also not easy to determine, so we use the radius upper limit of circular arc in trunk segmentation as threshold value, to prevent that some trunk segmentations from being isolated outside the tree cluster under own.
In clustering algorithm, we are first to segmentation set
in each trunk segmentation sort.The object of doing is like this that the trunk segmentation that guarantees to be subordinated to same one tree is all to enter corresponding cluster according to order from low to high, while that is to say that top trunk section can not occur enters cluster, the root trunk section of its below not yet enters the situation of cluster, branch can not occur yet and prior to trunk, enter the situation of cluster.Set after sequence is
,
.
The current cluster set of initialization
for empty set.Algorithm traversal trunk segmentation set
, be each trunk section
cluster under finding.Check whether each cluster meets (15) formula.If cluster
satisfy condition, will
insert
.If
do not find affiliated cluster, illustrated before this search, also do not run into
trunk segmentation in affiliated tree, therefore will
as in affiliated tree one section of trunk of close root start alone a new cluster
.Consider in the situation that some are extremely special,
may find a plurality of satisfying condition
.Now, chosen distance
nearest cluster is as its subordinate cluster.
Because the trunk segmentation in this algorithm requirement cluster is according to from low paramount sequential storage, therefore trunk segmentation is inserted to cluster relevant to the close structure of tree.According to the type of tree, will
be divided into Types Below, and with indicating
represent.
2)
,
in contain more than one segmentation, and
in each segmentation between there is bifurcated relation.Each trunk segmentation is from low paramount storage until bifurcated for the first time, and all the other segmentations enter cluster in turn.The trunk segmentation of crotch is for the first time designated as
.
3)
,
in contain more than one segmentation, and
in each segmentation between there is not bifurcated relation.
Analysis based on above, will
insert
time position may exert an influence to trunk type.The step of insertion algorithm is as follows:
Here, "
" illustrate that in cluster, existing trunk segmentation is arranged in turn by order from low to high, or only have a trunk segmentation in original cluster."
be arranged in cluster best result section
on " actual algorithm statement be
the barycenter z coordinate of upper minimum scan bow is more than or equal to
the minimum z coordinate of maximum scan arc.Use
the minimum z coordinate of upper maximum scan arc and not use the highest z coordinate be to consider that sweep segment out-of-level done a kind of condition relax.
For the cluster of bifurcated type, we enter each son field of sequential storage of cluster and the segmentation of higher height by it.Therefore, no matter whether the higher height of this tree exists bifurcated again, or whether branch be divided into multistage, all no longer changes the type of cluster.Once set bifurcated, we just think that it is all the time
type cluster.And, because we are only concerned about the trunk of root, for the position of branch or higher tree section, no longer distinguished.Now, trunk is Already in cluster, and crotch is also indicated for the first time, so as long as will
as tail element, add
, do not need to change cluster type.
3) if
and
be arranged in cluster best result section
under, travel through according to the order of sequence
find minimum bifurcated trunk.If
on minimum sweep segment be not less than
on minimum sweep segment, and also not higher than
on maximum scan segmentation, and this condition met for the first time, so
it is exactly the trunk segmentation of crotch for the first time.If
and
.
the integration of trunk segmentation
Cluster after a given sequence
, the object that trunk segmentation is integrated is that all segmentations are wherein combined into the dry section of a Naoki.If
, first will
after segmentation remove, be about to branch segment and remove.Then, the cluster that each segments is greater than to 1
, the segmentation in them is all joined in turn, so we change according to their direction of sequential search from low to high, and merges.
The method that direction checks is as follows, first segmentation from cluster
start, check successively whether the direction of a rear segmentation differs too large with previous comparing,
with
between angle whether be less than threshold value
.Here, define two trunk segmentations
with
between angle between the angle that the becomes axis that is them (span is
) according to experience, allow
.
Once too greatly direction difference stop attended operation, and no longer consider subsequent segment, by they all from
with
middle deletion.Only retain the segmentation consistent with first segmentation axis direction, and they are end to end.Specific practice is, by
each sweep segment is connected to current from low to high successively
maximum scan segmentation on.
After merging, obtain
to be the dry section of a Naoki, sweep segment therebetween perhaps has fracture but can not have obvious bending point.Next, right
axis carry out matching again, and calculate association attributes.
So far, solved the problem of extracting natural landmark from forest environment, robot can use these road signs to navigate in conjunction with a certain location algorithm.Fig. 6, Fig. 7 are respectively that a secondary original 3 D laser scanning and the road sign of scale Forest Scene extracts result.Through on-the-spot operation, prove that the technology of the present invention is reliable and stable, for forest environment natural landmark extracting method provides a kind of method of science.
Claims (1)
1. a natural landmark for mobile robot extracting method in forest environment, is characterized in that the method comprising the steps of:
The 1.th pair of forest environment carries out 3 D laser scanning detection, utilizes the circular feature of trunk profile that three-dimensional point cloud is divided into sweep segment, according to space adjacency, sweep segment is organized into trunk segmentation;
The inadequate trunk segmentation of the 2.th geological information obtaining from the 1st step, extract circle model, and according to the center of circle, simulate the axis of trunk segmentation;
The 3.th by analyzing the discontinuous type in space of trunk segmentation, carry out cluster and integration, the difference according to trunk segmentation in space distribution, combines the segmentation that belongs to same trunk, remove crotch, and it is dry as road sign finally to obtain unique Naoki that represents trees simultaneously.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310568954.8A CN103575272A (en) | 2013-11-15 | 2013-11-15 | Method for extracting natural landmarks for mobile robot in forest environment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310568954.8A CN103575272A (en) | 2013-11-15 | 2013-11-15 | Method for extracting natural landmarks for mobile robot in forest environment |
Publications (1)
Publication Number | Publication Date |
---|---|
CN103575272A true CN103575272A (en) | 2014-02-12 |
Family
ID=50047565
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310568954.8A Pending CN103575272A (en) | 2013-11-15 | 2013-11-15 | Method for extracting natural landmarks for mobile robot in forest environment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103575272A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108647321A (en) * | 2018-05-11 | 2018-10-12 | 长安大学 | A kind of intelligence multi-source heterogeneous manufacture big data integrated model in workshop and semantic computation method |
CN110780669A (en) * | 2019-07-29 | 2020-02-11 | 苏州博田自动化技术有限公司 | Forest robot navigation and information acquisition method |
CN111325796A (en) * | 2020-02-28 | 2020-06-23 | 北京百度网讯科技有限公司 | Method and apparatus for determining pose of vision device |
CN111461023A (en) * | 2020-04-02 | 2020-07-28 | 山东大学 | Method for quadruped robot to automatically follow pilot based on three-dimensional laser radar |
CN114295118A (en) * | 2021-12-30 | 2022-04-08 | 杭州海康机器人技术有限公司 | Multi-robot positioning method, device and equipment |
-
2013
- 2013-11-15 CN CN201310568954.8A patent/CN103575272A/en active Pending
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108647321A (en) * | 2018-05-11 | 2018-10-12 | 长安大学 | A kind of intelligence multi-source heterogeneous manufacture big data integrated model in workshop and semantic computation method |
CN108647321B (en) * | 2018-05-11 | 2021-10-01 | 长安大学 | Tree-shaped intelligent workshop manufacturing big data integrated modeling and semantic calculation method |
CN110780669A (en) * | 2019-07-29 | 2020-02-11 | 苏州博田自动化技术有限公司 | Forest robot navigation and information acquisition method |
CN111325796A (en) * | 2020-02-28 | 2020-06-23 | 北京百度网讯科技有限公司 | Method and apparatus for determining pose of vision device |
CN111325796B (en) * | 2020-02-28 | 2023-08-18 | 北京百度网讯科技有限公司 | Method and apparatus for determining pose of vision equipment |
CN111461023A (en) * | 2020-04-02 | 2020-07-28 | 山东大学 | Method for quadruped robot to automatically follow pilot based on three-dimensional laser radar |
CN111461023B (en) * | 2020-04-02 | 2023-04-18 | 山东大学 | Method for quadruped robot to automatically follow pilot based on three-dimensional laser radar |
CN114295118A (en) * | 2021-12-30 | 2022-04-08 | 杭州海康机器人技术有限公司 | Multi-robot positioning method, device and equipment |
CN114295118B (en) * | 2021-12-30 | 2024-01-26 | 杭州海康机器人股份有限公司 | Positioning method, device and equipment for multiple robots |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Sefati et al. | Improving vehicle localization using semantic and pole-like landmarks | |
CN105702151B (en) | A kind of indoor map construction method and device | |
Yu et al. | Semiautomated extraction of street light poles from mobile LiDAR point-clouds | |
Yang et al. | Automatic registration of large-scale urban scene point clouds based on semantic feature points | |
CN109144072A (en) | A kind of intelligent robot barrier-avoiding method based on three-dimensional laser | |
Sohn et al. | Using a binary space partitioning tree for reconstructing polyhedral building models from airborne lidar data | |
Li et al. | A dual growing method for the automatic extraction of individual trees from mobile laser scanning data | |
CN106092104A (en) | The method for relocating of a kind of Indoor Robot and device | |
CN106842231A (en) | A kind of road edge identification and tracking | |
CN109270544A (en) | Mobile robot self-localization system based on shaft identification | |
CN110838115B (en) | Ancient cultural relic three-dimensional model change detection method by contour line extraction and four-dimensional surface fitting | |
Guan et al. | A novel framework to automatically fuse multiplatform LiDAR data in forest environments based on tree locations | |
CN112184736B (en) | Multi-plane extraction method based on European clustering | |
CN103575272A (en) | Method for extracting natural landmarks for mobile robot in forest environment | |
CN108564650B (en) | Lane tree target identification method based on vehicle-mounted 2D LiDAR point cloud data | |
CN109584294A (en) | A kind of road surface data reduction method and apparatus based on laser point cloud | |
Bremer et al. | Eigenvalue and graph-based object extraction from mobile laser scanning point clouds | |
CN108830933B (en) | Method, system, medium and equipment for rebuilding tower body of electric tower | |
Fan et al. | Identifying man-made objects along urban road corridors from mobile LiDAR data | |
Yue et al. | Fast 3D modeling in complex environments using a single Kinect sensor | |
CN106525000A (en) | A road marking line automatic extracting method based on laser scanning discrete point strength gradients | |
CN105844224A (en) | Point cloud fast ordering method for on-vehicle LiDAR road points | |
CN113963259A (en) | Street view ground object multi-dimensional extraction method and system based on point cloud data | |
Schlichting et al. | Vehicle localization by lidar point correlation improved by change detection | |
Hammoudi et al. | Extracting building footprints from 3D point clouds using terrestrial laser scanning at street level |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20140212 |
|
WD01 | Invention patent application deemed withdrawn after publication |