CN106054208B - The anticollision device, collision-prevention device of the recognition methods of multi-line laser radar vehicle target and automobile - Google Patents
The anticollision device, collision-prevention device of the recognition methods of multi-line laser radar vehicle target and automobile Download PDFInfo
- Publication number
- CN106054208B CN106054208B CN201610674769.0A CN201610674769A CN106054208B CN 106054208 B CN106054208 B CN 106054208B CN 201610674769 A CN201610674769 A CN 201610674769A CN 106054208 B CN106054208 B CN 106054208B
- Authority
- CN
- China
- Prior art keywords
- grid
- cloud
- obstacle
- laser radar
- detecting object
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S17/00—Systems using the reflection or reradiation of electromagnetic waves other than radio waves, e.g. lidar systems
- G01S17/88—Lidar systems specially adapted for specific applications
- G01S17/93—Lidar systems specially adapted for specific applications for anti-collision purposes
- G01S17/931—Lidar systems specially adapted for specific applications for anti-collision purposes of land vehicles
Abstract
The embodiment of the invention provides the anticollision device, collision-prevention devices of a kind of recognition methods of multi-line laser radar vehicle target and automobile, laser radar is multi-line laser radar, this method comprises: the detection data of multi-line laser radar is obtained, if detection data includes the coordinate data for doing cloud of at least two scanning slices;If being clustered based on the coordinate data for doing cloud if doing cloud by establishing grating map pair, if will do cloud be divided at least one class;The coordinate data for forming any one object, each scanning slice point cloud is fitted to line segment;It extracts the feature of each scanning slice of each object and extracts the feature of line segment;Identify whether the object is vehicle by machine learning method based on the feature extracted, if so, sending a warning.The embodiment of the present invention determines shape, the classification of object near vehicle by the detection data of multi-line laser radar, to take effective driving strategy and safety measure, avoids the injury and loss of the outer personal property of interior and vehicle.
Description
Technical field
The present embodiments relate to intelligent automobile technical fields more particularly to a kind of multi-line laser radar vehicle target to identify
The anticollision device, collision-prevention device of method and automobile.
Background technique
Vehicle in the process of moving, detects and identifies the object of vehicle periphery, to vehicle safe driving important in inhibiting.
By the detection to vehicle periphery object, driver can be reminded to change driving strategy in time, for example, selection slow down, turn to,
Emergency stop vehicle etc. reduces traffic accident to reduce the risk of vehicle drive.
The prior art is usually using a kind of anti-dress strategy of active, specifically, being visited using laser radar, camera, infrared light
Survey or ultrasonic listening etc., detect the object of vehicle periphery, for example, the barrier to vehicle front detects,
To reduce Vehicular impact barrier, and reduce unexpected traffic accident.
During the realization present invention of inventor, the discovery prior art is had following defects that
For moving traffic, the classification of distant objects can determine the driving strategy of driver, for example, such as
Fruit objects in front is trees or people, then driver should to select suitable speed and distance to slow down or avoid vehicle,
It may cause occupant's injury because acutely slowing down;If objects in front is another vehicle, the row according to opponent vehicle is needed
It sails speed, slow down with appropriate degree of selection one such as the spacings of opponent vehicle.All in all, if it is possible in vehicle and surrounding objects
Object type information is known before colliding or contacting, then driving strategy and safety can be determined according to the type information etc.
Measure for example, determines whether to need to pop up air bag and tightens safety belt, the brake etc. of which kind of degree is needed, to have
The outer personal property injury of car and loss is reduced or avoided in effect.But although the prior art is able to detect that vehicle periphery object,
But for some moving objects, it is difficult to detect its movement velocity, in addition, the shape of vehicle periphery object, classification can not be detected,
To which driver cannot be reminded rapidly to take effective driving strategy and safety measure, to avoid property personal outside interior and vehicle
Injury and loss.
Summary of the invention
The embodiment of the present invention provides a kind of automobile anti-collision method of laser radar and the anticollision device, collision-prevention device of automobile, is able to detect vapour
The shape of object near vehicle, and remind driver that effective driving strategy and safety is taken to arrange according to the shape of object
It applies.
An embodiment according to the present invention provides a kind of multi-line laser radar vehicle target recognition methods, which comprises
The detection data of the multi-line laser radar is obtained, if the detection data includes the coordinate for doing cloud of at least two scanning slices
Data;If based on the coordinate data for doing cloud, if being clustered by establishing grating map to the cloud of doing, by institute
It is divided at least one detecting object if stating and doing cloud;The coordinate of any one detecting object, each scanning slice point cloud will be formed
Data are fitted to line segment;It extracts the feature of each scanning slice of each detecting object and extracts the line segment being fitted
Feature;Identify whether the detecting object is vehicle by machine learning method based on the feature extracted, if so, sounding an alarm letter
Breath.
On the other hand, a kind of anticollision device, collision-prevention device of automobile is provided, described device include multi-line laser radar, central processing unit and
Warning device, wherein the multi-line laser radar is mounted at least one direction of the automobile, for detecting its direction
Object, and detection data is exported, if the detection data includes the coordinate data for doing cloud of at least two scanning slices;It is described
Central processing unit, if the coordinate data for doing cloud of at least two scanning slice is received, and if based on the seat for doing cloud
Data are marked, if clustering by establishing grating map to the cloud of doing, if the cloud of doing is divided at least one spy
Object is surveyed, the coordinate data for forming any one detecting object, each scanning slice point cloud is fitted to line segment, extracts each spy
It surveys the feature of each scanning slice of object and extracts the feature for the line segment being fitted, passed through based on the feature extracted
Machine learning method identifies whether the detecting object is vehicle, if so, issuing control instruction to the alarm device;The alarm
Device, for sending a warning when receiving the control instruction that the central processing unit issues.
The anticollision device, collision-prevention device of a kind of recognition methods of multi-line laser radar vehicle target and automobile of the embodiment of the present invention, by more
The detection data of line laser radar, determines shape, the classification of object near vehicle, and is mentioned according to the shape etc. of object near vehicle
Awake driver takes effective driving strategy and safety measure, to avoid the injury and loss of property personal outside interior and vehicle.
The above description is only an overview of the technical scheme of the present invention, in order to better understand the technical means of the present invention,
And it can be implemented in accordance with the contents of the specification, and in order to allow above and other objects of the present invention, feature and advantage can
It is clearer and more comprehensible, the followings are specific embodiments of the present invention.
Detailed description of the invention
Fig. 1 shows a kind of one of the structural schematic diagram of automobile provided in an embodiment of the present invention;
Fig. 2 shows a kind of second structural representations of automobile provided in an embodiment of the present invention;
Fig. 3 shows a kind of flowage structure of multi-line laser radar vehicle target recognition methods provided in an embodiment of the present invention
One of schematic diagram;
Fig. 4 shows a kind of flowage structure of multi-line laser radar vehicle target recognition methods provided in an embodiment of the present invention
The two of schematic diagram;
Fig. 5 shows a kind of cluster signal of multi-line laser radar vehicle target recognition methods provided in an embodiment of the present invention
One of figure;
Fig. 6 shows a kind of cluster signal of multi-line laser radar vehicle target recognition methods provided in an embodiment of the present invention
The two of figure.
Specific embodiment
Exemplary embodiments of the present disclosure are described in more detail below with reference to accompanying drawings.Although showing the disclosure in attached drawing
Exemplary embodiment, it being understood, however, that may be realized in various forms the disclosure without should be by embodiments set forth here
It is limited.On the contrary, these embodiments are provided to facilitate a more thoroughly understanding of the present invention, and can be by the scope of the present disclosure
It is fully disclosed to those skilled in the art.
Laser radar is also laser scanner (Laser Scanner), laser range finder (Laser Range Finder),
It is designed based on the principle for calculating flight time (Time of Flight), may include that Laser emission end and laser connect
Receiving end.Laser emission end can issue beam of laser pulse or laser beam, when laser pulse or laser beam expose to a certain object
Meeting reflection echo signal when body, laser pick-off end receives reflection echo information, to accurately measure laser pulse or laser beam
The time difference for issuing and returning, and with this calculate the object and laser radar (or equipment or system where laser radar) it
Between distance.The measurement data of laser radar is that each is measured point at a distance of the distance at Laser emission end and the pole of angle information
Coordinate data.
If there is a laser radar multi-stripe laser scan line to be scanned measurement simultaneously, such laser radar is referred to as
Multi-line laser radar or multi-layer laser radar.Multi-line laser radar has the scanning field of view on both horizontally and vertically, in portion
Shunt excitation light beam scanning surface remains to the object on the periphery of system where detecting multi-line laser radar when being blocked.Further, since tool
Standby vertical direction visual field, multi-line laser radar can obtain the information in the short transverse of measured target.
Embodiment one
The embodiment of the present invention provides a kind of anticollision device, collision-prevention device of automobile comprising multi-line laser radar.It can be by multi-thread laser
Radar is installed on automobile, and makes the scanning direction of multi-line laser radar towards desired search coverage direction, for example, it is desirable to search coverage
Direction is the front of automobile, then multi-line laser radar can be mounted on to the headstock part of automobile.As depicted in figs. 1 and 2, may be used
Multi-line laser radar is mounted in front of the roof of automobile, and adjusts the front for being allowed to probe vehicle, or by multi-thread laser thunder
Up to being mounted near the front bumper of automobile, and adjust the front for being allowed to probe vehicle, illustratively, multi-line laser radar
Angle theta between scanning direction and ground is 0.
It will be appreciated by persons skilled in the art that multi-line laser radar may be mounted at least one direction of automobile
On, for example, at least one of the front of automobile, rear, left and right side, are mounted on multiple directions, car steering can be improved
Intelligence.Due to consideration that driver on the run, is primarily upon the object of vehicle front, further, it is contemplated that cost etc.
Multi-line laser radar is preferably mounted on the front of automobile by problem.Multi-line laser radar is installed on automobile, for detecting its institute
Object in direction, and export detection data, wherein if detection data includes the number of coordinates for doing cloud of at least two scanning slices
According to.
The anticollision device, collision-prevention device of the automobile of the embodiment of the present invention can also include central processing unit and alarm, and central processing unit connects
If receiving the coordinate data for doing cloud of at least two scanning slices, and if based on the coordinate data for doing cloud, by with establishing grid
If figure pair is done cloud and is clustered, if will do cloud be divided at least one detecting object, wherein any one will be formed and detect object
Body, each scanning slice point cloud coordinate data be fitted to line segment, extract each scanning slice of each object feature and
Whether the feature for extracting the line segment being fitted, identify the detecting object by machine learning method based on the feature extracted
It is vehicle, if so, issuing control instruction to the alarm device.Alarm device refers in the control for receiving central processing unit sending
When enabling, send a warning.Driver can take effective driving strategy according to warning information, and/or make central controller
(or central processing unit) control corresponding mechanism takes safety measures, to improve the safety of car steering.
Embodiment two
Fig. 3 shows a kind of flowage structure of multi-line laser radar vehicle target recognition methods provided in an embodiment of the present invention
One of schematic diagram.
As shown in figure 3, a kind of multi-line laser radar vehicle target recognition methods includes the following steps:
Step 200, the detection data of multi-line laser radar is obtained, wherein detection data includes at least two scanning slices
If doing the coordinate data of cloud.
In this step, in the laser beam flying to any object of multi-line laser radar, it can detect series of points
The polar data (m, r, φ) of cloud, the wherein number of plies of m expression laser radar scanning layer, such as m=1,2,3 or 4, it is possible to understand that
Be m value it is without being limited thereto, can be select and set according to the actual situation.Illustratively, it can be converted by coordinate
The polar data of arbitrary point cloud is converted to the three-dimensional rectangular coordinate (x, y, z) of the cloud by formula.Those skilled in the art can
With understanding, polar coordinates, which are converted into three-dimensional rectangular coordinate, can combine existing Formula of Coordinate System Transformation based on known parameters,
This is repeated no more.
Step 201, it if based on the coordinate data for doing cloud, is clustered if doing cloud by establishing grating map pair, with
If will do cloud be divided at least one detecting object.
Detecting object as mentioned in the embodiments of the present invention is any one detecting object that multi-line laser radar scans.It is multi-thread
For laser radar scanning to arbitrary objects, the data detected include the coordinate data of several clouds, for example, if doing the three of cloud
It ties up rectangular co-ordinate (x, y, z).If without clustering processing, be difficult distinguish detect object number and/or object classification,
Shape etc..
An embodiment according to the present invention, if directly being clustered using the three-dimensional rectangular coordinate for doing cloud detected.
Alternatively, since multi-line laser radar may scan at least one object, and the three of any one object is formed
The data for tieing up the point cloud of rectangular co-ordinate are larger, can be huge because of data volume if the three-dimensional rectangular coordinate of point cloud is directly clustered
And computation rate and low efficiency, it is difficult to ensure that the real-time of detection.Therefore, another embodiment according to the present invention, by establishing grid
If lattice map classifies to the cloud of doing detected.Illustratively, M × N grating map is established, if the three-dimensional straight that cloud will be done
Angular coordinate (x, y, z) projects in grating map plane (x, y).Indicate to fall into the point cloud in grid with the grid of a plane,
Illustratively, the elevation information for dropping into each point cloud in grid is recorded.It illustratively, will include the coordinate data for having a cloud
Grid is known as obstacle grid.
Step 202, the coordinate data for forming any one detecting object, each scanning slice point cloud is fitted to line segment.
If multi-line laser radar includes m laser beam, the arbitrary objects scanned have m layers of scanning slice, each scanning
If layer all includes to do cloud.If doing cloud for each scanning slice carries out data fitting, available at least one line segment
Cluster.
Step 203, it extracts the feature of each scanning slice of each detecting object and extracts the line segment being fitted
Feature.
In one example, when the point-cloud fitting of each scanning slice in step 202 is a line segment, in this step
The feature of extraction includes: the number for forming all the points cloud of the object, forms total scanning where all the points cloud of the object
It counts layer by layer, form all the points cloud of the object to the distance average of multi-line laser radar, the coordinate of the point cloud of each scanning slice
Length, the centre-height of the line segment of the coordinate data fitting of the point cloud of each scanning slice, each scanning of the line segment of data fitting
The slope of the line segment of the coordinate data fitting of the point cloud of layer.
Illustratively, if the starting point of line segment and the coordinate of terminal that fitting obtains are respectively as follows: (xp1,yp1,zp1) and (xq1,
yq1,zq1), then, the feature that can extract line segment in step 203 includes:
(1) length value of the line segment, such as can be calculated by following formula:
It is understood that if there is m scanning slice, and when the point-cloud fitting of each scanning slice is a line segment, then need
Extract the length value of m line segment;
(2) the centre-height value of line segment, such as can be calculated by following formula:If there is m
A scanning slice, and the point-cloud fitting of each scanning slice be a line segment when, then need to extract the centre-height value of m line segment;
(3) slope of line segment, wherein
If yp1=yq1, then slope k 1 is not present, and the slope k 1 of the line segment is labeled as k1=0, if yp1≠yq1, then slope k 1
For k1=(xp1-xq1)/(yp1-yq1).It is appreciated that if there is m scanning slice, and the point-cloud fitting of each scanning slice is one
When line segment, then need to extract the slope of m line segment.
By obtaining length value, centre-height value and the slope of the line segment of all scanning slices, can predict scanned
Length, height and shape of body etc. improve the accuracy of identification.
It in addition to the feature of line segment, as in the foregoing embodiment, also to extract: form of all the points cloud of the detecting object
Number forms total scanning slice number of plies where all the points cloud of the detecting object, forms all the points cloud of the detecting object to more
The distance average of line laser radar.Where it is assumed that the number for forming all the points cloud of the detecting object is n, and known each
The coordinate of cloud is put, then the distance average by the available all the points cloud for forming the detecting object to multi-line laser radar, example
Such as, it can be calculated by following formula:
In another example, when the point-cloud fitting of each scanning slice in step 202 is two lines section, in this step
The feature of middle extraction includes: the number for forming all the points cloud of the detecting object, all the points cloud place for forming the detecting object
Total scanning slice number of plies, the length of every line segment, the centre-height of two lines section average value, form the institute of the detecting object
Have a cloud to the distance average of multi-line laser radar, the lenth ratio of two lines section, two lines section length product, two lines
Angle of the section in floor projection.
Illustratively, if two lines section is respectively line segment 1 and line segment 2, and the starting point of line segment 1 and the coordinate difference of terminal are set
For (xp1,yp1,zp1) and (xq1,yq1,zq1), then the centre coordinate of line segment 1 is
And setting the starting point of line segment 2 and the coordinate of terminal is respectively (xp2,yp2,zp2) and (xq2,yq2,zq2), then the center of line segment 2 is sat
It is designated asIt is understood that multi-line laser radar has m scanning
Layer, so that the starting point of line segment 1, terminal and centre coordinate have m a and the starting point of line segment 2, terminal and centre coordinate are equal
There are m.Based on this, the feature of the line segment of extraction includes following several:
(1) length value of line segment 1, such as can be calculated by following formula:
And
The length value of line segment 2, such as can be calculated by following formula:
Wherein, the length value of line segment 1 and line segment 2 has m respectively;
(2) the centre-height mean value of line segment 1 and line segment 2, such as can be calculated by following formula:Wherein, which has m;
(3) ratio of the length of line segment 1 and line segment 2, such as can be calculated by following formula:Wherein, which has m;
(4) product of the length value of line segment 1 and line segment 2, such as can be calculated by following formula:Wherein, this multiplies
Product has m;
(5) line segment 1 and line segment 2 such as may include following several situations in the angle of floor projection: if yp1=yq1, then
Slope k 1 is not present, and the slope k 1 of the line segment is labeled as k1=0;If yp1≠yq1, then slope k 1 is k1=(xp1-xq1)/(yp1-
yq1);If yp2=yq2, then slope k 2 is not present, and the slope k 2 of the line segment is labeled as k2=0;If yp1≠yq1, then slope k 1 is k2
=(xp2-xq2)/(yp2-yq2).To sum up, two lines section the angle of floor projection can be with are as follows:Its
In, the number of angle is m.
It in addition to the feature of line segment, as in the foregoing embodiment, also to extract: form of all the points cloud of the detecting object
Number forms total scanning slice number of plies where all the points cloud of the detecting object, forms all the points cloud of the detecting object to more
The distance average of line laser radar.If forming the number of all the points cloud of the detecting object as n, and the seat of known each cloud
Mark, then can be calculated by following formula all the points cloud to multi-line laser radar distance average:
For some irregular objects, for example, in irregular shape and/or placement position is irregular (for example, relative to more
Line laser radar automobile for, the placement location of detecting object is tiltedly to put rather than face automobile is placed) object, often
The point cloud of a scanning slice may be fitted to two lines section, by obtain two lines section as previous embodiment feature, and combine
Point cloud data, scanning layer data etc., can more accurately identify shape, the type of object, improve vehicle security drive performance.
Illustratively, in step 203, the feature extracted can be expressed as vector.
Step 204, identify whether the detecting object is vehicle by machine learning method based on the feature extracted, if
It is to send a warning.
In this step, a large amount of vehicle images can be collected, and these images are labeled, based on these a large amount of warps
The image of mark obtains the neural network (classifier) of vehicle for identification by machine learning method training.Illustratively, refreshing
It can include but is not limited to radial basis function (RBF-Radial Basis Function) neural network, BP (Back through network
Propagation) at least one such as neural network.
An embodiment according to the present invention can indicate the form of the feature extracted in step 203 eigenmatrix.Show
Example property, will be fitted to when a line segment extracted feature as matrix element formation fisrt feature matrix Y1,
Wherein, the feature from line segment has 3m altogether, from object features 3 (for example, all the points cloud of composition object
Number, the number of plies of scanning slice and all the points cloud to multi-line laser radar distance value), therefore, this feature matrix comes to 3
+ 3m row.
And extracted feature will be fitted to when two lines section as matrix element and form second characteristic matrix Y2,
Based on previous embodiment, 6m are had altogether from line segment feature, whole object feature 3 are come from, so this feature square
Battle array comes to 3+6m row.
Merge fisrt feature matrix Y1 and second characteristic matrix Y2, to the fisrt feature matrix Y1 and second feature after merging
Matrix Y2 is normalized, and obtains merging obtained eigenmatrix Y, such as:
Further, eigenmatrix Y normalized merging obtained.For example, the maximum element of numerical value in Y is set
Be 1, the smallest element of numerical value is set as 0, between the maximum element between numerical value minimum of numerical value it is proportional be set as 0 and 1 it
Between numerical value, to obtain third feature matrix Y4.
After obtaining third feature matrix Y4, in step 204, third feature matrix Y4 can be inputted and pass through machine learning side
In the classifier that method training obtains, to identify whether the detecting object is vehicle.It is understood that returning to eigenmatrix
After one change processing, unified data representation can be obtained, reduces operation pressure, improves arithmetic speed.
An embodiment according to the present invention, in step 204 if it is determined that language can be passed through when detecting that detecting object is vehicle
Sound, video show etc. that at least one mode reminds current driver, the modes such as can also shows by language, video to work as
Preceding driver provides safe driving strategy instruction, alternatively, can be with the safety measure in automobile, such as air bag, system
The connections such as dynamic device control air bag, brake etc. by central controller etc. and take corresponding safety measure, so that protection is worked as
The safety of preceding vapour occupant etc..
Shape, the class of object near vehicle are determined by the detection data of multi-line laser radar using previous embodiment
Not, and according to shape of object near vehicle etc. driver is reminded to take effective driving strategy and safety measure, to avoid car
And the injury and loss of the outer personal property of vehicle.
Embodiment three
Fig. 4 shows a kind of flowage structure of multi-line laser radar vehicle target recognition methods provided in an embodiment of the present invention
The two of schematic diagram.
As shown in figure 4, a kind of multi-line laser radar vehicle target recognition methods includes the following steps:
Step 300, the detection data of multi-line laser radar is obtained, wherein detection data includes at least two scanning slices
If doing the coordinate data of cloud.
Step 300 realizes that process and principle are identical with step 200 in embodiment two, for sake of simplicity, details are not described herein.
Step 301, if projecting the coordinate data for doing cloud of at least two scanning slices to raster plane to establish grid
Map, wherein raster plane includes several blank grids.
Illustratively, M × N grating map is established, if with projecting to grid by the three-dimensional rectangular coordinate (x, y, z) for doing cloud
On plan (x, y).Indicate to fall into the point cloud in grid with the grid of a plane.Illustratively, it records and drops into grid
The elevation information of middle each point cloud, usual multi-line laser radar detect that object has certain height relative to ground, pass through foundation
Grating map, and the elevation information for forming each of any one object cloud is recorded, it can count each in each obstacle grid
The height of point cloud, finds obstacle grid highest point H_max and minimum point H_min, and calculate the maximum phase of each obstacle grid
To height Δ H=H_max-H_min.
Step 302, it determines in grating map including there is the grid of the coordinate data of a cloud for obstacle grid.
Step 303, the noise grid in obstacle grid is deleted, wherein noise grid is not inconsistent for the coordinate data of its cloud
Close the obstacle grid of preset condition.
Multi-line laser radar during the scanning process, it is possible that some erroneous reflections point (for example, acnode) and branch,
Hanging point caused by winged insect, in order to avoid these reflection points, hanging point etc. impact final classification, identification etc., in this reality
It applies and proposes the method for filtering out these noise spots in example, in other words the method for erased noise grid.
An embodiment according to the present invention, statistics determine that the height coordinate in all the points cloud in each obstacle grid is highest
Point cloud and the minimum point cloud of height coordinate determine corresponding barrier based on the minimum point cloud of highest cloud of height coordinate and height coordinate
Hinder the elevation information of grid, further, the grid in grating map is filtered according to the elevation information of each obstacle grid
Wave processing, with erased noise grid.Illustratively, judge whether the elevation information of each obstacle grid is greater than preset height, with
And in the obstacle grid whether include at least two non-same scanning slices point, if elevation information be greater than preset height and the barrier
Hinder the point of the non-same scanning slice in grid less than two, it is determined that the obstacle grid is noise grid, deletes the noise grid.
For example, each obstacle grid maximum relative altitude Δ H is counted, if being less than two non-same layers in the grid Δ H > 2m and grid
The obstacle grid is then considered as hanging point (or noise grid) and is filtered out by radar data point.
Another embodiment according to the present invention judges whether any obstacle grid has other within a preset range in grating map
Obstacle grid, and according to the coordinate data of the obstacle grid midpoint cloud calculate the obstacle grid to multi-line laser radar distance
Information, when, without other obstacle grids and when range information is greater than pre-determined distance, determining the obstacle grid in the preset range
Lattice are noise grid, delete the noise grid.For example, within the scope of multi-line laser radar 20m, single obstacle if it exists
Grid, around 5x5 grid within the scope of without other obstacle grids, then the grid can be regarded as to acnode (or noise grid)
It is filtered out.
Step 304, all grids in the grating map are traversed, when encountering any one described obstacle grid, determining should
Obstacle grid is benchmark grid.
Step 305, whether left side, upper left side, positive upside and the grid of upper right side four for judging reference grid are obstacle grid
Lattice.If so, step 306 is executed, conversely, executing step 307.
Step 306, the left side of reference grid, upper left side, positive upside and the grid of upper right side four and reference grid are classified as
Same detecting object, and the identical identifier of grid tag that same detecting object will be classified as.
Step 307, reference grid is labeled as new detecting object.
Due to multi-line laser radar obtain point cloud data and grating map data be all it is discrete, in order to detect
One object (such as vehicle), discrete point inter-related in every frame data is usually flocked together carry out clustering.
Clustering can cut apart several objects, that is, tell object mutually indepedent, unrelated in scanning range
Body.Abovementioned steps 304 are to the clustering method that step 307 is that the embodiment of the present invention improves, illustratively, using eight neighborhood
Zone marker method carries out clustering to the point cloud in grating map.
Clustering method provided in an embodiment of the present invention and process are described in detail below in conjunction with attached drawing.Fig. 5 is shown
A kind of one of cluster schematic diagram of multi-line laser radar vehicle target recognition methods provided in an embodiment of the present invention, Fig. 6 is shown
The two of a kind of cluster schematic diagram of multi-line laser radar vehicle target recognition methods provided in an embodiment of the present invention.
Illustratively, grating map can be traversed from row, column both direction, if encountering obstacle grid B, disturbance in judgement grid
Whether the left side (L) of lattice B, upper left (UL), top (U) and four grids of upper right (UR) are obstacle grid, if so, they are closed
And to together and assign it is identical label number be used as a same cluster (detecting object);If not being then labeled as a new obstacle grid
The lattice cluster new as one (detecting object).Illustratively, as shown in Figure 5, step is realized are as follows:
(1) all grating maps are traversed from left to right, from top to bottom, are not done and are located if not being obstacle grid if the grid taken out
Reason, directly takes next grid to be judged.If encountering obstacle grid B, its four neighborhood grids L, UL, U, UR are sentenced
It is disconnected, determine whether tetra- grids of L, UL, U, UR are obstacle grid.
(2) if L is obstacle grid, the label number of L is assigned to B, then detects UR.If UR is not obstacle grid, tie
Processing of the beam to B;If UR is obstacle grid, merge the label number of B, L, UR three, currently processed end.
(3) if L is not obstacle grid, UL is detected.If UL is obstacle grid, UL label number is assigned to B, is then detected
UR.If UR is not obstacle grid, terminate the processing to B;If UR is obstacle grid, merge the label number of B, UL, UR three, when
Pre-treatment terminates.
(4) if L and UL are not obstacle grids, U is detected.If U is obstacle grid, the label number of U is assigned to B, when
Pre-treatment terminates.
(5) if L, UL and U are not obstacle grids, UR is detected.The label number of UR is assigned to if UR is obstacle grid
B, currently processed end.
(6) if L, UL, U and RU are not obstacle grids, a new label number, currently processed end are created for B.
As shown in fig. 6, after carrying out eight neighborhood zone marker to grating map, the barrier of same object on grating map
Grid is hindered to have unique identical label number, i.e. the identical grid of label number indicates the same cluster, for example, figure acceptance of the bid mark
For 1 obstacle grid from an object, for the obstacle grid that figure acceptance of the bid mark is 2 from another object, figure acceptance of the bid mark is 3
Obstacle grid from other objects.In this way, each object is separated after cluster operation completion.
The data distribution for the point cloud that multi-line laser radar scanning object obtains is complicated, if do not carried out to these point cloud datas
Cluster, and be directly based upon a cloud and identified, will increase the complexity of calculating and calculate error rate, reduce recognition efficiency and identification
Accuracy rate.And previous embodiment is utilized, by clustering processing, by the point cloud data classification annotation of each object, in identification process
In, more targetedly and the data precision is high, complexity is low, reduces the complexity of calculating and improve accuracy rate, Jin Erti
High recognition efficiency and accuracy rate.
An embodiment according to the present invention, since the size of vehicle is all larger, size is lesser after can excluding cluster
Object, illustratively, statistics form the number of the obstacle grid of each object;If the number for forming the obstacle grid of the object is small
In predetermined number, then the coordinate data for the point cloud for forming the object is deleted.For example, the obstacle grid of jobbie after cluster
Lattice number is less than or equal to 3 obstacle grids, then directly deletes the point cloud data of the object.It is understood that deleting
The object of object to be identified feature is not met, operand can be reduced, reduces invalid operation.
Step 308, the coordinate data for forming any one detecting object, each scanning slice point cloud is fitted to line segment.
The step is consistent with the realization process of the step 202 in embodiment two and principle, for sake of simplicity, no longer superfluous herein
It states.
Step 309, it extracts the feature of each scanning slice of each detecting object and extracts the line segment being fitted
Feature.
The step is consistent with the realization process of the step 203 in embodiment two and principle, for sake of simplicity, no longer superfluous herein
It states.
Step 310, identify whether the detecting object is vehicle by machine learning method based on the feature extracted, if
It is to send a warning.
The step is consistent with the realization process of the step 204 in embodiment two and principle, for sake of simplicity, no longer superfluous herein
It states.
Shape, the class of object near vehicle are determined by the detection data of multi-line laser radar using previous embodiment
Not, and according to the shape etc. of object near vehicle effective driving strategy and safety measure are taken, to avoid interior and vehicle stranger
The injury and loss of body property.
Those of ordinary skill in the art will appreciate that: realize that all or part of the steps of above-mentioned each method embodiment can lead to
The relevant hardware of program instruction is crossed to complete.Program above-mentioned can be stored in a computer readable storage medium.The journey
When being executed, execution includes the steps that above-mentioned each method embodiment to sequence;And storage medium above-mentioned include: ROM, RAM, magnetic disk or
The various media that can store program code such as person's CD.
Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention., rather than its limitations;To the greatest extent
Pipe present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: its according to
So be possible to modify the technical solutions described in the foregoing embodiments, or to some or all of the technical features into
Row equivalent replacement;And these are modified or replaceed, various embodiments of the present invention technology that it does not separate the essence of the corresponding technical solution
The range of scheme.
Claims (4)
1. a kind of multi-line laser radar vehicle target recognition methods, which is characterized in that the described method includes:
The detection data of the multi-line laser radar is obtained, if the detection data includes that at least two scanning slices do cloud
Coordinate data;
If based on the coordinate data for doing cloud, if being clustered by establishing grating map to the cloud of doing, by institute
It is divided at least one detecting object if stating and doing cloud;
The coordinate data for forming any one detecting object, each scanning slice point cloud is fitted to line segment;
It extracts the feature of each scanning slice of each detecting object and extracts the feature for the line segment being fitted;
Identify whether the detecting object is vehicle by machine learning method based on the feature extracted, if so, sounding an alarm letter
Breath,
Establishing grating map includes:
If projecting the coordinate data for doing cloud of at least two scanning slice to raster plane to establish grating map,
In, the raster plane includes several blank grids;And
It determines including there is the grid of the coordinate data of a cloud for obstacle grid in the grating map,
The method also includes:
Delete the noise grid in the obstacle grid, wherein the noise grid does not meet pre- for the coordinate data of its cloud
If the obstacle grid of condition,
Statistics determines that highest cloud of height coordinate in all the points cloud in each obstacle grid and height coordinate are minimum
Point cloud;
The elevation information of corresponding obstacle grid is determined based on the minimum point cloud of highest cloud of the height coordinate and height coordinate;
And
The noise grid deleted in the obstacle grid, comprising:
The grid in the grating map is filtered according to the elevation information of each obstacle grid, is made an uproar with deleting
Acoustic grating lattice,
The elevation information according to each obstacle grid is filtered the grid in the grating map, to delete
Except noise grid, comprising:
Whether judge whether the elevation information of each obstacle grid is greater than in preset height and the obstacle grid includes extremely
The point of few two non-same scanning slices;
If the elevation information is greater than preset height and the point of the non-same scanning slice in the obstacle grid is less than two, really
The fixed obstacle grid is noise grid;
The noise grid is deleted,
The coordinate data of point cloud that form any one object, each scanning slice is at most fitted to two lines section,
When being fitted to a line segment, extracted feature includes: the number for forming all the points cloud of the detecting object, forms and be somebody's turn to do
Total scanning slice number of plies where all the points cloud of detecting object forms all the points cloud of the detecting object to the multi-thread laser
The distance average of radar, the length of line segment of coordinate data fitting of the point cloud of each scanning slice, each scanning slice point cloud
The centre-height of line segment of coordinate data fitting, each scanning slice point cloud coordinate data fitting line segment slope;
When being fitted to two lines section, extracted feature includes: the number for forming all the points cloud of the detecting object, forms and be somebody's turn to do
Total scanning slice number of plies where all the points cloud of detecting object, length, the centre-height of the two lines section of every line segment
Average value, form all the points cloud of the detecting object to the distance average of the multi-line laser radar, the two lines section
Lenth ratio, the length product of the two lines section, the two lines section floor projection angle;
The method also includes:
Extracted feature will be fitted to when a line segment as matrix element and form fisrt feature matrix, and two will be fitted to
Extracted feature forms second characteristic matrix as matrix element when line segment;
Merge the fisrt feature matrix and second characteristic matrix;
To after merging the fisrt feature matrix and second characteristic matrix be normalized, obtain third feature matrix;
The third feature Input matrix is trained in obtained classifier by machine learning method, to identify the detecting object
It whether is vehicle.
2. the method according to claim 1, wherein the method also includes:
Judge whether any obstacle grid there are other obstacle grids within a preset range in the grating map;
And
According to the coordinate data of the obstacle grid midpoint cloud calculate the obstacle grid to the multi-line laser radar range information;
And
The noise grid deleted in the obstacle grid, comprising:
When, without other obstacle grids and when the range information is greater than pre-determined distance, determining the barrier in the preset range
Hindering grid is noise grid;
Delete the noise grid.
If 3. the method according to claim 1, wherein it is described by establish grating map to it is described do cloud into
Row clusters
Traverse all grids in the grating map;
When encountering any one described obstacle grid, determine that the obstacle grid is benchmark grid, and judge a left side for the reference grid
Whether side, upper left side, positive upside and the grid of upper right side four are obstacle grid:
If so, the left side of the reference grid, upper left side, positive upside and the grid of upper right side four are returned with the reference grid
For same detecting object, and
By the mutually similar identifier of the grid tag for being classified as same detecting object;
Conversely, if it is not, the reference grid is then labeled as new detecting object.
4. the method according to claim 1, wherein the method also includes:
Statistics forms the number of the obstacle grid of each detecting object;
If the number for forming the obstacle grid of the detecting object is less than predetermined number, the point cloud of the detecting object will be formed
Coordinate data delete.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610674769.0A CN106054208B (en) | 2016-08-16 | 2016-08-16 | The anticollision device, collision-prevention device of the recognition methods of multi-line laser radar vehicle target and automobile |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610674769.0A CN106054208B (en) | 2016-08-16 | 2016-08-16 | The anticollision device, collision-prevention device of the recognition methods of multi-line laser radar vehicle target and automobile |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106054208A CN106054208A (en) | 2016-10-26 |
CN106054208B true CN106054208B (en) | 2019-02-19 |
Family
ID=57481583
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610674769.0A Active CN106054208B (en) | 2016-08-16 | 2016-08-16 | The anticollision device, collision-prevention device of the recognition methods of multi-line laser radar vehicle target and automobile |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106054208B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108345007A (en) * | 2017-01-23 | 2018-07-31 | 郑州宇通客车股份有限公司 | A kind of obstacle recognition method and device |
Families Citing this family (23)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106597472A (en) * | 2016-12-21 | 2017-04-26 | 深圳市镭神智能系统有限公司 | Intelligent vehicle collision avoidance system and method based on laser radar |
CN108268483A (en) * | 2016-12-30 | 2018-07-10 | 乐视汽车(北京)有限公司 | The method for the grid map that generation controls for unmanned vehicle navigation |
US10838067B2 (en) * | 2017-01-17 | 2020-11-17 | Aptiv Technologies Limited | Object detection system |
US10401866B2 (en) * | 2017-05-03 | 2019-09-03 | GM Global Technology Operations LLC | Methods and systems for lidar point cloud anomalies |
CN106970395B (en) * | 2017-05-08 | 2019-12-03 | 奇瑞汽车股份有限公司 | The method and apparatus for determining Obstacle Position |
US10416679B2 (en) * | 2017-06-27 | 2019-09-17 | GM Global Technology Operations LLC | Method and apparatus for object surface estimation using reflections delay spread |
CN110019609B (en) * | 2017-11-20 | 2021-05-25 | 北京京东乾石科技有限公司 | Map updating method, apparatus and computer readable storage medium |
CN107845095B (en) * | 2017-11-20 | 2020-10-02 | 维坤智能科技(上海)有限公司 | Moving object real-time detection method based on three-dimensional laser point cloud |
CN109840448A (en) * | 2017-11-24 | 2019-06-04 | 百度在线网络技术(北京)有限公司 | Information output method and device for automatic driving vehicle |
CN108710367B (en) * | 2018-05-23 | 2022-02-11 | 广州视源电子科技股份有限公司 | Laser data identification method and device, robot and storage medium |
CN109633676A (en) * | 2018-11-22 | 2019-04-16 | 浙江中车电车有限公司 | A kind of method and system based on the laser radar obstruction detection direction of motion |
CN116588083A (en) * | 2018-12-26 | 2023-08-15 | 北京图森智途科技有限公司 | Parking control method, device and system |
CN110441789B (en) * | 2019-06-27 | 2022-12-30 | 南京莱斯信息技术股份有限公司 | Non-uniform thread laser radar target identification method |
JP7259685B2 (en) * | 2019-09-30 | 2023-04-18 | トヨタ自動車株式会社 | Driving control device for automatic driving vehicle, stop target, driving control system |
CN110634297B (en) * | 2019-10-08 | 2020-08-07 | 交通运输部公路科学研究所 | Signal lamp state identification and passing control system based on vehicle-mounted laser radar |
CN110618420A (en) * | 2019-10-15 | 2019-12-27 | 广州小鹏汽车科技有限公司 | Ultrasonic data processing method and system, vehicle and storage medium |
CN111208530B (en) * | 2020-01-15 | 2022-06-17 | 北京四维图新科技股份有限公司 | Positioning layer generation method and device, high-precision map and high-precision map equipment |
CN111337898B (en) * | 2020-02-19 | 2022-10-14 | 北京百度网讯科技有限公司 | Laser point cloud processing method, device, equipment and storage medium |
CN111308499A (en) * | 2020-03-09 | 2020-06-19 | 中振同辂(江苏)机器人有限公司 | Obstacle detection method based on multi-line laser radar |
CN111665524A (en) * | 2020-04-29 | 2020-09-15 | 武汉光庭科技有限公司 | Method and system for ground rejection by utilizing multi-line laser radar |
CN113970758A (en) * | 2020-07-22 | 2022-01-25 | 上海商汤临港智能科技有限公司 | Point cloud data processing method and device |
CN112558045B (en) * | 2020-12-07 | 2024-03-15 | 福建(泉州)哈工大工程技术研究院 | Offline acceptance method for multi-line laser radar function of automatic driving equipment |
CN116380108B (en) * | 2023-06-02 | 2023-08-11 | 山东科技大学 | Track planning method and device based on laser radar |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103065151A (en) * | 2012-11-04 | 2013-04-24 | 北京工业大学 | Vehicle identification method based on depth information |
CN105404844A (en) * | 2014-09-12 | 2016-03-16 | 广州汽车集团股份有限公司 | Road boundary detection method based on multi-line laser radar |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
GB201318185D0 (en) * | 2013-10-14 | 2013-11-27 | Guidance Navigation Ltd | Tracking device |
CN104408443B (en) * | 2014-12-15 | 2017-07-18 | 长春理工大学 | The method for recognizing road surface types and device based on laser radar of multisensor auxiliary |
CN104597453B (en) * | 2015-01-27 | 2017-06-20 | 长春理工大学 | The vehicle safe driving method for detecting area and device of inertial measuring unit correction |
-
2016
- 2016-08-16 CN CN201610674769.0A patent/CN106054208B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103065151A (en) * | 2012-11-04 | 2013-04-24 | 北京工业大学 | Vehicle identification method based on depth information |
CN105404844A (en) * | 2014-09-12 | 2016-03-16 | 广州汽车集团股份有限公司 | Road boundary detection method based on multi-line laser radar |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108345007A (en) * | 2017-01-23 | 2018-07-31 | 郑州宇通客车股份有限公司 | A kind of obstacle recognition method and device |
CN108345007B (en) * | 2017-01-23 | 2020-10-20 | 郑州宇通客车股份有限公司 | Obstacle identification method and device |
Also Published As
Publication number | Publication date |
---|---|
CN106054208A (en) | 2016-10-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106054208B (en) | The anticollision device, collision-prevention device of the recognition methods of multi-line laser radar vehicle target and automobile | |
CN106371104A (en) | Vehicle targets recognizing method and anti-collision device using multi-line point cloud data machine learning | |
US11885910B2 (en) | Hybrid-view LIDAR-based object detection | |
CN106339669A (en) | Multiline point cloud data machine learning human target recognition method and anti-collision device | |
CN106371105A (en) | Vehicle targets recognizing method, apparatus and vehicle using single-line laser radar | |
CN106291506A (en) | Vehicle target recognition methods based on single line cloud data machine learning and device | |
CN106353769A (en) | Humanoid target identification method using multi-line laser radar and anti-collision device of automobile | |
CN106485950B (en) | The method and apparatus evaded for rear cross wagon flow | |
US10137890B2 (en) | Occluded obstacle classification for vehicles | |
US9121717B1 (en) | Collision avoidance for vehicle control | |
CN201918032U (en) | Low-altitude flying anti-collision device of aircraft | |
CN105549616B (en) | A kind of multiaxis unmanned plane obstacle avoidance system and its barrier-avoiding method based on laser array | |
Sato et al. | Multilayer lidar-based pedestrian tracking in urban environments | |
CN110764108A (en) | Obstacle detection method and device for port automatic driving scene | |
CN106295586A (en) | Humanoid target identification method based on single line cloud data machine learning and device | |
WO2021216348A1 (en) | Map consistency checker | |
CN106371106A (en) | People shape targets recognizing method, apparatus and vehicle using single-line laser radar | |
EP4077085A1 (en) | In-path obstacle detection and avoidance system | |
CN108509820A (en) | Method for obstacle segmentation and device, computer equipment and readable medium | |
CN109334563A (en) | A kind of anticollision method for early warning based on road ahead pedestrian and bicyclist | |
CN109271857A (en) | A kind of puppet lane line elimination method and device | |
JP2022548743A (en) | Obstacle information sensing method and device for mobile robot | |
CN110320531A (en) | Obstacle recognition method, map creating method and device based on laser radar | |
CN106680832A (en) | Obstacle detection method and device of mobile robot and mobile robot | |
CN108470174A (en) | Method for obstacle segmentation and device, computer equipment and readable medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
TR01 | Transfer of patent right |
Effective date of registration: 20210414 Address after: Room C313, 1759 Mingxi Road, Beihu science and Technology Development Zone, Changchun City, Jilin Province, 130000 Patentee after: CHANGCHUN CHANGGUANG PRECISION INSTRUMENT GROUP Co.,Ltd. Address before: 130022 No. 7089 Satellite Road, Jilin, Changchun Patentee before: CHANGCHUN University OF SCIENCE AND TECHNOLOGY |
|
TR01 | Transfer of patent right |