CN104197897B - A kind of downtown roads graticule automatic classification method based on Vehicle-borne Laser Scanning point cloud - Google Patents
A kind of downtown roads graticule automatic classification method based on Vehicle-borne Laser Scanning point cloud Download PDFInfo
- Publication number
- CN104197897B CN104197897B CN201410171408.5A CN201410171408A CN104197897B CN 104197897 B CN104197897 B CN 104197897B CN 201410171408 A CN201410171408 A CN 201410171408A CN 104197897 B CN104197897 B CN 104197897B
- Authority
- CN
- China
- Prior art keywords
- roadmarking
- line
- strip
- point
- road
- 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
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C11/00—Photogrammetry or videogrammetry, e.g. stereogrammetry; Photographic surveying
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Remote Sensing (AREA)
- Life Sciences & Earth Sciences (AREA)
- Radar, Positioning & Navigation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Road Repair (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention discloses a kind of downtown roads graticule automatic classification method based on Vehicle-borne Laser Scanning point cloud, comprise the following steps: S1, based on wheelpath data, original point cloud data is carried out road surface segmentation, obtain road surface cloud data；S2, described road surface cloud data is carried out binary conversion treatment, and extract road line point；S3, described road line point is clustered, isolate roadmarking target independent of each other；S4, size according to described roadmarking target, sort out large scale roadmarking and little yardstick roadmarking；S5, described large scale roadmarking is carried out classification based on wheelpath and curb line process；S6, described little yardstick roadmarking is carried out based on the degree of depth study and principal component analysis classification process.The present invention can extract and sort out downtown roads graticule quickly and accurately, greatly reduces time and labour cost that data process, the safety of traffic and the reliability of intelligent driving have been effectively ensured.
Description
Technical field
The present invention relates to intelligent transportation system and smart city construction field, particularly relate to a kind of based on car
Carry the downtown roads graticule automatic classification method of laser scanning point cloud.
Background technology
Downtown roads graticule, as the important composition element of traffic monitoring system, sets on numerous bases, city
Irreplaceable effect is played in executing.On the one hand, it can regulate and control the behavior of vehicle and pedestrian
Activity, effectively reduces the generation of vehicle accident, has safely provided guarantee for vehicle and pedestrian；Separately
On the one hand, its classification and positional information can contribute to carrying as the important input of intelligent driving system
The reliability of high intelligent driving.Therefore, traffic control department and intelligent transportation system be badly in need of a kind of quickly,
Real-time for extracting and the system of downtown roads graticule of classifying, thus ensure safety and the intelligence of traffic
The reliability driven.
At present, the extraction of downtown roads graticule and classification are based primarily upon image or video data.Main method
Including shape segmentations, Texture Segmentation and dividing method based on geometric properties.Common downtown roads mark
Line drawing method has: Markov model method, condition random field method and machine learning method.But,
These methods are mainly retrained by the following aspects: (1) downtown roads graticule shape and type are many
Sample；(2) illumination condition of data acquisition and the time；(3) move shade that vehicle caused and block；
(4) roadmarking shape distortion produced by image or video data.Therefore, based on image or video
Method also cannot meet the demand that downtown roads graticule automatically extracts and classifies.
In recent years, the development of Vehicle-borne Laser Scanning technology was the rapidest.It quickly, accurately obtains Three Dimensional Ground
The ability of spatial information, is increasingly paid much attention to by people.Only due to Vehicle-borne Laser Scanning system
Special design form and staggered form bidifly shaven head scan mode, Vehicle-borne Laser Scanning system not only possesses airborne sharp
Photo-scanning system can gather the characteristic of broad range of data, and can reach territorial laser scanning system institute
The data precision possessed and dot density.Therefore, Vehicle-borne Laser Scanning system is increasingly becoming city space information
A kind of important technical gathered.
But, how to automatically extract from high density, high-precision magnanimity Vehicle-borne Laser Scanning cloud data
Landform, characters of ground object are that a cloud post-processing technology researches and develops the challenge faced.From massive point cloud quickly,
Automatically extract and characters of ground object accurate, effective of classifying also is still within the starting stage.
Summary of the invention
It is an object of the invention to provide a kind of downtown roads graticule based on Vehicle-borne Laser Scanning point cloud automatic
Sorting technique.
For achieving the above object, the present invention is by the following technical solutions:
A kind of downtown roads graticule automatic classification method based on Vehicle-borne Laser Scanning point cloud, including following step
Rapid:
S1, based on wheelpath data, original point cloud data is carried out road surface segmentation, obtain road surface point cloud number
According to；
S2, described road surface cloud data is carried out binary conversion treatment, and extract road line point；
S3, described road line point is clustered, isolate roadmarking target independent of each other；
S4, size according to described roadmarking target, sort out large scale roadmarking and little yardstick road
Marking lines；
S5, described large scale roadmarking is carried out classification based on wheelpath and curb line process, point
Class goes out stop line, centrage and boundary line；
S6, described little yardstick roadmarking is carried out based on the degree of depth study and principal component analysis classification process,
Sort out direction cue mark, pedestrian's warning label, zebra crossing, centrage and boundary line.
Preferably, described step S1 specifically include following step by step:
S11, original point cloud data is evenly divided into one group of some cloud mass along the direction of wheelpath；
S12, for each some cloud mass, be syncopated as a some cloud along the direction being perpendicular to wheelpath and cut open sheet,
And cut open extraction curb point sheet from a cloud；
S13, all curb points are fitted, obtain curb line；
S14, original point cloud data is split along curb line, obtain road surface cloud data.
Preferably, described step S2 specifically include following step by step:
S21, described road surface cloud data is divided into one group to both sides with a certain division interval along wheelpath
Road surface point cloud burst, uses OTSU binarization method to carry out each road surface point cloud burst respectively at binaryzation
Reason, and extract road line point.
S22, calculate the spatial density value of each road line point, if the spatial density of certain road line point
When value is less than noise marginal value set in advance, then rejected.
Preferably, described step S5 specifically include following step by step:
S51, to each large scale roadmarking, if it comprises multiclass roadmarking, then perform step S52,
If it comprises a type of roadmarking, then it is directly designated as roadmarking unit；
S52, the large scale roadmarking comprising multiclass roadmarking is carried out normalization based on volume elements divide
Cut, obtain the most independent roadmarking unit；
S53, each roadmarking unit is sampled, and calculate the distribution characteristics of sampled point, it is judged that symbol
Whether close the sampled point quantity of direction verification condition more than the first predetermined threshold value, if so, Ze Jianggai road road sign
Line unit is labeled as stop line, if it is not, then perform step S54, described direction verification condition is sampled point
Distribution characteristics is perpendicular to the direction of wheelpath；
S54, the sampled point calculated on roadmarking unit and the distance of curb line, it is judged that meet position verification
Whether the sampled point quantity of condition is more than the second predetermined threshold value, the most then by this roadmarking unit labelling
For boundary line, if it is not, this roadmarking unit is then labeled as centrage, described position verification condition is
Sampled point is less than 1m to the distance of curb line.
Preferably, described step S52 specifically includes following sub-step:
S521, the large scale roadmarking comprising multiclass roadmarking is carried out volume elements, and with volume elements
After volume elements point cloud mass structure full-mesh weighted graph；
S522, based on full-mesh weighted graph, large scale roadmarking is normalized segmentation.
Preferably, described step S6 specifically includes following sub-step:
S61, utilize degree of deep learning method train a deep layer Boltzmann machine comprising two-layer hidden layer, and
Deep layer Boltzmann machine after training is configured to the grader of a multilamellar；
S62, using little yardstick roadmarking as the input of Multilayer Classifier, Multilayer Classifier exports little yardstick
The classification information of roadmarking, described classification information comprises strip roadmarking, direction cue mark, OK
People's warning label and other roadmarkings；
S63, to each strip roadmarking P, be w along its width to two-sided search widthside=1m
Neighborhood, if there is strip roadmarking Q in its neighborhood, then perform step S64, if not depositing in this neighborhood
At strip roadmarking, then perform step S65；
Whether S64, the strip roadmarking Q judged in strip roadmarking P and its neighborhood meet zebra crossing
Decision condition, if meeting, being then labeled as zebra crossing by this strip roadmarking P, if being unsatisfactory for, then holding
Row step S65, described zebra crossing decision condition is:
(vp∥vq)∧(vp⊥vpq)∧(vq⊥vpq)
Wherein, some p is strip roadmarking P geometric center in X/Y plane, vpDistribution for a p
Feature, some q is strip roadmarking Q geometric center in X/Y plane, vqFor the distribution characteristics of a q,
vpqDirection vector for a p to some q；
S65, it is w along the length direction of strip roadmarking P to two-sided search widthendThe neighborhood of=3m,
If there is strip roadmarking Q in its neighborhood, then judge strip roadmarking P and the strip road in its neighborhood
Whether marking lines Q meets intermittent line decision condition, if meet, then strip roadmarking P is carried out based on
The classification of wheelpath and curb line processes, and strip roadmarking P is labeled as boundary line or centrage,
Described intermittent line decision condition is:
(vp//vq)^(vp//vpq)^(vq//vpq)
Wherein, some p is strip roadmarking P geometric center in X/Y plane, vpDistribution for a p
Feature, some q is strip roadmarking Q geometric center in X/Y plane, vqFor the distribution characteristics of a q,
vpqDirection vector for a p to some q.
After using technique scheme, the present invention, compared with background technology, has the advantage that
1, the present invention can extract and sort out downtown roads graticule quickly and accurately, greatly reduces number
According to the time processed and labour cost, the safety of traffic and the reliability of intelligent driving are effectively ensured.
2, carry out burst binary conversion treatment based on wheelpath by road pavement cloud data, solve sharp
Luminous point cloud intensity data is excited the problem of angle of incidence and target range impact of light beam, is effectively improved
The quality that marking lines extracts.
3, by large scale roadmarking being carried out normalization dividing processing based on volume elements, effectively will weight
Folded stop line is separated with boundary line, improves the accuracy to roadmarking classification.
4, by little yardstick roadmarking being carried out classification process based on degree of depth study and principal component analysis,
It is effectively improved the nicety of grading of little yardstick roadmarking.
Accompanying drawing explanation
Fig. 1 is the workflow schematic diagram of the present invention.
Fig. 2 is the extraction result schematic diagram of curb point.
Fig. 3 is the fitting result schematic diagram of curb point.
Fig. 4 shows the result of road surface cloud data segmentation.
Fig. 5 shows the binaryzation result of road surface cloud data.
Fig. 6 shows the result after the road surface cloud data of binary conversion treatment is filtered.
Fig. 7 is the schematic diagram of Multilayer Classifier.
Fig. 8 is principal component analysis schematic diagram.
Detailed description of the invention
In order to make the purpose of the present invention, technical scheme and advantage clearer, below in conjunction with accompanying drawing and
Embodiment, is further elaborated to the present invention.Should be appreciated that described herein being embodied as
Example only in order to explain the present invention, is not intended to limit the present invention.
Embodiment
Refer to Fig. 1, the invention discloses a kind of downtown roads graticule based on Vehicle-borne Laser Scanning point cloud certainly
Moving sorting technique, it comprises the following steps:
S1, road surface cloud data are split
Based on wheelpath data, original point cloud data is carried out road surface segmentation, obtains road surface cloud data,
Referred to herein as wheelpath data be by being integrated in the inertial navigation system in Vehicle-borne Laser Scanning system
It is acquired.This step realizes especially by following steps:
S11, original point cloud data is evenly divided into one group of some cloud mass, in this reality along the direction of wheelpath
Executing in example, the segmentation of some cloud mass is spaced apart 3m.
S12, for each some cloud mass, be syncopated as a some cloud along the direction being perpendicular to wheelpath and cut open sheet,
And cut open extraction curb point (with reference to shown in Fig. 2) sheet from a cloud.According to prophet's experience, the curb of road leads to
Often it is perpendicular to road surface arrange and higher than road surface, as long as therefore finding the some cloud of height sudden change i.e. to have found road
Along point.The present embodiment extracts the method for curb point specifically, cut open intersecting of sheet and wheelpath from some cloud
The point cloud of height sudden change, to two-sided search, is found, as curb point in position.
S13, all curb points are fitted, obtain curb line (with reference to shown in Fig. 3).
S14, original point cloud data is split along curb line, obtain road surface cloud data (with reference to Fig. 4
Shown in).
It should be noted that for extraction result and fitting result, Fig. 2 of showing curb point more intuitively
Having omitted with the cloud data of road both sides atural object in Fig. 3, therefore, Fig. 2 and Fig. 3 show only
Curb point and the distribution situation of curb line, do not represent the result of step S12, S13.
S2, extraction road line point
Road pavement cloud data carries out binary conversion treatment, and extracts road line point.This step is specifically led to
Cross following steps to realize:
S21, road surface cloud data is divided into one group road surface to both sides with a certain division interval along wheelpath
Point cloud burst, then uses OTSU binarization method to carry out each road surface point cloud burst respectively at binaryzation
Reason, and extract road line point (with reference to shown in Fig. 5).In the present embodiment, the division of cloud burst is put
It is spaced apart 2m.
S22, calculate the spatial density value of each road line point, if the spatial density of certain road line point
When value is less than noise marginal value set in advance, then rejected (with reference to shown in Fig. 6).For certain together
Marking lines point p (x, y, z), its spatial density value is defined as:
Wherein, N (p) represents the local neighborhood of some p, dNRepresent the size of neighborhood, pi(xi,yi,zi) it is a some p
Local neighborhood in point.
In the present embodiment, dNTaking 0.1m, noise marginal value is 8.
S3, road line point cluster
Road line point is clustered, isolates roadmarking target independent of each other.At the present embodiment
In, road line point is clustered and specifically with the clustering distance of 0.05m, road line point is carried out Europe
Family name's distance cluster.In order to reduce time and the data space that subsequent step calculating processes, can be according to road
The dimensional characteristic of marking lines, carries out simple goal filtering, to reject to the roadmarking after cluster
The non-rice habitats graticule target that size is little is (as roadmarking construction wrong omits the paint vehicle on road surface and road
The inspection well cover etc. that marking lines color is close).
S4, roadmarking target size are classified
According to the size of roadmarking target, sort out large scale roadmarking and little yardstick roadmarking.
For each roadmarking target, comprised three-dimensional coordinate information a little according to it and calculated its bounding box,
And calculate the catercorner length of bounding box.The catercorner length roadmarking target label more than 8m is become big
Yardstick roadmarking, becomes little yardstick roadmarking by the catercorner length roadmarking target label less than 8m.
S5, large scale roadmarking are classified
Large scale roadmarking carries out classification based on wheelpath and curb line process, sort out stopping
Line, centrage and boundary line.This step realizes especially by following steps:
S51, to each large scale roadmarking, if it comprises multiclass roadmarking, then perform step S52,
If it comprises a type of roadmarking, then it is directly designated as roadmarking unit.Judge big chi
Whether degree roadmarking comprises multiclass roadmarking, can be by the diverse location on large scale roadmarking
Sample, and calculate the distribution characteristics of each sampled point.If the distribution characteristics of each sampled point is identical, then sentence
This large scale roadmarking fixed only comprises a type of roadmarking；If the distribution characteristics of each sampled point is not
With, then judge that this large scale roadmarking comprises multiclass roadmarking.Certainly, it is possible to use artificial means
Go to judge whether large scale roadmarking comprises multiclass roadmarking.
S52, the large scale roadmarking comprising multiclass roadmarking is carried out normalization based on volume elements divide
Cut, obtain the most independent roadmarking unit.This step realizes especially by following steps:
S521, the large scale roadmarking comprising multiclass roadmarking is carried out volume elements with certain volume elements width
Change, and construct full-mesh weighted graph G={V, E}, body in the present embodiment with the volume elements point cloud mass after volume elements
Unit's width takes 0.1m.Summit V on this full-mesh weighted graph G is by each the volume elements point cloud after volume elements
Block forms, and the limit E on G is connected to every pair of volume elements point cloud mass.For each volume elements point cloud mass, choose it
In the nearest point of in vitro unit point cloud mass geometric center as the characteristic point of this volume elements point cloud mass, and stored
In set PvoxelIn.Then, for set PvoxelIn any point q, utilize its k-nearest neighbor point
p1,p2,…pkCalculate its covariance matrix:
By above-mentioned covariance matrix is carried out Eigenvalues Decomposition, calculate three eigenvalues and corresponding three
Individual characteristic vector.Using eigenvalue of maximum characteristic of correspondence vector as the distribution characteristics of some q, i.e. put q institute
The distribution characteristics of corresponding volume elements point cloud mass.The distribution characteristics of note point q is vq.Then on full-mesh weighted graph
Connection any two summit (i, the weight on limit j) can be defined as follows:
Wherein, qi, qj ∈ Pvoxel；vqiAnd vqjDistribution characteristics for its correspondence；σDAnd σAFor standard deviation；ds
For distance constraints, it is used for determining the maximum coverage of two volume elements point cloud masses.
S522, based on full-mesh weighted graph, large scale roadmarking is normalized segmentation.Normalization is split
Method is intended to be divided by weighted graph G by maximization similar degree in the class and the method for similarity between class that minimizes
Become two disjoint volume elements set of blocks A and B.The cost function of normalization segmentation is defined as:
Wherein, cut (A, B) represents the weight sum on the limit between A and B；Assoc (A, V) represents whole limit
Fall the weight sum on the limit in A；Assoc (B, V) represents that whole limit falls the weight sum on the limit in B.
Realize comprising multiclass roadmarking finally by the generalized eigenvalue decomposition problem resolving cost function corresponding
The segmentation of large scale roadmarking, thus effectively by overlapping multiclass roadmarking (such as stop line and
Boundary line) separate, it is simple to the subsequent step classification to roadmarking, improve roadmarking classification
Accuracy.
S53, each roadmarking unit is sampled, and calculate the distribution characteristics of sampled point, it is judged that symbol
Whether close the sampled point quantity of direction verification condition more than the first predetermined threshold value, if so, Ze Jianggai road road sign
Line unit is labeled as stop line, if it is not, then perform step S54, direction verification condition is the distribution of sampled point
Feature is perpendicular to the direction of wheelpath.In the present embodiment, the hits to each roadmarking unit
Amount is 100, and the first predetermined threshold value is 95.
S54, the sampled point calculated on roadmarking unit and the distance of curb line, it is judged that meet position verification
Whether the sampled point quantity of condition is more than the second predetermined threshold value, the most then by this roadmarking unit labelling
For boundary line, if it is not, this roadmarking unit is then labeled as centrage, verification condition in position is sampling
Point is less than 1m to the distance of curb line.In the present embodiment, the second predetermined threshold value is 95.
S6, little yardstick roadmarking are classified
Little yardstick roadmarking carries out classification based on degree of depth study and principal component analysis process, sort out
Direction cue mark, pedestrian's warning label, zebra crossing, centrage and boundary line.This step especially by
Following steps realize:
S61, utilize degree of deep learning method train a deep layer Boltzmann machine comprising two-layer hidden layer, and
Deep layer Boltzmann machine after training is configured to the grader of a multilamellar.With reference to shown in Fig. 7, these are many
Layer grader comprises an input layer, two hidden layers and an output layer.Wherein, W1, W2And W3
Represent the model parameter of Multilayer Classifier respectively；V represents input data；h1And h2Represent ground floor respectively
With second layer abstract characteristics；F(h2| v) represent posterior probability based on input data v；Y is Multilayer Classifier
Output.
The energy function of deep layer Boltzmann machine is defined as:
E(v,h1,h2；θ)=-vTW1h1-(h1)TW2h2
Wherein, θ={ W1,W2Represent the model parameter of deep layer Boltzmann machine.This deep layer Boltzmann machine
The definition of probability of one data v of input is:
Wherein,Partition function for this deep layer Boltzmann machine.By right
Number maximum likelihood learning method, just can train the model parameter of this deep layer Boltzmann machine.Finally, with
This deep layer Boltzmann machine structure Multilayer Classifier as shown in Figure 7.
S62, using little yardstick roadmarking as the input of Multilayer Classifier, Multilayer Classifier exports little yardstick
The classification information of roadmarking, classification information comprises strip roadmarking, direction cue mark, pedestrian police
Accuse labelling and other roadmarkings.In the present embodiment, Multilayer Classifier output [1,0,0,0]T, [0,1,0,0]T,
[0,0,1,0]T, [0,0,0,1]TRepresent respectively strip roadmarking, direction cue mark, pedestrian's warning label and
Other roadmarkings.
S63, to each strip roadmarking P, be w along its width to two-sided search widthside=1m
Neighborhood (with reference to shown in Fig. 8 a), if there is strip roadmarking Q in its neighborhood, then perform step S64,
If there is not strip roadmarking in this neighborhood, then perform step S65.Referred to herein as " width "
Refer to the direction vertical with the long limit of strip roadmarking P.
Whether S64, the strip roadmarking Q judged in strip roadmarking P and its neighborhood meet zebra crossing
Decision condition, if meeting, being then labeled as zebra crossing by this strip roadmarking P, if being unsatisfactory for, then holding
Row step S65, zebra crossing decision condition is:
(vp∥vq)∧(vp⊥vpq)∧(vq⊥vpq)
Wherein, some p is strip roadmarking P geometric center in X/Y plane, vpDistribution for a p
Feature, some q is strip roadmarking Q geometric center in X/Y plane, vqFor the distribution characteristics of a q,
vpqDirection vector for a p to some q；
S65, it is w along the length direction of strip roadmarking P to two-sided search widthendNeighborhood (the ginseng of=3m
Examine shown in Fig. 8 b), referred to herein as " length direction " refer to the long limit with strip roadmarking P put down
The direction of row.If there is strip roadmarking Q in its neighborhood, then judge strip roadmarking P and its neighborhood
In strip roadmarking Q whether meet intermittent line decision condition:
(vp//vq)^(vp//vpq)^(vq//vpq)
Wherein, some p is strip roadmarking P geometric center in X/Y plane, vpDistribution for a p
Feature, some q is strip roadmarking Q geometric center in X/Y plane, vqFor the distribution characteristics of a q,
vpqDirection vector for a p to some q.
If meeting, then strip roadmarking P is carried out classification based on wheelpath and curb line and processes,
It is identical that it specifically processes step and step S53, S54, need to be using strip roadmarking P as roadmarking
Unit carries out process can sort out boundary line or centrage.
After above-mentioned steps, the automatic classification to downtown roads graticule can be realized.
The above, the only present invention preferably detailed description of the invention, but protection scope of the present invention is not
Being confined to this, any those familiar with the art, can in the technical scope that the invention discloses
The change readily occurred in or replacement, all should contain within protection scope of the present invention.Therefore, the present invention
Protection domain should be as the criterion with scope of the claims.
Claims (2)
1. a downtown roads graticule automatic classification method based on Vehicle-borne Laser Scanning point cloud, it is characterised in that include following step
Rapid:
S1, based on wheelpath data, original point cloud data is carried out road surface segmentation, obtain road surface cloud data；
S2, described road surface cloud data is carried out binary conversion treatment, and extract road line point；
S3, described road line point is clustered, isolate roadmarking target independent of each other；
S4, size according to described roadmarking target, sort out large scale roadmarking and little yardstick roadmarking, by diagonal angle
The line length roadmarking target label more than 8m becomes large scale roadmarking, by the catercorner length roadmarking mesh less than 8m
It is marked as little yardstick roadmarking；
S5, described large scale roadmarking is carried out classification based on wheelpath and curb line process, sort out stop line, in
Heart line and boundary line；
S6, described little yardstick roadmarking is carried out based on the degree of depth study and principal component analysis classification process, sort out direction and refer to
Indicating note, pedestrian's warning label, zebra crossing, centrage and boundary line；
Described step S1 specifically include following step by step:
S11, original point cloud data is evenly divided into one group of some cloud mass along the direction of wheelpath；
S12, for each some cloud mass, be syncopated as a some cloud along the direction being perpendicular to wheelpath and cut open sheet, and cut open sheet from a cloud
Extract curb point；
S13, all curb points are fitted, obtain curb line；
S14, original point cloud data is split along curb line, obtain road surface cloud data；
Described step S2 specifically include following step by step:
S21, described road surface cloud data is divided into one group road surface point cloud burst to both sides with a certain division interval along wheelpath,
Use OTSU binarization method that each road surface point cloud burst is carried out binary conversion treatment respectively, and extract road line point；
S22, calculate the spatial density value of each road line point, if the spatial density value of certain road line point is less than presetting
Noise marginal value time, then rejected；
Described step S5 specifically include following step by step:
S51, to each large scale roadmarking, if it comprises multiclass roadmarking, then perform step S52, if it comprises one
The roadmarking of type, then be directly designated as roadmarking unit by it；
S52, the large scale roadmarking comprising multiclass roadmarking is carried out normalization based on volume elements segmentation, obtain the most only
Vertical roadmarking unit；
S53, each roadmarking unit is sampled, and calculate the distribution characteristics of sampled point, it is judged that meet direction verification condition
Sampled point quantity whether more than the first predetermined threshold value, the most then this roadmarking unit is labeled as stop line, if it is not, then
Performing step S54, described direction verifies the distribution characteristics that condition is sampled point and is perpendicular to the direction of wheelpath；
S54, the sampled point calculated on roadmarking unit and the distance of curb line, it is judged that meet the sampling number of position verification condition
This roadmarking unit whether more than the second predetermined threshold value, is the most then labeled as boundary line by amount, if it is not, Ze Jianggai road road sign
Line unit is labeled as centrage, and described position verification condition is that the sampled point distance to curb line is less than 1m；
Described step S6 specifically includes following sub-step:
S61, utilize degree of deep learning method train a deep layer Boltzmann machine comprising two-layer hidden layer, and will training after deep layer
Boltzmann machine is configured to the grader of a multilamellar；
S62, using little yardstick roadmarking as the input of Multilayer Classifier, Multilayer Classifier exports the classification of little yardstick roadmarking
Information, described classification information comprises strip roadmarking, direction cue mark, pedestrian's warning label and other roadmarkings；
S63, to each strip roadmarking P, be w along its width to two-sided search widthsideThe neighborhood of=1m, if it is adjacent
There is strip roadmarking Q in territory, then performing step S64, if there is not strip roadmarking Q in this neighborhood, then performing step
S65；
Whether S64, the strip roadmarking Q judged in strip roadmarking P and its neighborhood meet zebra crossing decision condition, if full
Foot, then be labeled as zebra crossing by this strip roadmarking P, if being unsatisfactory for, then performs step S65, and described zebra crossing judge bar
Part is:
(vp∥vq)∧(vp⊥vpq)∧(vq⊥vpq)
Wherein, some p is strip roadmarking P geometric center in X/Y plane, vpFor the distribution characteristics of a p, some q is bar
Shape roadmarking Q geometric center in X/Y plane, vqFor the distribution characteristics of a q, vpqDirection vector for a p to some q；
S65, it is w along the length direction of strip roadmarking P to two-sided search widthendThe neighborhood of=3m, if existing in its neighborhood
Strip roadmarking Q, then judge whether strip roadmarking P and the strip roadmarking Q in its neighborhood meets intermittent line and judge
Condition, if meeting, then carries out classification based on wheelpath and curb line and processes, by strip road road sign strip roadmarking P
Line P is labeled as boundary line or centrage, and described intermittent line decision condition is:
(vp//vq)∧(vp//vpq)∧(vq//vpq)
Wherein, some p is strip roadmarking P geometric center in X/Y plane, vpFor the distribution characteristics of a p, some q is bar
Shape roadmarking Q geometric center in X/Y plane, vqFor the distribution characteristics of a q, vpqDirection vector for a p to some q.
A kind of downtown roads graticule automatic classification method based on Vehicle-borne Laser Scanning point cloud, it is special
Levying and be, described step S52 specifically includes following sub-step:
S521, the large scale roadmarking comprising multiclass roadmarking is carried out volume elements, and with the volume elements point cloud mass after volume elements
Structure full-mesh weighted graph；
S522, based on full-mesh weighted graph, large scale roadmarking is normalized segmentation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410171408.5A CN104197897B (en) | 2014-04-25 | 2014-04-25 | A kind of downtown roads graticule automatic classification method based on Vehicle-borne Laser Scanning point cloud |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410171408.5A CN104197897B (en) | 2014-04-25 | 2014-04-25 | A kind of downtown roads graticule automatic classification method based on Vehicle-borne Laser Scanning point cloud |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104197897A CN104197897A (en) | 2014-12-10 |
| CN104197897B true CN104197897B (en) | 2017-01-04 |
Family
ID=52083225
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410171408.5A Active CN104197897B (en) | 2014-04-25 | 2014-04-25 | A kind of downtown roads graticule automatic classification method based on Vehicle-borne Laser Scanning point cloud |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104197897B (en) |
Families Citing this family (39)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105069395B (en) * | 2015-05-17 | 2018-10-09 | 北京工业大学 | Roadmarking automatic identifying method based on Three Dimensional Ground laser scanner technique |
| US10482331B2 (en) * | 2015-11-20 | 2019-11-19 | GM Global Technology Operations LLC | Stixel estimation methods and systems |
| CN105551082B (en) * | 2015-12-02 | 2018-09-07 | 百度在线网络技术（北京）有限公司 | A kind of pavement identification method and device based on laser point cloud |
| CN106845321B (en) * | 2015-12-03 | 2020-03-27 | 高德软件有限公司 | Method and device for processing pavement marking information |
| CN105627938A (en) * | 2016-01-07 | 2016-06-01 | 厦门大学 | Pavement asphalt thickness detection method based on vehicle-mounted laser scanning spot cloud |
| CN105488498B (en) * | 2016-01-15 | 2019-07-30 | 武汉中海庭数据技术有限公司 | A kind of lane sideline extraction method and system based on laser point cloud |
| CN105957145A (en) * | 2016-04-29 | 2016-09-21 | 百度在线网络技术（北京）有限公司 | Road barrier identification method and device |
| CN106022381B (en) * | 2016-05-25 | 2020-05-22 | 厦门大学 | Automatic extraction method of street lamp pole based on vehicle-mounted laser scanning point cloud |
| CN106067003B (en) * | 2016-05-27 | 2020-05-19 | 山东科技大学 | Automatic extraction method for road vector identification line in vehicle-mounted laser scanning point cloud |
| CN106126816B (en) * | 2016-06-24 | 2019-04-05 | 南京林业大学 | Repeat the extensive ALS building point cloud modeling method under building automatic sensing |
| DE102016214027A1 (en) * | 2016-07-29 | 2018-02-01 | Volkswagen Aktiengesellschaft | Method and system for detecting landmarks in a traffic environment of a mobile unit |
| CN106503678A (en) * | 2016-10-27 | 2017-03-15 | 厦门大学 | Roadmarking automatic detection and sorting technique based on mobile laser scanning point cloud |
| CN106599915B (en) * | 2016-12-08 | 2017-11-28 | 立得空间信息技术股份有限公司 | A kind of vehicle-mounted laser point cloud classifications method |
| CN106650809B (en) * | 2016-12-20 | 2018-02-23 | 福州大学 | A kind of vehicle-mounted laser point cloud objective classification method and system |
| CN108241819B (en) * | 2016-12-23 | 2020-07-28 | 阿里巴巴（中国）有限公司 | Method and device for identifying pavement marker |
| CN107085710B (en) * | 2017-04-26 | 2020-06-02 | 长江空间信息技术工程有限公司(武汉) | Single-tree automatic extraction method based on multispectral LiDAR data |
| CN107679458B (en) * | 2017-09-07 | 2020-09-29 | 中国地质大学（武汉） | Method for extracting road marking lines in road color laser point cloud based on K-Means |
| CN109670376B (en) * | 2017-10-13 | 2021-05-25 | 神州优车股份有限公司 | Lane line identification method and system |
| CN110019627B (en) * | 2017-12-25 | 2022-04-12 | 北京京东乾石科技有限公司 | Method, system and computer system for identifying traffic diversion line |
| CN108615452B (en) * | 2018-04-03 | 2019-08-20 | 华南理工大学 | A kind of unknown method for extracting roads based on people's wheel paths point multi-resolution hierarchy |
| CN108845569A (en) * | 2018-04-27 | 2018-11-20 | 厦门维斯云景信息科技有限公司 | Generate semi-automatic cloud method of the horizontal bend lane of three-dimensional high-definition mileage chart |
| CN108931786A (en) * | 2018-05-17 | 2018-12-04 | 北京智行者科技有限公司 | Curb detection device and method |
| CN108960326B (en) * | 2018-07-04 | 2021-03-23 | 深圳先进技术研究院 | Point cloud fast segmentation method and system based on deep learning framework |
| CN109165549B (en) * | 2018-07-09 | 2021-03-19 | 厦门大学 | Road identification obtaining method based on three-dimensional point cloud data, terminal equipment and device |
| CN109166128B (en) * | 2018-07-31 | 2021-11-26 | 东华大学 | Non-equivalent three-dimensional point cloud segmentation method |
| CN109798903B (en) * | 2018-12-19 | 2021-03-30 | 广州文远知行科技有限公司 | Method and device for acquiring road information from map data |
| CN110008921B (en) * | 2019-04-12 | 2021-12-28 | 北京百度网讯科技有限公司 | Road boundary generation method and device, electronic equipment and storage medium |
| CN110210415B (en) * | 2019-06-05 | 2022-09-30 | 福州大学 | Vehicle-mounted laser point cloud road marking identification method based on graph structure |
| CN110349260B (en) * | 2019-07-11 | 2022-06-17 | 武汉中海庭数据技术有限公司 | Automatic pavement marking extraction method and device |
| CN110379006A (en) * | 2019-07-24 | 2019-10-25 | 厦门大学 | A kind of three-dimensional zebra stripes method for reconstructing based on mobile laser radar scanning point cloud |
| CN110866449A (en) * | 2019-10-21 | 2020-03-06 | 北京京东尚科信息技术有限公司 | Method and device for identifying target object in road |
| CN111783722B (en) * | 2020-07-13 | 2021-07-06 | 湖北亿咖通科技有限公司 | Lane line extraction method of laser point cloud and electronic equipment |
| CN112059407A (en) * | 2020-09-17 | 2020-12-11 | 苏州快光科技有限公司 | Picosecond ultrafast laser processing system and method based on deep learning |
| CN112070054B (en) * | 2020-09-17 | 2022-07-29 | 福州大学 | Vehicle-mounted laser point cloud marking classification method based on graph structure and attention mechanism |
| CN112686089B (en) * | 2020-10-10 | 2023-06-13 | 星际空间（天津）科技发展有限公司 | Intelligent automatic extraction method for road information |
| CN112435336B (en) * | 2020-11-13 | 2022-04-19 | 武汉中海庭数据技术有限公司 | Curve type identification method and device, electronic equipment and storage medium |
| CN112561944A (en) * | 2020-11-27 | 2021-03-26 | 中央财经大学 | Lane line extraction method based on vehicle-mounted laser point cloud |
| CN112200171B (en) * | 2020-12-07 | 2021-03-23 | 速度时空信息科技股份有限公司 | Road point cloud extraction method based on scanning lines |
| CN114998770B (en) * | 2022-07-06 | 2023-04-07 | 中国科学院地理科学与资源研究所 | Highway identifier extraction method and system |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP3690366B2 (en) * | 2001-12-27 | 2005-08-31 | 日産自動車株式会社 | Front object detection device |
| CN102445186A (en) * | 2011-09-28 | 2012-05-09 | 中交第二公路勘察设计研究院有限公司 | Method for generating road design surface information by laser radar scan |
| CN102779280A (en) * | 2012-06-19 | 2012-11-14 | 武汉大学 | Traffic information extraction method based on laser sensor |
| CN103500338A (en) * | 2013-10-16 | 2014-01-08 | 厦门大学 | Road zebra crossing automatic extraction method based on vehicle-mounted laser scanning point cloud |
| CN103530689A (en) * | 2013-10-31 | 2014-01-22 | 中国科学院自动化研究所 | Deep learning-based clustering method |
-
2014
- 2014-04-25 CN CN201410171408.5A patent/CN104197897B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP3690366B2 (en) * | 2001-12-27 | 2005-08-31 | 日産自動車株式会社 | Front object detection device |
| CN102445186A (en) * | 2011-09-28 | 2012-05-09 | 中交第二公路勘察设计研究院有限公司 | Method for generating road design surface information by laser radar scan |
| CN102779280A (en) * | 2012-06-19 | 2012-11-14 | 武汉大学 | Traffic information extraction method based on laser sensor |
| CN103500338A (en) * | 2013-10-16 | 2014-01-08 | 厦门大学 | Road zebra crossing automatic extraction method based on vehicle-mounted laser scanning point cloud |
| CN103530689A (en) * | 2013-10-31 | 2014-01-22 | 中国科学院自动化研究所 | Deep learning-based clustering method |
Non-Patent Citations (2)
| Title |
|---|
| "Automatic Road Object Extraction from Mobile Mapping Systems";A. Mancini et al.;《Mechatronics and Embedded Systems and Applicaions,2012 IEEE/ASME International Conference》;20120610;正文全文 * |
| "Using mobile laser scanning data for automated extraction of road markings";Haiyan Guan et al.;《ISPRS Journal of Photogrammetry and Remote Sensing》;20140131;第87卷;正文全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104197897A (en) | 2014-12-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104197897B (en) | A kind of downtown roads graticule automatic classification method based on Vehicle-borne Laser Scanning point cloud | |
| Yu et al. | Learning hierarchical features for automated extraction of road markings from 3-D mobile LiDAR point clouds | |
| CN101859382B (en) | License plate detection and identification method based on maximum stable extremal region | |
| CN104766046B (en) | One kind is detected using traffic mark color and shape facility and recognition methods | |
| CN102496034B (en) | High-spatial resolution remote-sensing image bag-of-word classification method based on linear words | |
| CN108133227A (en) | Laser radar point cloud data sorting technique and device | |
| CN103942546B (en) | Traffic marking identifying system and method are oriented in a kind of urban environment | |
| CN105335702B (en) | A kind of bayonet model recognizing method based on statistical learning | |
| CN103077389A (en) | Text detection and recognition method combining character level classification and character string level classification | |
| CN106778835A (en) | The airport target by using remote sensing image recognition methods of fusion scene information and depth characteristic | |
| CN105160309A (en) | Three-lane detection method based on image morphological segmentation and region growing | |
| CN103870803A (en) | Vehicle license plate recognition method and system based on coarse positioning and fine positioning fusion | |
| CN103971091B (en) | Automatic plane number recognition method | |
| CN104091321A (en) | Multi-level-point-set characteristic extraction method applicable to ground laser radar point cloud classification | |
| CN107247956A (en) | A kind of fast target detection method judged based on grid | |
| CN103500329B (en) | Street lamp automatic extraction method based on vehicle-mounted mobile laser scanning point cloud | |
| Guo et al. | Automatic UAV forced landing site detection using machine learning | |
| CN103034863A (en) | Remote-sensing image road acquisition method combined with kernel Fisher and multi-scale extraction | |
| CN115605777A (en) | Dynamic target point cloud rapid identification and point cloud segmentation method based on road side sensing unit | |
| CN107220638A (en) | A kind of car plate detection recognition methods based on deep learning convolutional neural networks | |
| Juranek et al. | Real-time pose estimation piggybacked on object detection | |
| CN109948690A (en) | A kind of high-speed rail scene perception method based on deep learning and structural information | |
| CN105005789A (en) | Vision lexicon based remote sensing image terrain classification method | |
| CN110032935A (en) | A kind of traffic signals label detection recognition methods based on deep learning cascade network | |
| CN110210415A (en) | Vehicle-mounted laser point cloud roadmarking recognition methods based on graph structure |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C53 | Correction of patent of invention or patent application | ||
| CB03 | Change of inventor or designer information |
Inventor after: Li Jun Inventor after: Yu Yongtao Inventor after: Guan Haiyan Inventor after: Wang Cheng Inventor before: Yu Yongtao Inventor before: Li Jun Inventor before: Guan Haiyan Inventor before: Wang Cheng |
|
| COR | Change of bibliographic data |
Free format text: CORRECT: INVENTOR; FROM: YU YONGTAO LI JUN GUAN HAIYAN WANG CHENG TO: LI JUN YU YONGTAO GUAN HAIYAN WANG CHENG |
|
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |