CN107167811B - The road drivable region detection method merged based on monocular vision with laser radar - Google Patents
The road drivable region detection method merged based on monocular vision with laser radar Download PDFInfo
- Publication number
- CN107167811B CN107167811B CN201710283453.3A CN201710283453A CN107167811B CN 107167811 B CN107167811 B CN 107167811B CN 201710283453 A CN201710283453 A CN 201710283453A CN 107167811 B CN107167811 B CN 107167811B
- Authority
- CN
- China
- Prior art keywords
- pixel
- point
- super
- road
- feature
- 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
- G01S17/86—Combinations of lidar systems with systems other than lidar, radar or sonar, e.g. with direction finders
Landscapes
- Engineering & Computer Science (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Physics & Mathematics (AREA)
- Computer Networks & Wireless Communication (AREA)
- Electromagnetism (AREA)
- General Physics & Mathematics (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention discloses a kind of road drivable region detection methods merged based on monocular vision with laser radar, belong to intelligent transportation field.The method that existing unmanned vehicle Approach for road detection is mainly based upon monocular vision, stereoscopic vision, laser sensor and Multi-sensor Fusion etc., exist to illumination not robust, three-dimensional matching is complicated, laser is sparse and overall fusion low efficiency the disadvantages of.Although some methods for having supervision obtain preferable precision, but training process is complicated, and extensive effect is poor.The road drivable region detection method proposed by the present invention merged based on monocular vision with laser radar, use super-pixel and point cloud data fusion, so that Machine self-learning is gone out road area using feature, the road information that each feature obtains is merged to determine final area by Bayesian frame.This method does not need strong assumption information and complicated training process, and generalization and robustness are superior, and speed is exceedingly fast, and precision is high, more easily promotes and uses in practical applications.
Description
Technical field
The invention belongs to the technique studies in intelligent transportation field, are related to one kind and are melted based on monocular vision with laser radar
The road drivable region detection method of conjunction.
Background technique
In recent years since, Road Detection is always the important content of unmanned area research.The road being widely used at present
Detection method has: monocular vision method, Stereo Vision, laser radar method and the method based on fusion.Wherein, monocular
Visible sensation method only considered the visual information of scene, easily be illuminated by the light condition, and weather conditions influence；The method of stereoscopic vision exists
Time consumption is huge on three-dimensional reconstruction, is not suitable for practice；The method of laser radar there are point cloud data it is sparse lack
Point.Based on the Approach for road detection that Pixel Information and depth information merge both taken full advantage of from camera provide about
The information such as texture, the color of scene, and the depth information of laser radar is combined to make up visual information robust does not lack to environment
Point overcomes non-fused method low efficiency, it is difficult to carry out real-time operation on efficiency of algorithm, it is difficult to the problem of practice,
Therefore the Approach for road detection based on fusion develops rapidly as the first choice of unmanned vehicle Road Detection.Road inspection based on fusion
Survey method is a kind of optimal road inspection to grow up on the basis of monocular vision, laser radar method, sensor fusion etc.
It surveys.To engineering in practice, especially unmanned vehicle driving in be widely used.
Unmanned vehicle Road Detection has been also divided into measure of supervision and unsupervised approaches.Due to the diversity of information of road surface, scene
The variability of the complexity and illumination weather condition of information, robustness and Generalization Capability of the unmanned vehicle for Approach for road detection
It is required that very high.Therefore unsupervised unmanned vehicle Approach for road detection is also the important content of unmanned area research.On the one hand,
Unsupervised Approach for road detection does not need a large amount of flag data and time-consuming training process, can be according to the feature of extraction
Automatically learn road information out, is a kind of method of high generalization ability.On the other hand, the traffic scene of real world is complicated
It is changeable, in the case where that can not be the unmanned training sample for providing all scenes, there is the method for supervision encountering and instructing
Risk is very big when practicing the scene difference biggish Driving Scene of sample, and unsupervised Approach for road detection is to nearly all
Scene robust is suitable for unpiloted practical application.
Summary of the invention
The purpose of the present invention is to provide a kind of roads merged based on monocular vision with laser radar can travel region inspection
Survey method.
In order to achieve the above objectives, the invention adopts the following technical scheme.
Firstly, the method for the fusion merging using super-pixel and laser point cloud data, point cloud data is joined according to laser
On picture after number labeling projection to super-pixel segmentation, super-pixel method takes full advantage of the textural characteristics of scene, greatly contracts
The range of small positioning road area, greatly improves efficiency of algorithm；Secondly, finding the spatial relationship of point, root with triangulation
According to the spatial relationship triangle obtained establish non-directed graph and and calculate the normal vector of every bit, classified barrier point according to non-directed graph；
Then, this method uses the method based on minimum filtering to define new feature (ray), and finds the initial alternative of road area
Region further reduces the detection range of road area, significant increase efficiency of algorithm；By defining new feature (level)
The travelable degree of numeralization each point, efficiently utilizes depth information in terms of depth information.In addition, fusion method is also sharp
Each feature (color characteristic, level are merged that is, based on the Bayesian frame of self study with a kind of unsupervised fusion method
Feature, normal direction measure feature, strength characteristic) probabilistic information of alternative road area that learns, this efficiency of algorithm is high, robust
Performance is strong.
The super-pixel merges that specific step is as follows with laser point cloud data:
The picture that camera acquires is carried out using the method for the improved linear iteraction cluster of existing jointing edge segmentation
Picture segmentation is N number of super-pixel, each super-pixel p by super-pixel segmentationc=(xc,yc,zc,1)TComprising several pixels,
Middle xc,yc,zcIndicate the average value of the location information in the super-pixel under the camera coordinates system of all pixels point, meanwhile, these
The RGB of pixel is unified for the average RGB of all pixels point in the super-pixel.Recycle existing calibration technique by laser
The every bit p that radar obtainsl=(xl,yl,zl,1T) project on the picture after super-pixel segmentation, finally obtain point setWherein Pi=(xi,yi,zi,ui,vi), xi,yi,ziIndicate the location information under the laser coordinate system of the point,
(ui,vi) indicate location information under the corresponding camera coordinates system of point.It is constrained finally by concurrent, only retains and be projected in
The laser point of super-pixel adjacent edges.
Method based on minimum filtering defines the initial alternative area that new feature (ray) finds road area；Specific step
It is rapid as follows:
Firstly, the initial alternative area for defining the road area isWherein, SiIt representsIt is super-pixel SiThe set for all pixels point for being included defines IDRMFor " direction ray map ", and point set will
Pi=(xi,yi,zi,ui,vi) (ui,vi) coordinate transformation is to the central point (P of picture last linebase) sat for the pole of origin
Under mark, then haveIt is point set(ui,vi) proper subclass, wherein Indicate i-th
A point belongs to h angle,Indicate beIn barrier point set, calculate
Method sees below text and flow chart.
Secondly, to solve the problems, such as laser beam leakage, the I handled using the method that minimum filtersDRMIt is expected that
IDRM, finally obtain
Define new feature (level)；Specific step is as follows:
Define every bitLevel feature beAlgorithm is shown in flow chart, and
Super-pixel S is obtained in conjunction with super-pixeliLevel feature L (Si):
Specific step is as follows to merge for the Bayesian frame using self study:
Firstly, being in conjunction with initial alternative area4 kinds of feature unsupervised learnings of the person of being utilized respectively alternative area it is general
Rate.
It is for initial alternative areaIn super-pixel point Si, SiEach pixel P for includingi=(xi,yi,zi,
ui,vi) rgb value unified, utilize Gaussian parameter μcWithSelf study color characteristic, formula are as follows:
θ=45 °
Utilize Gaussian parameter μlWithSelf study super-pixel SiLevel feature L (Si) formula are as follows:
Utilize Gaussian parameter μnWithSelf study super-pixel SiNormal direction measure feature N (Si) formula are as follows:
Define Sg (Si) it is across super-pixel SiRay quantity, self study super-pixel SiStrength characteristic Sg (Si)
Formula are as follows:
Finally, establishing Bayesian frame merges 4 kinds of features, formula is as follows:
Wherein, p (Si=R | Obs) indicate super-pixel SiBelong to the probability of road area, Obs indicates to be based on this 4 kinds of features
Observation.
The beneficial effects of the present invention are embodied in:
First, which greatly limits the practicability of these algorithms and meters since traditional fusion method is using global fusion
Calculate efficiency.The present invention proposes merging using super-pixel and laser point cloud data.This method greatly reduces the standby of road area
Select range, significant increase efficiency of algorithm.Second, therefore, new feature (ray) proposed by the present invention finds road area
Initial alternative area further reduces the detection range of road area, significant increase efficiency of algorithm.Third, proposed by the present invention
Level feature quantizes the travelable degree of each point in terms of depth information, overcomes the Sparse Problems of depth information, effectively
Depth information is utilized in ground, contributes arithmetic accuracy very big.Fourth, proposed by the present invention quantify super-pixel using strength characteristic
With the syncretic relation of depth information, the near big and far smaller problem of visual information has been fully considered, arithmetic accuracy has been contributed very big.Therefore
Algorithm has more important research significance and extensive engineering application value.Fifth, the Bayesian frame of self study, fusion
The probabilistic information for the alternative road area that each feature learning arrives, this efficiency of algorithm is high, and robust performance is strong
Detailed description of the invention
Fig. 1 is the road drivable region detection method functional block diagram merged based on monocular vision with laser radar；
Fig. 2 is to obtain the algorithm flow chart of ray feature；
Fig. 3 be by not using minimum filtering processing ray leakage (under) with after use (on) initial alternative area effect
Figure；
Fig. 4 is to obtain the algorithm flow chart of level feature；
Fig. 5 is the alternative road area probability distribution effect picture obtained by color characteristic self study；
Fig. 6 is the alternative road area probability distribution effect picture obtained by level feature self study；
Fig. 7 is the alternative road area probability distribution effect picture obtained by normal direction measure feature self study；
Fig. 8 is the alternative road area probability distribution effect picture obtained by strength characteristic self study；
Fig. 9 is the probability distribution graph for the final area that the fusion of the Bayesian frame of self study obtains；
Specific embodiment
Shown in referring to Fig.1, camera is adopted using the method for the improved linear iteraction cluster of existing jointing edge segmentation
The picture of collection carries out super-pixel segmentation, is N number of super-pixel, each super-pixel p by picture segmentationc=(xc,yc,zc,1)TIf comprising
A pixel is done, wherein xc,yc,zcIndicate being averaged for the location information in the super-pixel under the camera coordinates system of all pixels point
Value, meanwhile, the RGB of these pixels is unified for the average RGB of all pixels point in the super-pixel.Recycle existing mark
Determine technology and spin matrixAnd transformed matrixTransformed matrix is obtained according to formula (1)
Utilize spin matrixWithThe transforming relationship for establishing 2 coordinate systems, such as formula (2):
The every bit p that laser radar is obtainedl=(xl,yl,zl,1)TIt is such as public on picture after projecting to super-pixel segmentation
Formula (3):
Obtain point setWherein Pi=(xi,yi,zi,ui,vi), xi,yi,ziUnder the laser coordinate system for indicating the point
Location information, (ui,vi) indicate location information under the corresponding camera coordinates system of point.Finally, only retaining super-pixel side
Laser point near edge.
Using data fusion classification barrier point, mapping relations ob (P is obtainedi), ob (Pi)=1 indicates PiFor barrier point, instead
Be 0.For PiCoordinate system (ui,vi) triangulation (Delaunay triangulation) is used, obtain numerous spaces
Triangle and generation non-directed graphE indicate figure in node PiThere are the set at the edge of incidence relation.It rejects and sits
Mark system (ui,vi) under Euclidean distance be unsatisfactory for the edge (P of formula (4)i,Pj):
||Pi-Pj| | < ε ... ... ... ... ... ... ... ... ... ... ... ... ... (4)
It is defined as and PiThe collection of the point of connection is combined into Nb (Pi), then it is { (u with the surface of related spatial trianglej,
vj) | j=iorPj∈Nb(Pi)}.Calculate the normal vector of each spatial triangle, it is clear that work as PiThe spatial triangle of surrounding is got over
It is flat close to ground, PiA possibility that as non-barrier point, is bigger, we take PiThe normal vector of the spatial triangle of surrounding is put down
Mean value is as PiNormal vectorFormula (5) indicates ob (Pi) judgment method:
Method based on minimum filtering defines ray feature and finds the initial alternative area of road areaFirstly, root
Obtained according to algorithm flow chart as shown in Figure 2 " direction ray map " -- IDRM, whereinWhat expression was calculated according to previous step barrier point classification methodIn barrier point
Set, Indicate PiBelong to h angle.Algorithm is by Pi=(xi,yi,zi,ui,vi) (ui,vi) sit
Mark is transformed into the central point (P with picture last linebase) be origin polar coordinates under, then haveIt is point set(ui,vi) proper subclass.Secondly, as shown in figure 3, due to laser data sparsity, need to handle ray leakage
The problem of, this method I that innovatively the minimum method filtered of use is handledDRMObtain desired IDRM.In conjunction with super-pixel point
It cuts, the initial alternative area for defining the road area isWherein, SiIt representsIt is super picture
Plain SiThe set for all pixels point for being included, the final super-pixel that merges obtain
Level feature is defined, Fig. 4 provides the every bit for calculating and belonging to h angle
Level featureFormula (6) indicates, super-pixel S is obtained in conjunction with super-pixeliLevel feature L (Si):
Such as Fig. 5, the travelable degree probabilistic information for the alternative road area that color characteristic obtains is utilized.For initial alternative
RegionIn super-pixel point Si, SiEach pixel P for includingi=(xi,yi,zi,ui,vi) rgb value unified
, since RGB color is to illumination condition and weather conditions not robust, therefore color space conversion method is utilized, it will be empty in RBG
Between original image I be converted into the image I under illuminant-invariant color spacelog, such as formula (7):
Wherein Ilog(u, v) is in IlogPixel value under coordinate system (u, v), IR,IG,IBIndicate that the rgb value of I, θ indicate just
Meet at the not varied angle of illumination variation line.Formula (8) utilizes Gaussian parameter μcWithSelf study color characteristic is to obtain alternative road
The travelable degree probabilistic information in road region:
Such as Fig. 6, the travelable degree probabilistic information for the alternative road area that level feature obtains is utilized.Formula (9) utilizes
Gaussian parameter μlWithSelf study super-pixel SiLevel feature L (Si):
Such as Fig. 7, the travelable degree probabilistic information for the alternative road area that normal direction measure feature obtains is utilized.It calculatesIn
Super-pixel SiNormal direction measure feature N (Si), i.e. SiIn the minimum point of normal vector height coordinate valueSuch as formula (10):
Formula (11) utilizes Gaussian parameter μnWithSelf study super-pixel SiNormal direction measure feature N (Si):
Such as Fig. 8, the travelable degree probabilistic information for the alternative road area that strength characteristic obtains is utilized.Sg(Si) be across
Super-pixel SiRay quantity, self study super-pixel SiStrength characteristic Sg (Si) such as formula (12):
Finally, the fusion for establishing the Bayesian frame that 4 kinds of features of Bayesian frame fusion obtain self study obtains such as Fig. 9
Final area probability distribution graph, such as formula (13):
Wherein, p (Si=R | Obs) indicate super-pixel SiBelong to the probability of road area, Obs indicates to be based on this 4 kinds of features
Observation, this method completes Road Detection task well as can be seen from Figure 9.
In order to prove the advantage of this method, we utilize the number of varying environment in 3 on ROAD-KITTI benchmark
According to collection, the urban environment (Urban Marked, UM) of label, multiple labeling urban environment (Urban Multiple Marked,
UMM) and unlabelled urban environment (Urban Unmarked, UU) tests this method, from maximum F-measure (Max F-
Measure, MaxF), mean accuracy (Average Precision, AP), precision (Precision, PRE), recall rate
(Recall, REC), false positive rate (False Positive Rate, FPR) and false negative rate (False Negative
Rate, FNR) this six indexs are analyzed.While analysis, experiment is compared, and has been announced at present in ROAD-
The method MixedCRF and fusion method RES3D- of best effects are achieved on KITTI benchmark data set using laser
Velo comparison, comparing result are shown in Table 1--4.
Table 1 is this method (Ours Test), the comparative experiments knot of MixedCRF, RES3D-Velo on UM data set
Fruit:
Contrast and experiment of the table 1 on UM data set
Table 2 is this method (Ours Test), the comparative experiments knot of MixedCRF, RES3D-Velo on UMM data set
Fruit:
Contrast and experiment of the table 2 on UMM data set
Table 3 is this method (Ours Test), the comparative experiments knot of MixedCRF, RES3D-Velo on UU data set
Fruit:
Contrast and experiment of the table 3 on UU data set
Table 4 is this method (Ours Test), and (i.e. UM, UMM, UU are integrally examined in URBAN by MixedCRF, RES3D-Velo
Consider) comparing result of the average value of experimental result on data set:
Contrast and experiment of the table 4 in URBAN data set
MixedCRF is the method for needing training, and this method has obtained similar under conditions of not needing any training
Precision, and highest precision is achieved in this index of AP, illustrate the superiority of this method.
In order to show the superiority of the fusion of self study Bayesian frame used by this method, in ROAD-KITTI
Using the data set of varying environment in 3 on benchmark, the urban environment (Urban Marked, UM) of label, multiple labeling
Urban environment (Urban Multiple Marked, UMM) and unlabelled urban environment (Urban Unmarked, UU) test
This method, from maximum F-measure (Max F-measure, MaxF), mean accuracy (Average Precision, AP), essence
It spends (Precision, PRE), recall rate (Recall, REC), false positive rate (False Positive Rate, FPR), and false
Negative rate (False Negative Rate, FNR) this six indexs, analyze it is single obtained using ray feature it is initial alternative
Region (Initial), color characteristic (Color), strength characteristic (Strength), level feature and normal direction measure feature
(Normal) precision, (Fusion) accuracy comparison is merged with Bayesian frame, and comparing result is shown in Table 5--8.
Table 5 is the single initial alternative area (Initial) obtained using ray feature, color characteristic (Color), intensity
Feature (Strength), level feature and normal direction measure feature (Normal), (Fusion) is merged with Bayesian frame in UM number
According to the comparing result on collection:
Table 5Comparison on UM Training Set (BEV)
Table 6 is the single initial alternative area (Initial) obtained using ray feature, color characteristic (Color), intensity
Feature (Strength), level feature and normal direction measure feature (Normal), (Fusion) is merged with Bayesian frame in UMM number
According to the comparing result on collection:
Table 6Comparison on UMM Training Set (Bev)
Table 7 is the single initial alternative area (Initial) obtained using ray feature, color characteristic (Color), intensity
Feature (Strength), level feature and normal direction measure feature (Normal), (Fusion) is merged with Bayesian frame in UU number
According to the comparing result on collection:
Table 7Comparison on UU Training Set (Bev)
Table 8 is the single initial alternative area (Initial) obtained using ray feature, color characteristic (Color), intensity
Feature (Strength), level feature and normal direction measure feature (Normal), (Fusion) are merged with Bayesian frame in UM,
The comparing result of the average value of experimental result on UMM, UU data set:
Table 8Comparison on URBAN Training Set (BEV)
By table 4 and table 8 it is found that being taken based on the road drivable region detection method that monocular vision is merged with laser radar
Obtaining current full accuracy AP, AP is also the most important index for measuring detection method, is also achieved in other indexs good
Advantage, therefore this method is suitable for practical application.
The above content is a further detailed description of the present invention in conjunction with specific preferred embodiments, cannot recognize
Determine a specific embodiment of the invention and be only limitted to this, for those of ordinary skill in the art to which the present invention belongs, not
Be detached from present inventive concept under the premise of, several simple deduction or replace can also be made, all shall be regarded as belonging to the present invention by
The claims submitted determine scope of patent protection.
Claims (2)
1. a kind of road drivable region detection method merged based on monocular vision with laser radar, it is characterised in that:
Firstly, the method for the fusion merging using super-pixel and laser point cloud data, by point cloud data according to laser parameter mark
Surely on the picture after projecting to super-pixel segmentation；
The super-pixel merges that specific step is as follows with laser point cloud data:
Super-pixel segmentation is carried out to the picture that camera acquires using linear iteraction cluster, is N number of super-pixel by picture segmentation, each
Super-pixel pc=(xc,yc,zc,1)TComprising several pixels, wherein xc,yc,zcIndicate the phase of all pixels point in the super-pixel
The average value of location information under machine coordinate system, meanwhile, the RGB of these pixels is unified for all pixels point in the super-pixel
Average RGB, recycle the calibration technique every bit p that obtains laser radarl=(xl,yl,zl,1)TProject to super-pixel segmentation
On picture afterwards, laser merge with point cloud, proposes concurrent constraint, and concurrent, which constrains, to be referred to, in super-pixel and laser point cloud data
It is only retained in the laser point of super-pixel adjacent edges during fusion, finally obtains point setWherein Pi=(xi,
yi,zi,ui,vi), xi,yi,ziIndicate the location information under the laser coordinate system of the point, (ui,vi) indicate the corresponding phase of point
Location information under machine coordinate system.
Secondly, finding the spatial relationship of point with triangulation, non-directed graph and simultaneously is established according to the spatial relationship triangle obtained
The normal vector for calculating every bit, according to non-directed graph classification barrier point；
Then, the initial alternative area that road area is found using the method based on minimum filtering, further reduces road area
Detection range；It is quantized in terms of depth information the travelable degree of each point by defining new feature level, in addition, melting
Conjunction method also utilizes a kind of unsupervised fusion method to merge each feature, i.e. color that is, based on the Bayesian frame of self study
Feature, level feature, normal direction measure feature, the probabilistic information for the alternative road area that strength characteristic learns；Level mark sheet
Show that the travelable degree of corresponding points, calculating process are as follows:
1) forIn i-th point, initialize h angle all the points level featureWhereinIt is point setProper subclass, wherein Indicate that belong to h angle at i-th point；
2) forIn i-th point, define ob (Pi) it is point PiThe planarization of circumferential surface, thenTable
Show that the point is barrier point, otherwise be 0, if ob (Pi (h))=1 is initializingOn the basis of constantly updateFor
Value originally adds adjacent 2 point Pi (h)WithDifference in height, i.e.,
If 3) i≤N(h), return to 2；
If 4) h=H, terminate；Otherwise, 1 is returned；
And super-pixel S is obtained in conjunction with super-pixeliLevel feature L (Si):
Wherein SiIt is super-pixel SiThe set for all pixels point for being included, definitionIt is described to use self study Bayes frame
Frame merges；Specific step is as follows:
Self study -- color characteristics are carried out using four kinds of features, level feature, normal direction measure feature, strength characteristic, firstly, in conjunction with
Initially alternative area isIt is utilized respectively the probability that these four features learn alternative area unsupervisedly；
It is for initial alternative areaIn super-pixel point Si, SiEach pixel P for includingi=(xi,yi,zi,ui,vi)
Rgb value unified, utilize Gaussian parameter μcWithSelf study color characteristic, formula are as follows:
Utilize Gaussian parameter μlWithSelf study super-pixel SiLevel feature L (Si) formula are as follows:
Utilize Gaussian parameter μnWithSelf study super-pixel SiNormal direction measure feature N (Si) formula are as follows:
Define Sg (Si) it is across super-pixel SiRay ray quantity, self study super-pixel SiStrength characteristic Sg (Si) public affairs
Formula are as follows:
Finally, establishing Bayesian frame merges 4 kinds of features, formula is as follows:
Wherein, p (Si=R | Obs) indicate super-pixel SiBelong to the probability of road area, Obs indicates the sight based on these four features
It surveys.
2. the road drivable region detection method merged according to claim 1 based on monocular vision with laser radar,
Be characterized in that: the initial alternative area of road area is found in the method definition based on minimum filtering；Specific step is as follows:
Firstly, the initial alternative area for defining the road area isDefine IDRMFor " direction is penetrated
Line chart (direction ray map) ", and by point set Pi=(xi,yi,zi,ui,vi) (ui,vi) coordinate transformation to picture most
The central point P of bottom a linebaseUnder the polar coordinates of origin, then to haveIt is point setIt is very sub
Collection, wherein Indicate that belong to h angle at i-th point,It indicates
It isIn barrier point set, calculating process is as follows:
1) I is initializedDRMFor size full 0 matrix identical with picture is originally inputted；
2) for the set of all the points of h angleFind obstacle point set therein
If 3)Construct containerOtherwise,
4) willIt is included into IDRM；
If 5) h=H, terminate；Otherwise, 2 are returned；
Secondly, the problem of being " ray leakage ", carry out subsequent processing using the method for minimum filtering, obtain desired IDRM, finally
It obtains
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710283453.3A CN107167811B (en) | 2017-04-26 | 2017-04-26 | The road drivable region detection method merged based on monocular vision with laser radar |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710283453.3A CN107167811B (en) | 2017-04-26 | 2017-04-26 | The road drivable region detection method merged based on monocular vision with laser radar |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107167811A CN107167811A (en) | 2017-09-15 |
| CN107167811B true CN107167811B (en) | 2019-05-03 |
Family
ID=59813240
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710283453.3A Active CN107167811B (en) | 2017-04-26 | 2017-04-26 | The road drivable region detection method merged based on monocular vision with laser radar |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107167811B (en) |
Families Citing this family (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107992850B (en) * | 2017-12-20 | 2020-01-14 | 大连理工大学 | Outdoor scene three-dimensional color point cloud classification method |
| CN108519773B (en) * | 2018-03-07 | 2020-01-14 | 西安交通大学 | Path planning method for unmanned vehicle in structured environment |
| CN108509918B (en) * | 2018-04-03 | 2021-01-08 | 中国人民解放军国防科技大学 | Target detection and tracking method fusing laser point cloud and image |
| CN108932475B (en) * | 2018-05-31 | 2021-11-16 | 中国科学院西安光学精密机械研究所 | Three-dimensional target identification system and method based on laser radar and monocular vision |
| CN110738223B (en) * | 2018-07-18 | 2022-04-08 | 宇通客车股份有限公司 | Point cloud data clustering method and device of laser radar |
| FR3085022B1 (en) * | 2018-08-16 | 2020-10-16 | Psa Automobiles Sa | PROCESS FOR DETERMINING A TRUST INDEX ASSOCIATED WITH AN OBJECT DETECTED BY A SENSOR IN THE ENVIRONMENT OF A MOTOR VEHICLE. |
| CN109358335B (en) * | 2018-09-11 | 2023-05-09 | 北京理工大学 | Range finder combining solid-state area array laser radar and double CCD cameras |
| CN109239727B (en) * | 2018-09-11 | 2022-08-05 | 北京理工大学 | Distance measurement method combining solid-state area array laser radar and double CCD cameras |
| CN109444911B (en) * | 2018-10-18 | 2023-05-05 | 哈尔滨工程大学 | Unmanned ship water surface target detection, identification and positioning method based on monocular camera and laser radar information fusion |
| DK180774B1 (en) * | 2018-10-29 | 2022-03-04 | Motional Ad Llc | Automatic annotation of environmental features in a map during navigation of a vehicle |
| CN109543600A (en) * | 2018-11-21 | 2019-03-29 | 成都信息工程大学 | A kind of realization drivable region detection method and system and application |
| CN109858460B (en) * | 2019-02-20 | 2022-06-10 | 重庆邮电大学 | Lane line detection method based on three-dimensional laser radar |
| CN109696173A (en) * | 2019-02-20 | 2019-04-30 | 苏州风图智能科技有限公司 | A kind of car body air navigation aid and device |
| CN109917419B (en) * | 2019-04-12 | 2021-04-13 | 中山大学 | Depth filling dense system and method based on laser radar and image |
| CN110378196B (en) * | 2019-05-29 | 2022-08-02 | 电子科技大学 | Road visual detection method combining laser point cloud data |
| CN110488320B (en) * | 2019-08-23 | 2023-02-03 | 南京邮电大学 | Method for detecting vehicle distance by using stereoscopic vision |
| CN110781720B (en) * | 2019-09-05 | 2022-08-19 | 国网江苏省电力有限公司 | Object identification method based on image processing and multi-sensor fusion |
| CN111582280B (en) * | 2020-05-11 | 2023-10-17 | 吉林省森祥科技有限公司 | Data deep fusion image segmentation method for multispectral rescue robot |
| CN111898687B (en) * | 2020-08-03 | 2021-07-02 | 成都信息工程大学 | Radar reflectivity data fusion method based on Dilongnie triangulation |
| CN112633326B (en) * | 2020-11-30 | 2022-04-29 | 电子科技大学 | Unmanned aerial vehicle target detection method based on Bayesian multi-source fusion |
| CN112749662B (en) * | 2021-01-14 | 2022-08-05 | 东南大学 | Method for extracting travelable area in unstructured environment based on laser radar |
| CN113284163B (en) * | 2021-05-12 | 2023-04-07 | 西安交通大学 | Three-dimensional target self-adaptive detection method and system based on vehicle-mounted laser radar point cloud |
| CN115984583B (en) * | 2022-12-30 | 2024-02-02 | 广州沃芽科技有限公司 | Data processing method, apparatus, computer device, storage medium, and program product |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103645480A (en) * | 2013-12-04 | 2014-03-19 | 北京理工大学 | Geographic and geomorphic characteristic construction method based on laser radar and image data fusion |
| CN105989334A (en) * | 2015-02-12 | 2016-10-05 | 中国科学院西安光学精密机械研究所 | Monocular vision-based road detection method |
| CN106529417A (en) * | 2016-10-17 | 2017-03-22 | 北海益生源农贸有限责任公司 | Visual and laser data integrated road detection method |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6753876B2 (en) * | 2001-12-21 | 2004-06-22 | General Electric Company | Method for high dynamic range image construction based on multiple images with multiple illumination intensities |
| CN103760569B (en) * | 2013-12-31 | 2016-03-30 | 西安交通大学 | A kind of drivable region detection method based on laser radar |
| CN104569998B (en) * | 2015-01-27 | 2017-06-20 | 长春理工大学 | The detection method and device in the vehicle safe driving region based on laser radar |
-
2017
- 2017-04-26 CN CN201710283453.3A patent/CN107167811B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103645480A (en) * | 2013-12-04 | 2014-03-19 | 北京理工大学 | Geographic and geomorphic characteristic construction method based on laser radar and image data fusion |
| CN105989334A (en) * | 2015-02-12 | 2016-10-05 | 中国科学院西安光学精密机械研究所 | Monocular vision-based road detection method |
| CN106529417A (en) * | 2016-10-17 | 2017-03-22 | 北海益生源农贸有限责任公司 | Visual and laser data integrated road detection method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107167811A (en) | 2017-09-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107167811B (en) | The road drivable region detection method merged based on monocular vision with laser radar | |
| Caraffi et al. | Off-road path and obstacle detection using decision networks and stereo vision | |
| Zhou et al. | Self‐supervised learning to visually detect terrain surfaces for autonomous robots operating in forested terrain | |
| Sun et al. | Aerial 3D building detection and modeling from airborne LiDAR point clouds | |
| Kühnl et al. | Monocular road segmentation using slow feature analysis | |
| CN110956651A (en) | Terrain semantic perception method based on fusion of vision and vibrotactile sense | |
| Alaba et al. | Deep learning-based image 3-d object detection for autonomous driving | |
| CN107491071B (en) | Intelligent multi-robot cooperative mapping system and method thereof | |
| CN103632167B (en) | Monocular vision space recognition method under class ground gravitational field environment | |
| CN114724120B (en) | Vehicle target detection method and system based on radar vision semantic segmentation adaptive fusion | |
| CN104850850A (en) | Binocular stereoscopic vision image feature extraction method combining shape and color | |
| CN110675415B (en) | Road ponding area detection method based on deep learning enhanced example segmentation | |
| CN105869178A (en) | Method for unsupervised segmentation of complex targets from dynamic scene based on multi-scale combination feature convex optimization | |
| Gao et al. | Fine-grained off-road semantic segmentation and mapping via contrastive learning | |
| CN106529417A (en) | Visual and laser data integrated road detection method | |
| CN111914615A (en) | Fire-fighting area passability analysis system based on stereoscopic vision | |
| CN103198475A (en) | Full-focus synthetic aperture perspective imaging method based on multilevel iteration visualization optimization | |
| CN114782729A (en) | Real-time target detection method based on laser radar and vision fusion | |
| Wang et al. | Multi-cue road boundary detection using stereo vision | |
| Yan et al. | Sparse semantic map building and relocalization for UGV using 3D point clouds in outdoor environments | |
| CN116597122A (en) | Data labeling method, device, electronic equipment and storage medium | |
| Laupheimer et al. | The importance of radiometric feature quality for semantic mesh segmentation | |
| Sanberg et al. | Extending the stixel world with online self-supervised color modeling for road-versus-obstacle segmentation | |
| CN105160324B (en) | A kind of vehicle checking method based on space of components relationship | |
| CN103646397A (en) | Real-time synthetic aperture perspective imaging method based on multi-source data fusion |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |