CN107248171A - A kind of monocular vision odometer yardstick restoration methods based on triangulation - Google Patents
A kind of monocular vision odometer yardstick restoration methods based on triangulation Download PDFInfo
- Publication number
- CN107248171A CN107248171A CN201710346708.6A CN201710346708A CN107248171A CN 107248171 A CN107248171 A CN 107248171A CN 201710346708 A CN201710346708 A CN 201710346708A CN 107248171 A CN107248171 A CN 107248171A
- Authority
- CN
- China
- Prior art keywords
- triangle
- mrow
- yardstick
- road surface
- monocular vision
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/50—Depth or shape recovery
- G06T7/55—Depth or shape recovery from multiple images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30248—Vehicle exterior or interior
- G06T2207/30252—Vehicle exterior; Vicinity of vehicle
- G06T2207/30256—Lane; Road marking
Abstract
The present invention relates to a kind of monocular vision odometer yardstick restoration methods based on triangulation, comprise the following steps：1) the ith feature point that two frames have been matched before and after obtainingWithAnd corresponding 3d space coordinate；2) estimated values theta at road surface inclination angle is calculated*；3) position (u by the characteristic point matched in latter two field picture1,v1) carry out triangulation；4) according to the 3d space coordinate of each characteristic point, the normal vector and height of the space plane where each triangle of calculating；5) pitching angle theta of the space plane where calculating trianglej, screen triangle；6) the height h on road surface*；7) by camera true altitude h and the height h on road surface*The scale parameter s of monocular vision odometer is obtained as business, recovers visual odometry absolute measure.Compared with prior art, the present invention has the advantages that speed is fast, accuracy rate is high, need not do image segmentation and road surface identification, strong robustness.
Description
Technical field
The present invention relates to a kind of yardstick restoration methods, more particularly, to a kind of monocular vision odometer based on triangulation
Yardstick restoration methods.
Background technology
The conventional method for calculating monocular vision odometer yardstick is broadly divided into two kinds at present, and one kind needs other sensings of fusion
Device such as laser radar or inertial sensor, another is to consider the known yardstick in environment.Former approach can be more smart
Accurate reply monocular yardstick, but use cost is added, while needing accurately to demarcate before use；The conventional ring of later approach
Known yardstick is generally camera to the height on road surface in border, is then recovered by calculating the geometrical model on road surface.Conventional meter
The method for calculating road surface geometrical model is broadly divided into two kinds, and a kind of is the 3D positions by characteristic point, is proposed with RANSAC method
Noise proposes noise dependent on road surface identification, and another is by single by minimizing on the premise of known road surface region
Answering property error calculates road surface geometrical model；Former approach often due to characteristic point is less on road surface accurately to calculate,
Later approach is higher, it is necessary to substantial amounts of calculating time or computing resource due to algorithm computation complexity.
The content of the invention
It is an object of the present invention to overcome the above-mentioned drawbacks of the prior art and provide a kind of speed is fast, accuracy rate
It is high, image segmentation and road surface identification, the monocular vision odometer yardstick recovery side based on triangulation of strong robustness need not be done
Method.
The purpose of the present invention can be achieved through the following technical solutions：
A kind of monocular vision odometer yardstick restoration methods based on triangulation, comprise the following steps：
1) the ith feature point that two frames have been matched before and after being obtained according to monocular vision odometerWithAnd it is right
The 3d space coordinate answered；
2) the spin matrix R that is obtained according to the monocular vision odometer and motion vector t for not possessing true yardstick*Calculate
The estimated values theta at road surface inclination angle*；
3) position (u by the characteristic point matched in latter two field picture1,v1) triangulation is carried out, met
The triangle sets Tri of condition；
4) according to the 3d space coordinate of each characteristic point, calculate space plane where each triangle normal vector and
Highly；
5) space plane where the normal vector of space plane according to where each triangle calculates triangle is bowed
Elevation angle thetaj, screen triangle；
6) the height h on road surface is obtained using RANSAC methods according to the vertex of a triangle filtered out*；
7) by camera true altitude h and the height h on road surface*The scale parameter s of monocular vision odometer is obtained as business, will not
Possesses the motion vector t of true yardstick*It is multiplied with scale parameter s, you can recover visual odometry absolute measure.
Described step 4) in, the normal vector n of j-th of trianglejWith height hjExpression formula be：
nj T*Xji+hj=0 j=1,2 ..., m i=1,2,3
Wherein, XjiFor i-th of summit of j-th of triangle, njFor the normal vector of j-th of triangle, hjFor j-th three
Angular height.
The estimated values theta at described road surface inclination angle*Calculating formula be：
Wherein, t*Not possess the motion vector of true yardstick, t*[2] it is not possess the motion vector t of true yardstick*'s
Second value,For t*Mould it is long.
The pitching angle theta of space plane where described trianglejCalculating formula be：
Wherein, nj[2] it is triangulation method to vector njSecond value,For njMould it is long.
Described step 5) in, filter out the pitching angle theta of the space plane where trianglejWith the estimate at road surface inclination angle
θ*Difference be less than 5 degree when corresponding triangle.
Compared with prior art, the present invention has advantages below：
First, image segmentation and road surface identification need not be done, calculating characteristic point 3D coordinates according only to visual odometry calculates road
Face geometrical model, strong robustness；
2nd, road surface whether is belonged to using triangulation methodology judging characteristic point, speed is fast and accuracy rate is high.
Brief description of the drawings
Fig. 1 is method schematic of the invention.
Embodiment
The present invention is described in detail with specific embodiment below in conjunction with the accompanying drawings.
Embodiment
As shown in figure 1, the calculation procedure of the present invention is as follows：
1) the ith feature point that two frames have been matched before and after being obtained according to monocular vision odometerWithAnd it is right
The 3d space coordinate answered；
2) the spin matrix R and motion vector t obtained according to monocular vision odometer*Calculate the estimate at road surface inclination angle
θ*(calculation formula sees below)；
3) position (u by the characteristic point matched in latter two field picture1,v1) triangulation is carried out, met
The triangle sets Tri of condition；
4) according to the 3d space coordinate of each characteristic point, calculate space plane where each triangle normal vector and
Highly；
5) according to the estimated values theta at road surface inclination angle*And by each triangle by normal vector calculate triangle belong to road surface
Inclination angle screening triangle, retain triangle of the error within 5 degree；
6) using step 5) in qualified vertex of a triangle using RANSAC methods obtain road surface estimation model and
Meet the characteristic point of road surface model
7) according to the geometrical model for selecting the corresponding characteristic point acquisition road surface of triangle, and by camera true altitude with calculating
Height out, which is divided by, obtains the scale parameter of monocular vision odometer, by this motion vector t*Multiply this this parameter, recovery
Recover visual odometry absolute measure.
This method describes in detail as follows：
Extracted first with prior art (ORBSLAM, LIBVISO etc.) and characteristic point in matching camera two field picturesWithWherein i=1,2 ..., n, for the number of the characteristic point matched, then calculate pose of the camera between front and rear two frame
Change R, t*, wherein R is spin matrix, t*For motion vector, because monocular vision have lost depth information, calculate herein
Displacement do not possess real yardstick, the variable of all relative scalars has superscript * in text.Then obtained by triangulation
The 3D coordinates of the characteristic point of all successful match
Using Triangulation Technique by the characteristic point in latter two field pictureIt is divided into legal triangle
Group.
For each triangle, according to the 3D coordinates for each point obtained before, each three are obtained according to equation below
The normal vector of angular place plane
nj T*Xji+hj=0 j=1,2 ..., m i=1,2,3
Wherein m is the number of triangle sets intermediate cam shape, and j is triangle sequence number, and i is the sequence number on three summits in triangle
njFor the normal direction of j-th of triangle, hjFor the height of j-th of triangle, for each triangle,
The mould for ignoring normal direction is long,
Try to achieve njAfterwards, the angle of pitch of this plane in camera coordinates system can just be obtained
Ground can be estimated according to the monocular VO spin matrix R and motion vector t for calculating obtained camera motion relative
In the angle of pitch of camera.When running car is on level road, the angle of pitch for the displacement that VO is calculated substantially is exactly road
The angle of pitch, it is as shown below.Level road herein refers to automobile and the road of pitch orientation rotation does not occur, including meets the slope of this condition
Road.Computational methods are as follows：
First according to spin matrix R, calculate whether automobile has pitching
θR=arctan (- R32/R33)
Wherein RijThe element of R midsole i rows jth row, if θRLess than 5 degree, then it is assumed that automobile does not have elevating movement, according to following
Formula calculates the road surface angle of pitch, if θRLess than 5 degree, then the road surface angle of pitch is set to the angle of pitch of last moment calculating
Compare the plane angle of pitch and the road angle of pitch, differ smaller, then illustrate that probability of this plane on road is bigger.
Then consider the centre-height of each triangle, all points above camera are rejected first, in remaining point, select road
The plane angle of pitch and the road angle of pitch differ the triangle within 5 degree.As a result probably as shown below, green area is estimation road
Face region, then belongs to vertex of a triangle using RANSAC methods according to road and obtains road surface normal direction and height, RANSAC methods
Concretely comprise the following steps：
Three summits are randomly selected in the triangular apex after screening, and its place is calculated using these three summits and are put down
The geometrical model in face：
NX+h=0
(computational methods are consistent with calculating triangle geometry model method above), then calculates all in addition to this at 3 points
Triangular apex is to this plane apart from d=| nXi+ h |, the point that distance is less than height 0.1h is considered interior point, and counts interior
Point quantity；Then this step is repeated, is randomly selected repeatedly, it is the final geometry mould in road surface to take the most road surface geometrical model of interior point
Type.
Finally true altitude is compared with the height in this geometrical model and obtains true yardstick, i.e.,：
T=st*
Wherein h is camera true altitude, h*The pavement-height (i.e. camera heights) estimated for algorithm, t*For without yardstick
Motion vector, t*To recover the motion vector after yardstick.
Claims (5)
1. a kind of monocular vision odometer yardstick restoration methods based on triangulation, it is characterised in that comprise the following steps：
1) the ith feature point that two frames have been matched before and after being obtained according to monocular vision odometerWithAnd it is corresponding
3d space coordinate；
2) the spin matrix R that is obtained according to the monocular vision odometer and motion vector t for not possessing true yardstick*Calculate road surface
The estimated values theta at inclination angle*；
3) position (u by the characteristic point matched in latter two field picture1,v1) triangulation is carried out, obtain eligible
Triangle sets Tri；
4) according to the 3d space coordinate of each characteristic point, the normal vector and height of the space plane where each triangle of calculating
Degree；
5) angle of pitch of the space plane where the normal vector calculating triangle of the space plane according to where each triangle
θj, screen triangle；
6) the height h on road surface is obtained using RANSAC methods according to the vertex of a triangle filtered out*；
7) by camera true altitude h and the height h on road surface*The scale parameter s of monocular vision odometer is obtained as business, will not be possessed
The motion vector t of true yardstick*It is multiplied with scale parameter s, you can recover visual odometry absolute measure.
2. a kind of monocular vision odometer yardstick restoration methods based on triangulation according to claim 1, its feature
It is, described step 4) in, the normal vector n of j-th of trianglejWith height hjExpression formula be：
nj T*Xji+hj=0 j=1,2 ..., m i=1,2,3
Wherein, XjiFor i-th of summit of j-th of triangle, njFor the normal vector of j-th of triangle, hjFor j-th of triangle
Height.
3. a kind of monocular vision odometer yardstick restoration methods based on triangulation according to claim 1, its feature
It is, the estimated values theta at described road surface inclination angle*Calculating formula be：
<mrow>
<msup>
<mi>&theta;</mi>
<mo>*</mo>
</msup>
<mo>=</mo>
<mi>a</mi>
<mi>r</mi>
<mi>c</mi>
<mi>s</mi>
<mi>i</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<msup>
<mi>t</mi>
<mo>*</mo>
</msup>
<mo>&lsqb;</mo>
<mn>2</mn>
<mo>&rsqb;</mo>
<mo>/</mo>
<msqrt>
<mrow>
<msup>
<mi>t</mi>
<mrow>
<mo>*</mo>
<mi>T</mi>
</mrow>
</msup>
<msup>
<mi>t</mi>
<mo>*</mo>
</msup>
</mrow>
</msqrt>
<mo>)</mo>
</mrow>
</mrow>
Wherein, t*Not possess the motion vector of true yardstick, t*[2] it is not possess the motion vector t of true yardstick*Second
Individual value,For t*Mould it is long.
4. a kind of monocular vision odometer yardstick restoration methods based on triangulation according to claim 2, its feature
It is, the pitching angle theta of the space plane where described trianglejCalculating formula be：
<mrow>
<msub>
<mi>n</mi>
<mi>j</mi>
</msub>
<mo>=</mo>
<mi>a</mi>
<mi>r</mi>
<mi>c</mi>
<mi>s</mi>
<mi>i</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>n</mi>
<mi>j</mi>
</msub>
<mo>&lsqb;</mo>
<mn>2</mn>
<mo>&rsqb;</mo>
<mo>/</mo>
<msqrt>
<mrow>
<msup>
<msub>
<mi>n</mi>
<mi>j</mi>
</msub>
<mi>T</mi>
</msup>
<msub>
<mi>n</mi>
<mi>j</mi>
</msub>
</mrow>
</msqrt>
<mo>)</mo>
</mrow>
</mrow>
Wherein, nj[2] it is triangulation method to vector njSecond value,For njMould it is long.
5. a kind of monocular vision odometer yardstick restoration methods based on triangulation according to claim 2, its feature
It is, described step 5) in, filter out the pitching angle theta of the space plane where trianglejWith the estimated values theta at road surface inclination angle*'s
Corresponding triangle when difference is less than 5 degree.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710346708.6A CN107248171B (en) | 2017-05-17 | 2017-05-17 | Triangulation-based monocular vision odometer scale recovery method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710346708.6A CN107248171B (en) | 2017-05-17 | 2017-05-17 | Triangulation-based monocular vision odometer scale recovery method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107248171A true CN107248171A (en) | 2017-10-13 |
| CN107248171B CN107248171B (en) | 2020-07-28 |
Family
ID=60016998
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710346708.6A Active CN107248171B (en) | 2017-05-17 | 2017-05-17 | Triangulation-based monocular vision odometer scale recovery method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107248171B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110388917A (en) * | 2018-04-23 | 2019-10-29 | 北京京东尚科信息技术有限公司 | Aircraft monocular vision Scale Estimation Method and device, aircraft guidance system and aircraft |
| CN117409071A (en) * | 2023-12-12 | 2024-01-16 | 知行汽车科技(苏州)股份有限公司 | Spatial scale recovery method, device, equipment and medium for monocular VSLAM |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100202657A1 (en) * | 2008-10-22 | 2010-08-12 | Garbis Salgian | System and method for object detection from a moving platform |
| CN102779340A (en) * | 2012-06-12 | 2012-11-14 | 华中科技大学 | Automatic corresponding method of feature point coordinates based on Delaunay triangulation |
| CN103093479A (en) * | 2013-03-01 | 2013-05-08 | 杭州电子科技大学 | Target positioning method based on binocular vision |
| CN103292804A (en) * | 2013-05-27 | 2013-09-11 | 浙江大学 | Monocular natural vision landmark assisted mobile robot positioning method |
| CN103542846A (en) * | 2012-07-16 | 2014-01-29 | 苏州科瓴精密机械科技有限公司 | Locating system and method of mobile robot |
| CN104180818A (en) * | 2014-08-12 | 2014-12-03 | 北京理工大学 | Monocular vision mileage calculating device |
| CN104766345A (en) * | 2015-04-08 | 2015-07-08 | 欧剑 | Body scanning and movement capturing method based on clothes feature points |
| CN105045263A (en) * | 2015-07-06 | 2015-11-11 | 杭州南江机器人股份有限公司 | Kinect-based robot self-positioning method |
| WO2016062996A1 (en) * | 2014-10-20 | 2016-04-28 | Bae Systems Plc | Apparatus and method for multi-camera visual odometry |
| CN105976402A (en) * | 2016-05-26 | 2016-09-28 | 同济大学 | Real scale obtaining method of monocular vision odometer |
| CN106326892A (en) * | 2016-08-01 | 2017-01-11 | 西南科技大学 | Visual landing pose estimation method of rotary wing type unmanned aerial vehicle |
| CN106406338A (en) * | 2016-04-14 | 2017-02-15 | 中山大学 | Omnidirectional mobile robot autonomous navigation apparatus and method based on laser range finder |
-
2017
- 2017-05-17 CN CN201710346708.6A patent/CN107248171B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100202657A1 (en) * | 2008-10-22 | 2010-08-12 | Garbis Salgian | System and method for object detection from a moving platform |
| CN102779340A (en) * | 2012-06-12 | 2012-11-14 | 华中科技大学 | Automatic corresponding method of feature point coordinates based on Delaunay triangulation |
| CN103542846A (en) * | 2012-07-16 | 2014-01-29 | 苏州科瓴精密机械科技有限公司 | Locating system and method of mobile robot |
| CN103093479A (en) * | 2013-03-01 | 2013-05-08 | 杭州电子科技大学 | Target positioning method based on binocular vision |
| CN103292804A (en) * | 2013-05-27 | 2013-09-11 | 浙江大学 | Monocular natural vision landmark assisted mobile robot positioning method |
| CN104180818A (en) * | 2014-08-12 | 2014-12-03 | 北京理工大学 | Monocular vision mileage calculating device |
| WO2016062996A1 (en) * | 2014-10-20 | 2016-04-28 | Bae Systems Plc | Apparatus and method for multi-camera visual odometry |
| CN104766345A (en) * | 2015-04-08 | 2015-07-08 | 欧剑 | Body scanning and movement capturing method based on clothes feature points |
| CN105045263A (en) * | 2015-07-06 | 2015-11-11 | 杭州南江机器人股份有限公司 | Kinect-based robot self-positioning method |
| CN106406338A (en) * | 2016-04-14 | 2017-02-15 | 中山大学 | Omnidirectional mobile robot autonomous navigation apparatus and method based on laser range finder |
| CN105976402A (en) * | 2016-05-26 | 2016-09-28 | 同济大学 | Real scale obtaining method of monocular vision odometer |
| CN106326892A (en) * | 2016-08-01 | 2017-01-11 | 西南科技大学 | Visual landing pose estimation method of rotary wing type unmanned aerial vehicle |
Non-Patent Citations (6)
| Title |
|---|
| DINGFU ZHOU 等: "Reliable scale estimation and correction for monocular Visual Odometry", 《2016IEEE INTERNATIONAL VEHICLES SYMPOSIUM》 * |
| SHIYU SONG 等: "High Accuracy Monocular SFM and Scale Correction for Autonomous Driving", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS YSIS AND MACHINE INTELLIGENCE》 * |
| 吴子章: "基于全景视觉的机器人SLAM方法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 孙作雷 等: "基于RANSAC的奇异值剔除的单目视觉里程计", 《上海海事大学学报》 * |
| 王彩玲 等: "基于掩模和俯仰角补偿的视觉自运动估计", 《系统仿真学报》 * |
| 袁小翠 等: "Delaunay三角剖分算法改进与对比分析", 《计算机应用与软件》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110388917A (en) * | 2018-04-23 | 2019-10-29 | 北京京东尚科信息技术有限公司 | Aircraft monocular vision Scale Estimation Method and device, aircraft guidance system and aircraft |
| CN117409071A (en) * | 2023-12-12 | 2024-01-16 | 知行汽车科技(苏州)股份有限公司 | Spatial scale recovery method, device, equipment and medium for monocular VSLAM |
| CN117409071B (en) * | 2023-12-12 | 2024-03-01 | 知行汽车科技(苏州)股份有限公司 | Spatial scale recovery method, device, equipment and medium for monocular VSLAM |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107248171B (en) | 2020-07-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2021004548A1 (en) | Vehicle speed intelligent measurement method based on binocular stereo vision system | |
| CN110223348B (en) | Robot scene self-adaptive pose estimation method based on RGB-D camera | |
| Zhuang et al. | Rolling-shutter-aware differential sfm and image rectification | |
| JP6211263B2 (en) | Road vertical contour detection | |
| CN110569704A (en) | Multi-strategy self-adaptive lane line detection method based on stereoscopic vision | |
| CN110910453B (en) | Vehicle pose estimation method and system based on non-overlapping view field multi-camera system | |
| CN111210477B (en) | Method and system for positioning moving object | |
| WO2013133129A1 (en) | Moving-object position/attitude estimation apparatus and method for estimating position/attitude of moving object | |
| CN105335955A (en) | Object detection method and object detection apparatus | |
| CN107636679A (en) | A kind of obstacle detection method and device | |
| CN105976402A (en) | Real scale obtaining method of monocular vision odometer | |
| CN112365549B (en) | Attitude correction method and device for vehicle-mounted camera, storage medium and electronic device | |
| CN111998862B (en) | BNN-based dense binocular SLAM method | |
| EP3876142A1 (en) | Map building method, apparatus and system, and storage medium | |
| CN113744315B (en) | Semi-direct vision odometer based on binocular vision | |
| CN103136525A (en) | Hetero-type expanded goal high-accuracy positioning method with generalized Hough transposition | |
| CN103458261A (en) | Video scene variation detection method based on stereoscopic vision | |
| CN109917359A (en) | Robust vehicle distances estimation method based on vehicle-mounted monocular vision | |
| CN103985154A (en) | Three-dimensional model reestablishment method based on global linear method | |
| CN114325634A (en) | Method for extracting passable area in high-robustness field environment based on laser radar | |
| CN108106613A (en) | The localization method and system of view-based access control model auxiliary | |
| CN101980292B (en) | Regular octagonal template-based board camera intrinsic parameter calibration method | |
| CN111443704A (en) | Obstacle positioning method and device for automatic driving system | |
| CN110992424B (en) | Positioning method and system based on binocular vision | |
| CN107248171A (en) | A kind of monocular vision odometer yardstick restoration methods based on triangulation |
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 |