CN109949355A - The method of half fan-shaped equidistant line model is established in monocular vision pedestrian's distance estimations - Google Patents
The method of half fan-shaped equidistant line model is established in monocular vision pedestrian's distance estimations Download PDFInfo
- Publication number
- CN109949355A CN109949355A CN201910194195.0A CN201910194195A CN109949355A CN 109949355 A CN109949355 A CN 109949355A CN 201910194195 A CN201910194195 A CN 201910194195A CN 109949355 A CN109949355 A CN 109949355A
- Authority
- CN
- China
- Prior art keywords
- image
- landmark point
- pedestrian
- half fan
- calibration
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Abstract
The method of half fan-shaped equidistant line model is established in monocular vision pedestrian's distance estimations, belong to the distance estimations field in computer vision application, it is fast in order to solve raising arithmetic speed, the problem of enhancing exploitativeness, landmark point is chosen in half fan-shaped distribution, obtains to calibration maps and carries out level correction and obtain correction image, further according to the principle displacement images that image pickup point pixel abscissa is constant, after calibration, landmark point location of pixels equal space line is about x=0.5lxSymmetrically, lx, lyRespectively picture traverse and length, left side landmark point directly utilize right side landmark point mapping to obtain.Effect is that use half fan-shaped equidistant line model be the strong real-time based on image information.
Description
Technical field
The invention belongs to the distance estimations fields in computer vision application, are a kind of using vehicle-mounted monocular specifically
Video camera is as headend equipment, the foot position stood using pedestrian, extrapolates pedestrian and list according to people-video camera equal space line
The method of distance between lens camera.
Background technique
With the rapid development of intelligent automobile, the correlative study of various automobile assistant driving system modules is also constantly complete
It is kind, and pedestrian's distance estimations module also increasingly obtains the concern of people as the important component that auxiliary drives.
Method for estimating distance can be mainly divided into the distance estimations based on visible light, the distance estimations based on infrared light at present
And distance estimations are carried out by various distance mearuring equipments.The above method for estimating distance leads to using image as the main source of information
The depth excavation to image information is crossed, the distance between target pedestrian and video camera result are obtained.
In method for estimating distance now, high-precision hardware is largely depended on unduly, improve the overall cost of system,
It is difficult to be widely popularized.And to the lower method for estimating distance of hardware requirement, largely there is a problem of that calculation amount is excessive,
The application requirement of Autonomous Vehicle is unable to reach in terms of real-time.Therefore, Autonomous Vehicle DAS (Driver Assistant System) is badly in need of a kind of calculating speed
Spend fast, the lower method for estimating distance of cost.
In the environment of actual vehicle operation, vehicle and pedestrian timely obtain pedestrian and vehicle all in motion state
Accurate distance, driver can be helped tentatively to prejudge the degree of danger of pedestrian, to avoid dangerous generation.China
Part traffic condition now mixes the mode of row still in people's vehicle, locating for pedestrian in the present context is compared to vehicle
State is more dangerous, therefore is the problem that can not avoid in vehicle-mounted DAS (Driver Assistant System) to the protection of pedestrains safety.
In general, usually guarantee distance estimations system by improving the requirement to hardware or enhancing the complexity of algorithm
Accuracy and precision.In patent " pedestrian detection and distance measuring method based on stereoscopic vision " (publication number: CN106952274A)
It proposes to obtain left image and right image that stereoscopic vision constructs scene, the predicted value of left image is calculated using convolutional neural networks,
Matching cost is calculated after matching right image predicted value, matching cost reckling is optimal parallax value, to obtain distance.Patent
" a kind of robot location algorithm based on Multi-sensor Fusion " (publication number: CN106226774A) merge ultrasonic measurement with
Infrared survey value measures the distance between barrier and robot.By dividing distance level scale and speed class, according to
Ultrasonic measurement and infrared survey value weight are calculated according to fuzzy control rule, to obtain distance.
At home in document, it can be seen that a variety of methods by image information acquisition distance.Generally use dependence hardware
The good characteristic of itself improves the precision of algorithm by increasing algorithm complexity.The mode of hardware is relied on by lifting system
Cost, it is difficult to be widely applied；The method for increasing algorithm complexity then extends operation time, is unable to satisfy in DAS (Driver Assistant System)
Requirement to pedestrian's distance estimations real-time.
Summary of the invention
In order to solve the problem of to improve the fast enhancing exploitativeness of arithmetic speed, the present invention passes through half fan-shaped equidistant line model
Chinese herbaceous peony pedestrian distance is obtained, this method arithmetic speed can be enhanced, and enhance the exploitativeness in Distance Judgment.On reaching
Purpose is stated, the following technical solutions are proposed by the present invention:
A kind of method that half fan-shaped equidistant line model is established in monocular vision pedestrian distance estimations, video camera is to pedestrian image
It obtains, after carrying out level correction to image, half fan-shaped equidistant line model is established with image acceptance of the bid fixed position, in acquisition calibration maps
In the process, using video camera position D as origin, rjFor radius, j=1,2,3 ..., n establishes calibration range, rjIt is greater than
Equal to distance estimations range maximum value, i once linear equation is established within the scope of terrestrial reference, so that landmark point (xm,ym) meet and close
System
ym=bi·xm, i=1,2,3 ..., n (1)
Wherein biFor linear equation coefficient of first order, m=1,2,3 ..., n, m indicate that the label of landmark point, n indicate ground
Punctuate quantity；
Landmark point is chosen in half fan-shaped distribution, obtains to calibration maps and carries out level correction and obtain correction image, further according to
The constant principle displacement images of image pickup point pixel abscissa, after calibration, landmark point location of pixels equal space line is about x=
0.5·lxSymmetrically, lx, lyRespectively picture traverse and length, left side landmark point directly utilize right side landmark point mapping to obtain.
The utility model has the advantages that
Present invention only requires images as the input information of basic calculation, by obtaining target pedestrian foot in the picture
Location of pixels can be widely applied to Autonomous Vehicle auxiliary and drive to judge the distance between pedestrian and video camera
Field.The present invention is low in cost and be easily achieved only using monocular-camera as image capture device；Obtain measured image
Afterwards, by having algorithm to the calibration on image progress angle and position.Image can accurately reflect image information after calibration, really
Protect the accuracy of half fan-shaped equidistant line model；Half fan-shaped equidistant line model is based on image information, by inputting foot part
Location of pixels obtains distance results, the model strong real-time, can be avoided and causes arithmetic speed slow because method is excessively complicated, nothing
Method reaches the problem of engineering demand；Body part is pedestrian's inherently sign, is carried out by the information extraction to body part
Distance estimations do not cause added burden for mobile pedestrian, with its superior simplicity be suitable in mobile device for
The distance estimations of pedestrian.The present invention only relies on monocular-camera in hardware aspect and carries out image acquisition, quickly establishes deployment, is applicable in
Industry is assembled in batch production and fast custom vehicle.By the fan-shaped equidistant line model of multiple repairing weld building half, can quickly estimate
Chinese herbaceous peony pedestrian's distance, the half fan-shaped equidistant line model energy in the case where camera height, angle remain unchanged, after accurate foundation
The distance estimations being enough in a variety of Different Planes.It is half-and-half fan-shaped under line after fixed vehicle-mounted vidicon when applied in vehicle
Equidistant line model is demarcated, and can permanently be used.It is driven for intelligence auxiliary and necessary pedestrian's estimated capacity is provided.Calculating essence
On degree, the valid data of protection pedestrains safety can be provided for driver by control errors in 1m within the scope of 30m.
The present invention obtains image by monocular cam, and is corrected using existing algorithm to image, avoids image
Foot's location of pixels is caused to obtain the problem of inaccuracy because of the offset in angle and position；The foundation of half fan-shaped equidistant line model
The acquisition of foot part location of pixels is relied primarily on, quickly and effectively Chinese herbaceous peony pedestrian's distance can be estimated.
The present invention is suitable for Autonomous Vehicle and assists driving field, by half fan-shaped equidistant line model obtain Chinese herbaceous peony pedestrian away from
From arithmetic speed is fast, and exploitativeness is strong.Image is obtained using monocular-camera as hardware foundation, it is easy to accomplish, it is assisting
Driving field has extremely strong promotional value；Half fan-shaped equidistant line model repeatedly obtains pedestrian foot location of pixels by original image
It establishes, the fixed camera of height equal angular can be used same half model estimated distance, be customized for vehicle, complete mould
Distance can be obtained without secondary change after type building；Partly fan-shaped equal space line model method is easy, calculates quickly, can satisfy certainly
Requirement of the master vehicle to pedestrian's Distance Judgment real-time.
Detailed description of the invention
Fig. 1 is the principle logic chart using the method for estimating distance based on pedestrian's foot position；
Fig. 2 is to utilize the result after mark point calibration for initial pictures；
Fig. 3 is for half fan-shaped equal space line model foundation schematic diagram；
Fig. 4 is the testing result figure of embodiment 1；
Fig. 5 is the testing result figure of embodiment 2；
Fig. 6 is a testing result figure.
Specific embodiment
Present invention is further described in detail with specific embodiment with reference to the accompanying drawing:
A kind of principle logic chart of equal space line monocular vision pedestrian's method for estimating distance is as shown in Figure 1, this method is specifically real
Apply that steps are as follows:
Step 1: calibration image is obtained by existing algorithm；
Step 2: half fan-shaped equidistant line model is constructed using image after calibration；
Step 3: using monocular cam as image capture device, the foot with Chinese herbaceous peony target pedestrian in range images is obtained
Pixel coordinate；
Step 4: foot's pixel coordinate of target pedestrian is input in half fan-shaped equidistant line model, model estimation is obtained
Distance.
The concrete scheme that the present invention is implemented is as follows:
The present invention proposes the method for estimating distance based on pedestrian's foot position, the two dimension obtained mainly for monocular-camera
Flat image carries out relevant calculation processing, is a kind of method by pedestrian's foot position acquisition of information target pedestrian's distance.This
Method complexity is low, strong real-time, obtains information by two-dimensional image and carries out distance estimations, and it is convenient that information source obtains,
Cost is low, convenient for promoting in auxiliary driving field, carries out distance estimations by body part, auxiliary without other external devices
Progress is helped, the distance estimations to mobile pedestrian are extremely conducive to.
Emphasis of the present invention is by pedestrian foot location of pixels in testing image come between target pedestrian and video camera
Distance carries out estimation operation.Image acquisition is carried out by monocular-camera, horizontal school is carried out to image according to existing other methods
After just, half fan-shaped equidistant line model is established using figure acceptance of the bid fixed position.When testing image inputs again, by obtaining foot
Part location of pixels coordinate, that is, can reach the purpose of the distance between determining pedestrian and video camera.Principle logic chart is as shown in Figure 1.
Image calibration angle is determined that image can be half fan-shaped equal space line mould after calibration by the deviation angle of camera in the horizontal direction
The levels of precision of type lays the foundation, and the two has codetermined the accuracy of distance estimations result.During acquiring calibration maps, take the photograph
Camera obtains pedestrian image, after carrying out level correction to image, establishes half fan-shaped equal space line mould with image acceptance of the bid fixed position
Type, during acquiring calibration maps, using video camera position D as origin, rjFor radius, j=1,2,3 ..., n is established
Calibration range, rjMore than or equal to distance estimations range maximum value, i once linear equation is established within the scope of terrestrial reference, so that ground
Punctuate (xm,ym) meet relationship
ym=bi·xm, i=1,2,3 ..., n (1)
Wherein biFor linear equation coefficient of first order, m=1,2,3 ..., n, m indicate that the label of landmark point, n indicate ground
Punctuate quantity；
Landmark point is chosen should be in half fan-shaped distribution, as shown in Figure 3.It obtains to calibration maps and carries out level correction and corrected
Image, further according to the principle displacement images that image pickup point pixel abscissa is constant.Signal such as Fig. 2 institute of image calibration operation
Show.After calibration, landmark point location of pixels equal space line is about x=0.5lxSymmetrically, lx, lyRespectively picture traverse and length, it is left
Side landmark point directly utilizes right side landmark point mapping to obtain.I equal space line is established by curve matching, establishes half fan-shaped equal space line
Model, model foundation schematic diagram is as shown in figure 3, dash area landmark point is that mapping obtains landmark point.
Further, formula is carried out by equal space line fitting according to landmark point location of pixels coordinate, is fitted equal space line equation are as follows:
Wherein, k1j, k2jAnd k3jFor curve matching coefficient, xjAnd yjFor image landmark point location of pixels coordinate, as ten in Fig. 3
Shown in word point mark.Set pedestrian's bipod arch of the foot central point P (xP,yP) it is foot part coordinate points, and xP∈xj,yP∈yj.Meanwhile
There are one to cross originWith the once linear function of P point:
yj=kxj, j=1,2,3 ..., n (3)
By formula (2) and (3) Simultaneous Equations, the intersecting point coordinate of straight line OP and whole equal space lines are found out are as follows:
Further, pixel distance point set of pedestrian's foot parts P point to origin O are as follows:
S={ sj|sj(xj,yj)}(6)
To gained intersection point point set S, i.e. straight line OP and equal space line whole intersecting point coordinate.Specifically, straight line OP and equal space line
Whole intersecting point coordinates are represented by whole points on OP, and OP indicates that origin O and pedestrian's foot part P institute is at an arbitrary position in image
Line, arbitrary point can be included in straight line OP in image.So arbitrary point is all straight line OP and equal space line in image
Intersection point, as point set S.
Each element application formula (7) Lai Jinhang Lagrange's interpolation calculates, and H (s) is interpolation polynomial, can be derived that opposite
The pedestrian target range estimation E answered, whereinFor interpolation coefficient.
Embodiment 1:
This embodiment example is to be shot using vehicle-mounted vidicon (3000*4000@30Hz) to pedestrian on street in spring
Simulation example, target line people's quantity are 4, and pedestrian shows to walk towards camera lens respectively, walk and cross camera lens away from camera lens
Preceding state.Through present invention application in the distance estimations of target pedestrian, processing result as shown, true value and estimated value it
Between error be no more than 1m.
Embodiment 2:
This embodiment example is to be shot using vehicle-mounted vidicon (3000*4000@30Hz) to pedestrian on street in spring
Simulation example, target line people's quantity be 5, pedestrian shows to bow respectively to play mobile phone, towards camera lens walk, cross zebra stripes and
The state gone in group.Through the present invention application in the distance estimations of target pedestrian, processing result as shown, true value with estimate
Error between evaluation is no more than 1m.
Claims (1)
1. establishing the method for half fan-shaped equidistant line model in a kind of monocular vision pedestrian distance estimations, which is characterized in that
Video camera obtains pedestrian image, after carrying out level correction to image, establishes half sector etc. with image acceptance of the bid fixed position
Away from line model, during acquiring calibration maps, using video camera position D as origin, rjFor radius, j=1,2,3 ...,
N establishes calibration range, rjMore than or equal to distance estimations range maximum value, i once linear equation is established within the scope of terrestrial reference,
So that landmark point (xm,ym) meet relationship
ym=bi·xm, i=1,2,3 ..., n (1)
Wherein biFor linear equation coefficient of first order, m=1,2,3 ..., n, m indicate that the label of landmark point, n indicate terrestrial reference points
Amount；
Landmark point is chosen in half fan-shaped distribution, obtains to calibration maps and carries out level correction and obtain correction image, further according to image
The constant principle displacement images of camera point pixel abscissa, after calibration, landmark point location of pixels equal space line is about x=0.5lx
Symmetrically, lx, lyRespectively picture traverse and length, left side landmark point directly utilize right side landmark point mapping to obtain.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910194195.0A CN109949355A (en) | 2019-03-14 | 2019-03-14 | The method of half fan-shaped equidistant line model is established in monocular vision pedestrian's distance estimations |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910194195.0A CN109949355A (en) | 2019-03-14 | 2019-03-14 | The method of half fan-shaped equidistant line model is established in monocular vision pedestrian's distance estimations |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109949355A true CN109949355A (en) | 2019-06-28 |
Family
ID=67008900
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910194195.0A Pending CN109949355A (en) | 2019-03-14 | 2019-03-14 | The method of half fan-shaped equidistant line model is established in monocular vision pedestrian's distance estimations |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109949355A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111366917A (en) * | 2020-03-13 | 2020-07-03 | 北京百度网讯科技有限公司 | Method, device and equipment for detecting travelable area and computer readable storage medium |
| CN111380503A (en) * | 2020-05-29 | 2020-07-07 | 电子科技大学 | Monocular camera ranging method adopting laser-assisted calibration |
| CN112033351A (en) * | 2020-08-05 | 2020-12-04 | 青岛聚好联科技有限公司 | Monocular camera-based distance measuring method and electronic equipment |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102096803A (en) * | 2010-11-29 | 2011-06-15 | 吉林大学 | Safe state recognition system for people on basis of machine vision |
| CN106909929A (en) * | 2015-12-22 | 2017-06-30 | 比亚迪股份有限公司 | Pedestrian's distance detection method and device |
-
2019
- 2019-03-14 CN CN201910194195.0A patent/CN109949355A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102096803A (en) * | 2010-11-29 | 2011-06-15 | 吉林大学 | Safe state recognition system for people on basis of machine vision |
| CN106909929A (en) * | 2015-12-22 | 2017-06-30 | 比亚迪股份有限公司 | Pedestrian's distance detection method and device |
Non-Patent Citations (1)
| Title |
|---|
| 程凡等: ""单目视觉行人身体多部件距离融合估计算法"", 《大连民族大学学报》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111366917A (en) * | 2020-03-13 | 2020-07-03 | 北京百度网讯科技有限公司 | Method, device and equipment for detecting travelable area and computer readable storage medium |
| CN111366917B (en) * | 2020-03-13 | 2022-07-15 | 北京百度网讯科技有限公司 | Method, device and equipment for detecting travelable area and computer readable storage medium |
| CN111380503A (en) * | 2020-05-29 | 2020-07-07 | 电子科技大学 | Monocular camera ranging method adopting laser-assisted calibration |
| CN112033351A (en) * | 2020-08-05 | 2020-12-04 | 青岛聚好联科技有限公司 | Monocular camera-based distance measuring method and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109270534B (en) | Intelligent vehicle laser sensor and camera online calibration method | |
| CN110031829B (en) | Target accurate distance measurement method based on monocular vision | |
| CN110567469B (en) | Visual positioning method and device, electronic equipment and system | |
| CN107704821B (en) | Vehicle pose calculation method for curve | |
| WO2018196391A1 (en) | Method and device for calibrating external parameters of vehicle-mounted camera | |
| TW202001924A (en) | Systems and methods for updating a high-resolution map based on binocular images | |
| WO2012043045A1 (en) | Image processing device and image capturing device using same | |
| CN102519434B (en) | Test verification method for measuring precision of stereoscopic vision three-dimensional recovery data | |
| CN110146099B (en) | Synchronous positioning and map construction method based on deep learning | |
| CA3066341A1 (en) | Map data correcting method and device | |
| EP3032818B1 (en) | Image processing device | |
| CN111263960B (en) | Apparatus and method for updating high definition map | |
| CN110307791B (en) | Vehicle length and speed calculation method based on three-dimensional vehicle boundary frame | |
| CN104848858A (en) | Two-dimensional code and vision-inert combined navigation system and method for robot | |
| CN104700414A (en) | Rapid distance-measuring method for pedestrian on road ahead on the basis of on-board binocular camera | |
| CN109949355A (en) | The method of half fan-shaped equidistant line model is established in monocular vision pedestrian's distance estimations | |
| WO2015024407A1 (en) | Power robot based binocular vision navigation system and method based on | |
| CN112037159B (en) | Cross-camera road space fusion and vehicle target detection tracking method and system | |
| CN110288659B (en) | Depth imaging and information acquisition method based on binocular vision | |
| CN112347205B (en) | Updating method and device for vehicle error state | |
| CN105551020A (en) | Method and device for detecting dimensions of target object | |
| JP2017181476A (en) | Vehicle location detection device, vehicle location detection method and vehicle location detection-purpose computer program | |
| CN110597249A (en) | Robot and recharging positioning method and device thereof | |
| CN114413958A (en) | Monocular vision distance and speed measurement method of unmanned logistics vehicle | |
| CN111998862A (en) | Dense binocular SLAM method based on BNN |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190628 |