CN109993133A - Automobile-used obstacle detection method based on longitudinal binocular vision - Google Patents
Automobile-used obstacle detection method based on longitudinal binocular vision Download PDFInfo
- Publication number
- CN109993133A CN109993133A CN201910271836.8A CN201910271836A CN109993133A CN 109993133 A CN109993133 A CN 109993133A CN 201910271836 A CN201910271836 A CN 201910271836A CN 109993133 A CN109993133 A CN 109993133A
- Authority
- CN
- China
- Prior art keywords
- camera
- characteristic point
- rearmounted
- pair
- image
- 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.)
- Withdrawn
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 25
- 230000004888 barrier function Effects 0.000 claims abstract description 38
- 238000009434 installation Methods 0.000 claims abstract description 23
- 238000004321 preservation Methods 0.000 claims abstract description 4
- 238000000034 method Methods 0.000 claims description 23
- 238000002405 diagnostic procedure Methods 0.000 claims description 3
- 239000011521 glass Substances 0.000 claims description 3
- 230000007547 defect Effects 0.000 abstract description 2
- 239000011159 matrix material Substances 0.000 description 12
- 238000004458 analytical method Methods 0.000 description 4
- 208000010877 cognitive disease Diseases 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 239000012491 analyte Substances 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/40—Image enhancement or restoration using histogram techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
-
- 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
- G06T2207/30261—Obstacle
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of unmanned automobile-used obstacle detection methods, and in particular to a kind of automobile-used obstacle detection method based on longitudinal binocular vision, comprising: one, camera selection and installation;Two, the acquisition and preservation of the inside and outside portion's parameter of camera;Three, Image Acquisition;Four, image characteristic point is detected and is matched;Five, barrier determines.Front camera of the present invention and rearmounted camera select the monocular-camera of model of the same race, specifications parameter is completely the same, detecting barrier, cost is relatively low, Front camera and rearmounted camera are for acquiring obstructions chart picture, and image transmitting is handled to barrier Feature Points Matching module by USB interface, last barrier determination module calculates the range index of each pair of characteristic point matched to judge whether there is steric hindrance object, solve whether other obstacle detection methods based on binocular vision directly there cannot be the defect of height by object point corresponding to judging characteristic point, make unmanned safer reliable.
Description
Technical field
The present invention relates to a kind of unmanned automobile-used obstacle detection methods, and in particular to one kind is based on longitudinal binocular vision
Automobile-used obstacle detection method.
Background technique
Existing automobile-used obstacle detection method and system are based primarily upon system laser radar, millimetre-wave radar or binocular vision
Feel technology.The Chinese patent of Publication No. CN105866790A proposes a kind of laser radar cognitive disorders object space method;It is open
Number a kind of three-dimensional barrier based on binocular camera shooting head system and three-dimensional laser radar is proposed for the Chinese patent of CN104899855A
Hinder object detecting method;The Chinese patent of Publication No. CN103411536A proposes a kind of side of environment sensing cognitive disorders object
Method, this method joined tangential distortion correction during the distortion correction of two CCD camera acquisition images, further effectively
Raising image coordinate obtain accuracy, while steric hindrance object it is matched during joined epipolar-line constraint condition, contract
Small Feature Points Matching range, reduces the calculation amount of Stereo matching process, improves the matched matching precision of stereoscopic features;It is public
The number of opening is that the Chinese patent of CN101852609A discloses a kind of earth bulging analyte detection based on binocular stereo vision of robot
Method, using binocular stereo visual sensor cognitive disorders object, the patent is according to the baseline length and focal length of binocular, using known
The geometrical configuration of image parses the ground parallax values of each row in image;On the basis of ground parallax values, pass through back projection's mould
Type calculates the three-dimensional coordinate that certain pixel corresponds to scene point, to tentatively judge that the pixel belongs to barrier or ground point.On
Although the technology of stating can judge barrier, the sensors such as laser, radar are at high cost, have binocular vision detection barrier skill
Art cannot directly judge whether object point corresponding to characteristic point has height.
Summary of the invention
In order to solve the above technical problems, it is an object of the invention to: a kind of automobile-used barrier based on longitudinal binocular vision is provided
Hinder object detecting method, can determine whether object point corresponding to characteristic point has height, to judge whether object point belongs to three-dimensional barrier
Hinder object, makes unmanned safer reliable.
The present invention is technical solution used by solving its technical problem are as follows:
The automobile-used obstacle detection method based on longitudinal binocular vision, comprising the following steps:
Step 1: camera selection and installation: Front camera is installed on vehicle engine hatch cover, in shield glass
Rear is installed by rearmounted camera;
Step 2: the acquisition and preservation of camera parameter: obtaining and save Front camera and rearmounted camera inner parameter and outer
Portion's parameter;
Step 3: Image Acquisition: Front camera and rearmounted camera carry out Image Acquisition to barrier simultaneously, save image letter
It ceases and is sent to barrier Feature Points Matching module;
Step 4: image characteristic point is detected and is matched: before barrier Feature Points Matching module detects and matches synchronization
It sets the characteristic point in the characteristic point and the acquired image of rearmounted camera in the acquired image of camera, marks that forward and backward to set camera same
The characteristic point at moment is a pair of of characteristic point;
Step 5: barrier determines: the range index of each pair of characteristic point after the calculating matching of barrier determination module, and according to
Whether object point corresponding to range index judging characteristic point belongs to steric hindrance object.
Front camera and rearmounted camera select the monocular-camera of model of the same race, and specifications parameter is completely the same, Front camera
With rearmounted camera for acquiring obstructions chart picture, and by USB interface by image transmitting to barrier Feature Points Matching module into
Row processing, last barrier determination module calculate the range index of each pair of characteristic point matched to judge whether there is three-dimensional barrier
Hinder object.
Wherein, preferred embodiment are as follows:
The forward and backward mounting means for setting camera of step 1 are as follows:
The Front camera and rearmounted camera are longitudinally mounted on same straight line, Front camera lens axis and ground and
Automobile axis parallel, rearmounted camera installation direction and Front camera installation side machine installation direction are to make camera lens pitch angle 0
When spending, mirror is to consistent.
The camera internal parameter of Front camera and rearmounted camera described in step 2 includes camera resolution, frame per second and parallactic angle
Degree;The external parameter of Front camera and rearmounted camera described in step 2 includes Front camera mounting height h1, rearmounted camera installation
Vertical equity distance △ d when height h2, Front camera and rearmounted camera are installed, front/rear camera focus f, Front camera pitching are set
Angle θ 1, rearmounted camera pitching angle theta 2, front/rear camera pixel dimension μ is set.
The detection of each pair of characteristic point uses Harris angle point diagnostic method in step 4, utilizes horizontal, vertical difference operator pair
The each pixel of Front camera shooting image and the rearmounted camera shooting each pixel of image are filtered in the hope of taking horizontal gradient,
Assuming that the pixel coordinate (x, y) of Front camera shooting image present analysis, it is assumed that the pixel of rearmounted camera present analysis is sat
It marks (u, v), calculates pixel (x, y) in the horizontal gradient Ix and Iy in the direction x and the direction y, calculate pixel (u, v) in the direction x
With the horizontal gradient Iu and Iv in the direction y, Ix 2, Iy 2And IxIyInner product two-by-two between respectively Ix and Iy.Iu 2, Iv 2And IuIvRespectively
Inner product two-by-two between Iu and Iv.Metzler matrix is initially set up, for pixel (x, y) and pixel (u, v), matrix M (x, y)
It is calculated with M (u, v) by following formula:
If the matrix M characteristic value of certain point is very big it can be seen from the expression formula of Metzler matrix, in any shifting of point
The distance of dynamic very little, will cause the variation of biggish gray value, illustrate that the point is a Harris angle point.
The value and pixel that the Harris angle point value c (x, y) of pixel (x, y) is matrix M (x, y) are in the direction x and the side y
To horizontal gradient Ix and Iy sum ratio, the Harris angle point value c (u, v) of pixel (u, v) is matrix M (u, v)
Value and pixel the direction x and the direction y horizontal gradient Iu and Iv's and ratio, be calculate by the following formula:
When the value of c (x, y) and c (u, v) are greater than given threshold value, then it is assumed that the pixel is a Harris angle point,
Each Harris angle point consecutive points all with it are compared, with it with eight consecutive points of scale and neighbouring scale
Totally 26 points compare corresponding 18 points, to ensure all to detect characteristic point, a Harris in scale space and image space
If angle point is maximum in 26 points or minimum, which is a characteristic point.
Each pair of Feature Points Matching uses histogram comparison match algorithm, detailed process in step 4 are as follows:
It detects and is matched in the characteristic point and the acquired image of rearmounted camera in certain acquired image of moment Front camera
Total n pairs of characteristic point, n is positive integer, and the acquired image characteristic point of Front camera is denoted as Pi, i=1,2 ..., n, rearmounted camera
Acquired image characteristic point is denoted as Qi, i=1,2 ..., n, HPi(k), k=0,1,2 ..., L-1 indicate characteristic point PiPart
Feature, HQi(k), k=0,1,2 ..., L-1 indicate characteristic point QiLocal feature, L indicate histogram dimension, histogram
With D (Pi, Qi) it is expressed from the next:
As matching distance D (Pi, Qi) when being less than its preset threshold, it is believed that two characteristic points be it is matched, otherwise for not
Match.
The calculating process that step 5 barrier determines are as follows:
It is detected in step 4 and is matched to characteristic point and rearmounted camera in certain acquired image of moment Front camera and adopted
The characteristic point collected in image is n pairs total, and n is positive integer, and the range index for defining i-th pair characteristic point is li, i=1,2 ..., n,
The distance d of object point corresponding to Front camera to i-th pair characteristic point is calculatediAnd rearmounted camera is to i-th pair characteristic point institute
The distance T of corresponding object pointi, according to formula li=| Ti-di- △ d |, △ d is longitudinal direction water when Front camera and rearmounted camera are installed
Flat distance calculates the range index l of i-th pair characteristic pointiIf the range index of i-th pair characteristic point is greater than preset threshold, judge
Object point corresponding to i-th pair characteristic point has height, which belongs to steric hindrance object, otherwise judges that i-th pair characteristic point institute is right
The object point answered is located on level ground, which is not belonging to steric hindrance object.
The distance d of object point corresponding to the Front camera to i-th pair characteristic pointiCalculating process are as follows:
It is rectangular coordinate system of the origin foundation as unit of pixel by the Front camera acquisition image upper left corner, which is sat
Image coordinate system of the mark system as the acquired image of Front camera, if the midpoint coordinates of Front camera acquisition image is (x0,y0),
The coordinate of the characteristic point of the acquired image of Front camera is (x in i-th pair characteristic pointi,yi), according to formula
Calculate the distance d of object point corresponding to Front camera to i-th pair characteristic pointi, in above formula, h1 is Front camera installation
Highly, f be it is front/rear set camera focus, θ 1 is Front camera pitch angle, μ be it is front/rear set camera pixel dimension, Front camera and
Rearmounted camera model is identical.
The distance T of object point corresponding to the rearmounted camera to i-th pair characteristic pointiCalculating process are as follows:
It is rectangular coordinate system of the origin foundation as unit of pixel by the rearmounted camera acquisition image upper left corner, which is sat
Image coordinate system of the mark system as the acquired image of rearmounted camera, if the midpoint coordinates of rearmounted camera acquisition image is (u0,v0),
The coordinate of the characteristic point of the acquired image of rearmounted camera is (u in i-th pair characteristic pointi,vi), according to formula
Calculate the distance T of object point corresponding to rearmounted camera to i-th pair characteristic pointi, in above formula, h2 is rearmounted camera installation
Highly, θ 2 be rearmounted camera pitch angle, f be it is front/rear set camera focus, μ be it is front/rear set camera pixel dimension, Front camera and
Rearmounted camera model is identical.
Whether object point corresponding to the point of judging characteristic described in step 5 belongs to steric hindrance object, judgment method are as follows: sets
The range index of i-th pair characteristic point is liIf threshold value is a, if li> a then judges that object point corresponding to i-th pair characteristic point has
Highly, which belongs to steric hindrance object, if li≤ a then judges that object point corresponding to i-th pair characteristic point is located at level ground
On, which is not belonging to steric hindrance object.
Compared with prior art, the invention has the following advantages:
Front camera of the present invention and rearmounted camera select the monocular-camera of model of the same race, and specifications parameter is completely the same, inspection
Surveying barrier, cost is relatively low, and Front camera and rearmounted camera are passed image by USB interface for acquiring obstructions chart picture
Barrier Feature Points Matching module is defeated by be handled, last barrier determination module calculate each pair of characteristic point matched away from
From index to judge whether there is steric hindrance object, solving other obstacle detection methods based on binocular vision cannot be straight
The defect whether object point corresponding to judging characteristic point has height is connect, is made unmanned safer reliable.
Detailed description of the invention
Fig. 1 is flow chart of the present invention.
Fig. 2, which is that the present invention is forward and backward, sets camera scheme of installation.
In figure: 1, automobile;2, Front camera;3, rearmounted camera.
Specific embodiment
The embodiment of the present invention is described further with reference to the accompanying drawing:
Embodiment 1:
As shown in Figs. 1-2, the automobile-used obstacle detection method of the present invention based on longitudinal binocular vision, including following step
It is rapid:
Step 1: camera selection and installation: installation Front camera 2 is covered in 1 enging cabin of automobile, in 1 front windshield of automobile
Glass rear is installed by rearmounted camera 3;
Step 2: the acquisition and preservation of camera parameter: obtain and save Front camera 2 and 3 inner parameter of rearmounted camera and
External parameter;
Step 3: Image Acquisition: Front camera 2 and rearmounted camera 3 carry out Image Acquisition to barrier simultaneously, save image
Information is simultaneously sent to barrier Feature Points Matching module;
Step 4: image characteristic point is detected and is matched: before barrier Feature Points Matching module detects and matches synchronization
It sets the characteristic point in the characteristic point and the acquired image of rearmounted camera 3 in the acquired image of camera 2, marks that forward and backward to set camera same
The characteristic point at one moment is a pair of of characteristic point;
Step 5: barrier determines: the range index of each pair of characteristic point after the calculating matching of barrier determination module, and according to
Whether object point corresponding to range index judging characteristic point belongs to steric hindrance object.
Front camera 2 and rearmounted camera 3 pass through USB interface for image transmitting to obstacle for acquiring obstructions chart picture
Object Feature Points Matching module is handled, and Front camera 2 and rearmounted camera 3 select the monocular-camera of model of the same race, specification ginseng
Number is completely the same, such as the small of entitled MYNT EYE looks for monocular-camera, and the resolution ratio of camera is 752*480, and frame per second is
60FPS, parallax angle are 150 °, and barrier Feature Points Matching module uses a high-performance GPU server, such as model
The GPU of NVIDIAGTX1070, video memory 8GB are detected and are matched for barrier characteristic point captured by two cameras, barrier
Computer can be used in determination module, such as CPU is i5-6500, the computer of memory 12GB.
The forward and backward mounting means for setting camera 3 in step 1 are as follows:
The Front camera 2 is longitudinally mounted on same straight line with rearmounted camera 3,2 lens axis of Front camera and ground
1 axis parallel of face and automobile, 3 installation direction of rearmounted camera and 2 side's of installation machine installation direction of Front camera are to make camera lens pitching
When angle is 0 degree, mirror is to consistent.
The camera internal parameter of Front camera 2 and rearmounted camera 3 described in step 2 includes camera resolution, frame per second and parallax
Angle;Front camera 2 and the external parameter of rearmounted camera 3 described in step 2 include 2 mounting height h1 of Front camera, rearmounted camera
Vertical equity distance △ d when 3 mounting height h2, Front camera 2 and rearmounted camera 3 are installed, front/rear camera focus f, preposition phase are set
2 pitching angle theta 1 of machine, front/rear sets camera pixel dimension μ at 3 pitching angle theta 2 of rearmounted camera.
The detection of each pair of characteristic point uses Harris angle point diagnostic method in step 4, utilizes horizontal, vertical difference operator pair
The each pixel of Front camera shooting image and the rearmounted camera shooting each pixel of image are filtered in the hope of taking horizontal gradient,
Assuming that the pixel coordinate (x, y) of Front camera shooting image present analysis, it is assumed that the pixel of rearmounted camera present analysis is sat
It marks (u, v), calculates pixel (x, y) in the horizontal gradient Ix and Iy in the direction x and the direction y, calculate pixel (u, v) in the direction x
With the horizontal gradient Iu and Iv in the direction y, Ix 2, Iy 2And IxIyInner product two-by-two between respectively Ix and Iy.Iu 2, Iv 2And IuIvRespectively
Inner product two-by-two between Iu and Iv.Metzler matrix is initially set up, for pixel (x, y) and pixel (u, v), matrix M (x, y)
It is calculated with M (u, v) by following formula:
If the matrix M characteristic value of certain point is very big it can be seen from the expression formula of Metzler matrix, in any shifting of point
The distance of dynamic very little, will cause the variation of biggish gray value, illustrate that the point is a Harris angle point.
The value and pixel that the Harris angle point value c (x, y) of pixel (x, y) is matrix M (x, y) are in the direction x and the side y
To horizontal gradient Ix and Iy sum ratio, the Harris angle point value c (u, v) of pixel (u, v) is matrix M (u, v)
Value and pixel the direction x and the direction y horizontal gradient Iu and Iv's and ratio, be calculate by the following formula:
When the value of c (x, y) and c (u, v) are greater than given threshold value, then it is assumed that the pixel is a Harris angle point,
Each Harris angle point consecutive points all with it are compared, with it with eight consecutive points of scale and neighbouring scale
Totally 26 points compare corresponding 18 points, to ensure all to detect characteristic point, a Harris in scale space and image space
If angle point is maximum in 26 points or minimum, which is a characteristic point.
Each pair of Feature Points Matching uses histogram comparison match algorithm, detailed process in step 4 are as follows:
The characteristic point and the acquired image of rearmounted camera 3 for detecting and being matched in certain acquired image of moment Front camera 2
In total n pairs of characteristic point, n is positive integer, and the acquired image characteristic point of Front camera 2 is denoted as Pi, i=1,2 ..., n, postposition
The acquired image characteristic point of camera 3 is denoted as Qi, i=1,2 ..., n, HPi(k), k=0,1,2 ..., L-1 indicate characteristic point Pi's
Local feature, HQi(k), k=0,1,2 ..., L-1 indicate characteristic point QiLocal feature, L indicate histogram dimension, histogram
Figure matching D (Pi, Qi) it is expressed from the next:
As matching distance D (Pi, Qi) when being less than its preset threshold, it is believed that two characteristic points be it is matched, otherwise for not
Match, characteristic point local feature function HPi(k)、HQi(k) pixel value or gray value function generally are taken, local feature function belongs to
Common sense well known to those skilled in the art, does not repeat herein.
The calculating process that step 5 barrier determines are as follows:
It is detected in step 4 and is matched to 3 institute of characteristic point and rearmounted camera in certain acquired image of moment Front camera 2
The characteristic point acquired in image is n pairs total, and n is positive integer, and the range index for defining i-th pair characteristic point is li, i=1,2 ...,
The distance d of object point corresponding to Front camera 2 to i-th pair characteristic point is calculated in niAnd rearmounted camera 3 arrives i-th pair feature
The distance T of the corresponding object point of pointi, according to formula li=| Ti-di- △ d |, calculate the range index l of i-th pair characteristic pointiIf the
I is greater than preset threshold to the range index of characteristic point, then judges that object point corresponding to i-th pair characteristic point has height, the object point
Belong to steric hindrance object, otherwise judges that object point corresponding to i-th pair characteristic point is located on level ground, which is not belonging to solid
Barrier.
The distance d of object point corresponding to the Front camera 2 to i-th pair characteristic pointiCalculating process are as follows:
Acquiring the image upper left corner by Front camera 2 is rectangular coordinate system of the origin foundation as unit of pixel, by the right angle
Image coordinate system of the coordinate system as the acquired image of Front camera 2, if the midpoint coordinates that Front camera 2 acquires image is (x0,
y0), the coordinate of the characteristic point of the acquired image of Front camera 2 is (x in i-th pair characteristic pointi,yi), according to formula
Calculate the distance d of object point corresponding to Front camera to i-th pair characteristic pointi, in above formula, h1 is Front camera installation
Highly, f be it is front/rear set camera focus, θ 1 is Front camera pitch angle, μ be it is front/rear set camera pixel dimension, Front camera and
Rearmounted camera model is identical.
The distance T of object point corresponding to the rearmounted camera 3 to i-th pair characteristic pointiCalculating process are as follows:
Acquiring the image upper left corner by rearmounted camera 3 is rectangular coordinate system of the origin foundation as unit of pixel, by the right angle
Image coordinate system of the coordinate system as the acquired image of rearmounted camera 3, if the midpoint coordinates that rearmounted camera 3 acquires image is (u0,
v0), the coordinate of the characteristic point of the acquired image of rearmounted camera 3 is (u in i-th pair characteristic pointi,vi), according to formula
Calculate the distance T of object point corresponding to rearmounted camera 3 to i-th pair characteristic pointi, in above formula, h2 is rearmounted camera peace
Dress height, θ 2 are rearmounted camera pitch angle, f be it is front/rear set camera focus, μ is front/rear to set camera pixel dimension, Front camera
It is identical with rearmounted camera model.
Whether object point corresponding to the point of judging characteristic described in step 5 belongs to steric hindrance object, judgment method are as follows: sets
The range index of i-th pair characteristic point is liIf threshold value is a, if li> a then judges that object point corresponding to i-th pair characteristic point has
Highly, which belongs to steric hindrance object, if li≤ a then judges that object point corresponding to i-th pair characteristic point is located at level ground
On, which is not belonging to steric hindrance object.
Claims (9)
1. a kind of automobile-used obstacle detection method based on longitudinal binocular vision, which comprises the following steps:
Step 1: camera selection and installation: installation Front camera (2) is covered in automobile (1) enging cabin, in automobile (1) front
Wind glass rear is installed by rearmounted camera (3);
Step 2: the acquisition and preservation of camera parameter: obtain and save Front camera (2) and rearmounted camera (3) inner parameter and
External parameter;
Step 3: Image Acquisition: Front camera (2) and rearmounted camera (3) carry out Image Acquisition to barrier simultaneously, save image
Information is simultaneously sent to barrier Feature Points Matching module;
Step 4: image characteristic point is detected and matched: barrier Feature Points Matching module detects and matches the preposition phase of synchronization
The characteristic point in characteristic point and rearmounted camera (3) acquired image in machine (2) acquired image, marks that forward and backward to set camera same
The characteristic point at one moment is a pair of of characteristic point;
Step 5: barrier determines: barrier determination module calculates the range index of each pair of characteristic point after matching, and according to distance
Whether object point corresponding to index judging characteristic point belongs to steric hindrance object.
2. the automobile-used obstacle detection method according to claim 1 based on longitudinal binocular vision, which is characterized in that step
The one forward and backward mounting means for setting camera are as follows:
The Front camera (2) and rearmounted camera (3) are longitudinally mounted on same straight line, Front camera (2) lens axis with
Ground and automobile (1) axis parallel, rearmounted camera (3) installation direction are to make with Front camera (2) installation side's machine installation direction
When camera lens pitch angle is 0 degree, mirror is to consistent.
3. the automobile-used obstacle detection method according to claim 1 based on longitudinal binocular vision, which is characterized in that step
The camera internal parameter of two Front cameras (2) and rearmounted camera (3) includes camera resolution, frame per second and parallax angle;Step
The external parameter of rapid two Front camera (2) and rearmounted camera (3) includes Front camera mounting height h1, rearmounted camera installation
Vertical equity distance △ d when height h2, Front camera (2) and rearmounted camera (3) are installed, front/rear camera focus f, preposition phase are set
Machine pitching angle theta 1, front/rear sets camera pixel dimension μ at rearmounted camera pitching angle theta 2.
4. the automobile-used obstacle detection method according to claim 1 based on longitudinal binocular vision, which is characterized in that step
The detection of each pair of characteristic point uses Harris angle point diagnostic method in four.
5. the automobile-used obstacle detection method according to claim 1 based on longitudinal binocular vision, which is characterized in that step
Each pair of Feature Points Matching uses histogram comparison match algorithm, detailed process in four are as follows:
The characteristic point and rearmounted camera (3) acquired image for detecting and being matched in certain moment Front camera (2) acquired image
In total n pairs of characteristic point, n is positive integer, and Front camera (2) acquired image characteristic point is denoted as Pi, i=1,2 ..., n, after
It sets camera (3) acquired image characteristic point and is denoted as Qi, i=1,2 ..., n, HPi(k), k=0,1,2 ..., L-1 indicate characteristic point
PiLocal feature, HQi(k), k=0,1,2 ..., L-1 indicate characteristic point QiLocal feature, L indicate histogram dimension,
Histogram Matching D (Pi, Qi) it is expressed from the next:
As matching distance D (Pi, Qi) be less than its preset threshold when, it is believed that two characteristic points be it is matched, otherwise be mismatch.
6. the automobile-used obstacle detection method according to claim 5 based on longitudinal binocular vision, which is characterized in that step
The calculating process that five barriers determine are as follows:
It is detected in step 4 and the characteristic point being matched in certain moment Front camera (2) acquired image and rearmounted camera (3) institute
The characteristic point acquired in image is n pairs total, and n is positive integer, and the range index for defining i-th pair characteristic point is li, i=1,2 ...,
The distance d of object point corresponding to Front camera (2) to i-th pair characteristic point is calculated in niAnd rearmounted camera (3) arrives i-th pair
The distance T of object point corresponding to characteristic pointi, according to formula li=| Ti-di- △ d |, △ d is Front camera (2) and rearmounted camera
(3) vertical equity distance when installing calculates the range index l of i-th pair characteristic pointiIf the range index of i-th pair characteristic point is greater than
Preset threshold then judges that object point corresponding to i-th pair characteristic point has height, which belongs to steric hindrance object, otherwise judge
Object point corresponding to i-th pair characteristic point is located on level ground, which is not belonging to steric hindrance object.
7. the automobile-used obstacle detection method according to claim 6 based on longitudinal binocular vision, which is characterized in that described
The distance d of object point corresponding to Front camera to i-th pair characteristic pointiCalculating process are as follows:
It is rectangular coordinate system of the origin foundation as unit of pixel by Front camera (2) acquisition image upper left corner, which is sat
Image coordinate system of the mark system as Front camera (2) acquired image, if the midpoint coordinates of Front camera (2) acquisition image is
(x0,y0), the coordinate of the characteristic point of Front camera (2) acquired image is (x in i-th pair characteristic pointi,yi), according to formula
Calculate the distance d of object point corresponding to Front camera (2) to i-th pair characteristic pointi, in above formula, h1 is that Front camera installation is high
Degree, f be it is front/rear set camera focus, θ 1 is Front camera pitch angle, μ be it is front/rear set camera pixel dimension, Front camera (2) and
Rearmounted camera (3) model is identical.
8. the automobile-used obstacle detection method according to claim 6 based on longitudinal binocular vision, which is characterized in that described
The distance T of object point corresponding to rearmounted camera (3) to i-th pair characteristic pointiCalculating process are as follows:
It is rectangular coordinate system of the origin foundation as unit of pixel by rearmounted camera (3) acquisition image upper left corner, which is sat
Image coordinate system of the mark system as rearmounted camera (3) acquired image, if the midpoint coordinates of rearmounted camera (3) acquisition image is
(u0,v0), the coordinate of the characteristic point of rearmounted camera (3) acquired image is (u in i-th pair characteristic pointi,vi), according to formula
Calculate the distance T of object point corresponding to rearmounted camera (3) to i-th pair characteristic pointi, in above formula, h2 is that rearmounted camera installation is high
Degree, θ 2 be rearmounted camera pitch angle, f be it is front/rear set camera focus, μ be it is front/rear set camera pixel dimension, Front camera is with after
It is identical to set camera model.
9. the automobile-used obstacle detection method according to claim 6 based on longitudinal binocular vision, which is characterized in that step
Whether object point corresponding to the point of judging characteristic described in five belongs to steric hindrance object, judgment method are as follows: sets i-th pair characteristic point
Range index is liIf threshold value is a, if li> a then judges that object point corresponding to i-th pair characteristic point has height, the object point category
In steric hindrance object, if li≤ a then judges that object point corresponding to i-th pair characteristic point is located on level ground, which is not belonging to
Steric hindrance object.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910271836.8A CN109993133A (en) | 2019-04-04 | 2019-04-04 | Automobile-used obstacle detection method based on longitudinal binocular vision |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910271836.8A CN109993133A (en) | 2019-04-04 | 2019-04-04 | Automobile-used obstacle detection method based on longitudinal binocular vision |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109993133A true CN109993133A (en) | 2019-07-09 |
Family
ID=67132448
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910271836.8A Withdrawn CN109993133A (en) | 2019-04-04 | 2019-04-04 | Automobile-used obstacle detection method based on longitudinal binocular vision |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109993133A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113191238A (en) * | 2021-04-22 | 2021-07-30 | 开易(北京)科技有限公司 | Blind area detection method and system based on binocular camera and storage medium |
-
2019
- 2019-04-04 CN CN201910271836.8A patent/CN109993133A/en not_active Withdrawn
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113191238A (en) * | 2021-04-22 | 2021-07-30 | 开易(北京)科技有限公司 | Blind area detection method and system based on binocular camera and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105225482B (en) | Vehicle detecting system and method based on binocular stereo vision | |
CN106650708B (en) | Automatic driving obstacle vision detection method and system | |
CN110942449B (en) | Vehicle detection method based on laser and vision fusion | |
US8385595B2 (en) | Motion detection method, apparatus and system | |
US9378553B2 (en) | Stereo image processing device for vehicle | |
JP4956452B2 (en) | Vehicle environment recognition device | |
JP4876080B2 (en) | Environment recognition device | |
CN109410264B (en) | Front vehicle distance measuring method based on laser point cloud and image fusion | |
JP6743882B2 (en) | Image processing device, device control system, imaging device, image processing method, and program | |
US20210241492A1 (en) | Vehicular vision system with camera calibration using calibration target | |
KR101709317B1 (en) | Method for calculating an object's coordinates in an image using single camera and gps | |
CN111723778B (en) | Vehicle distance measuring system and method based on MobileNet-SSD | |
JP6035620B2 (en) | On-vehicle stereo camera system and calibration method thereof | |
CN109299696B (en) | Face detection method and device based on double cameras | |
CN113920183A (en) | Monocular vision-based vehicle front obstacle distance measurement method | |
KR100508763B1 (en) | System for detecting and recognizing a car plate area by using stereo vision and method thereof | |
CN103673977B (en) | The method and apparatus of rear dead zone of vehicle detection | |
CN113838111A (en) | Road texture feature detection method and device and automatic driving system | |
CN109993133A (en) | Automobile-used obstacle detection method based on longitudinal binocular vision | |
JP2018088217A (en) | Information processing device, imaging device, apparatus control system, information processing method, and program | |
CN206421414U (en) | A kind of automatic Pilot barrier vision detection system | |
CN111986248B (en) | Multi-vision sensing method and device and automatic driving automobile | |
CN113793388A (en) | Stereoscopic vision interpersonal safe distance detection method based on deep learning | |
CN114762019A (en) | Camera system | |
JP7180739B2 (en) | Information processing device, information processing system, information processing method and information processing program |
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 | ||
WW01 | Invention patent application withdrawn after publication | ||
WW01 | Invention patent application withdrawn after publication |
Application publication date: 20190709 |