CN101789123B - Method for creating distance map based on monocular camera machine vision - Google Patents
Method for creating distance map based on monocular camera machine vision Download PDFInfo
- Publication number
- CN101789123B CN101789123B CN2010101020004A CN201010102000A CN101789123B CN 101789123 B CN101789123 B CN 101789123B CN 2010101020004 A CN2010101020004 A CN 2010101020004A CN 201010102000 A CN201010102000 A CN 201010102000A CN 101789123 B CN101789123 B CN 101789123B
- Authority
- CN
- China
- Prior art keywords
- image
- boundary
- line
- distance
- machine 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.)
- Active
Links
Images
Abstract
The invention discloses a method for creasing a distance map based on monocular camera machine vision, which comprises the following steps: carrying out red component extraction on an image collected by a car camera for obtaining a baseline image; utilizing an edge detection operator for carrying out boundary detection on the baseline image, and further carrying out Hough transform on discontinuity points in the boundary for detecting a boundary line; carrying out the Hough transform on the boundary detection result for finding out straight lines in the image, then carrying out statistical weighting on end points of a plurality of straight lines approaching a vanishing point, taking the average value of detected column coordinates of a group of the straight lines as the column coordinate of the vanishing point of an optical axis of the camera, and taking the average value of row coordinates as the row coordinate position of a vanishing line; and respectively filling distance information from a projection line of the optical axis to the boundaries on the two sides, carrying out approximate treatment on scenes being far away from the vanishing line, eliminating the impacts of long-distance errors on defogging treatment and obtaining the distance-depth map of the scenes. The utilization of the method can solve the problems of difficult creation of the distance map and expensive distance measuring equipment during model defogging.
Description
Technical field
The present invention relates to vehicle-mounted DAS (Driver Assistant System) technical field of image processing, relate in particular to the method for a kind of establishment based on the monocular camera machine vision distance map.
Background technology
The research of current driver's DAS (Driver Assistant System) focuses mostly on aspect the Vehicular automatic driving under fine weather, and the research that recovers for the scene under the inclement weather also seldom; Image mist elimination research both domestic and external mainly concentrates on following several mode: 1, histogram equalization algorithm; 2, based on the constant theory of color image is carried out the mist elimination processing and carry out color and contrast recovery; 3, based on the restored method of atmospheric scattering model.
Wherein, histogram equalization algorithm is a kind of simply and effectively Enhancement Method commonly used, and it comes the histogram of correction image based on theory of probability by the mapping of gray scale, makes it to have smooth distribution.The thin general contrast of cloud degraded image is lower, can be dynamic range expanded by histogram equalization.The histogram Enhancement Method that increases contrast can be described as classical image processing method, this method is simply effective, under a lot of situations, image strengthens the effect that just can reach satisfied by histogram, and algorithm is realized than being easier to, but because the degree of degeneration of degraded image and the depth of field exponent function relation of scene, so the effect that histogram is handled is not ideal enough.
Based on the constant theory of color image is carried out the mist elimination processing and carry out color and contrast recovery.The constant theory of color also can improve the contrast of Misty Image to a certain extent to image mist elimination algorithm, but scenery image contrast's reduction and object are the relation of non-linear increasing in the Misty Image to the distance of camera, the depth of field difference of scene point, the degree of degeneration of corresponding picture point are also different.Therefore, in actual treatment, there are defectives such as image local overtreating or undertreatment, can't the scenery of different scene distances in the piece image be recovered accordingly.
Restored method based on the atmospheric scattering model.Handle image based on the method for model and be referred to as the image recovery, this class algorithm is set about from the Physical Mechanism of image degradation, by the true picture after the recovery of damping capacity can really be restored in various degree to scenery under the different distance.There is the difficulty of obtaining that key issue is exactly a subject distance information in method based on atmospheric scattering model mist elimination.Existent method is the distance-measuring equipments such as radar by costliness at present, therefore has significant limitation on system's popularization, so the acquisition mode of scene depth information becomes the key factor that Misty Image is recovered.
In addition, under active at night lighting condition, the particulate back scattering produces serious influence to imaging, and the atmospheric scattering model no longer is applicable to the situation at night between daytime.
Summary of the invention
(1) technical matters that will solve
Special construction at trap for automobile, the present invention proposes a kind of distance map creation method,, set up the corresponding relation of scene and CCD target surface by demarcation to vehicle-mounted vidicon based on monocular camera machine vision, distance map is created difficulty, the problem of distance measuring equipment costliness when solving the model mist elimination.
(2) technical scheme
For achieving the above object, the technical solution used in the present invention is as follows:
A kind of method of creating based on the monocular camera machine vision distance map, this method comprises:
Image to vehicle-mounted camera collection carries out the red component extraction, obtains the benchmark image based on the monocular camera machine vision distance map;
Utilize edge detection operator that this benchmark image is carried out Boundary Detection, again the discontinuous point in the border is made the Hough change detection and go out the boundary line;
The result of above-mentioned Boundary Detection is carried out the Hough conversion find the straight line that exists in the image, then the end points near many straight lines of end point is carried out statistical weight, as camera optical axis disappearance point range coordinate, row-coordinate mean value is as the row-coordinate position of vanishing line with the row coordinate mean value of detected one group of straight line;
Begin to fill range information from the optical axis projection line respectively, vanishing line is carried out approximate processing with scenery far away, get rid of the influence that remote error is handled mist elimination, obtain the scene distance depth map to the border, both sides.
In the such scheme, the described edge detection operator that utilizes carries out this benchmark image further comprising before the step of Boundary Detection: this benchmark image is carried out The disposal of gentle filter eliminate noise.
In the such scheme, the described process of utilizing edge detection operator that this benchmark image is carried out Boundary Detection is the process with original image and smothing filtering impulse response single order differential convolution algorithm.
In the such scheme, described discontinuous point in the border is done in the step that the Hough change detection goes out the boundary line, the determining positions of boundary line the size of area-of-interest.
In the such scheme, described end points near many straight lines of end point is carried out in the step of statistical weight, the weights near the row-coordinate of picture centre and row coordinate are big more more.
In the such scheme, describedly begin to fill the step of range information to the border, both sides from the optical axis projection line respectively, the zone of suppose vehicle front is a road area, from the centre, bottom of image, search respectively to the left and right sides, with the borderline region that runs at first as the road boundary zone.
(3) beneficial effect
From technique scheme as can be seen, the present invention has following beneficial effect:
1, establishment provided by the invention by the demarcation to vehicle-mounted vidicon, has been set up the corresponding relation of scene and CCD target surface based on the method for monocular camera machine vision distance map, and distance map is created difficulty, the problem of distance measuring equipment costliness when having solved the model mist elimination.
If 2, this paper highway distance model can be estimated leading vehicle distance in conjunction with vehicle recongnition technique, the prompting driver remains a safe distance behind the car in front.
Description of drawings
Fig. 1 is the method flow diagram of establishment provided by the invention based on the monocular camera machine vision distance map;
Fig. 2 is an image coordinate, camera coordinates, the corresponding relation between the world coordinate system;
Fig. 3 is the camera calibration template;
Fig. 4 is the pinhole imaging system model, and wherein, H is the video camera setting height(from bottom), and f is a focal length of camera, when y, z represent road surface distance for z respectively on the CCD face of correspondence distance be y;
Fig. 5 is the scene depth figure that generates, x, and y axle presentation video pixel coordinate, the z axle is represented the respective distances value.
Embodiment
For making the purpose, technical solutions and advantages of the present invention clearer, below in conjunction with specific embodiment, and with reference to accompanying drawing, the present invention is described in more detail.
Distance map creation method based on monocular camera machine vision provided by the invention is finished structure to highway scene distance depth map by the monocular camera machine vision principle, has solved the key issue of atmospheric scattering model mist elimination.For obtaining the depth information of object scene, at first experiment is demarcated the acquisition calibrating parameters with video camera, use linear imaging modelling scene depth figure then, for driving the road surface, position, video camera axis x value is zero, and projection equation is reduced to the relation between z and the y, as shown in Figure 4:
As shown in Figure 1, Fig. 1 is the method flow diagram of establishment provided by the invention based on the monocular camera machine vision distance map, and this method specifically may further comprise the steps:
Step 1: image pre-service: under the greasy weather situation, owing to the selectivity of particulate to wavelength absorbs, different channel image difference in qualitys are very big.The characteristics of image of red component is preserved best, helps subsequent detection and handles, so utilize the red component gray level image to carry out the Boundary Extraction computing.The image that vehicle-mounted camera is collected carries out the red component extraction, as the benchmark image of treatment step such as step 2 grade.
Step 2: the boundary line is detected: edge detection operator can relate to a problem when using, if promptly directly adopt rim detection (Canny) operator, the edge that obtains will comprise the pseudo-edge point that noise causes, the way that addresses this problem is earlier image to be carried out The disposal of gentle filter to eliminate noise, detects with the Canny operator then.And the testing process of Canny operator is exactly the process of original image and smothing filtering impulse response single order differential convolution algorithm.Utilize the Canny operator that the image that obtains in the step 1 is carried out Boundary Detection, again the discontinuous point in the border is made the Hough change detection and go out the boundary line.The determining positions of boundary line the size of area-of-interest in the later step.
Step 3: area-of-interest extracts: in order to improve algorithm accuracy and real-time, only area-of-interest is handled.According to the projection theory as can be known, when camera optical axis was parallel to the ground, end point was positioned at the center of the plane of delineation, and promptly the zone, road surface is positioned at the lower half-plane of image.In order to reduce operand, the influence of avoiding natural scene complicated in the road background that lane line is detected as far as possible, with road at a distance the zone that surrounds of vanishing line and both sides road boundary as area-of-interest, so both can improve the robustness of detection, can significantly reduce calculated amount again, improve the real-time of system.Under the actual conditions, more near end point, kerb line is fuzzy more, and straight-line detection has very big difficulty, and prolonging the end point of back acquisition and the vanishing line of human eye custom has certain deviation.Do not need to find the exact position of end point in the distance map constructive process, only go out the vanishing line position and get final product according to detected straight line Information Statistics.Utilize Boundary Detection result in the step 2 to carry out the Hough conversion and find the straight line that exists in the image, will carry out statistical weight near the end points of many straight lines of end point then, the weights near the row-coordinate of picture centre and row coordinate are big more more.As camera optical axis disappearance point range coordinate, row-coordinate mean value is as the row-coordinate position of vanishing line with the row coordinate mean value of detected one group of straight line.
Step 4: distance parameter is filled: road distance model construction last needs concrete range information is filled among the result that step 3 obtains.Precision of information requires relatively low because the recovery of Misty Image is adjusted the distance, both sides of highway scenery is single, therefore, 2 hypothesis are proposed: one, with the lane line be the separatrix, edge when theoretical modeling, two, both sides scenery equates with the road boundary distance, according to the mapping relations of pinhole imaging system model, draw the scene distance model of approximate U type.During Image Acquisition, variable quantity and x on the z direction, the variable quantity of y direction differ an order of magnitude, and the Misty Image accuracy requirement that recovers to adjust the distance is not high, and therefore, the every bit distance is approximately:
After step 3 is finished, need to begin to fill range information from the optical axis projection line respectively to the border, both sides.The zone of supposing vehicle front is a road area, in the middle of the bottom of image, search for respectively to the left and right sides, with the borderline region that runs at first as the road boundary zone.Vanishing line is carried out approximate processing with scenery far away, got rid of the influence that remote error is handled mist elimination.The scene distance depth map that obtains at last is Fig. 5.
There are 4 canonical coordinates systems in single camera vision system: objective world coordinate system, camera coordinate system, plane of delineation coordinate system and frame are deposited coordinate system.The target location of real world is joined the frame that matrix rotational transform and translation transformation, pinhole imaging system projective transformation, the conversion of confidential reference items equation obtain in the image outward by video camera successively and is deposited coordinate, when these a series of conversion are mainly used in off-line calibration, based on camera calibration, can utilize reflection to penetrate and set up road surface model.When system used, what at first obtain was that frame in the computing machine is deposited coordinate, then through confidential reference items inverse transformation, pinhole imaging system inverse transformation and outside join inverse transformation and obtain the real world coordinates that frame is deposited the coordinate correspondence.
In the pinhole imaging system model, the image space of any point, space P in image can represent with the pinhole imaging system model approximation, and promptly any some projected position p of P in image is photocentre O and the P line OP of ordering and the intersection point of the plane of delineation.This relation is also referred to as central projection or perspective projection.As Fig. 2, xyz is a camera coordinate system among the figure, and XOY is an image coordinate system, X
WY
WZ
WRepresent world coordinate system.
Experiment is used as shown in Figure 3 chequer with calibrating template, and the size of each square lattice all is 30mm * 30mm.After calibrating template printed, stick on the very high panel of flatness.Calibrating template is placed on the used video camera of experiment front, and the angle between conversion stencil plane and the picture plane is gathered the set of diagrams picture successively, is used for follow-up demarcation.
According to above analysis as can be known: for any spatial point P, if know the inside and outside parameter of video camera, can determine the image point position of this spatial point correspondence, but because any picture point that is positioned at the spatial point on the ray OP all is the p point, therefore, the projection of picture point p is not unique.But for driving the road surface, position, video camera axis x value is zero, and projection equation is reduced to the relation between z and the y, as shown in Figure 4:
Precision of information requires relatively low because the recovery of Misty Image is adjusted the distance, both sides of highway scenery is single, therefore, when theoretical modeling, suppose: the one, be the separatrix, edge with the lane line, two is that two lateral extents equate with the road position of intersecting point, according to the mapping relations of the linear imaging model of aperture, draw the scene distance model of approximate U type.
During Image Acquisition, video camera height overhead is H=1.2m, the focal distance f=16mm of camera, and resolution is 640 * 480.12 ° at camera field of view angle, variable quantity and x on the z direction, the variable quantity of y direction differ an order of magnitude, and the recovery of the Misty Image accuracy requirement of adjusting the distance is not high, therefore, the every bit distance is approximately:
Because this paper recovers at the Misty Image of road surface scenery, so the distance beyond the hypothesis lane line edge is all identical with corresponding edge, corresponding scene distance figure is Fig. 5.
Above-described specific embodiment; purpose of the present invention, technical scheme and beneficial effect are further described; institute is understood that; the above only is specific embodiments of the invention; be not limited to the present invention; within the spirit and principles in the present invention all, any modification of being made, be equal to replacement, improvement etc., all should be included within protection scope of the present invention.
Claims (5)
1. an establishment is characterized in that based on the method for monocular camera machine vision distance map this method comprises:
Image to vehicle-mounted camera collection carries out the red component extraction, obtains the benchmark image based on the monocular camera machine vision distance map;
Utilize edge detection operator that this benchmark image is carried out Boundary Detection, again the discontinuous point in the border is done the Hough conversion, detect the boundary line;
Result to above-mentioned Boundary Detection carries out the Hough conversion, find the straight line that exists in the image, then the end points near many straight lines of end point is carried out statistical weight, as camera optical axis disappearance point range coordinate, row-coordinate mean value is as the row-coordinate position of vanishing line with the row coordinate mean value of detected one group of straight line;
Begin to fill range information from the optical axis projection line respectively, vanishing line is carried out approximate processing with scenery far away, get rid of the influence that remote error is handled mist elimination, obtain the scene distance depth map to the border, both sides;
Wherein, describedly begin to fill the step of range information to the border, both sides from the optical axis projection line respectively, the zone of suppose vehicle front is a road area, from the centre, bottom of image, search respectively to the left and right sides, with the borderline region that runs at first as the road boundary zone.
2. establishment according to claim 1 is based on the method for monocular camera machine vision distance map, it is characterized in that, the described edge detection operator that utilizes carries out this benchmark image further comprising before the step of Boundary Detection: this benchmark image is carried out The disposal of gentle filter eliminate noise.
3. establishment according to claim 1 is based on the method for monocular camera machine vision distance map, it is characterized in that the described process of utilizing edge detection operator that this benchmark image is carried out Boundary Detection is the process with original image and smothing filtering impulse response single order differential convolution algorithm.
4. establishment according to claim 1 is characterized in that based on the method for monocular camera machine vision distance map, described discontinuous point in the border is done in the step that the Hough change detection goes out the boundary line, the determining positions of boundary line the size of area-of-interest.
5. establishment according to claim 1 is based on the method for monocular camera machine vision distance map, it is characterized in that, described end points near many straight lines of end point is carried out in the step of statistical weight, the weights near the row-coordinate of picture centre and row coordinate are big more more.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2010101020004A CN101789123B (en) | 2010-01-27 | 2010-01-27 | Method for creating distance map based on monocular camera machine vision |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2010101020004A CN101789123B (en) | 2010-01-27 | 2010-01-27 | Method for creating distance map based on monocular camera machine vision |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101789123A CN101789123A (en) | 2010-07-28 |
| CN101789123B true CN101789123B (en) | 2011-12-07 |
Family
ID=42532326
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2010101020004A Active CN101789123B (en) | 2010-01-27 | 2010-01-27 | Method for creating distance map based on monocular camera machine vision |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101789123B (en) |
Families Citing this family (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102122344A (en) * | 2011-01-07 | 2011-07-13 | 南京理工大学 | Road border detection method based on infrared image |
| CN102324017B (en) * | 2011-06-09 | 2013-04-03 | 中国人民解放军国防科学技术大学 | FPGA (Field Programmable Gate Array)-based lane line detection method |
| CN102750702B (en) * | 2012-06-21 | 2014-10-15 | 东华大学 | Monocular infrared image depth estimation method based on optimized BP (Back Propagation) neural network model |
| KR102003776B1 (en) * | 2013-08-13 | 2019-07-26 | 한화테크윈 주식회사 | Method and apparatus of detecting posture of the camera for surveillance |
| CN103617412B (en) * | 2013-10-31 | 2017-01-18 | 电子科技大学 | Real-time lane line detection method |
| CN105303522B (en) * | 2014-06-30 | 2018-09-28 | 高德软件有限公司 | A kind of image defogging processing method and processing device |
| JP6548893B2 (en) * | 2014-11-18 | 2019-07-24 | 日立オートモティブシステムズ株式会社 | Roadway recognition device and travel support system using the same |
| CN104506872B (en) * | 2014-11-26 | 2017-09-29 | 深圳凯澳斯科技有限公司 | A kind of method and device of converting plane video into stereoscopic video |
| CN105260993A (en) * | 2015-10-23 | 2016-01-20 | 哈尔滨工业大学 | Hyperspectral image stripe missing restoring method based on edge constraint and self-adaptive morphological filter |
| CN106204572B (en) * | 2016-07-06 | 2020-12-04 | 合肥工业大学 | Road target depth estimation method based on scene depth mapping |
| CN106874842B (en) * | 2016-12-30 | 2020-07-03 | 长安大学 | Digital image-based automobile and road edge stone distance detection method |
| CN109086650B (en) * | 2017-06-14 | 2022-04-12 | 现代摩比斯株式会社 | Calibration method and calibration apparatus |
| CN108668096B (en) * | 2018-05-18 | 2021-08-20 | 京东方科技集团股份有限公司 | Video data management method and device and video recording equipment |
| CN109146948B (en) * | 2018-07-27 | 2021-07-30 | 内蒙古大学 | Crop growth phenotype parameter quantification and yield correlation analysis method based on vision |
| CN111931638B (en) * | 2020-08-07 | 2023-06-20 | 华南理工大学 | Pedestrian re-identification-based local complex area positioning system and method |
| CN113607058B (en) * | 2021-07-21 | 2023-11-17 | 杭州电子科技大学 | Straight blade size detection method and system based on machine vision |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5963664A (en) * | 1995-06-22 | 1999-10-05 | Sarnoff Corporation | Method and system for image combination using a parallax-based technique |
| US6791540B1 (en) * | 1999-06-11 | 2004-09-14 | Canon Kabushiki Kaisha | Image processing apparatus |
| CN101236653A (en) * | 2008-03-03 | 2008-08-06 | 华为技术有限公司 | Image correction method and system |
| CN101276415A (en) * | 2008-03-03 | 2008-10-01 | 北京航空航天大学 | Apparatus and method for realizing multi-resolutions image acquisition with multi-focusing video camera |
| CN101630406A (en) * | 2008-07-14 | 2010-01-20 | 深圳华为通信技术有限公司 | Camera calibration method and camera calibration device |
-
2010
- 2010-01-27 CN CN2010101020004A patent/CN101789123B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5963664A (en) * | 1995-06-22 | 1999-10-05 | Sarnoff Corporation | Method and system for image combination using a parallax-based technique |
| US6791540B1 (en) * | 1999-06-11 | 2004-09-14 | Canon Kabushiki Kaisha | Image processing apparatus |
| CN101236653A (en) * | 2008-03-03 | 2008-08-06 | 华为技术有限公司 | Image correction method and system |
| CN101276415A (en) * | 2008-03-03 | 2008-10-01 | 北京航空航天大学 | Apparatus and method for realizing multi-resolutions image acquisition with multi-focusing video camera |
| CN101630406A (en) * | 2008-07-14 | 2010-01-20 | 深圳华为通信技术有限公司 | Camera calibration method and camera calibration device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101789123A (en) | 2010-07-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101789123B (en) | Method for creating distance map based on monocular camera machine vision | |
| WO2021259344A1 (en) | Vehicle detection method and device, vehicle, and storage medium | |
| CN108647638B (en) | Vehicle position detection method and device | |
| US20200065594A1 (en) | Path sensing using structured lighting | |
| CN102722705B (en) | Method for detecting multi-lane line on basis of random sample consensus (RANSAC) algorithm | |
| KR102267562B1 (en) | Device and method for recognition of obstacles and parking slots for unmanned autonomous parking | |
| CN101750049B (en) | Monocular vision vehicle distance measuring method based on road and vehicle information | |
| US11373532B2 (en) | Pothole detection system | |
| CN103473762B (en) | A kind of method for detecting lane lines and device | |
| EP1403615B1 (en) | Apparatus and method for processing stereoscopic images | |
| WO2014070448A1 (en) | Vehicular path sensing system and method | |
| CN104916163A (en) | Parking space detection method | |
| CN110307791B (en) | Vehicle length and speed calculation method based on three-dimensional vehicle boundary frame | |
| CN103204104B (en) | Monitored control system and method are driven in a kind of full visual angle of vehicle | |
| CN103196418A (en) | Measuring method of vehicle distance at curves | |
| CN202134079U (en) | Unmanned vehicle lane marker line identification and alarm device | |
| CN107145825A (en) | Ground level fitting, camera calibration method and system, car-mounted terminal | |
| Kellner et al. | Road curb detection based on different elevation mapping techniques | |
| CN107290738A (en) | A kind of method and apparatus for measuring front vehicles distance | |
| US10108866B2 (en) | Method and system for robust curb and bump detection from front or rear monocular cameras | |
| CN105632180A (en) | System and method of recognizing tunnel entrance vehicle type based on ARM | |
| CN112308913A (en) | Vision-based vehicle positioning method and device and vehicle-mounted terminal | |
| Lion et al. | Smart speed bump detection and estimation with kinect | |
| CN107220632B (en) | Road surface image segmentation method based on normal characteristic | |
| CN106803073B (en) | Auxiliary driving system and method based on stereoscopic vision target |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |