CN108182376A - A kind of vehicle window localization method based on vehicle window Corner Detection - Google Patents
A kind of vehicle window localization method based on vehicle window Corner Detection Download PDFInfo
- Publication number
- CN108182376A CN108182376A CN201711133210.8A CN201711133210A CN108182376A CN 108182376 A CN108182376 A CN 108182376A CN 201711133210 A CN201711133210 A CN 201711133210A CN 108182376 A CN108182376 A CN 108182376A
- Authority
- CN
- China
- Prior art keywords
- width
- vehicle window
- platerect
- height
- left corner
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/59—Context or environment of the image inside of a vehicle, e.g. relating to seat occupancy, driver state or inner lighting conditions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
Abstract
A kind of vehicle window localization method based on vehicle window Corner Detection of the present invention, it is proposed that a kind of that vehicle window following two angle point, the method that window edge positioning is then carried out according to corner location are detected by SVM.Positioning action is carried out to vehicle window by using the method for the present invention, it must can effectively navigate to window edge, and there are inclined situations to position for vehicle window lower edges.
Description
Technical field
It is specifically a kind of that vehicle window following two is detected by SVM the present invention relates to computer vision and intelligent transportation field
Angle point, the method that window edge positioning is then carried out according to corner location.
Background technology
With the continuous development of intelligent transportation field, at present for driving behavior in traffic safety monitoring system
Detection and analysis be increasingly taken seriously.And the analysis that carry out driving behavior just must first navigate to position of driver,
The accurate positionin of vehicle window then can substantially determine the position of driver, therefore vehicle window positioning is closed very much in driving behavior detection
One step of key.
It is current to propose different vehicle window localization methods there are many scholar, wherein the technical solution being closer to the present invention
For：Document (Pan's generation Ji intelligent transportation peccancy detection algorithm researches and software systems realize [D] Harbin Institute of Technology,
2016.) horizontal edge above and below vehicle window is carried out using Canny edge detections and the method for linear structure element opening operation to position, it should
Method does not account for vehicle window when carrying out opening operation using linear structure element there are inclined situation, once vehicle window is upper and lower
Edge somewhat tilts, and the horizontal edge up and down of opening operation vehicle rear window is carried out by linear structure element and can be also removed；Text
It offers (marquis hall good fortune vehicle window detection technique research [D] Beijing Jiaotong University, 2012.) to propose first to position vehicle, Ran Hougen
Window locations are tentatively obtained according to the ratio of headstock and vehicle window, Canny edge detections are carried out in vehicle window approximate region, then using mould
Plate matches to carry out horizontal linear filtering, and the detection of vehicle window up-and-down boundary is then carried out using Hough transform, and this method uses
Template matching method come to carry out horizontal linear filtering be also not account for vehicle window lower edges there are inclined situation, tilt
Straight line by can equally be filtered out after template matches；Document (Yao Dongming, Han Anhua, the vehicle body that is waited to be detected based on vehicle window
Color identification method research [J] information communications, 2017 (2):Canny edge detections 87-88.) is equally used to combine horizontal straight
Line template filters and the method for Hough straight-line detection vehicle window horizontal stripes, there is also detection when the inclination of vehicle window lower edges not
Situation about arriving, and window locations are not carried out with coarse localization and begins to horizontal stripes above and below detection vehicle window, this is likely to
It detects the horizontal linear of vehicle roof, causes the situation of flase drop.
In conclusion there is following deficiencies for current vehicle window localization method：(1) exist for vehicle window lower edges and tilt
Situation can't detect；(2) it easily detects the horizontal linear of vehicle roof, causes window edge flase drop.
Invention content
For the above problem present in existing vehicle window localization method, the present invention proposes one kind and detects vehicle by SVM
Window following two angle point, the method that window edge positioning is then carried out according to corner location.
A kind of vehicle window localization method based on vehicle window Corner Detection, it is characterised in that include the following steps：
Step 1：License Plate is carried out to the coloured image image containing vehicle, the car plate position navigated to is denoted as square
Shape region plateRect；
Step 2：According to car plate position plateRect define vehicle window lower left corner angle point detection zone detectRoiL and
The detection zone detectRoiR of vehicle window lower right corner angle point；
Step 3：The candidate region of all vehicle window lower left corners angle point is detected using SVM in detectRoiL, is collected
Close leftCornersList={ RLi| i=1,2 ..., n }, RLiRepresent i-th of vehicle window lower left corner angle point candidate region, n is represented
The quantity of vehicle window lower left corner angle point candidate region；In detectRoiR all vehicle window lower right corner angle point is detected using SVM
Candidate region obtains set rightCornersList={ RRj| j=1,2 ..., m }, RRjRepresent j-th of vehicle window lower right corner angle
Point candidate region, m represent the quantity of vehicle window lower right corner angle point candidate region；
Step 4：It is determined respectively from leftCornersList and rightCornersList using scan matching method
Final vehicle window lower left corner angle point region and vehicle window lower right corner angle point region, and be denoted as respectively winCornerRectL and
winCornerRectR；
Step 5：Vehicle window lower edge detection zone is defined according to the position of winCornerRectL and winCornerRectR,
It is denoted as winDownRect；
Step 6：WinDownRect regions part is intercepted in image image, obtains vehicle window lower edge detection zone figure
Picture, and the obtained vehicle window lower edge detection zone image of interception is carried out image gray processing, the detection of Sobel horizontal edges and
The binary image finally obtained is denoted as winDownImg by OTSU binarization operations；
Step 7：Floor projection is carried out to winDownImg, finds the vertical seat of floor projection value maximum in winDownImg
Scale value is denoted as winDownLine；
Step 8：Vehicle window edge detection region is defined according to the position of winDownRect, is denoted as winUpRect；
Step 9：WinUpRect regions part is intercepted in image image, obtains vehicle window edge detection area image,
And image gray processing, the detection of Sobel horizontal edges and OTSU are carried out to the vehicle window edge detection area image that interception obtains
The binary image finally obtained is denoted as winUpImg by binarization operation；
Step 10：Floor projection, the bottom-up scanning in winUpImg, until scanning to a certain are carried out to winUpImg
The floor projection value of position meets vehicle window top edge condition and then stops scanning, and the ordinate value of the position in winUpImg is denoted as
WinUpLine if not finding the floor projection value for meeting vehicle window top edge condition, enables winUpLine=
WinUpImg.heiht/2, wherein winUpImg.heiht are the height of winUpImg；
Step 11：According to the position and winDownLine of the position of winCornerRectL and winCornerRectR
Position of the vehicle window in image image is calculated with winUpLine, is denoted as rectangle winRect.
A kind of vehicle window localization method based on vehicle window Corner Detection, it is characterised in that the vehicle window lower left corner in step 2
The detection zone detectRoiL of angle point and the detection zone detectRoiR of vehicle window lower right corner angle point are defined as follows：
2.1)：DetectRoiL.x=plateRect.x-plateRect.width × 3；
2.2)：DetectRoiL.y=plateRect.y-plateRect.width × 3.5；
2.3)：DetectRoiL.width=plateRect.x+plateRect.width/2-detectRoi L.x；
2.4)：DetectRoiL.height=plateRect.y-plateRect.width-detectRoiL .y；
2.5)：DetectRoiR.x=plateRect.x+plateRect.width/2；
2.6)：DetectRoiR.y=plateRect.y-plateRect.width × 3.5；
2.7)：DetectRoiR.width=plateRect.width × 3.5；
2.8)：DetectRoiR.height=plateRect.y-plateRect.width-detectRoiR .y.
Wherein plateRect.x is the upper left corner abscissa of plateRect, and plateRect.y is the upper left of plateRect
Angle ordinate, plateRect.width are the width of plateRect, and the upper left corner that detectRoiL.x is detectRoiL is horizontal
Coordinate, detectRoiL.y are the upper left corner ordinate of detectRoiL, and detectRoiL.width is the width of detectRoiL
Degree, detectRoiL.height are the height of detectRoiL, and detectRoiR.x is the horizontal seat in the upper left corner of detectRoiR
Mark, detectRoiR.y are the upper left corner ordinate of detectRoiR, and detectRoiR.width is the width of detectRoiR,
DetectRoiR.height is the height of detectRoiR.
A kind of vehicle window localization method based on vehicle window Corner Detection, it is characterised in that in step 4 from
Determine that final vehicle window lower left corner angle point region and vehicle window are right in leftCornersList and rightCornersList respectively
The step of scan matching method in inferior horn angle point region, is as follows：
4.1)：For each element RR in rightCornersListj, one is found from leftCornersList
A element RLiSo that (RRj.y+RRj.height/2)–(RLi.y+RLi.height/2 absolute value) is minimum, and this is absolutely
Value is less than plateRect.width/2, if finding the RL of the condition of satisfactioni, then by RRjAnd RLiIt is stored as a pair of of candidate angular
Get off；
4.2)：For each element RL in leftCornersListi, one is found from rightCornersList
A element RRjSo that (RLi.y+RLi.height/2)–(RRj.y+RRj.height/2 absolute value) is minimum, and this is absolutely
Value is less than plateRect.width/2, if finding the RR of the condition of satisfactionj, then by RRjAnd RLiIt is stored as a pair of of candidate angular
Get off；
4.3)：For the candidate angular pair stored in step 4.1) and step 4.2), only retain those in step
4.1) candidate angular pair stored and in step 4.2) is repeated；
4.4)：For the candidate angular pair that step 4.3) remains, only choose one pair of which and cause (RRj.x+
RRj.width/2)–(RLi.x+RLi.width/2 maximum absolute value) then enables winCornerRectL be equal to what is chosen
RLi, winCornerRectR is equal to the RR chosenj。
Wherein RLi.x it is RLiUpper left corner abscissa, RLi.y it is RLiUpper left corner ordinate, RLi.width it is RLi's
Width, RLi.height it is RLiHeight, RRj.x it is RRjUpper left corner abscissa, RRj.y it is RRjUpper left corner ordinate,
RRj.width it is RRjWidth, RRj.height it is RRjHeight.
A kind of vehicle window localization method based on vehicle window Corner Detection, it is characterised in that vehicle window lower edge in step 5
Detection zone winDownRect is defined as follows：
5.1)：WinDownRect.x=winCornerRectL.x；
5.2)：WinDownRect.y=Top-(Buttom-Top)/2；
5.3)：WinDownRect.width=winCornerRectR.x+winCornerRectR.width-
winCornerRectL.x；
5.4)：WinDownRect.height=(Buttom-Top) × 2；
Wherein winDownRect.x is the upper left corner abscissa of winDownRect, and winDownRect.y is
The upper left corner ordinate of winDownRect, winDownRect.width are the width of winDownRect,
Height of the winDownRect.height for winDownRect, Top=min (winCornerRectL.y,
WinCornerRectR.y), Buttom=max (winCornerRectL.buttom, winCornerRectR.buttom),
WinCornerRectL.x is the upper left corner abscissa of winCornerRectL, and winCornerRectL.y is
The upper left corner ordinate of winCornerRectL, the lower left corner that winCornerRectL.buttom is winCornerRectL are indulged
Coordinate, winCornerRectR.x are the upper left corner abscissa of winCornerRectR, and winCornerRectR.y is
The upper left corner ordinate of winCornerRectR, the lower left corner that winCornerRectR.buttom is winCornerRectR are indulged
Coordinate, winCornerRectR.width are the width of winCornerRectR, and min expressions take smaller, max from two numbers
Expression takes the greater from two numbers.
A kind of vehicle window localization method based on vehicle window Corner Detection, it is characterised in that vehicle window top edge in step 8
Detection zone winUpRect is defined as follows：
8.1)：WinUpRect.x=winDownRect.x；
8.2)：WinUpRect.y=winDownRect.y-winDownRect.height；
8.3)：WinUpRect.width=winDownRect.width；
8.4)：WinUpRect.height=winDownRect.height；
Wherein winUpRect.x is the upper left corner abscissa of winUpRect, and winUpRect.y is the upper left of winUpRect
Angle ordinate, winUpRect.width are the width of winUpRect, and winUpRect.height is the height of winUpRect.
A kind of vehicle window localization method based on vehicle window Corner Detection, it is characterised in that a certain position in step 10
The condition that floor projection value meets vehicle window top edge is as follows：
10.1)：project[h]>projectThresh；
10.2)：abs((projectLeft[h]-projectRight[h])/(winUpImg.width/2))<0.15.
Wherein h is the ordinate of image winUpImg, and project [h] represents that ordinate is at h in image winUpImg
The sum of the floor projection value of floor projection value and ordinate at h-1, h-2, h-3, h-4, h-5, projectThresh represent pre-
The projection value threshold value first set, projectLeft [h] represent that ordinate is at h in half rim portion of a left side of image winUpImg
The sum of the floor projection value of floor projection value and ordinate at h-1, h-2, h-3, h-4, h-5, projectRight [h] are represented
Floor projection value and ordinate of the ordinate at h are in h-1, h-2, h-3, h-4, h- in half rim portion of the right side of image winUpImg
The sum of floor projection value at 5, winUpImg.width represent the width of image winUpImg, and abs expressions take absolute value.
A kind of vehicle window localization method based on vehicle window Corner Detection, it is characterised in that vehicle window is in image in step 11
The calculating of position winRect in image is as follows：
11.1)：WinRect.x=winCornerRectL.x+winCornerRectL.width/2；
11.2)：WinRect.y=winUpLine+winUpRect.y；
11.3)：WinRect.width=winCornerRectR.x+winCornerRectR.width/2-
winRect.x；
11.4)：WinRect.height=winDownLine+winDownRect.y-winRect.y；
Wherein winRect.x is the upper left corner abscissa of winRect, and seat is indulged in the upper left corner that winRect.y is winRect
Mark, winRect.width are the width of winRect, and winRect.height is the height of winRect,
WinCornerRectL.width is the width of winCornerRectL.
Positioning action is carried out to vehicle window by using the method for the present invention, it must can effectively navigate to window edge, and
And there are inclined situations to position for vehicle window lower edges.
Description of the drawings
Fig. 1 is the image image chosen in the embodiment of the present invention；
Fig. 2 be the embodiment of the present invention in plateRect, detectRoiL, detectRoiR, winCornerRectL and
The position view of winCornerRectR；
Fig. 3 is the position view of winDownRect, winUpRect in the embodiment of the present invention；
Fig. 4 is the image winDownImg in the embodiment of the present invention；
Fig. 5 is the image winUpImg in the embodiment of the present invention；
Fig. 6 is vehicle window positioning result figure in the embodiment of the present invention.
Specific embodiment
It is specifically real that the vehicle window localization method based on vehicle window Corner Detection of the present invention is elaborated with reference to embodiment
Apply mode.It should be appreciated that specific embodiment described herein is only used for explaining the present invention, it is not intended to limit the present invention.
A kind of vehicle window localization method based on vehicle window Corner Detection of the present invention, includes the following steps：
Step 1：License Plate is carried out to the coloured image image containing vehicle, the car plate position navigated to is denoted as square
Shape region plateRect, the image image chosen in the present embodiment is as shown in Figure 1, the plateRect that License Plate obtains
Position is as shown in Figure 2；
Step 2：According to car plate position plateRect define vehicle window lower left corner angle point detection zone detectRoiL and
The detection zone detectRoiR of vehicle window lower right corner angle point, defines in the present embodiment according to plateRect positions
DetectRoiL and detectRoiR regions are as shown in Figure 2；
Step 3：The candidate region of all vehicle window lower left corners angle point is detected using SVM in detectRoiL, is collected
Close leftCornersList={ RLi| i=1,2 ..., n }, RLiRepresent i-th of vehicle window lower left corner angle point candidate region, n is represented
The quantity of vehicle window lower left corner angle point candidate region；In detectRoiR all vehicle window lower right corner angle point is detected using SVM
Candidate region obtains set rightCornersList={ RRj| j=1,2 ..., m }, RRjRepresent j-th of vehicle window lower right corner angle
Point candidate region, m represent the quantity of vehicle window lower right corner angle point candidate region；
Step 4：It is determined respectively from leftCornersList and rightCornersList using scan matching method
Final vehicle window lower left corner angle point region and vehicle window lower right corner angle point region, and be denoted as respectively winCornerRectL and
WinCornerRectR, winCornerRectL the and winCornerRectR positions chosen in the present embodiment are as shown in Figure 2；
Step 5：Vehicle window lower edge detection zone is defined according to the position of winCornerRectL and winCornerRectR,
WinDownRect is denoted as, the position of winDownRect is as shown in Figure 3 in the present embodiment；
Step 6：WinDownRect regions part is intercepted in image image, obtains vehicle window lower edge detection zone figure
Picture, and the obtained vehicle window lower edge detection zone image of interception is carried out image gray processing, the detection of Sobel horizontal edges and
The binary image finally obtained is denoted as winDownImg, obtained in the present embodiment by OTSU binarization operations
WinDownImg images are as shown in Figure 4；
Step 7：Floor projection is carried out to winDownImg, finds the vertical seat of floor projection value maximum in winDownImg
Scale value is denoted as winDownLine, and the winDownLine obtained in the present embodiment is equal to 129；
Step 8：Vehicle window edge detection region is defined according to the position of winDownRect, is denoted as winUpRect, at this
The position of winUpRect is as shown in Figure 3 in embodiment；
Step 9：WinUpRect regions part is intercepted in image image, obtains vehicle window edge detection area image,
And image gray processing, the detection of Sobel horizontal edges and OTSU are carried out to the vehicle window edge detection area image that interception obtains
The binary image finally obtained is denoted as winUpImg by binarization operation, the winUpImg images obtained in the present embodiment
As shown in Figure 5；
Step 10：Floor projection, the bottom-up scanning in winUpImg, until scanning to a certain are carried out to winUpImg
The floor projection value of position meets vehicle window top edge condition and then stops scanning, and the ordinate value of the position in winUpImg is denoted as
WinUpLine if not finding the floor projection value for meeting vehicle window top edge condition, enables winUpLine=
WinUpImg.heiht/2, wherein winUpImg.heiht are the height of winUpImg, are obtained in the present embodiment
The value of winUpLine is equal to 173；
Step 11：According to the position and winDownLine of the position of winCornerRectL and winCornerRectR
Position of the vehicle window in image image is calculated with winUpLine, is denoted as rectangle winRect, calculates in the present embodiment
The winRect regions arrived are as shown in Figure 6.
A kind of vehicle window localization method based on vehicle window Corner Detection, it is characterised in that the vehicle window lower left corner in step 2
The detection zone detectRoiL of angle point and the detection zone detectRoiR of vehicle window lower right corner angle point are defined as follows：
2.1)：DetectRoiL.x=plateRect.x-plateRect.width × 3；
2.2)：DetectRoiL.y=plateRect.y-plateRect.width × 3.5；
2.3)：DetectRoiL.width=plateRect.x+plateRect.width/2-detectRoi L.x；
2.4)：DetectRoiL.height=plateRect.y-plateRect.width-detectRoiL .y；
2.5)：DetectRoiR.x=plateRect.x+plateRect.width/2；
2.6)：DetectRoiR.y=plateRect.y-plateRect.width × 3.5；
2.7)：DetectRoiR.width=plateRect.width × 3.5；
2.8)：DetectRoiR.height=plateRect.y-plateRect.width-detectRoiR .y.
Wherein plateRect.x is the upper left corner abscissa of plateRect, and plateRect.y is the upper left of plateRect
Angle ordinate, plateRect.width are the width of plateRect, and the upper left corner that detectRoiL.x is detectRoiL is horizontal
Coordinate, detectRoiL.y are the upper left corner ordinate of detectRoiL, and detectRoiL.width is the width of detectRoiL
Degree, detectRoiL.height are the height of detectRoiL, and detectRoiR.x is the horizontal seat in the upper left corner of detectRoiR
Mark, detectRoiR.y are the upper left corner ordinate of detectRoiR, and detectRoiR.width is the width of detectRoiR,
DetectRoiR.height is the height of detectRoiR.
A kind of vehicle window localization method based on vehicle window Corner Detection, it is characterised in that in step 4 from
Determine that final vehicle window lower left corner angle point region and vehicle window are right in leftCornersList and rightCornersList respectively
The step of scan matching method in inferior horn angle point region, is as follows：
4.1)：For each element RR in rightCornersListj, one is found from leftCornersList
A element RLiSo that (RRj.y+RRj.height/2)–(RLi.y+RLi.height/2 absolute value) is minimum, and this is absolutely
Value is less than plateRect.width/2, if finding the RL of the condition of satisfactioni, then by RRjAnd RLiIt is stored as a pair of of candidate angular
Get off；
4.2)：For each element RL in leftCornersListi, one is found from rightCornersList
A element RRjSo that (RLi.y+RLi.height/2)–(RRj.y+RRj.height/2 absolute value) is minimum, and this is absolutely
Value is less than plateRect.width/2, if finding the RR of the condition of satisfactionj, then by RRjAnd RLiIt is stored as a pair of of candidate angular
Get off；
4.3)：For the candidate angular pair stored in step 4.1) and step 4.2), only retain those in step
4.1) candidate angular pair stored and in step 4.2) is repeated；
4.4)：For the candidate angular pair that step 4.3) remains, only choose one pair of which and cause (RRj.x+
RRj.width/2)–(RLi.x+RLi.width/2 maximum absolute value) then enables winCornerRectL be equal to what is chosen
RLi, winCornerRectR is equal to the RR chosenj。
Wherein RLi.x it is RLiUpper left corner abscissa, RLi.y it is RLiUpper left corner ordinate, RLi.width it is RLi's
Width, RLi.height it is RLiHeight, RRj.x it is RRjUpper left corner abscissa, RRj.y it is RRjUpper left corner ordinate,
RRj.width it is RRjWidth, RRj.height it is RRjHeight.
A kind of vehicle window localization method based on vehicle window Corner Detection, it is characterised in that vehicle window lower edge in step 5
Detection zone winDownRect is defined as follows：
5.1)：WinDownRect.x=winCornerRectL.x；
5.2)：WinDownRect.y=Top-(Buttom-Top)/2；
5.3)：WinDownRect.width=winCornerRectR.x+winCornerRectR.width-
winCornerRectL.x；
5.4)：WinDownRect.height=(Buttom-Top) × 2；
Wherein winDownRect.x is the upper left corner abscissa of winDownRect, and winDownRect.y is
The upper left corner ordinate of winDownRect, winDownRect.width are the width of winDownRect,
Height of the winDownRect.height for winDownRect, Top=min (winCornerRectL.y,
WinCornerRectR.y), Buttom=max (winCornerRectL.buttom, winCornerRectR.buttom),
WinCornerRectL.x is the upper left corner abscissa of winCornerRectL, and winCornerRectL.y is
The upper left corner ordinate of winCornerRectL, the lower left corner that winCornerRectL.buttom is winCornerRectL are indulged
Coordinate, winCornerRectR.x are the upper left corner abscissa of winCornerRectR, and winCornerRectR.y is
The upper left corner ordinate of winCornerRectR, the lower left corner that winCornerRectR.buttom is winCornerRectR are indulged
Coordinate, winCornerRectR.width are the width of winCornerRectR, and min expressions take smaller, max from two numbers
Expression takes the greater from two numbers.
A kind of vehicle window localization method based on vehicle window Corner Detection, it is characterised in that vehicle window top edge in step 8
Detection zone winUpRect is defined as follows：
8.1)：WinUpRect.x=winDownRect.x；
8.2)：WinUpRect.y=winDownRect.y-winDownRect.height；
8.3)：WinUpRect.width=winDownRect.width；
8.4)：WinUpRect.height=winDownRect.height；
Wherein winUpRect.x is the upper left corner abscissa of winUpRect, and winUpRect.y is the upper left of winUpRect
Angle ordinate, winUpRect.width are the width of winUpRect, and winUpRect.height is the height of winUpRect.
A kind of vehicle window localization method based on vehicle window Corner Detection, it is characterised in that a certain position in step 10
The condition that floor projection value meets vehicle window top edge is as follows：
10.1)：project[h]>projectThresh；
10.2)：abs((projectLeft[h]-projectRight[h])/(winUpImg.width/2))<0.15.
Wherein h is the ordinate of image winUpImg, and project [h] represents that ordinate is at h in image winUpImg
The sum of the floor projection value of floor projection value and ordinate at h-1, h-2, h-3, h-4, h-5, projectThresh represent pre-
The projection value threshold value first set is set as left one side of something that 750, projectLeft [h] represents image winUpImg in the present embodiment
The sum of the floor projection value of floor projection value and ordinate of the ordinate at h at h-1, h-2, h-3, h-4, h-5 in part,
Floor projection value and ordinate of the ordinate at h in half rim portion of the right side of projectRight [h] expression images winUpImg
The sum of floor projection value at h-1, h-2, h-3, h-4, h-5, winUpImg.width represent the width of image winUpImg,
Abs expressions take absolute value.
A kind of vehicle window localization method based on vehicle window Corner Detection, it is characterised in that vehicle window is in image in step 11
The calculating of position winRect in image is as follows：
11.1)：WinRect.x=winCornerRectL.x+winCornerRectL.width/2；
11.2)：WinRect.y=winUpLine+winUpRect.y；
11.3)：WinRect.width=winCornerRectR.x+winCornerRectR.width/2-
winRect.x；
11.4)：WinRect.height=winDownLine+winDownRect.y-winRect.y；
Wherein winRect.x is the upper left corner abscissa of winRect, and seat is indulged in the upper left corner that winRect.y is winRect
Mark, winRect.width are the width of winRect, and winRect.height is the height of winRect,
WinCornerRectL.width is the width of winCornerRectL.
In the present embodiment, by handling above, it can be seen that the vehicle window region in Fig. 6 is accurately positioned out.
Claims (7)
1. a kind of vehicle window localization method based on vehicle window Corner Detection, includes the following steps：
Step 1：License Plate is carried out to the coloured image image containing vehicle, the car plate position navigated to is denoted as rectangle region
Domain plateRect；
Step 2：The detection zone detectRoiL and vehicle window of vehicle window lower left corner angle point are defined according to car plate position plateRect
The detection zone detectRoiR of lower right corner angle point；
Step 3：The candidate region of all vehicle window lower left corners angle point is detected using SVM in detectRoiL, is gathered
LeftCornersList={ RLi| i=1,2 ..., n }, RLiRepresent i-th of vehicle window lower left corner angle point candidate region, n represents vehicle
The quantity of window lower left corner angle point candidate region；The time of all vehicle window lower right corner angle point is detected using SVM in detectRoiR
Favored area obtains set rightCornersList={ RRj| j=1,2 ..., m }, RRjRepresent j-th of vehicle window lower right corner angle point
Candidate region, m represent the quantity of vehicle window lower right corner angle point candidate region；
Step 4：It is determined respectively finally from leftCornersList and rightCornersList using scan matching method
Vehicle window lower left corner angle point region and vehicle window lower right corner angle point region, and be denoted as respectively winCornerRectL and
winCornerRectR；
Step 5：Vehicle window lower edge detection zone is defined according to the position of winCornerRectL and winCornerRectR, is denoted as
winDownRect；
Step 6：WinDownRect regions part is intercepted in image image, obtains vehicle window lower edge detection zone image, and
Image gray processing, the detection of Sobel horizontal edges and OTSU bis- are carried out to the vehicle window lower edge detection zone image that interception obtains
Value operates, and the binary image finally obtained is denoted as winDownImg；
Step 7：Floor projection is carried out to winDownImg, finds the ordinate value of floor projection value maximum in winDownImg
It is denoted as winDownLine；
Step 8：Vehicle window edge detection region is defined according to the position of winDownRect, is denoted as winUpRect；
Step 9：WinUpRect regions part is intercepted in image image, obtains vehicle window edge detection area image, and right
It intercepts obtained vehicle window edge detection area image and carries out image gray processing, the detection of Sobel horizontal edges and OTSU two-values
Change operation, the binary image finally obtained is denoted as winUpImg；
Step 10：Floor projection, the bottom-up scanning in winUpImg, until scanning to a certain position are carried out to winUpImg
Floor projection value meet vehicle window top edge condition and then stop scanning, the ordinate value of the position in winUpImg is denoted as
WinUpLine if not finding the floor projection value for meeting vehicle window top edge condition, enables winUpLine=
WinUpImg.heiht/2, wherein winUpImg.heiht are the height of winUpImg；
Step 11：According to the position and winDownLine of the position of winCornerRectL and winCornerRectR and
Position of the vehicle window in image image is calculated in winUpLine, is denoted as rectangle winRect.
2. a kind of vehicle window localization method based on vehicle window Corner Detection as described in claim 1, it is characterised in that：In step 2
The detection zone detectRoiL's of the vehicle window lower left corner angle point and detection zone detectRoiR of vehicle window lower right corner angle point determines
Justice is as follows：
2.1)：DetectRoiL.x=plateRect.x-plateRect.width × 3；
2.2)：DetectRoiL.y=plateRect.y-plateRect.width × 3.5；
2.3)：DetectRoiL.width=plateRect.x+plateRect.width/2-detectRoi L.x；
2.4)：DetectRoiL.height=plateRect.y-plateRect.width-detectRoiL .y；
2.5)：DetectRoiR.x=plateRect.x+plateRect.width/2；
2.6)：DetectRoiR.y=plateRect.y-plateRect.width × 3.5；
2.7)：DetectRoiR.width=plateRect.width × 3.5；
2.8)：DetectRoiR.height=plateRect.y-plateRect.width-detectRoiR .y.
Wherein plateRect.x is the upper left corner abscissa of plateRect, and the upper left corner that plateRect.y is plateRect is indulged
Coordinate, plateRect.width are the width of plateRect, and detectRoiL.x is the upper left corner abscissa of detectRoiL,
DetectRoiL.y is the upper left corner ordinate of detectRoiL, and detectRoiL.width is the width of detectRoiL,
DetectRoiL.height is the height of detectRoiL, and detectRoiR.x is the upper left corner abscissa of detectRoiR,
DetectRoiR.y is the upper left corner ordinate of detectRoiR, and detectRoiR.width is the width of detectRoiR,
DetectRoiR.height is the height of detectRoiR.
3. a kind of vehicle window localization method based on vehicle window Corner Detection as described in claim 1, it is characterised in that：In step 4
Determine final vehicle window lower left corner angle point region and vehicle window respectively from leftCornersList and rightCornersList
The step of scan matching method in lower right corner angle point region, is as follows：
4.1)：For each element RR in rightCornersListj, an element is found from leftCornersList
RLiSo that (RRj.y+RRj.height/2)–(RLi.y+RLi.height/2 absolute value) is minimum, and the absolute value is small
In plateRect.width/2, if finding the RL of the condition of satisfactioni, then by RRjAnd RLiIt is stored as a pair of of candidate angular；
4.2)：For each element RL in leftCornersListi, an element is found from rightCornersList
RRjSo that (RLi.y+RLi.height/2)–(RRj.y+RRj.height/2 absolute value) is minimum, and the absolute value is small
In plateRect.width/2, if finding the RR of the condition of satisfactionj, then by RRjAnd RLiIt is stored as a pair of of candidate angular；
4.3)：For the candidate angular pair stored in step 4.1) and step 4.2), only retain those in step 4.1) and
The candidate angular pair stored is repeated in step 4.2)；
4.4)：For the candidate angular pair that step 4.3) remains, only choose one pair of which and cause (RRj.x+RRj.width/
2)–(RLi.x+RLi.width/2 maximum absolute value) then enables winCornerRectL be equal to the RL choseni,
WinCornerRectR is equal to the RR chosenj；
Wherein RLi.x it is RLiUpper left corner abscissa, RLi.y it is RLiUpper left corner ordinate, RLi.width it is RLiWidth,
RLi.height it is RLiHeight, RRj.x it is RRjUpper left corner abscissa, RRj.y it is RRjUpper left corner ordinate,
RRj.width it is RRjWidth, RRj.height it is RRjHeight.
4. a kind of vehicle window localization method based on vehicle window Corner Detection as described in claim 1, it is characterised in that：In step 5
Vehicle window lower edge detection zone winDownRect is defined as follows：
5.1)：WinDownRect.x=winCornerRectL.x；
5.2)：WinDownRect.y=Top-(Buttom-Top)/2；
5.3)：WinDownRect.width=winCornerRectR.x+winCornerRectR.width-
winCornerRectL.x；
5.4)：WinDownRect.height=(Buttom-Top) × 2；
Wherein winDownRect.x is the upper left corner abscissa of winDownRect, and winDownRect.y is winDownRect's
Upper left corner ordinate, winDownRect.width are the width of winDownRect, and winDownRect.height is
The height of winDownRect, Top=min (winCornerRectL.y, winCornerRectR.y), Buttom=max
(winCornerRectL.buttom, winCornerRectR.buttom), winCornerRectL.x are
The upper left corner abscissa of winCornerRectL, winCornerRectL.y are the upper left corner ordinate of winCornerRectL,
WinCornerRectL.buttom is the lower left corner ordinate of winCornerRectL, and winCornerRectR.x is
The upper left corner abscissa of winCornerRectR, winCornerRectR.y are the upper left corner ordinate of winCornerRectR,
WinCornerRectR.buttom is the lower left corner ordinate of winCornerRectR, and winCornerRectR.width is
The width of winCornerRectR, min expressions take smaller from two numbers, and max expressions take the greater from two numbers.
5. a kind of vehicle window localization method based on vehicle window Corner Detection as described in claim 1, it is characterised in that：In step 8
Vehicle window edge detection region winUpRect is defined as follows：
8.1)：WinUpRect.x=winDownRect.x；
8.2)：WinUpRect.y=winDownRect.y-winDownRect.height；
8.3)：WinUpRect.width=winDownRect.width；
8.4)：WinUpRect.height=winDownRect.height；
Wherein winUpRect.x is the upper left corner abscissa of winUpRect, and the upper left corner that winUpRect.y is winUpRect is indulged
Coordinate, winUpRect.width are the width of winUpRect, and winUpRect.height is the height of winUpRect.
6. a kind of vehicle window localization method based on vehicle window Corner Detection as described in claim 1, it is characterised in that：In step 10
The condition that the floor projection value of a certain position meets vehicle window top edge is as follows：
10.1)：project[h]>projectThresh；
10.2)：abs((projectLeft[h]-projectRight[h])/(winUpImg.width/2))<0.15.
Wherein h is the ordinate of image winUpImg, and project [h] represents level of the ordinate at h in image winUpImg
The sum of the floor projection value of projection value and ordinate at h-1, h-2, h-3, h-4, h-5, projectThresh expressions are set in advance
Fixed projection value threshold value is set as half rim portion of a left side that 750, projectLeft [h] represents image winUpImg in the present embodiment
The sum of the floor projection value of floor projection value and ordinate of the middle ordinate at h at h-1, h-2, h-3, h-4, h-5,
Floor projection value and ordinate of the ordinate at h in half rim portion of the right side of projectRight [h] expression images winUpImg
The sum of floor projection value at h-1, h-2, h-3, h-4, h-5, winUpImg.width represent the width of image winUpImg,
Abs expressions take absolute value.
7. a kind of vehicle window localization method based on vehicle window Corner Detection as described in claim 1, it is characterised in that：In step 11
The calculating of position winRect of the vehicle window in image image is as follows：
11.1)：WinRect.x=winCornerRectL.x+winCornerRectL.width/2；
11.2)：WinRect.y=winUpLine+winUpRect.y；
11.3)：WinRect.width=winCornerRectR.x+winCornerRectR.width/2-wi nRect.x；
11.4)：WinRect.height=winDownLine+winDownRect.y-winRect.y；
Wherein winRect.x is the upper left corner abscissa of winRect, and winRect.y is the upper left corner ordinate of winRect,
WinRect.width is the width of winRect, and winRect.height is the height of winRect,
WinCornerRectL.width is the width of winCornerRectL.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711133210.8A CN108182376B (en) | 2017-11-15 | 2017-11-15 | Vehicle window corner detection-based vehicle window positioning method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711133210.8A CN108182376B (en) | 2017-11-15 | 2017-11-15 | Vehicle window corner detection-based vehicle window positioning method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108182376A true CN108182376A (en) | 2018-06-19 |
| CN108182376B CN108182376B (en) | 2020-12-08 |
Family
ID=62544971
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711133210.8A Active CN108182376B (en) | 2017-11-15 | 2017-11-15 | Vehicle window corner detection-based vehicle window positioning method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108182376B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110059623A (en) * | 2019-04-18 | 2019-07-26 | 北京字节跳动网络技术有限公司 | Method and apparatus for generating information |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102779267A (en) * | 2011-05-12 | 2012-11-14 | 株式会社理光 | Method and device for detection of specific object region in image |
| US20130101227A1 (en) * | 2011-10-19 | 2013-04-25 | Lee F. Holeva | Identifying and evaluating possible horizontal and vertical lines intersecting potential pallet features |
| CN103324935A (en) * | 2013-06-27 | 2013-09-25 | 中山大学 | Method for vehicle positioning and region segmenting in image |
| CN103854029A (en) * | 2014-02-21 | 2014-06-11 | 杭州奥视图像技术有限公司 | Detection method for front automobile window top right corner point |
| CN103871052A (en) * | 2014-02-21 | 2014-06-18 | 杭州奥视图像技术有限公司 | Straight line detection-based safety belt detecting algorithm |
| CN104376561A (en) * | 2014-11-19 | 2015-02-25 | 青岛海信网络科技股份有限公司 | Safety belt detection method |
| CN105373783A (en) * | 2015-11-17 | 2016-03-02 | 高新兴科技集团股份有限公司 | Seat belt not-wearing detection method based on mixed multi-scale deformable component model |
| CN105469057A (en) * | 2015-11-27 | 2016-04-06 | 成都神州数码索贝科技有限公司 | Hough line detection and projection-based automobile window detection method |
| CN106250824A (en) * | 2016-07-21 | 2016-12-21 | 乐视控股（北京）有限公司 | Vehicle window localization method and system |
| CN106845482A (en) * | 2017-01-23 | 2017-06-13 | 南京邮电大学盐城大数据研究院有限公司 | A kind of license plate locating method |
-
2017
- 2017-11-15 CN CN201711133210.8A patent/CN108182376B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102779267A (en) * | 2011-05-12 | 2012-11-14 | 株式会社理光 | Method and device for detection of specific object region in image |
| US20130101227A1 (en) * | 2011-10-19 | 2013-04-25 | Lee F. Holeva | Identifying and evaluating possible horizontal and vertical lines intersecting potential pallet features |
| CN103324935A (en) * | 2013-06-27 | 2013-09-25 | 中山大学 | Method for vehicle positioning and region segmenting in image |
| CN103854029A (en) * | 2014-02-21 | 2014-06-11 | 杭州奥视图像技术有限公司 | Detection method for front automobile window top right corner point |
| CN103871052A (en) * | 2014-02-21 | 2014-06-18 | 杭州奥视图像技术有限公司 | Straight line detection-based safety belt detecting algorithm |
| CN104376561A (en) * | 2014-11-19 | 2015-02-25 | 青岛海信网络科技股份有限公司 | Safety belt detection method |
| CN105373783A (en) * | 2015-11-17 | 2016-03-02 | 高新兴科技集团股份有限公司 | Seat belt not-wearing detection method based on mixed multi-scale deformable component model |
| CN105469057A (en) * | 2015-11-27 | 2016-04-06 | 成都神州数码索贝科技有限公司 | Hough line detection and projection-based automobile window detection method |
| CN106250824A (en) * | 2016-07-21 | 2016-12-21 | 乐视控股（北京）有限公司 | Vehicle window localization method and system |
| CN106845482A (en) * | 2017-01-23 | 2017-06-13 | 南京邮电大学盐城大数据研究院有限公司 | A kind of license plate locating method |
Non-Patent Citations (3)
| Title |
|---|
| HONGJIN ZHU ET AL: "Moving Vehicle Detection and Tracking in Traffic Images Based on Horizontal Edges", 《TELKOMNIKA INDONESIAN JOURNAL OF ELECTRICAL ENGINEERING》 * |
| 徐志刚 等: "基于角点检测和边缘提取的车牌定位方法", 《科学技术与工程》 * |
| 高飞 等: "基于灰度跳变与字符间隔模式的车牌定位方法研究", 《计算机测量与控制》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110059623A (en) * | 2019-04-18 | 2019-07-26 | 北京字节跳动网络技术有限公司 | Method and apparatus for generating information |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108182376B (en) | 2020-12-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US9158979B2 (en) | Vehicle vicinity monitoring device | |
| KR101517181B1 (en) | System and method for warning lane departure | |
| EP1671216B1 (en) | Moving object detection using low illumination depth capable computer vision | |
| JP5900623B2 (en) | Three-dimensional object detection apparatus and three-dimensional object detection method | |
| JP4988786B2 (en) | Boundary line recognition device | |
| CN103984950B (en) | A kind of moving vehicle brake light status recognition methods for adapting to detection on daytime | |
| CN109948552B (en) | Method for detecting lane line in complex traffic environment | |
| CN105216797B (en) | Method of overtaking and system | |
| JP5787024B2 (en) | Three-dimensional object detection device | |
| CN107991671A (en) | A kind of method based on radar data and vision signal fusion recognition risk object | |
| JP4692344B2 (en) | Image recognition device | |
| CN105740782A (en) | Monocular vision based driver lane-changing process quantization method | |
| CN105260705B (en) | A kind of driver's making and receiving calls behavioral value method suitable under multi-pose | |
| Zou et al. | Robust nighttime vehicle detection by tracking and grouping headlights | |
| US9773174B2 (en) | Vehicle detection method based on thermal imaging | |
| CN104915642B (en) | Front vehicles distance measuring method and device | |
| CN108108656A (en) | A kind of vehicle window accurate positioning method based on vehicle window Corner Detection and multi-direction projection | |
| CN103593981A (en) | Vehicle model identification method based on video | |
| CN104700066A (en) | Method and device for detecting whether driver wears safety belt or not | |
| KR101809088B1 (en) | Apparatus and method for forward collision warning | |
| CN108182376A (en) | A kind of vehicle window localization method based on vehicle window Corner Detection | |
| KR101850794B1 (en) | Parking assist appratus and method for assisting parking | |
| KR101347886B1 (en) | Method and Apparatus for Road Lane Recognition by Surface Region and Geometry Information | |
| CN109101980B (en) | Automobile weightlessness detection and control method based on machine learning | |
| CN106355159A (en) | Method for rapidly detecting zebra crossing based on vertical projection integration |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |