CN109635733A - View-based access control model conspicuousness and the modified parking lot of queue and vehicle target detection method - Google Patents
View-based access control model conspicuousness and the modified parking lot of queue and vehicle target detection method Download PDFInfo
- Publication number
- CN109635733A CN109635733A CN201811517627.9A CN201811517627A CN109635733A CN 109635733 A CN109635733 A CN 109635733A CN 201811517627 A CN201811517627 A CN 201811517627A CN 109635733 A CN109635733 A CN 109635733A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- bbsm
- notable
- value
- parking lot
- 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
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/66—Analysis of geometric attributes of image moments or centre of gravity
-
- 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
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/08—Detecting or categorising vehicles
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Geometry (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
- Traffic Control Systems (AREA)
Abstract
View-based access control model conspicuousness and the modified parking lot of queue and vehicle target detection method, it belongs to technical field of vehicle detection in parking lot.The present invention solves the problems, such as that existing remote sensing images vehicle target detection method processing speed is slow, vehicle target detection effect is poor.The present invention designs a kind of notable figure BBSM based on brightness for car park areas coarse extraction according to car park areas brightness, and the color characteristic of car park areas and region feature is recycled accurately to extract parking lot profile；In the parking lot profile that each essence is extracted, extracting may be comprising the suspicious region of vehicle, design corrects vehicle platoon arragement direction based on the calculation method of the vehicle platoon arragement direction of edge statistics model, finally vehicle platoon is cut using sliding window method to be sliced to extract doubtful vehicle, after the HOG feature for extracting slice, two classification are carried out using SVM classifier, the target label that would be classified as vehicle returns original image, realizes vehicle detection；The present invention is applied to technical field of vehicle detection in parking lot.
Description
Technical field
The invention belongs to technical field of vehicle detection in parking lot, and in particular to a kind of parking lot and vehicle target detection side
Method.
Background technique
In remote sensing images the detection of vehicle target urban planning, in terms of all with important application meaning
Justice.But existing vehicle detection studies the more detection for concentrating on road vehicle, for the vehicle detection being parked in parking lot
Study relatively fewer, method can be roughly divided into two kinds: the objective classification method based on template matching；Based on feature extraction
Objective classification method.
Method based on template matching is one of the basic skills of target classification, passes through calculation template image and mesh to be identified
Its similarity of the Euclidean distance isometry of logo image pixel, to judge the classification of region to be identified or target.But it can only
It is matched with the sample contained in template library, causes the generalization ability of algorithm lower, do not have good illumination invariant, rotation
Invariance and view transformation invariance, and point-by-point computational complexity is high, and operation time is long, it is impossible to be used in processing in real time.It is based on
The classification method of feature is more common and effective method to remote sensing extracted region and target detection identification.By analyzing target
With difference of the false-alarm in some features, such as Scale invariant features transform feature, histograms of oriented gradients (Histogram of
Oriented Gradient, HOG) feature, geometric invariant moment, length-width ratio, textural characteristics etc., it then selects wherein to be conducive to point
The feature of class describes slice, finally using the methods of machine learning to tagsort.
The main problem of the vehicle target detection of existing remote sensing images is that region span is wide, and data volume is big and in Asia
Target detail feature is unobvious under meter level resolution ratio.It is big that this will lead to processing system EMS memory occupation, and treatment process is slower, and
The feature that vehicle sample slice can extract is less, influences the classifying quality of classifier, and the effect for causing vehicle target to detect is poor.
Therefore, a kind of processing that not only can be reduced hash is studied, but also the method that can accurately detect vehicle target is just shown
It obtains necessary.
Summary of the invention
The purpose of the present invention is cause to handle greatly for the data volume of the existing remote sensing images vehicle target detection method of solution
The feature that speed is slow and vehicle sample slice can extract leads to the problem of vehicle target detection effect difference less.
The technical solution adopted by the present invention to solve the above technical problem is: view-based access control model conspicuousness and queue is modified stops
Parking lot and vehicle target detection method, method includes the following steps:
Step 1: input sub-meter grade high-resolution optical remote sensing image, brightness and vision based on parking lot are significant
Property, calculates the BBSM notable figure of the sub-meter grade high-resolution optical remote sensing image of input, and by calculated BBSM notable figure
Binaryzation；
Step 2: carrying out the segmentation of super-pixel to the BBSM notable figure after binaryzation according to the region feature in parking lot, obtain
Whole super-pixel block after segmentation, setting screening conditions screen whole super-pixel block after segmentation, what calculating sifting went out
The mass center figure CDM of each super-pixel block(i′,j′), utilize the mass center figure CDM of calculated each super-pixel block(i′,j′)Update two-value
BBSM notable figure after change, the mass center figure CDM of the BBSM notable figure after obtaining binaryzation；
The mass center Density Distribution index CDDI figure of BBSM notable figure after calculating binaryzation according to obtained CDM, and according to
CDDI figure obtains ROI image, i.e. acquisition coarse extraction car park areas image；
Step 3: obtaining essence according to the car park areas image of coarse extraction and the color characteristic and region feature in parking lot and mentioning
The car park areas image taken completes parking field detecting；
Step 4: calculating the SR notable figure for the car park areas image that essence is extracted, and extract the doubtful vehicle in SR notable figure
Region；
The angle for calculating the queue arragement direction of each doubtful vehicle region, according to the angle acquired by all doubtful vehicles
Region rotates the queue amendment that doubtful vehicle region is completed to the direction of horizontal arrangement；
Step 5: postrotational all doubtful vehicle regions are cut into slice using sliding window cutting method, and to cutting
At slice carry out HOG feature extraction and using SVM classifier to the HOG tagsort of extraction would be classified as the slice mark of vehicle
Remember Hui Yuantu, completes vehicle detection.
The beneficial effects of the present invention are: the invention proposes view-based access control model conspicuousness and the modified parking lot of queue and vehicles
Object detection method, the present invention is high according to car park areas brightness first and Luminance Distribution section is small relative to input figure accounting
Brightness devises the coarse extraction that a kind of notable figure BBSM based on brightness is used for car park areas, then recycles
Car park areas is colourless and color characteristic of high brightness and the big region feature of region area, further accurate to extract parking lot wheel
Exterior feature completes parking field detecting；In the parking lot profile that each essence is extracted, being extracted using SR notable figure may be comprising vehicle
Suspicious region, and a kind of calculation method of vehicle platoon arragement direction based on edge statistics model is devised for correcting vehicle
The arragement direction of queue finally cuts vehicle platoon using sliding window method, extracts the slice of a large amount of doubtful vehicles, these are cut
Piece is classified after extracting HOG feature using the true and false vehicle slice of SVM classifier progress, and the target label that would be classified as vehicle returns original
Figure, finally realizes vehicle detection；The car park areas extracting method and corrected and SR notable figure based on queue that the present invention designs
Vehicle checking method can guarantee that the accuracy of vehicle detection reaches 85% or more.
And be compared with the traditional method, the calculation method for the BBSM notable figure that the present invention designs can substantially speed up vehicle inspection
The processing speed of survey.
Detailed description of the invention
Fig. 1 is a kind of view-based access control model conspicuousness of the invention and the modified parking lot of queue and vehicle target detection method
Flow chart；
Fig. 2 is the signal of the sub-meter grade high-resolution optical remote sensing image of input described in the specific embodiment of the invention two
Figure；
Fig. 3 is the calculated BBSM notable figure of the specific embodiment of the invention two；
Fig. 4 is the schematic diagram of the car park areas image of coarse extraction described in the specific embodiment of the invention three；
Fig. 5 is the schematic diagram for the car park areas image that essence described in the specific embodiment of the invention four is extracted；
Fig. 6 is the SR notable figure of certain car park areas described in the specific embodiment of the invention five；
Fig. 7 is the binaryzation SR notable figure of certain car park areas described in the specific embodiment of the invention five；
Fig. 8 is that doubtful vehicle region direction described in the specific embodiment of the invention five the case where deviation schematic diagram occurs；
Fig. 9 is the edge detection results figure of doubtful vehicle region described in the specific embodiment of the invention five；
Figure 10 is AngleNum statistical results chart described in the specific embodiment of the invention five；
Wherein: abscissa is angle value, and ordinate is point to quantity；
Figure 11 is 1 figure of vehicle target testing result of the invention；
Figure 12 is 2 figure of vehicle target testing result of the invention；
Figure 13 is 3 figure of vehicle target testing result of the invention；
Figure 14 is 4 figure of vehicle target testing result of the invention；
Specific embodiment
Specific embodiment 1: as shown in Figure 1, view-based access control model conspicuousness described in present embodiment and queue is modified stops
Parking lot and vehicle target detection method, method includes the following steps:
Step 1: input sub-meter grade high-resolution optical remote sensing image, brightness and vision based on parking lot are significant
Property, calculates the BBSM notable figure of the sub-meter grade high-resolution optical remote sensing image of input, and by calculated BBSM notable figure
Binaryzation；
Step 2: carrying out the segmentation of super-pixel to the BBSM notable figure after binaryzation according to the region feature in parking lot, obtain
Whole super-pixel block after segmentation, setting screening conditions screen whole super-pixel block after segmentation, what calculating sifting went out
The mass center figure CDM of each super-pixel block(i′,j′), utilize the mass center figure CDM of calculated each super-pixel block(i′,j′)Update two-value
BBSM notable figure after change, the mass center figure CDM of the BBSM notable figure after obtaining binaryzation；
The mass center Density Distribution index CDDI figure of BBSM notable figure after calculating binaryzation according to obtained CDM, and according to
CDDI figure obtains ROI image, i.e. acquisition coarse extraction car park areas image；
Step 3: obtaining essence according to the car park areas image of coarse extraction and the color characteristic and region feature in parking lot and mentioning
The car park areas image taken completes parking field detecting；
Step 4: calculating the SR notable figure for the car park areas image that essence is extracted, and extract the doubtful vehicle in SR notable figure
Region；
The angle for calculating the queue arragement direction of each doubtful vehicle region, according to the angle acquired by all doubtful vehicles
Region rotates the queue amendment that doubtful vehicle region is completed to the direction of horizontal arrangement；
Step 5: postrotational all doubtful vehicle regions are cut into slice using sliding window cutting method, and to cutting
At slice carry out HOG feature extraction and using SVM classifier to the HOG tagsort of extraction would be classified as the slice mark of vehicle
Remember Hui Yuantu, completes vehicle detection.
Specific embodiment 2: the present embodiment is different from the first embodiment in that: the specific mistake of the step 1
Journey are as follows:
Step 1 one, as shown in Fig. 2, input sub-meter grade high-resolution optical remote sensing image, by the sub-meter grade high score of input
After resolution remote sensing image is transformed into luminance channel, luminance picture I is obtained, the size of luminance picture I is M × N, M and N difference
Represent the height and width of luminance picture I；
Count all pixels point corresponding brightness value u, u=0 ..., 255 in luminance picture I；
Statistical result PuIndicate the pixel number that brightness value is u in luminance picture I, such as P14Indicate that brightness value is in I
14 pixel number then calculates the BBSM value of each pixel in luminance picture I are as follows:
Wherein: (i, j) represents any one pixel in luminance picture I, I(i,j)It represents any one in luminance picture I
The brightness value of a pixel (i, j), BBSM(i,j)Represent the value of the BBSM of pixel (i, j)；
Then the BBSM value of each pixel forms BBSM notable figure；The BBSM notable figure is as shown in Figure 3；
Step 1 two chooses binarization threshold T1, doing binaryzation to BBSM notable figure, to obtain the BBSM after binaryzation significant
Scheme BBSM':
T1=0.3 × max (BBSM)+0.7 × min (BBSM)
Wherein: max (BBSM) is the maximum value of BBSM notable figure, and min (BBSM) is the minimum value of BBSM notable figure；
The then BBSM value after the binaryzation of pixel (i, j) are as follows:
Calculate the conventional method of BBSM notable figure are as follows: input sub-meter grade high-resolution optical remote sensing image, by the Asia of input
After meter level high-resolution optical remote sensing image is transformed into luminance channel, luminance picture I is obtained, the size of luminance picture I is M × N,
Wherein: i=1 ..., M, j=1 ..., N, then in luminance picture I pixel BBSM value are as follows:
Wherein: M and N respectively represents the height and width of luminance picture I, and (i, j) represents any one in luminance picture I
Pixel, Ii,jRepresent the brightness value of any one pixel in luminance picture I, BBSM(i,j)For the BBSM value of pixel (i, j),
D(I(i,j),I(m,n)) it is brightness value I(i,j)With brightness value I(m,n)Absolute difference, i.e.,
D(I(i,j),I(m,n))=| I(i,j)-I(m,n)|
The BBSM value of each pixel forms BBSM notable figure；
Due in luminance picture I, the pixel of same brightness should BBSM value having the same, so and conventional method
It compares, the BBSM remarkable picture capturing method of present embodiment accelerates processing speed.
Specific embodiment 3: the present embodiment is different from the first embodiment in that: the specific mistake of the step 2
Journey are as follows:
BBSM has only used the brightness in parking lot in step 2 one, step 1, next to utilize the face in parking lot
Feature；
In BBSM notable figure after binarization, the cutting step-length of short transverse is taken to be W, take the cutting step-length of width direction
For H, the BBSM notable figure after binaryzation is cut into using W and HA super-pixel block SP, each super-pixel block its
It is in fact a small binary map, when W cannot be divided exactly by M or H cannot be evenly divisible by N, then removes a small amount of fringe region of image,
To guarantee that W can be divided exactly by M and H can be evenly divisible by N.
Define the weight W of each super-pixel blockSPAre as follows:
Wherein: SP(i′,j′)For the BBSM value after the binaryzation of pixel in super-pixel block (i ', j ')；
Step 2 two, setting screening conditions are WSP> β WH, in which: β is sieveing coeffecient, utilizes the screening item of setting
Part is divided obtained super-pixel block to step 2 one and is screened, and the super-pixel block that weight is greater than screening conditions is filtered out, and rejects
Weight is not more than the super-pixel block of screening conditions；
The mass center figure CDM for each super-pixel block that step 2 three, calculating sifting go out(i′,j′): CDM refers to identifying each
The binary map of a super-pixel block mass center,
The calculation method of CDM is
Wherein:WithIndicate the mass center of super-pixel block, that is, each super-pixel block intermediate value be 1 all the points it is flat
Equal abscissa value and ordinate value.Exist in a certain size the region put centered on indicating the point that i.e. mass center figure intermediate value is 1
Parking lot.Utilize the mass center figure CDM of calculated each super-pixel block(i′,j′)BBSM notable figure after updating binaryzation, obtains
The mass center figure CDM of BBSM notable figure after binaryzation；
Step 2 four, with the unit matrix Te having a size of k × kk×kWith the mass center figure CDM of the BBSM notable figure after binaryzation
Convolution obtains the mass center Density Distribution index CDDI figure of the BBSM notable figure after binaryzation；
Wherein:Wherein the value of k is 120, this is allowed under 4 meters of resolution ratio, 120 ×
120 pixels are corresponding, and having a size of 480m × 480m, the size than most of parking lots is big, therefore in convolution, can ensure that
The ROI image of generation can cover all parking lots and omit without having, to prevent occurring excessive false dismissal when final vehicle detection.
Step 2 five chooses binarization threshold T2, by the mass center Density Distribution index of the BBSM notable figure after binaryzation
ROI image is obtained after CDDI figure binaryzation, obtains the car park areas image of coarse extraction.The car park areas image of coarse extraction is such as
Shown in Fig. 4, that is to say, that as long as CDDI is not 0 point, all take 1.
Specific embodiment 4: the present embodiment is different from the first embodiment in that: the specific mistake of the step 3
Journey are as follows:
Step 3 one rejects non-road surface extraneous areas, parking lot color characteristic body using the color characteristic that parking lot has
Neutral grey white colour is presented in present place, and brightness is higher,
In the car park areas image of coarse extraction, is screened by setting the value of tri- wave bands of R, G, B and meet screening item
The pixel of part, in which: screening conditions are as follows: the value of tri- wave bands of R, G, B is in the range of (130,250), and any two
The difference of wave band must not exceed 40；
Step 3 two rejects road area using the region feature that parking lot has, and region feature is in parking lot relative to it
His road surface and floor area have biggish area；
Holes filling is carried out to the selection result of step 3 one, rejects road of the width less than 10 meters using etching operation
Afterwards, then the area of each connected domain is calculated, to remove connected domain of the area less than 800 square metres；It is obtained after expansion reduction
The car park areas essence of sub-meter grade high-resolution optical remote sensing image extracts the car park areas image as a result, acquisition essence extraction,
Complete parking field detecting.It is as shown in Figure 5 that car park areas essence extracts result；
Specific embodiment 5: the present embodiment is different from the first embodiment in that: the specific mistake of the step 4
Journey are as follows:
The car park areas image that essence is extracted is carried out Fourier transformation to calculate amplitude spectrum A (f) and phase by step 4 one
It composes P (f):
Wherein: I ' (x) is the luminance picture in the car park areas that essence is extracted；Indicate that I ' (x) takes in Fu
The amplitude of leaf transformation,Indicate that I ' (x) takes the phase value of Fourier transformation；Represent Fourier transformation；
Amplitude spectrum A (f) is become into logarithmic spectrum L (f), then (filtering mode is 3 × 3 to logarithmic spectrum progress linear space filtering
Mean filter), make the difference logarithmic spectrum L (f) and linear space filter result to obtain residual spectra R (f)；
L (f)=log (A (f))
R (f)=L (f)-hn(f)*L(f)
Wherein: hnIt (f) is mean filter operator；
Residual spectra R (f) and phase spectrum P (f) is carried out inverse Fourier transform and obtains inverse Fourier transform as a result, sharp again
Carry out linear space to inverse Fourier transform result with Gaussian filter to filter to obtain SR notable figure: obtained SR notable figure is such as
Shown in Fig. 6,
Wherein: g (x) is gaussian filtering operator；SR (x) represents SR notable figure；Represent inverse Fourier transform；
It is normalized to SR notable figure is obtained, then binaryzation, setting two is carried out to the SR notable figure after normalization
Value threshold value T3The maximum value of=0.12max (SR), max (SR) for SR notable figure, the SR notable figure after obtaining binaryzation, then
The region that SR notable figure intermediate value after binaryzation is 1 is doubtful vehicle region；I.e. doubtful vehicle region is that a width identifies institute
There is the binary map of vehicle suspicious region, as shown in Figure 7.
On parking lot, the most of equal Assembled distribution of vehicle, i.e. more vehicles are parked into a pile, are with the same direction
Region arragement direction in the image of the suspicious region made containing vehicle is horizontal, and the invention proposes object in a kind of region
Body arragement direction calculation method.Fig. 8 is one of the image of suspicious region in parking lot, as seen from the figure, vehicle in the suspicious region
There is another vehicle beside queue, the significance of the independent vehicle in SR notable figure is higher, therefore obtained when binaryzation
Vehicle platoon is connected with the mark figure of separate vehicle in binary map, leads to the longitudinal direction of its minimum circumscribed rectangle and the row of script
Cloth direction is inconsistent, and reason is that vehicle platoon is closer with separate vehicle, leads to its SR notable figure binarization result phase
Even, if the external quadrangle of minimum for directly taking the connected domain, allow for that last direction value is subject to from queue
The same influence with pinpoint target.And the judgement of people has comprehensively considered the influence of all vehicles in fact, that is to say, that if most
This angle afterwards regards single ballot, then 15 vehicles should respectively have a franchise in queue, and independent vehicle is same
Sample has a franchise, eventually votes, and selects the highest angle of poll.Based on this principle, present embodiment is devised
A kind of angle computation method.
Step 4 two obtains queue parked vehicle using the edge of the doubtful vehicle region of Sobel operator extraction
The edge graph of doubtful vehicle region, the result of the edge graph of doubtful vehicle region is as shown in figure 9, the method for recycling statistics, meter
The angle value for calculating any two points line and the angle of horizontal direction in the edge graph of doubtful vehicle region records all angle values
Corresponding point is denoted as AngleNum to quantityu′, u '=0 ..., 179, (any two points connect in the edge graph of doubtful vehicle region
Line and the angle of horizontal direction is defined as: using the extended line of any two points line and the intersection point of horizontal direction as the center of circle, the center of circle
Right side is that horizontal direction is positive, is horizontal direction negative sense on the left of the center of circle, definition is by horizontal direction forward direction, is connected to any two points
The angle passed through until line is " angle of any two points line and horizontal direction in the edge graph of doubtful vehicle region ", angle
Value range is 0-179 degree) wherein: AngleNumu′The corresponding point of angle value u ' is represented to quantity；Such as AngleNum5It indicates
All line angles are the total number of the point pair of 5 (± 0.5) degree in the edge graph of doubtful vehicle region；
Figure 10 is the AngleNum statistical results chart of Fig. 9, and the corresponding angle of the maximum value of AngleNum is 6 degree here,
By maximum AngleNumu′Angle of the corresponding angle value u ' as the queue arragement direction of doubtful vehicle region,
Doubtful vehicle region is obtained to the doubtful vehicle region of horizontal direction arrangement according to the angle rotation acquired；
Such as: the corresponding angle of some doubtful vehicle region is 8 °, then by the doubtful vehicle region to the edge side of image
8 ° are rotated to (i.e. horizontal direction), makes the sides aligned parallel of doubtful vehicle region and image, i.e., the arrangement of acquisition horizontal direction is doubtful
Vehicle region.
Step 4 three, for non-queue parked vehicle, (non-queue parked vehicle refers to the area of doubtful vehicle region
Region no more than 25 square metres), i.e., an only vehicle and the case where almost take whole region in doubtful vehicle region, due to
Its area is smaller, and the longitudinal direction of minimum circumscribed rectangle generally can accurately indicate the direction of vehicle, then directly takes doubtful vehicle
The angle of the broadside in region and the angle value of horizontal direction angle as queue arragement direction, by doubtful vehicle region according to asking
The angle value obtained rotates to obtain the doubtful vehicle region of horizontal direction arrangement；
Step 4 four, the process for repeating step 4 two and step 4 three, calculate all doubtful vehicles in entire parking lot
All doubtful vehicle regions are obtained horizontal direction row according to the corresponding angle rotation acquired by the angle of the arragement direction in region
All doubtful vehicle regions of cloth complete the queue amendment of doubtful vehicle region.
Specific embodiment 6: the present embodiment is different from the first embodiment in that: the specific mistake of the step 5
Journey are as follows:
Since vehicle dimension is small under sub-meter grade resolution ratio, it is difficult to utilize feature Primary Location vehicle, allow also for
False dismissal as few as possible；
The window size that sliding window is arranged is 20 × 12 pixels, and step-length is 2 pixels, by postrotational all doubtful vehicle regions
It is cut into slice, and HOG feature extraction is carried out to the slice being cut into, carries out two using HOG feature of the SVM classifier to extraction
Classification, the biopsy marker that would be classified as vehicle return original image, complete vehicle detection.
The window size of sliding window will be big than all conventional vehicles, this is allowed for when in window including vehicle, sample
Should there are complete vehicle and its edge in this slice, vehicle details itself can be obtained simultaneously when extracting HOG feature in this way
The HOG feature of HOG feature and vehicle's contour can more fully utilize the characteristic information of vehicle in this way, be conducive to improve classification
Accuracy.
The SVM classifier of two classification employed in the present invention is trained SVM classifier, wherein for training
The positive sample of SVM classifier is the extracted HOG feature of slice of 400 real vehicles manually chosen, and negative sample is 500
The extracted HOG feature of the slice for the non-vehicle manually chosen.
Specific embodiment 7: present embodiment is unlike specific embodiment one or six: the HOG feature extraction
When parameter setting are as follows: the size of cell is 4 pixels, and the angle block gauge number of division is 5, and step-length is 2 pixels.It is final to extract
Obtained HOG characteristic dimension is 420 dimensions.
Specific embodiment 8: present embodiment is unlike specific embodiment three: the cutting of the short transverse
The value range of step-length W and the cutting step-length H of width direction are [10,100].
Specific embodiment 9: present embodiment is unlike specific embodiment three: the value of the sieveing coeffecient β
Range is [0.3,0.8].
Specific embodiment 10: present embodiment is unlike specific embodiment three: the binarization threshold T2's
Value range is [0.1,0.9].
Beneficial effects of the present invention are verified using following embodiment:
Experimental image is the Texas, USA large parking lot region obtained from Google Earth
Large scene image, image size are 13000*13000 pixel, and it includes 5145 vehicle targets that resolution ratio, which is 0.5 meter,.Test
The result shows that the parking lot extracting method based on BBSM notable figure, 93.8% parking can be extracted according to pixel calculating
Field pixel, can relatively accurately extract car park areas；Based on the vehicle detecting algorithm of queue amendment and SR notable figure, press
It is calculated according to vehicle target, accuracy 85.12%, recall rate 89.76%, verification and measurement ratio with higher.As Figure 11, Figure 12,
It is partial detection figure shown in Figure 13 and Figure 14.
Above-mentioned example of the invention only explains computation model and calculation process of the invention in detail, and is not to this
The restriction of the embodiment of invention.It for those of ordinary skill in the art, on the basis of the above description can be with
It makes other variations or changes in different ways, all embodiments can not be exhaustive here, it is all to belong to the present invention
The obvious changes or variations extended out of technical solution still in the scope of protection of the present invention.
Claims (10)
1. view-based access control model conspicuousness and the modified parking lot of queue and vehicle target detection method, which is characterized in that this method packet
Include following steps:
Step 1: input sub-meter grade high-resolution optical remote sensing image, brightness and vision significance based on parking lot, meter
The BBSM notable figure of the sub-meter grade high-resolution optical remote sensing image of input is calculated, and by calculated BBSM notable figure two-value
Change；
Step 2: carrying out the segmentation of super-pixel to the BBSM notable figure after binaryzation according to the region feature in parking lot, divided
Each of whole super-pixel block afterwards, setting screening conditions screen whole super-pixel block after segmentation, and calculating sifting goes out
The mass center figure CDM of super-pixel block(i′,j′), utilize the mass center figure CDM of calculated each super-pixel block(i′,j′)After updating binaryzation
BBSM notable figure, the mass center figure CDM of the BBSM notable figure after obtaining binaryzation；
The mass center Density Distribution index CDDI figure of BBSM notable figure after calculating binaryzation according to obtained CDM, and according to CDDI
Figure obtains ROI image, i.e. acquisition coarse extraction car park areas image；
Step 3: obtaining what essence was extracted according to the car park areas image of coarse extraction and the color characteristic and region feature in parking lot
Car park areas image completes parking field detecting；
Step 4: calculating the SR notable figure for the car park areas image that essence is extracted, and extract the doubtful vehicle area in SR notable figure
Domain；
The angle for calculating the queue arragement direction of each doubtful vehicle region, according to the angle acquired by all doubtful vehicle areas
Domain rotates the queue amendment that doubtful vehicle region is completed to the direction of horizontal arrangement；
Step 5: postrotational all doubtful vehicle regions are cut into slice using sliding window cutting method, and to being cut into
Slice carries out HOG feature extraction, and using SVM classifier to the HOG tagsort of extraction, the biopsy marker that would be classified as vehicle is returned
Original image completes vehicle detection.
2. view-based access control model conspicuousness according to claim 1 and the modified parking lot of queue and vehicle target detection method,
It is characterized in that, the detailed process of the step 1 are as follows:
Step 1 one, input sub-meter grade high-resolution optical remote sensing image, by the sub-meter grade high-resolution optical remote sensing figure of input
As after being transformed into luminance channel, obtaining luminance picture I, the size of luminance picture I is that M × N, M and N respectively represent luminance picture I
Height and width；
Count all pixels point corresponding brightness value u, u=0 ..., 255 in luminance picture I；
Statistical result PuIt indicates the pixel number that brightness value is u in luminance picture I, then calculates each pixel in luminance picture I
The BBSM value of point are as follows:
Wherein: (i, j) represents any one pixel in luminance picture I, I(i,j)Represent any one picture in luminance picture I
The brightness value of vegetarian refreshments (i, j), BBSM(i,j)Represent the value of the BBSM of pixel (i, j)；
Then the BBSM value of each pixel forms BBSM notable figure；
Step 1 two chooses binarization threshold T1, the BBSM notable figure after binaryzation obtains binaryzation is done to BBSM notable figure
BBSM':
T1=0.3 × max (BBSM)+0.7 × min (BBSM)
Wherein: max (BBSM) is the maximum value of BBSM notable figure, and min (BBSM) is the minimum value of BBSM notable figure；
Then the BBSM value after the binaryzation of pixel (i, j) is
3. view-based access control model conspicuousness according to claim 1 and the modified parking lot of queue and vehicle target detection method,
It is characterized in that, the detailed process of the step 2 are as follows:
In step 2 one, BBSM notable figure after binarization, the cutting step-length of short transverse is taken to be W, take cutting for width direction
Cutting step-length is H, is cut into the BBSM notable figure after binaryzation using W and HA super-pixel block SP；
Define the weight W of each super-pixel blockSPAre as follows:
Wherein: SP(i′,j′)For the BBSM value after the binaryzation of pixel in super-pixel block (i ', j ')；
Step 2 two, setting screening conditions are WSP> β WH, in which: β is sieveing coeffecient, utilizes the screening conditions pair of setting
The super-pixel block that the segmentation of step 2 one obtains is screened, and the super-pixel block that weight is greater than screening conditions is filtered out, and rejects weight
No more than the super-pixel block of screening conditions；
The mass center figure CDM for each super-pixel block that step 2 three, calculating sifting go out(i′,j′):
Wherein:WithThe mass center for indicating super-pixel block, utilizes the mass center figure CDM of calculated each super-pixel block(i′,j′)More
BBSM notable figure after new binaryzation, the mass center figure CDM of the BBSM notable figure after obtaining binaryzation；
Step 2 four, with the unit matrix Te having a size of k × kk×kWith the mass center figure CDM convolution of the BBSM notable figure after binaryzation
The mass center Density Distribution index CDDI of BBSM notable figure after obtaining binaryzation schemes；
Wherein:
Step 2 five chooses binarization threshold T2, by the mass center Density Distribution index CDDI figure two of the BBSM notable figure after binaryzation
ROI image is obtained after value, obtains the car park areas image of coarse extraction.
4. view-based access control model conspicuousness according to claim 1 and the modified parking lot of queue and vehicle target detection method,
It is characterized in that, the detailed process of the step 3 are as follows:
Step 3 one, in the car park areas image of coarse extraction, by set tri- wave bands of R, G, B value come screen meet sieve
Select the pixel of condition, in which: screening conditions are as follows: the value of tri- wave bands of R, G, B is and any in the range of (130,250)
The difference of two wave bands is less than or equal to 40；
Step 3 two carries out holes filling to the selection result of step 3 one, rejects width less than 10 meters using etching operation
After road, then the area of each connected domain is calculated, to remove connected domain of the area less than 800 square metres；After expansion reduction
The car park areas essence for obtaining sub-meter grade high-resolution optical remote sensing image is extracted as a result, obtaining the car park areas figure that essence is extracted
Picture completes parking field detecting.
5. view-based access control model conspicuousness according to claim 1 and the modified parking lot of queue and vehicle target detection method,
It is characterized in that, the detailed process of the step 4 are as follows:
The car park areas image that essence is extracted is carried out Fourier transformation to calculate amplitude spectrum A (f) and phase spectrum P by step 4 one
(f):
Wherein: I ' (x) is the luminance picture in the car park areas that essence is extracted；Indicate that I ' (x) takes Fourier transformation
Amplitude,Indicate that I ' (x) takes the phase value of Fourier transformation,Represent Fourier transformation；
Amplitude spectrum A (f) is become into logarithmic spectrum L (f), then linear space filtering is carried out to logarithmic spectrum, by logarithmic spectrum L (f) and linearly
Space filtering result makes the difference to obtain residual spectra R (f)；
L (f)=log (A (f))
R (f)=L (f)-hn(f)*L(f)
Wherein: hnIt (f) is mean filter operator；
Residual spectra R (f) and phase spectrum P (f) is carried out inverse Fourier transform and obtains inverse Fourier transform as a result, recycling high
This filter carries out linear space to inverse Fourier transform result and filters to obtain SR notable figure:
Wherein: g (x) is gaussian filtering operator；SR (x) represents SR notable figure,Represent inverse Fourier transform；
Obtained SR notable figure is normalized, then binaryzation is carried out to the SR notable figure after normalization, sets two-value
Change threshold value T3=0.12max (SR), max (SR) are the maximum value of SR notable figure, the SR notable figure after obtaining binaryzation, then two
The region that SR notable figure intermediate value after value is 1 is doubtful vehicle region；
Step 4 two obtains doubtful queue parked vehicle using the edge of the doubtful vehicle region of Sobel operator extraction
The edge graph of vehicle region calculates the angle of any two points line and the angle of horizontal direction in the edge graph of doubtful vehicle region
Value, records the corresponding point of all angle values to quantity, is denoted as AngleNumu′, u '=0 ..., 179, in which: AngleNumu′
The corresponding point of angle value u ' is represented to quantity；
By maximum AngleNumu′Angle of the corresponding angle value u ' as the queue arragement direction of doubtful vehicle region, will doubt
The doubtful vehicle region of horizontal direction arrangement is obtained according to the angle rotation acquired like vehicle region；
Step 4 three, for non-queue parked vehicle, then directly take the broadside and horizontal direction angle of doubtful vehicle region
Doubtful vehicle region is obtained horizontal direction according to the angle value rotation acquired by angle of the angle value as queue arragement direction
The doubtful vehicle region of arrangement；
Step 4 four, the process for repeating step 4 two and step 4 three, calculate all doubtful vehicle regions in entire parking lot
Arragement direction angle, by all doubtful vehicle regions according to acquire corresponding angle rotation come obtain horizontal direction arrangement
All doubtful vehicle regions complete the queue amendment of doubtful vehicle region.
6. view-based access control model conspicuousness according to claim 1 and the modified parking lot of queue and vehicle target detection method,
It is characterized in that, the detailed process of the step 5 are as follows:
The window size that sliding window is arranged is 20 × 12 pixels, and step-length is 2 pixels, and postrotational all doubtful vehicle regions are cut
HOG feature extraction is carried out at slice, and to the slice being cut into, carries out two points using HOG feature of the SVM classifier to extraction
Class, the biopsy marker that would be classified as vehicle return original image, complete vehicle detection.
7. view-based access control model conspicuousness according to claim 1 or 6 and the modified parking lot of queue and vehicle target detection side
Method, which is characterized in that the parameter setting when HOG feature extraction are as follows: the size of cell is 4 pixels, the angle block gauge of division
Number is 5, and step-length is 2 pixels.
8. view-based access control model conspicuousness according to claim 3 and the modified parking lot of queue and vehicle target detection method,
It is characterized in that, the value range of the cutting step-length H of the cutting step-length W and width direction of the short transverse be [10,
100]。
9. view-based access control model conspicuousness according to claim 3 and the modified parking lot of queue and vehicle target detection method,
It is characterized in that, the value range of the sieveing coeffecient β is [0.3,0.8].
10. view-based access control model conspicuousness according to claim 3 and the modified parking lot of queue and vehicle target detection method,
It is characterized in that, the binarization threshold T2Value range be [0.1,0.9].
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811517627.9A CN109635733B (en) | 2018-12-12 | 2018-12-12 | Parking lot and vehicle target detection method based on visual saliency and queue correction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811517627.9A CN109635733B (en) | 2018-12-12 | 2018-12-12 | Parking lot and vehicle target detection method based on visual saliency and queue correction |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109635733A true CN109635733A (en) | 2019-04-16 |
| CN109635733B CN109635733B (en) | 2020-10-27 |
Family
ID=66073158
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811517627.9A Active CN109635733B (en) | 2018-12-12 | 2018-12-12 | Parking lot and vehicle target detection method based on visual saliency and queue correction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109635733B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110488280A (en) * | 2019-08-29 | 2019-11-22 | 广州小鹏汽车科技有限公司 | A kind of modification method and device, vehicle, storage medium of parking stall profile |
| CN111444806A (en) * | 2020-03-19 | 2020-07-24 | 成都云盯科技有限公司 | Commodity touch information clustering method, device and equipment based on monitoring video |
| CN111753692A (en) * | 2020-06-15 | 2020-10-09 | 珠海格力电器股份有限公司 | Target object extraction method, product detection method, device, computer and medium |
| CN113033408A (en) * | 2021-03-26 | 2021-06-25 | 北京百度网讯科技有限公司 | Data queue dynamic updating method and device, electronic equipment and storage medium |
| CN115438702A (en) * | 2022-10-18 | 2022-12-06 | 国网山东省电力公司营销服务中心(计量中心) | Power line carrier channel noise detection method and system |
| CN118135506A (en) * | 2024-04-30 | 2024-06-04 | 东莞市杰瑞智能科技有限公司 | Electronic guideboard based on visual unit structure and road condition target self-identification method thereof |
| CN118196730A (en) * | 2024-05-13 | 2024-06-14 | 深圳金语科技有限公司 | Method, device, equipment and storage medium for processing vehicle image data |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130028468A1 (en) * | 2011-07-29 | 2013-01-31 | International Business Machines Corporation | Example-Based Object Retrieval for Video Surveillance |
| CN103310195A (en) * | 2013-06-09 | 2013-09-18 | 西北工业大学 | LLC-feature-based weak-supervision recognition method for vehicle high-resolution remote sensing images |
| CN107133558A (en) * | 2017-03-13 | 2017-09-05 | 北京航空航天大学 | A kind of infrared pedestrian's conspicuousness detection method based on probability propagation |
-
2018
- 2018-12-12 CN CN201811517627.9A patent/CN109635733B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130028468A1 (en) * | 2011-07-29 | 2013-01-31 | International Business Machines Corporation | Example-Based Object Retrieval for Video Surveillance |
| CN103310195A (en) * | 2013-06-09 | 2013-09-18 | 西北工业大学 | LLC-feature-based weak-supervision recognition method for vehicle high-resolution remote sensing images |
| CN107133558A (en) * | 2017-03-13 | 2017-09-05 | 北京航空航天大学 | A kind of infrared pedestrian's conspicuousness detection method based on probability propagation |
Non-Patent Citations (2)
| Title |
|---|
| ZONGJIE CAO 等: ""Fast target detection method for high-resolution"", 《EURASIP JOURNAL ON ADVANCES IN SIGNAL PROCESSING》 * |
| 韩现伟: ""大幅面可见光遥感图像典型目标识别关键技术研究"", 《中国博士论文全数据库》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110488280A (en) * | 2019-08-29 | 2019-11-22 | 广州小鹏汽车科技有限公司 | A kind of modification method and device, vehicle, storage medium of parking stall profile |
| CN111444806A (en) * | 2020-03-19 | 2020-07-24 | 成都云盯科技有限公司 | Commodity touch information clustering method, device and equipment based on monitoring video |
| CN111753692A (en) * | 2020-06-15 | 2020-10-09 | 珠海格力电器股份有限公司 | Target object extraction method, product detection method, device, computer and medium |
| CN111753692B (en) * | 2020-06-15 | 2024-05-28 | 珠海格力电器股份有限公司 | Target object extraction method, product detection method, device, computer and medium |
| CN113033408A (en) * | 2021-03-26 | 2021-06-25 | 北京百度网讯科技有限公司 | Data queue dynamic updating method and device, electronic equipment and storage medium |
| CN113033408B (en) * | 2021-03-26 | 2023-10-20 | 北京百度网讯科技有限公司 | Data queue dynamic updating method and device, electronic equipment and storage medium |
| CN115438702A (en) * | 2022-10-18 | 2022-12-06 | 国网山东省电力公司营销服务中心(计量中心) | Power line carrier channel noise detection method and system |
| CN118135506A (en) * | 2024-04-30 | 2024-06-04 | 东莞市杰瑞智能科技有限公司 | Electronic guideboard based on visual unit structure and road condition target self-identification method thereof |
| CN118135506B (en) * | 2024-04-30 | 2024-07-16 | 东莞市杰瑞智能科技有限公司 | Electronic guideboard based on visual unit structure and road condition target self-identification method thereof |
| CN118196730A (en) * | 2024-05-13 | 2024-06-14 | 深圳金语科技有限公司 | Method, device, equipment and storage medium for processing vehicle image data |
| CN118196730B (en) * | 2024-05-13 | 2024-08-06 | 深圳金语科技有限公司 | Method, device, equipment and storage medium for processing vehicle image data |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109635733B (en) | 2020-10-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109635733A (en) | View-based access control model conspicuousness and the modified parking lot of queue and vehicle target detection method | |
| CN111080629B (en) | Method for detecting image splicing tampering | |
| CN108918532B (en) | System and method for detecting damage of expressway traffic sign | |
| CN103034863B (en) | The remote sensing image road acquisition methods of a kind of syncaryon Fisher and multiple dimensioned extraction | |
| CN103699905B (en) | Method and device for positioning license plate | |
| CN106651872A (en) | Prewitt operator-based pavement crack recognition method and system | |
| Choi et al. | Vehicle detection from aerial images using local shape information | |
| CN102708356A (en) | Automatic license plate positioning and recognition method based on complex background | |
| CN110390255A (en) | High-speed rail environmental change monitoring method based on various dimensions feature extraction | |
| CN106971397B (en) | Based on the city high-resolution remote sensing image dividing method for improving JSEG algorithms | |
| CN108491498A (en) | A kind of bayonet image object searching method based on multiple features detection | |
| EP2447884B1 (en) | Method for detecting and recognising an object in an image, and an apparatus and a computer program therefor | |
| CN110070545B (en) | Method for automatically extracting urban built-up area by urban texture feature density | |
| CN108537169A (en) | A kind of high-resolution remote sensing image method for extracting roads based on center line and detection algorithm of having a lot of social connections | |
| Yamazaki et al. | Vehicle extraction and speed detection from digital aerial images | |
| CN113537211A (en) | Deep learning license plate frame positioning method based on asymmetric IOU | |
| CN114299247A (en) | Rapid detection and problem troubleshooting method for road traffic sign lines | |
| CN111611858B (en) | Multi-angle discrimination-based automatic detection method and device for tilting track surface | |
| CN108447045A (en) | A kind of SAR remote sensing images waters detection method based on SAT integrograms | |
| CN112036246B (en) | Construction method of remote sensing image classification model, remote sensing image classification method and system | |
| CN114677428A (en) | Power transmission line icing thickness detection method based on unmanned aerial vehicle image processing | |
| CN114862883A (en) | Target edge extraction method, image segmentation method and system | |
| CN114998876A (en) | Sea-land transition phase shale streak layer structure identification method based on rock slice image | |
| Huang et al. | Classification of very high spatial resolution imagery based on the fusion of edge and multispectral information | |
| CN107239754B (en) | Automobile logo identification method based on sparse sampling intensity profile and gradient distribution |
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 |