CN107330432A - A kind of various visual angles vehicle checking method based on weighting Hough ballot - Google Patents
A kind of various visual angles vehicle checking method based on weighting Hough ballot Download PDFInfo
- Publication number
- CN107330432A CN107330432A CN201710554766.8A CN201710554766A CN107330432A CN 107330432 A CN107330432 A CN 107330432A CN 201710554766 A CN201710554766 A CN 201710554766A CN 107330432 A CN107330432 A CN 107330432A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- ballot
- image
- munder
- 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
- G06V20/60—Type of objects
- G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/254—Fusion techniques of classification results, e.g. of results related to same input data
-
- 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
- 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
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
- G06V10/507—Summing image-intensity values; Histogram projection analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/259—Fusion by voting
-
- 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)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a kind of various visual angles vehicle checking method based on weighting Hough ballot, comprise the following steps：Step A：Define training sample image collection；Step B：Visual angle subclass division is carried out to the positive sample set that training sample image is concentrated；Step C：Calculate contribution weight of each positive sample to different visual angles subclass；Step D：Ballot score value of the image block in position candidate is determined using Hough voting method is weighted；Step E：Vehicle detection frame is determined in test image；The present invention realizes the automatic division of vehicle different visual angles subclass using LLE and k means, using the Definition of Division, positive sample is integrated into the ballot weight under different visual angles in Hough voting process, with reference to ballot weight, pinpoint Hough ballot is carried out, so as to realize the accurate detection of the lower vehicle of various visual angles；Compared to prior art, the present invention substantially increases detection speed, and effectively utilizes the shared information between different visual angles subclass, so as to further increase the accuracy of vehicle detection.
Description
Technical field
The present invention relates to field of vehicle detection under video traffic environment, more particularly to it is a kind of based on many of weighting Hough ballot
Visual angle vehicle checking method.
Background technology
As automobile is increasingly becoming most important instrument in daily life, vehicle detection also becomes in smart city and intelligently handed over
The important component of way system, but in reality scene, the vehicle detection of various visual angles is still the difficulties of vehicle detection,
When this is due to vehicle movement or when camera site is different, vehicle will be caused to be presented in the picture with different visual angles, outside vehicle
See feature and occur in that very big difference, so as to result in the drastically decline of vehicle detection accuracy rate.
The vehicle detection for various visual angles is broadly divided into three major types, first in the prior art：Using manual method or it is based on
Training set of images is divided into different subclasses by sample length-width ratio (Aspect Ratio), and each subclass includes the visual angle of a certain scope
Change, and set up detection model for each subclass independence；Second：The method divided based on automatic subclass, or in study detection
Embedded Unsupervised clustering process during device；3rd：3D Viewing-angle informations are embedded in a model, and estimate aspect；Above-mentioned three
Class method solves the problems, such as various visual angles vehicle detection from different perspectives, but each there is obvious defect or limitation, such as neglects
Slightly various visual angles clarification of objective general character or 3D Viewing-angle informations are difficult collection etc., so as to cause various visual angles vehicle detection testing result not
Accurately.
The content of the invention
, can be effective it is an object of the invention to provide a kind of various visual angles vehicle checking method based on weighting Hough ballot
Solution due to prior art ignores various visual angles clarification of objective general character or ignores 3D Viewing-angle informations and cause various visual angles vehicle examine
Survey the problem of testing result is inaccurate.
To achieve these goals, the present invention uses following technical scheme：
A kind of various visual angles vehicle checking method based on weighting Hough ballot, comprises the following steps：
Step A：Define training sample image collectionImage size is 128 × 64；Wherein, fiFor image
Feature representation, is carrying out using HOG features when various visual angles are divided, multichannel pixel is used when training vision word and Hough ballot
Feature；yi∈ { -1 ,+1 } is training sample label, yiI when=- 1iFor background sample, yiI when=+ 1iFor target sample；N is instruction
Practice sample set size；
Step B：To training sample image collectionIn positive sample setCarry out
Visual angle subclass is divided, wherein, N+Represent positive sample number；
Step C：Calculate contribution weight of each positive sample to different visual angles subclass；
Step D：Ballot score value of the image block in position candidate is determined using Hough voting method is weighted；
Step E：Vehicle detection frame is determined in test image.
Described step B comprises the following steps：
Step B1：Using LLE algorithms by positive sample set D+The positive sample image of middle HOG feature representations is embedded into two-dimentional sky
Between；
Step B2：The central point for the ring-type for selecting to be formed by the probability distribution of samples points in two-dimensional space, based on sample point in this
The relative angle of heart point, by all sample rules a to circle；
Step B3：The upper sample of circle is clustered using k-means algorithms, positive sample set D+It is divided into K visual angle
Class.
Described step C specifically uses following methods：
In LLE embedded spaces, the cluster centre for defining k-th of visual angle subclass sample set is ok, k ∈ { 1,2 ..., K },
Positive sample set D+Middle positive sample image ΓjF is expressed as in the LLE embedded spacesj`, then positive sample ΓjTo visual angle subclass k's
Contribute weight wjkIt is defined as:
Wherein, d (fj',ok) it is f in LLE embedded spacesj' and okBetween distance；In order to ensure calculate correctness, it is necessary to
Align sample ΓjContribution weight w under each visual angle subclassjkIt is normalized, to ensure weight and value as 1, i.e.,
Described step D comprises the following steps：
Step D1：Definition and image block ptThe vision word of matching is L, comprising positive sample image block offset vector in L
Collection is combined into EL, class probability C is drawn comprising positive sample image block proportion by counting in LL, then image block ptIn candidate bit
The ballot score value for putting h is：
Wherein, ELIn each ballot unit e to position candidate h ballot using the estimation of Gauss Parzen windows, | EL| represent collection
Close size, qtFor image block ptCenter；After vision word L generations, its corresponding class probability CLIt has been determined that image
Block ptOffset vector set E is depended primarily in position candidate h ballot score valueLIn ballot unit e, utilize Hough ballot point
Be worth linear superposition characteristic, can by V (h | pt) definition by add up ELIn each ballot unit position candidate h is voted the form of score value
It is rewritten as cumulative and ELThe associated positive sample image Γ of middle ballot unitjTo position candidate h vote score value form, i.e.,：
Wherein,
Wherein,Represent ELIn ballot unit e come from positive sample set D+Middle positive sample image Γj；Then traversal is surveyed
Attempt as image block p in GtIt is in position candidate h final vote score value：
Step D2：Due to positive sample set D+In each sample image aspects difference it is excessive, can cause final vote generate
Hough figure is chaotic, and bright spot is not enough concentrated, it is impossible to accurate to determine position candidate h, therefore this programme exists
Introduce visual angle variable k ∈ { 1,2 ..., K } in the Voting Model of definition to be limited under same view angle ballot, to ensure to candidate
The visual angle uniformity of position h ballots, i.e.,：
The formula calculation process is as follows：It is positive sample set D first with the various visual angles subclass division methods described in step B+
In one visual angle subclass k ∈ of every sample image mark { 1,2 ..., K }；Then above-mentioned formula is utilizedCalculate every
Open the visual angle contribution weight w of sample imagejk,j∈{1,2,...,N+}；Contribute the positive sample of weight in final visual angle and the visual angle demarcated
This set D+, thenIt is redefined：
Wherein, W is wjkWhat is constituted is written as N greatly+× K weight matrix.
Described step E comprises the following steps：
Step E1：Test image is subjected to Scale Decomposition first, defining test image metric space isM is
The number of discrete yardstick；
Step E2：It is λ in yardstickmTest image in intensive sampling with the image block of size, find and each image block
The vision word of matching, then utilizes above-mentioned formulaHough under the yardstick is obtained to throw
Ticket figure；
Step E3：In (h, λm) constitute three-dimensional hough space (h=(hx,hy), include horizontal, ordinate position in image)
In, using mean-shift algorithms and decision threshold, determine final goal center (h ', λ 'm), and in former test chartPosition mark size isDetection block.
Beneficial effects of the present invention：
Compared with prior art, a kind of various visual angles vehicle checking method based on weighting Hough ballot of the present invention,
Using (LLE) is locally linear embedding into and k-means realizes the automatic division of vehicle different visual angles subclass, the Definition of Division is utilized
Positive sample is integrated into the ballot weight under different visual angles in Hough voting process, and with reference to ballot weight, progress is pinpoint suddenly
Husband votes, so as to realize the accurate detection of the lower vehicle of various visual angles；Compared to prior art, the present invention substantially increases detection speed
Degree, and the shared information between different visual angles subclass is effectively utilized, so as to further increase the accuracy of vehicle detection.
Brief description of the drawings
, below will be to specific in order to illustrate more clearly of the specific embodiment of the invention or technical scheme of the prior art
The accompanying drawing used required in embodiment or description of the prior art is briefly described, it should be apparent that, in describing below
Accompanying drawing is some embodiments of the present invention, for those of ordinary skill in the art, before creative work is not paid
Put, other accompanying drawings can also be obtained according to these accompanying drawings.
Fig. 1 is schematic flow sheet of the invention；
Fig. 2 is the partial detection schematic diagram detected using the present invention.
Embodiment
Technical scheme is clearly and completely described below in conjunction with accompanying drawing, it is clear that described implementation
Example is a part of embodiment of the invention, rather than whole embodiments.Based on the embodiment in the present invention, ordinary skill
The every other embodiment that personnel are obtained under the premise of creative work is not made, belongs to the scope of protection of the invention.
A kind of various visual angles vehicle checking method based on weighting Hough ballot of the present invention, comprises the following steps：
Step A：Define training sample image collectionImage size is 128 × 64；Wherein, fiFor image
Feature representation (is carrying out using HOG features when various visual angles are divided, multichannel pixel is used when training vision word and Hough ballot
Feature)；yi∈ { -1 ,+1 } is training sample label (yiI when=- 1iFor background sample, yiI when=+ 1iFor target sample)；N is
Training sample set size；
During collecting training sample, background image is consistent with target image size, and quantity is approached, and the guarantee that should try one's best
The diversity of training sample, i.e. background image should include the various scenes that be likely to occur of target, and target image should can including target
The various visual angle forms that can be presented.
Step B：To training sample image collectionIn positive sample setCarry out
Various visual angles subclass is divided, wherein, N+Represent positive sample number；Specifically include following steps：
Step B1：Using LLE algorithms by positive sample set D+Positive sample image (the various visual angles vehicle of middle HOG feature representations
Image) it is embedded into two-dimensional space；As a result show, two-dimensional space is embedded in using the various visual angles vehicle image of HOG character representations
Afterwards, the probability distribution of samples points formation ring-type, along the ring, vehicle visual angle smooth change；
Step B2：According to embedded sample (sample set D+The positive sample image of middle HOG feature representations) distribution, in two-dimensional space
The central point of ring-type formed by the probability distribution of samples points is selected, based on the relative angle of sample point to the central point, by all samples
Regularization is in a circle, and the sample of proximate region has close visual angle on circle；
Step B3：As shown in Figure 1：The upper sample of circle is clustered using k-means algorithms, by the close sample quilt in visual angle
It is divided on same section of round arc, every section of arc represents a subclass at close visual angle on circle, K visual angle is divided into altogether
Class；
Step C：Contribution weight of each sample to different visual angles subclass is calculated, specifically using following methods：
As positive sample set D in training sample image collection D+K visual angle subclass divide and determine after, calculate each positive sample
To the contribution weight of each visual angle subclass, to make full use of the sharing and otherness of information between each visual angle subclass；Computational methods
It is as follows：
In LLE embedded spaces, the cluster centre for defining k-th of the visual angle subclass sample set divided is ok, k ∈ 1,
2 ..., K }, positive sample set D+Middle positive sample image ΓjF is expressed as in the LLE embedded spacesj`, then positive sample ΓjTo regarding
Silver coin class k contribution weight wjkIt is defined as:
Wherein, d (fj',ok) it is f in LLE embedded spacesj' and okBetween distance；In order to ensure calculate correctness, it is necessary to
To positive sample ΓjContribution weight w under each visual angle subclassjkIt is normalized, to ensure weight and value as 1, i.e.,
Step D：Ballot score value of the image block in position candidate is determined using Hough voting method is weighted, is specifically included following
Step：
Step D1：Definition and test topography block ptThe vision word of matching is L, inclined comprising positive sample image block in L
The collection for the amount of shifting to is combined into EL, class probability C is drawn comprising positive sample image block proportion by counting in LL, then image block pt
It is in position candidate h ballot score value
Wherein, ELIn each offset vector e to position candidate h ballot using the estimation of Gauss Parzen windows, | EL| represent collection
Close size, qtFor image block ptCenter；After vision word L generations, its corresponding class probability CLIt has been determined that local
Image block ptOffset vector set E is depended primarily in position candidate h ballot score valueLIn ballot unit e, thrown using Hough
Ticket score value linear superposition characteristic, can by V (h | pt) definition by add up ELIn each ballot unit position candidate h is voted score value
Form is rewritten as cumulative ELThe associated positive sample image Γ of middle ballot unitjTo position candidate h vote score value form, i.e.,：
Wherein,
Wherein,Represent ELIn ballot unit e come from positive sample image Γ in positive sample set D+j；Traversal is surveyed
Attempt as all topography's block p in Gt, position candidate h final vote score value is：
Step D2：Because each sample image aspects difference is excessive in positive sample set D+, final vote can be caused to generate
Hough figure is chaotic, and bright spot is not enough concentrated, it is impossible to accurate to determine position candidate h, therefore this programme exists
Introduce visual angle variable k ∈ { 1,2 ..., K } in the Voting Model of definition to be limited under same view angle ballot, to ensure to candidate
The visual angle uniformity of position h ballots, i.e.,：
The formula calculation process is as follows：It is positive sample set D first with visual angle subclass division methods+In every image tagged
One visual angle subclass k ∈ 1,2 ..., K }；Then above-mentioned formula is utilized
Calculate the visual angle contribution weight w of every imagejk,j∈{1,2,...,N+}；Eventually through having demarcated visual angle and visual angle
Contribute the sample set positive sample set D of weight+, thenIt is redefined：
Wherein, W is wjkThe big weight matrix for being written as N+ × K constituted.
Step E：Final target detection frame is determined in vehicle detection image under various visual angles, step is described in detail below：
Step E1：Test image is subjected to Scale Decomposition first, defining test image metric space isM is
The number of discrete yardstick；
Step E2：It is λ in yardstickmTest image in intensive sampling with the image block of size, find and each image block
The vision word of matching, then utilizes above-mentioned formulaHough under the yardstick is obtained to throw
Ticket figure；
Step E3：In (h,λm) constitute three-dimensional hough space (h=(hx,hy), include horizontal, ordinate position in image)
In, using mean-shift algorithms and decision threshold, determine final goal center (h ', λ 'm), and in former test chartPosition mark size isDetection block.
Embodiment one：
The present embodiment includes multiple images sample, utilizes a kind of various visual angles based on weighting Hough ballot of the present invention
In vehicle checking method detection image sample during vehicle location, using following steps：
Step A：Define training sample image collectionUsed sample image size is all standardized
For 128 × 64；Wherein, fiFor image feature representation (carry out various visual angles divide when use HOG features, training vision word and
Multichannel pixel characteristic is used when Hough is voted)；yi∈ { -1 ,+1 } is training sample label (yiI when=- 1iFor background sample,
yiI when=+ 1iFor target sample)；N is training sample set size；
During collecting training sample image, background image is consistent with target image size, and quantity is approached, and should be tried one's best
Ensure that the diversity of training sample image, i.e. background image should include the various scenes that target is likely to occur, target image should be wrapped
Include the various forms that target may be presented.
Step B：To training sample image collectionIn positive sample setCarry out
Various visual angles subclass is divided, wherein, N+Represent positive sample number；In the present embodiment, visual angle subclass number is defined as 8；It is specific use with
Lower step：
Step B1：Using LLE algorithms by positive sample set D+Positive sample image (the various visual angles vehicle of middle HOG feature representations
Image) it is embedded into two-dimensional space, the probability distribution of samples points formation ring-type, along the ring, vehicle visual angle smooth change；
Step B2：The central point of the ring-type that the probability distribution of samples points is formed in step B1 is selected, based on sample point to the center
The relative angle of point, by all sample rules to circle O, the sample of proximate region has close visual angle on circle O；
Step B3：Sample on circle O is clustered using k-means algorithms, the close sample in visual angle is divided into round O
Same section of circular arc on, every section of circular arc represents a subclass at close visual angle, is divided into 8 visual angle subclasses；
Step C：Contribution weight of each sample to 8 different visual angles subclasses is calculated, specifically using following methods：
In LLE embedded spaces, the cluster centre for defining k-th of the visual angle subclass sample set divided is ok, k ∈ 1,
2 ..., 8 }, positive sample image Γ in positive sample set D+jF is expressed as in LLE embedded spacesj`, then positive sample ΓjTo visual angle
Subclass k contribution weight wjkDefinition be:
Wherein, d (fj',ok) it is f in LLE embedded spacesj' and okBetween distance；In order to ensure calculate correctness, it is necessary to
To positive sample ΓjContribution weight w under each visual angle subclassjkIt is normalized, to ensure weight and value as 1, i.e.,
Step D：Image block p is determined using Hough voting method is weightedtIn position candidate h ballot score value, specifically include
Following steps：
Step D1：Definition and image block ptThe vision word of matching is L, comprising positive sample image block offset vector in L
Collection is combined into EL, class probability C is drawn comprising positive sample image block proportion by counting in LL, then image block ptIn candidate bit
The ballot score value for putting h is
Wherein, ELIn each offset vector e to position candidate h ballot using the estimation of Gauss Parzen windows, | EL| represent collection
Close size, qtFor image block ptCenter；Topography block ptSkew is depended primarily in position candidate h ballot score value
Vectorial set ELIn ballot unit e, using Hough vote score value linear superposition characteristic, can by V (h | pt) definition by add up EL
In vote the position candidate h form of score value of each ballot unit be rewritten as cumulative and ELThe associated positive sample figure of middle ballot unit
As ΓjTo position candidate h vote score value form, i.e.,：
Wherein,
Wherein,Represent ELIn ballot unit e come from positive sample set D+Middle positive sample image Γj；Traversal test
All topography's block p in image GtIt is to position candidate h final vote score value：
Step D2：Due to positive sample set D+In each sample image aspects difference it is excessive, can cause final vote generate
Hough figure is chaotic, and bright spot is not enough concentrated, it is impossible to accurate to determine position candidate h, therefore this programme exists
Introduce visual angle variable k ∈ { 1,2 ..., K } in the Voting Model of definition to be limited under same view angle ballot, to ensure to candidate
The visual angle uniformity of position h ballots, i.e.,：
The formula calculation process is as follows：It is positive sample set D first with visual angle subclass division methods+In every image tagged
One visual angle subclass k ∈ 1,2 ..., K }；Then above-mentioned formula is utilizedCalculate the visual angle contribution power of every image
Weight wjk,j∈{1,2,...,N+}；The sample set positive sample set D of weight is contributed eventually through visual angle and visual angle has been demarcated+, thenIt is redefined：
Wherein, W is wjkWhat is constituted is written as N greatly+× K weight matrix；
Step E：Final target detection frame is determined in vehicle detection image under various visual angles, following steps are specifically included：
Step E1：Test image is subjected to Scale Decomposition first, defining test image metric space isM is
The number of discrete yardstick；
Step E2：It is λ in yardstickmTest image in intensive sampling with size image block, in the present embodiment, in yardstick
For λmTest image in randomly select 50 sizes on each sample image and be 16 × 16 image block, and be using step B
Each image block demarcation visual angle classification；When generating vision word L using random forest, the number set in forest is 20, forest
Depth capacity is 15, in burl dot splitting end condition, and minimum image block number is 20 in split vertexes, and maximum kind purity is
99.5%, the deviation difference of two squares minimum 30 of positive sample image block offset vector in node；To the metric space of every test chart
It is set on 0.1 to 0.8 20 yardsticks；Then formula described in step D is utilizedObtain
Obtain Hough ballot figure under the yardstick；
Step E3：In (h,λm) constitute three-dimensional hough space (h=(hx,hy), include horizontal, ordinate position in image)
In, using mean-shift algorithms and decision threshold, determine final goal center (h ', λ 'm), and in former test chartPosition mark size isDetection block, detection block position is the position of vehicle in target image
Put；Partial detection is as shown in Figure 2.
Compared with prior art, a kind of various visual angles vehicle checking method based on weighting Hough ballot of the present invention,
Using (LLE) is locally linear embedding into and k-means realizes the automatic division of vehicle different visual angles subclass, the Definition of Division is utilized
Positive sample is integrated into the ballot weight under different visual angles in Hough voting process, and with reference to ballot weight, progress is pinpoint suddenly
Husband votes, so as to realize the accurate detection of the lower vehicle of various visual angles；Compared to prior art, the present invention substantially increases detection speed
Degree, and the shared information between different visual angles subclass is effectively utilized, so as to further increase the accuracy of vehicle detection.
Finally it should be noted that：Various embodiments above is merely illustrative of the technical solution of the present invention, rather than its limitations；To the greatest extent
The present invention is described in detail with reference to foregoing embodiments for pipe, it will be understood by those within the art that：Its according to
The technical scheme described in foregoing embodiments can so be modified, or which part or all technical characteristic are entered
Row equivalent substitution；And these modifications or replacement, the essence of appropriate technical solution is departed from various embodiments of the present invention technology
The scope of scheme.
Claims (5)
1. a kind of various visual angles vehicle checking method based on weighting Hough ballot, it is characterised in that comprise the following steps：
Step A：Define training sample image collectionImage size is 128 × 64；Wherein, fiFor characteristics of image
Expression, is carrying out using HOG features when various visual angles are divided, special using multichannel pixel when training vision word and Hough ballot
Levy；yi∈ { -1 ,+1 } is training sample label, yiI when=- 1iFor background sample, yiI when=+ 1iFor target sample；N is training
Sample set size；
Step B：To training sample image collectionIn positive sample setCarry out visual angle
Subclass is divided, wherein, N+Represent positive sample number；
Step C：Calculate contribution weight of each positive sample to different visual angles subclass；
Step D：Ballot score value of the image block in position candidate is determined using Hough voting method is weighted；
Step E：Vehicle detection frame is determined in test image.
2. a kind of various visual angles vehicle checking method based on weighting Hough ballot according to claim 1, it is characterised in that
Described step B comprises the following steps：
Step B1：Using LLE algorithms by positive sample set D+The positive sample image of middle HOG feature representations is embedded into two-dimensional space；
Step B2：The central point for the ring-type for selecting to be formed by the probability distribution of samples points in two-dimensional space, based on sample point to the central point
Relative angle, by all sample rules a to circle；
Step B3：The upper sample of circle is clustered using k-means algorithms, positive sample set D+It is divided into K visual angle subclass.
3. a kind of various visual angles vehicle checking method based on weighting Hough ballot according to claim 1, it is characterised in that
Described step C specifically uses following methods：
In LLE embedded spaces, the cluster centre for defining k-th of visual angle subclass sample set is ok, k ∈ { 1,2 ..., K }, positive sample
This set D+Middle positive sample image ΓjIt is expressed as in the LLE embedded spacesThen positive sample ΓjContribution to visual angle subclass k
Weight wjkIt is defined as:
<mrow>
<msub>
<mi>w</mi>
<mrow>
<mi>j</mi>
<mi>k</mi>
</mrow>
</msub>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mrow>
<mi>d</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>f</mi>
<mi>j</mi>
<mo>&prime;</mo>
</msubsup>
<mo>,</mo>
<msub>
<mi>o</mi>
<mi>k</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
<mo>;</mo>
</mrow>
Wherein, d (f'j,ok) it is f' in LLE embedded spacesjAnd okBetween distance；In order to ensure the correctness calculated, it is necessary to align
Sample ΓjContribution weight w under each visual angle subclassjkIt is normalized, to ensure weight and value as 1, i.e.,
4. a kind of various visual angles vehicle checking method based on weighting Hough ballot according to claim 1, it is characterised in that
Described step D comprises the following steps：
Step D1：Definition and image block ptThe vision word of matching is L, and the collection comprising positive sample image block offset vector is combined into L
EL, class probability C is drawn comprising positive sample image block proportion by counting in LL, then image block ptIn position candidate h throwing
Ticket score value is：
<mrow>
<mi>V</mi>
<mrow>
<mo>(</mo>
<mi>h</mi>
<mo>|</mo>
<msub>
<mi>p</mi>
<mi>t</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mrow>
<mo>|</mo>
<msub>
<mi>E</mi>
<mi>L</mi>
</msub>
<mo>|</mo>
</mrow>
</mfrac>
<munder>
<mo>&Sigma;</mo>
<mrow>
<mi>e</mi>
<mo>&Element;</mo>
<msub>
<mi>E</mi>
<mi>L</mi>
</msub>
</mrow>
</munder>
<mrow>
<mo>(</mo>
<mfrac>
<mn>1</mn>
<msqrt>
<mrow>
<mn>2</mn>
<mi>&pi;</mi>
<mi>&delta;</mi>
</mrow>
</msqrt>
</mfrac>
<mo>&CenterDot;</mo>
<mi>exp</mi>
<mo>(</mo>
<mrow>
<mo>-</mo>
<mfrac>
<mrow>
<mo>|</mo>
<mo>|</mo>
<mrow>
<mo>(</mo>
<mi>h</mi>
<mo>-</mo>
<msub>
<mi>q</mi>
<mi>t</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mi>e</mi>
<mo>|</mo>
<msup>
<mo>|</mo>
<mn>2</mn>
</msup>
</mrow>
<mrow>
<mn>2</mn>
<msup>
<mi>&delta;</mi>
<mn>2</mn>
</msup>
</mrow>
</mfrac>
</mrow>
<mo>)</mo>
<mo>&CenterDot;</mo>
<msub>
<mi>C</mi>
<mi>L</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>;</mo>
</mrow>
Wherein, ELIn each ballot unit e to position candidate h ballot using the estimation of Gauss Parzen windows, | EL| represent that set is big
It is small, qtFor image block ptCenter；After vision word L generations, its corresponding class probability CLIt has been determined that image block pt
Offset vector set E is depended primarily in position candidate h ballot score valueLIn ballot unit e, utilize Hough ballot score value line
Property cumulative characteristics, can by V (h | pt) definition by add up ELIn vote the position candidate h form of score value of each ballot unit rewrite
For cumulative and ELThe associated positive sample image Γ of middle ballot unitjTo position candidate h vote score value form, i.e.,：
<mrow>
<mi>V</mi>
<mrow>
<mo>(</mo>
<mi>h</mi>
<mo>|</mo>
<msub>
<mi>p</mi>
<mi>t</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munder>
<mo>&Sigma;</mo>
<mrow>
<msub>
<mi>&Gamma;</mi>
<mi>j</mi>
</msub>
<mo>&Element;</mo>
<msup>
<mi>D</mi>
<mo>+</mo>
</msup>
</mrow>
</munder>
<mi>&Phi;</mi>
<mrow>
<mo>(</mo>
<mi>h</mi>
<mo>|</mo>
<msub>
<mi>&Gamma;</mi>
<mi>j</mi>
</msub>
<mo>,</mo>
<msub>
<mi>p</mi>
<mi>t</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>;</mo>
</mrow>
Wherein,
<mrow>
<mi>&Phi;</mi>
<mrow>
<mo>(</mo>
<mi>h</mi>
<mo>|</mo>
<msub>
<mi>&Gamma;</mi>
<mi>j</mi>
</msub>
<mo>,</mo>
<msub>
<mi>p</mi>
<mi>t</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mrow>
<mo>|</mo>
<msub>
<mi>E</mi>
<mi>L</mi>
</msub>
<mo>|</mo>
</mrow>
</mfrac>
<munder>
<mo>&Sigma;</mo>
<mrow>
<mi>e</mi>
<mo>&Element;</mo>
<msub>
<mi>E</mi>
<mi>L</mi>
</msub>
<mo>,</mo>
<mi>e</mi>
<mo>&CirclePlus;</mo>
<msub>
<mi>&Gamma;</mi>
<mi>j</mi>
</msub>
</mrow>
</munder>
<mrow>
<mo>(</mo>
<mfrac>
<mn>1</mn>
<msqrt>
<mrow>
<mn>2</mn>
<mi>&pi;</mi>
<mi>&delta;</mi>
</mrow>
</msqrt>
</mfrac>
<mo>&CenterDot;</mo>
<mi>exp</mi>
<mo>(</mo>
<mrow>
<mo>-</mo>
<mfrac>
<mrow>
<mo>|</mo>
<mo>|</mo>
<mrow>
<mo>(</mo>
<mi>h</mi>
<mo>-</mo>
<msub>
<mi>q</mi>
<mi>t</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mi>e</mi>
<mo>|</mo>
<msup>
<mo>|</mo>
<mn>2</mn>
</msup>
</mrow>
<mrow>
<mn>2</mn>
<msup>
<mi>&delta;</mi>
<mn>2</mn>
</msup>
</mrow>
</mfrac>
</mrow>
<mo>)</mo>
<mo>&CenterDot;</mo>
<msub>
<mi>C</mi>
<mi>L</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>;</mo>
</mrow>
Wherein,Represent ELIn ballot unit e come from positive sample set D+Middle positive sample image Γj；Then travel through test chart
As image block p in GtIt is in position candidate h final vote score value：
<mrow>
<mi>S</mi>
<mrow>
<mo>(</mo>
<mi>h</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munder>
<mo>&Sigma;</mo>
<mrow>
<msub>
<mi>&Gamma;</mi>
<mi>j</mi>
</msub>
<mo>&Element;</mo>
<msup>
<mi>D</mi>
<mo>+</mo>
</msup>
</mrow>
</munder>
<munder>
<mo>&Sigma;</mo>
<mrow>
<msub>
<mi>p</mi>
<mi>t</mi>
</msub>
<mo>&Element;</mo>
<mi>G</mi>
</mrow>
</munder>
<mi>&Phi;</mi>
<mrow>
<mo>(</mo>
<mi>h</mi>
<mo>|</mo>
<mi>&Gamma;</mi>
<mo>,</mo>
<msub>
<mi>p</mi>
<mi>t</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>;</mo>
</mrow>
Step D2：Due to positive sample set D+In each sample image aspects difference it is excessive, can cause final vote generate Hough figure
Confusion, bright spot is not enough concentrated, it is impossible to accurate to determine position candidate h, therefore this programme existsDefinition
Introduce visual angle variable k ∈ { 1,2 ..., K } in Voting Model to be limited under same view angle ballot, to ensure to position candidate h
The visual angle uniformity of ballot, i.e.,：
<mrow>
<mi>S</mi>
<mrow>
<mo>(</mo>
<mi>h</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munder>
<mrow>
<mi>m</mi>
<mi>a</mi>
<mi>x</mi>
</mrow>
<mrow>
<mi>k</mi>
<mo>&Element;</mo>
<mi>K</mi>
</mrow>
</munder>
<munder>
<mo>&Sigma;</mo>
<mrow>
<msub>
<mi>&Gamma;</mi>
<mi>j</mi>
</msub>
<mo>&Element;</mo>
<msup>
<mi>D</mi>
<mo>+</mo>
</msup>
</mrow>
</munder>
<munder>
<mo>&Sigma;</mo>
<mrow>
<msub>
<mi>p</mi>
<mi>t</mi>
</msub>
<mo>&Element;</mo>
<mi>G</mi>
</mrow>
</munder>
<mi>&Phi;</mi>
<mrow>
<mo>(</mo>
<mi>h</mi>
<mo>,</mo>
<mi>k</mi>
<mo>|</mo>
<msub>
<mi>&Gamma;</mi>
<mi>j</mi>
</msub>
<mo>,</mo>
<msub>
<mi>p</mi>
<mi>t</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>;</mo>
</mrow>
The formula calculation process is as follows：It is positive sample set D first with the various visual angles subclass division methods described in step B+In every
One visual angle subclass k ∈ of sample image mark 1,2 ..., K }；Then above-mentioned formula is utilizedCalculate every sample
The visual angle contribution weight w of imagejk,j∈{1,2,...,N+}；Contribute the positive sample set of weight in final visual angle and the visual angle demarcated
D+, thenIt is redefined：
<mrow>
<mi>S</mi>
<mrow>
<mo>(</mo>
<mi>h</mi>
<mo>,</mo>
<mi>W</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munder>
<mrow>
<mi>m</mi>
<mi>a</mi>
<mi>x</mi>
</mrow>
<mrow>
<mi>k</mi>
<mo>&Element;</mo>
<mi>K</mi>
</mrow>
</munder>
<munder>
<mo>&Sigma;</mo>
<mrow>
<msub>
<mi>&Gamma;</mi>
<mi>j</mi>
</msub>
<mo>&Element;</mo>
<msup>
<mi>D</mi>
<mo>+</mo>
</msup>
</mrow>
</munder>
<msub>
<mi>w</mi>
<mrow>
<mi>j</mi>
<mi>k</mi>
</mrow>
</msub>
<munder>
<mo>&Sigma;</mo>
<mrow>
<msub>
<mi>p</mi>
<mi>t</mi>
</msub>
<mo>&Element;</mo>
<mi>G</mi>
</mrow>
</munder>
<mi>&Phi;</mi>
<mrow>
<mo>(</mo>
<mi>h</mi>
<mo>|</mo>
<msub>
<mi>&Gamma;</mi>
<mi>j</mi>
</msub>
<mo>,</mo>
<msub>
<mi>p</mi>
<mi>t</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>;</mo>
</mrow>
Wherein, W is wjkWhat is constituted is written as N greatly+× K weight matrix.
5. a kind of various visual angles vehicle checking method based on weighting Hough ballot according to claim 1, it is characterised in that
Described step E comprises the following steps：
Step E1：Test image is subjected to Scale Decomposition first, defining test image metric space isM is discrete
The number of yardstick；
Step E2：It is λ in yardstickmTest image in intensive sampling with the image block of size, find and each image Block- matching
Vision word, then utilizes above-mentioned formulaObtain Hough ballot figure under the yardstick；
Step E3：In (h, λm) constitute three-dimensional hough space (h=(hx,hy), include horizontal, ordinate position in image) in, profit
With mean-shift algorithms and decision threshold, determine final goal center (h ', λ 'm), and in former test chart
Position mark size isDetection block.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710554766.8A CN107330432B (en) | 2017-07-07 | 2017-07-07 | Multi-view vehicle detection method based on weighted Hough voting |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710554766.8A CN107330432B (en) | 2017-07-07 | 2017-07-07 | Multi-view vehicle detection method based on weighted Hough voting |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107330432A true CN107330432A (en) | 2017-11-07 |
| CN107330432B CN107330432B (en) | 2020-08-18 |
Family
ID=60197190
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710554766.8A Active CN107330432B (en) | 2017-07-07 | 2017-07-07 | Multi-view vehicle detection method based on weighted Hough voting |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107330432B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109726761A (en) * | 2018-12-29 | 2019-05-07 | 青岛海洋科学与技术国家实验室发展中心 | CNN evolvement method, AUV cluster working method, device and storage medium based on CNN |
| CN109948692A (en) * | 2019-03-16 | 2019-06-28 | 四川大学 | Picture detection method is generated based on the computer of multiple color spaces convolutional neural networks and random forest |
| CN111723721A (en) * | 2020-06-15 | 2020-09-29 | 中国传媒大学 | Three-dimensional target detection method, system and device based on RGB-D |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104112282A (en) * | 2014-07-14 | 2014-10-22 | 华中科技大学 | A method for tracking a plurality of moving objects in a monitor video based on on-line study |
| US8953888B2 (en) * | 2011-02-10 | 2015-02-10 | Microsoft Corporation | Detecting and localizing multiple objects in images using probabilistic inference |
| US20150131873A1 (en) * | 2013-11-14 | 2015-05-14 | Adobe Systems Incorporated | Exemplar-based feature weighting |
| CN106529461A (en) * | 2016-11-07 | 2017-03-22 | 湖南源信光电科技有限公司 | Vehicle model identifying algorithm based on integral characteristic channel and SVM training device |
-
2017
- 2017-07-07 CN CN201710554766.8A patent/CN107330432B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8953888B2 (en) * | 2011-02-10 | 2015-02-10 | Microsoft Corporation | Detecting and localizing multiple objects in images using probabilistic inference |
| US20150131873A1 (en) * | 2013-11-14 | 2015-05-14 | Adobe Systems Incorporated | Exemplar-based feature weighting |
| CN104112282A (en) * | 2014-07-14 | 2014-10-22 | 华中科技大学 | A method for tracking a plurality of moving objects in a monitor video based on on-line study |
| CN106529461A (en) * | 2016-11-07 | 2017-03-22 | 湖南源信光电科技有限公司 | Vehicle model identifying algorithm based on integral characteristic channel and SVM training device |
Non-Patent Citations (4)
| Title |
|---|
| DOUMANOGLOU A, KOUSKOURIDAS R, MALASSIOTIS S, ET AL.: "Recovering 6D object pose and predicting next-best-view in the crowd", 《PROCEEDINGS OF THE IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
| LEIBE B, LEONARDIS A, SCHIELE B: "Robust object detection with interleaved categorization and segmentation", 《 INTERNATIONAL JOURNAL OF COMPUTER VISION》 * |
| 向涛: "复杂场景下目标检测算法研究", 《中国博士学位论文全文数据库》 * |
| 杜本汉: "基于霍夫变换和条件随机场模型的目标检测", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109726761A (en) * | 2018-12-29 | 2019-05-07 | 青岛海洋科学与技术国家实验室发展中心 | CNN evolvement method, AUV cluster working method, device and storage medium based on CNN |
| CN109948692A (en) * | 2019-03-16 | 2019-06-28 | 四川大学 | Picture detection method is generated based on the computer of multiple color spaces convolutional neural networks and random forest |
| CN109948692B (en) * | 2019-03-16 | 2020-12-15 | 四川大学 | Computer-generated picture detection method based on multi-color space convolutional neural network and random forest |
| CN111723721A (en) * | 2020-06-15 | 2020-09-29 | 中国传媒大学 | Three-dimensional target detection method, system and device based on RGB-D |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107330432B (en) | 2020-08-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111259809B (en) | Unmanned aerial vehicle coastline floating garbage inspection system based on DANet | |
| CN112084869B (en) | Compact quadrilateral representation-based building target detection method | |
| Prathap et al. | Deep learning approach for building detection in satellite multispectral imagery | |
| Changzhen et al. | A traffic sign detection algorithm based on deep convolutional neural network | |
| CN103034863B (en) | The remote sensing image road acquisition methods of a kind of syncaryon Fisher and multiple dimensioned extraction | |
| CN104156734B (en) | A kind of complete autonomous on-line study method based on random fern grader | |
| CN103366602B (en) | Method of determining parking lot occupancy from digital camera images | |
| CN107730905A (en) | Multitask fake license plate vehicle vision detection system and method based on depth convolutional neural networks | |
| CN106127204A (en) | A kind of multi-direction meter reading Region detection algorithms of full convolutional neural networks | |
| CN106991382A (en) | A kind of remote sensing scene classification method | |
| Ghorbanzadeh et al. | Optimizing Sample Patches Selection of CNN to Improve the mIOU on Landslide Detection. | |
| CN109214308A (en) | A kind of traffic abnormity image identification method based on focal loss function | |
| CN103996027B (en) | Space-based space target recognizing method | |
| CN107330432A (en) | A kind of various visual angles vehicle checking method based on weighting Hough ballot | |
| CN109299673A (en) | The green degree spatial extraction method of group of cities and medium | |
| CN106558051A (en) | A kind of improved method for detecting road from single image | |
| CN104463240B (en) | A kind of instrument localization method and device | |
| CN108647721A (en) | Ground mulching verification method based on network picture automatic identification and system | |
| CN107301649A (en) | A kind of region merging technique SAR image coastline Detection Method algorithm based on super-pixel | |
| CN104331711B (en) | SAR image recognition methods based on multiple dimensioned fuzzy mearue and semi-supervised learning | |
| Li et al. | 3D map system for tree monitoring in hong kong using google street view imagery and deep learning | |
| CN116416626B (en) | Method, device, equipment and storage medium for acquiring circular seal data | |
| CN106295556A (en) | A kind of Approach for road detection based on SUAV Aerial Images | |
| Xia et al. | Geographically local representation learning with a spatial prior for visual localization | |
| CN112241844A (en) | Method and device for determining and updating environmental risk source background list of drinking water source area |
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 | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20200722 Address after: 224000 in Jiangsu Province in the south of Yancheng City District Xindu street landscape Avenue branch building 22 North Building (CND) Applicant after: YANCHENG CHANTU INTELLIGENT TECHNOLOGY Co.,Ltd. Address before: 450016, Zhengzhou City, Henan Province, Second West Avenue, South Road, one South Road Xinghua science and Technology Industrial Park Building 2, 9, 908, -37 room Applicant before: ZHENGZHOU CHANTU INTELLIGENT TECHNOLOGY Co.,Ltd. |
|
| TA01 | Transfer of patent application right | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |