CN106295716A - A kind of movement of traffic objective classification method based on video information and device - Google Patents
A kind of movement of traffic objective classification method based on video information and device Download PDFInfo
- Publication number
- CN106295716A CN106295716A CN201610709135.4A CN201610709135A CN106295716A CN 106295716 A CN106295716 A CN 106295716A CN 201610709135 A CN201610709135 A CN 201610709135A CN 106295716 A CN106295716 A CN 106295716A
- Authority
- CN
- China
- Prior art keywords
- field picture
- image sequence
- target
- gauss distribution
- movement
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of movement of traffic objective classification method based on video information and device.The method comprises the following steps: obtain the image sequence in Video Information files；For image sequence, set up background model, it is thus achieved that the target background of each two field picture；According to the target background obtained, the movement of traffic target of each two field picture in detection image sequence；Extract the movement of traffic clarification of objective point of each two field picture in image sequence, it is thus achieved that corresponding Feature Descriptor；The Feature Descriptor of each two field picture of image sequence is clustered, forms visual dictionary, a vision word in each cluster centre correspondence visual dictionary；The frequency occurred according to each vision word in visual dictionary, determines the classification of movement of traffic target in each two field picture.The technical scheme that the application embodiment of the present invention is provided, relatively accurately can classify to movement of traffic target, have preferable classifying quality.
Description
Technical field
The present invention relates to technical field of image processing, particularly relate to a kind of movement of traffic target based on video information and divide
Class method and device.
Background technology
Along with the progress of science and technology, the technological means used in traffic administration gets more and more.
Traffic administration, is primarily referred to as the requirement according to traffic law, regulation and the actual state of road traffic, uses religion
Educate, the means such as technology reasonably limit and scientifically organize, directing traffic.In correct process road traffic between people, car, road
Relation, makes that traffic is the safest, unobstructed, public hazards are little and less energy consumption.
Traffic administration includes the management to traffic control and the management etc. processed violating the regulations, during implementing traffic administration,
How accurately identifying the movement of traffic target in video information is currently to need badly to solve the technical problem that.
Summary of the invention
It is an object of the invention to provide a kind of movement of traffic objective classification method based on video information and device, Ke Yiyou
The movement of traffic targets such as motor vehicles, bicycle, pedestrian are classified by effect ground, it is simple to vehicle supervision department implements traffic pipe
Reason.
A kind of movement of traffic objective classification method based on video information, including:
Obtain the image sequence in Video Information files；
For described image sequence, set up background model, it is thus achieved that the target background of each two field picture；
According to the target background obtained, detect the movement of traffic target of each two field picture in described image sequence；
Extract the movement of traffic clarification of objective point of each two field picture in described image sequence, it is thus achieved that corresponding feature description
Son；
The Feature Descriptor of each two field picture of described image sequence is clustered, forms visual dictionary, each cluster
A vision word in the corresponding described visual dictionary in center；
The frequency occurred according to each vision word in described visual dictionary, determines movement of traffic target in each two field picture
Classification.
In a kind of detailed description of the invention of the present invention, described for described image sequence, set up background model, it is thus achieved that every
The target background of one two field picture, including:
For each pixel of each two field picture of described image sequence, according to the pixel value of this pixel, determine
The current Gauss distribution of this pixel；
Each in the mixed Gauss model corresponding with this pixel pre-build by described current Gauss distribution is gone through
History Gauss distribution is mated；
If any one the history Gauss distribution in described current Gauss distribution and described mixed Gauss model is the most not
Join, and the number of described history Gauss distribution is less than preset first threshold value, then described current Gauss distribution is increased to described mixed
Close in Gauss model；
If any one the history Gauss distribution in described current Gauss distribution and described mixed Gauss model is the most not
Join, and the number of described history Gauss distribution is equal to described preset first threshold value, then described current Gauss distribution is replaced described
The history Gauss distribution that mixed Gauss model medium priority is minimum；
If described current Gauss distribution is mated, then with at least one the history Gauss distribution in described mixed Gauss model
The more relevant parameter of new historical Gauss distribution；
According to the priority of Gauss distribution in the mixed Gauss model after updating, right to choose weight values sum is more than presetting second
The Gauss distribution of threshold value, sets up background model, it is thus achieved that target background.
In a kind of detailed description of the invention of the present invention, described according to the target background obtained, detect described image sequence
In the movement of traffic target of each two field picture, including:
For each two field picture of described image sequence, according to the target background of this two field picture obtained, pass through background subtraction
Point-score obtains the sport foreground region of this two field picture；
Movement of traffic target detection is carried out in the sport foreground region of this two field picture.
In a kind of detailed description of the invention of the present invention, described in the sport foreground region of this two field picture, carry out traffic fortune
Moving-target detects, including:
For continuous print N two field picture in described image sequence, two two field pictures being separated by a frame are done difference, obtains multiple frame poor
As a result, described N is preset value；
Each frame difference result is done and computing with the frame difference of intermediate frame and background frames, it is thus achieved that multiple and operation result；
Carry out all of or computing with operation result, incite somebody to action or operation result is defined as movement of traffic target.
In a kind of detailed description of the invention of the present invention, the traffic fortune of each two field picture in the described image sequence of described extraction
The characteristic point of moving-target, it is thus achieved that corresponding Feature Descriptor, including:
Gaussian pyramid iconic model is built, it is thus achieved that difference of Gaussian pyramid diagram picture according to described image sequence；
Calculate the Local Extremum in described difference of Gaussian pyramid diagram picture, calculated Local Extremum is defined as
Candidate's extreme point, and calculate corresponding yardstick；
Candidate's extreme point described in matching, it is thus achieved that the position of the characteristic point of moving target described in described image sequence and chi
Degree；
Position according to described characteristic point and yardstick, determine the Feature Descriptor of described Feature point correspondence.
A kind of movement of traffic target classification device based on video information, including:
Image sequence obtains module, for obtaining the image sequence in Video Information files；
Target background obtains module, for for described image sequence, sets up background model, it is thus achieved that the mesh of each two field picture
Mark background；
Movement of traffic module of target detection, for according to the target background obtained, detecting each frame in described image sequence
The movement of traffic target of image；
Feature Descriptor obtains module, for extracting the spy of the movement of traffic target of each two field picture in described image sequence
Levy a little, it is thus achieved that corresponding Feature Descriptor；
Visual dictionary forms module, for the Feature Descriptor of each two field picture of described image sequence is clustered,
Form visual dictionary, a vision word in the corresponding described visual dictionary of each cluster centre；
Movement of traffic target classification module, for the frequency occurred according to each vision word in described visual dictionary, really
The classification of movement of traffic target in fixed each two field picture.
In a kind of detailed description of the invention of the present invention, described target background obtains module, specifically for:
For each pixel of each two field picture of described image sequence, according to the pixel value of this pixel, determine
The current Gauss distribution of this pixel；
Each in the mixed Gauss model corresponding with this pixel pre-build by described current Gauss distribution is gone through
History Gauss distribution is mated；
If any one the history Gauss distribution in described current Gauss distribution and described mixed Gauss model is the most not
Join, and the number of described history Gauss distribution is less than preset first threshold value, then described current Gauss distribution is increased to described mixed
Close in Gauss model；
If any one the history Gauss distribution in described current Gauss distribution and described mixed Gauss model is the most not
Join, and the number of described history Gauss distribution is equal to described preset first threshold value, then described current Gauss distribution is replaced described
The history Gauss distribution that mixed Gauss model medium priority is minimum；
If described current Gauss distribution is mated, then with at least one the history Gauss distribution in described mixed Gauss model
The more relevant parameter of new historical Gauss distribution；
According to the priority of Gauss distribution in the mixed Gauss model after updating, right to choose weight values sum is more than presetting second
The Gauss distribution of threshold value, sets up background model, it is thus achieved that target background.
In a kind of detailed description of the invention of the present invention, described movement of traffic module of target detection, specifically for:
For each two field picture of described image sequence, according to the target background of this two field picture obtained, pass through background subtraction
Point-score obtains the sport foreground region of this two field picture；
Movement of traffic target detection is carried out in the sport foreground region of this two field picture.
In a kind of detailed description of the invention of the present invention, described movement of traffic module of target detection, specifically for:
For continuous print N two field picture in described image sequence, two two field pictures being separated by a frame are done difference, obtains multiple frame poor
As a result, described N is preset value；
Each frame difference result is done and computing with the frame difference of intermediate frame and background frames, it is thus achieved that multiple and operation result；
Carry out all of or computing with operation result, incite somebody to action or operation result is defined as movement of traffic target.
In a kind of detailed description of the invention of the present invention, described Feature Descriptor obtains module, specifically for:
Gaussian pyramid iconic model is built, it is thus achieved that difference of Gaussian pyramid diagram picture according to described image sequence；
Calculate the Local Extremum in described difference of Gaussian pyramid diagram picture, calculated Local Extremum is defined as
Candidate's extreme point, and calculate corresponding yardstick；
Candidate's extreme point described in matching, it is thus achieved that the position of the characteristic point of moving target described in described image sequence and chi
Degree；
Position according to described characteristic point and yardstick, determine the Feature Descriptor of described Feature point correspondence.
The technical scheme that the application embodiment of the present invention is provided, it is thus achieved that after the image sequence in Video Information files, for
Image sequence, sets up background model, it is thus achieved that the target background of each two field picture, according to target background, detects in each two field picture
Movement of traffic target, extract movement of traffic clarification of objective point, it is thus achieved that corresponding Feature Descriptor, Feature Descriptor carried out
Cluster, can form visual dictionary, the frequency occurred according to visual dictionary, it may be determined that movement of traffic target in each two field picture
Classification, relatively accurately movement of traffic target can be classified, there is preferable classifying quality.
Accompanying drawing explanation
For the clearer explanation embodiment of the present invention or the technical scheme of prior art, below will be to embodiment or existing
In technology description, the required accompanying drawing used is briefly described, it should be apparent that, the accompanying drawing in describing below is only this
Some bright embodiments, for those of ordinary skill in the art, on the premise of not paying creative work, it is also possible to root
Other accompanying drawing is obtained according to these accompanying drawings.
Fig. 1 is the implementing procedure of a kind of movement of traffic objective classification method based on video information in the embodiment of the present invention
Figure；
Fig. 2 a is a kind of schematic diagram determining characteristic point direction in the embodiment of the present invention；
Fig. 2 b is the another kind of schematic diagram determining characteristic point direction in the embodiment of the present invention.
Fig. 3 is characteristic point neighborhood gradient information figure in the embodiment of the present invention；
Fig. 4 is embodiment of the present invention neutron region gradient hum pattern；
Fig. 5 is the schematic diagram extracting SIFT feature sample in the embodiment of the present invention；
Fig. 6 is the structural representation of a kind of movement of traffic target classification device based on video information in the embodiment of the present invention
Figure.
Detailed description of the invention
In order to make those skilled in the art be more fully understood that the present invention program, below in conjunction with the accompanying drawings and detailed description of the invention
The present invention is described in further detail.Obviously, described embodiment be only a part of embodiment of the present invention rather than
Whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art are not making creative work premise
Lower obtained every other embodiment, broadly falls into the scope of protection of the invention.
Embodiments provide a kind of movement of traffic objective classification method based on video information, can be at traffic pipe
The scenes such as the enforcement of reason are applied.For the Video Information files obtained, detect movement of traffic target, according to movement of traffic mesh
Target local feature determines the classification of movement of traffic target.Movement of traffic target is the traffic such as motor vehicles, bicycle or pedestrian
Primary Actor.
A kind of based on video information movement of traffic target classification shown in Figure 1, that provided by the embodiment of the present invention
The implementing procedure figure of method, the method may comprise steps of:
S110: obtain the image sequence in Video Information files.
In embodiments of the present invention, video information can be the traffic video letter that the photographic head by specific region captures
Breath.Such as, certain crossing or other traffic conditions monitoring regions are carried out seizure and obtain.Video Information files can be from traffic
Supervision department corresponding to administration section or specific region obtains.
S120: for image sequence, sets up background model, it is thus achieved that the target background of each two field picture.
In a kind of detailed description of the invention of the present invention, step S120 may comprise steps of:
Step one: for each pixel of each two field picture of image sequence, according to the pixel value of this pixel, really
The current Gauss distribution of this pixel fixed；
Step 2: each in the mixed Gauss model corresponding with this pixel pre-build by current Gauss distribution
History Gauss distribution is mated；
Step 3: if current Gauss distribution and any one the history Gauss distribution in mixed Gauss model are the most not
Join, and the number of history Gauss distribution is less than preset first threshold value, then current Gauss distribution increased in mixed Gauss model；
Step 4: if current Gauss distribution and any one the history Gauss distribution in mixed Gauss model are the most not
Join, and the number of history Gauss distribution is equal to preset first threshold value, then current Gauss distribution is replaced in mixed Gauss model excellent
The history Gauss distribution that first level is minimum；
Step 5: if current Gauss distribution is mated, then with at least one the history Gauss distribution in mixed Gauss model
The more relevant parameter of new historical Gauss distribution；
Step 6: according to the priority of Gauss distribution in the mixed Gauss model after updating, right to choose weight values sum is more than
Preset the Gauss distribution of Second Threshold, set up background model, it is thus achieved that target background.
For ease of describing, above-mentioned six steps are combined and illustrates.
In step S110, it is thus achieved that Video Information files, Video Information files comprises the image being made up of a frame two field picture
Sequence.For each pixel of each two field picture of image sequence, this picture can be determined according to the pixel value of this pixel
The current Gauss distribution of vegetarian refreshments.
In embodiments of the present invention, can pre-build a mixed Gauss model for each pixel, this mixing is high
This model comprises multiple history Gauss distribution of this pixel.
Shown in the specific descriptions such as formula (1) of the gray value of each pixel of t two field picture:
Wherein, It(x is y) that ((x, y) is space coordinates, and x is this frame for x, pixel value y) for the pixel of t two field picture
Image is the value coordinate of any pixel in x-axis, and y is the value coordinate of this two field picture any pixel on the y axis, ωi,tFor
The weight of i-th Gauss distribution in mixed Gauss model, K is the number of Gauss distribution in mixed Gauss model, for positive integer, leads to
Often take 3-5, ui,tFor the average of i-th Gauss distribution,∑i,tFor the covariance of i-th Gauss distribution, η (It
(x, y), uI, t, ΣI, t) it is the probability density function of background, specifically describe as shown in formula (2):
Shown in the specific descriptions of the priority of each Gauss distribution such as formula (3):
Comprising k history Gauss distribution in the mixed Gauss model pre-build, history Gauss distribution is according to priority
Sequence arranges.
Current Gauss distribution is mated with each history Gauss distribution.
Matching condition specifically describes as shown in formula (4):
|It(x,y)-ui,t-1| ＜ 2.5* δI, t-1Formula (4)
Wherein, δi,t-1For standard deviation.
If current Gauss distribution is not mated with any one the history Gauss distribution in mixed Gauss model, and history
The number of Gauss distribution is less than preset first threshold value, then increase in mixed Gauss model by current Gauss distribution, i.e. increases new
Gauss distribution.Preset first threshold value is K.
If current Gauss distribution is not mated with any one the history Gauss distribution in mixed Gauss model, and history
The number of Gauss distribution is equal to preset first threshold value, then current Gauss distribution is replaced mixed Gauss model medium priority minimum
One history Gauss distribution, i.e. replaces, by new Gauss distribution, the Gauss distribution that priority is minimum.
If current Gauss distribution is mated with at least one the history Gauss distribution in mixed Gauss model, the most more new historical
The relevant parameter of Gauss distribution.Concrete, first can be updated according to formula (5), formula (6), formula (7) and formula (8)
The relevant parameter of the history Gauss distribution that the match is successful, updates the weight of other history Gauss distribution according to formula (9), and other are gone through
Other parameters of history Gauss distribution, such as average and variance etc., keep constant.
ωi,t=(1-α) ωi,t-1+αMI, tFormula (5)
β=α η (It(x,y)ui,t,δi,t) formula (8)
ωi,t=(1-α) ωi,t-1Formula (9)
Wherein, α is learning rate, represents the speed of context update, 0≤α≤1；β is parameter turnover rate；Mi,tRepresent a boundary
Limit, if the current Gauss distribution of t is mated with i-th history Gauss distribution, then Mi,t=1.
After the relevant parameter of each Gauss distribution in mixed Gauss model is updated, can regain each high
The priority of this distribution.According to priority, the Gauss distribution of each pixel is ranked up, will most possibly represent background
Before Gauss distribution comes.Right to choose weight values sum, more than the Gauss distribution of default Second Threshold, sets up background model, it is thus achieved that mesh
Mark background.
Concrete, background model can be set up according to formula (10):
Wherein, ε is Second Threshold, generally takes the number that 0.3-0.9, x are the Gauss distribution taken.
S130: according to the target background obtained, the movement of traffic target of each two field picture in detection image sequence.
In a kind of detailed description of the invention of the present invention, step S130 may comprise steps of:
First step: for each two field picture of image sequence, according to the target background of this two field picture obtained, pass through
Background subtraction obtains the sport foreground region of this two field picture.
In step S120, it is thus achieved that target background, according to the target background obtained, fortune can be obtained by background subtraction
Dynamic foreground area.
Concrete, formula (11) can be passed through and obtain:
DT (x, y)=| IT (x, y)-Bt(x,y)| formula (11)
Wherein, BT (x, y)For the target background obtained by Gaussian modeling method.
Second step: carry out movement of traffic target detection in the sport foreground region of this two field picture.
In a kind of detailed description of the invention of the present invention, this second step specifically may comprise steps of:
Step one: for continuous print N two field picture in image sequence, two two field pictures being separated by a frame are done difference, obtains multiple
Frame difference result, N is preset value；
Step 2: each frame difference result is done and computing with the frame difference of intermediate frame and background frames, it is thus achieved that multiple and computing is tied
Really；
Step 3: carry out all of or computing with operation result, incites somebody to action or operation result is defined as movement of traffic target.
For convenience of understanding, it is illustrated with five frame difference methods.
Assume that in image sequence, continuous print 5 two field picture is: T1, T2, T3, T4, T5.Wherein T3For intermediate frame.To being separated by a frame
Two two field pictures do difference, as shown in formula (12), formula (13) and formula (14), obtain multiple frame difference result:
DifT31=| T3-T1| formula (12)
DifT42=| T4-T2| formula (13)
DifT53=| T5-T3| formula (14)
3 the frame difference results obtained are respectively as follows: DifT31, DifT42, DifT53.By each frame difference result with intermediate frame T3With
Frame difference Dif of background framesTMDo AND operation, as shown in formula (15), formula (16) and formula (17), it is thus achieved that multiple tie with computing
Really:
D1=DifT31∩DifTMFormula (15)
D2=DifT42∩DifTMFormula (16)
D3=DifT53∩DifTMFormula (17)
Carry out all of or computing, i.e. to D with operation result1、D2、D3Carry out inclusive-OR operation to obtain or operation result D,
To or operation result is defined as movement of traffic target.
S140: extract the movement of traffic clarification of objective point of each two field picture in image sequence, it is thus achieved that corresponding feature is retouched
State son.
In embodiments of the present invention, it is possible to use SIFT (Scale-invariant feature transform, yardstick
Invariant features converts) algorithm movement of traffic Objective extraction characteristic point to each two field picture, and use characteristic vector to be described.
One characteristic vector represents a characteristic point, can extract several characteristic points in a two field picture.
In a kind of detailed description of the invention of the present invention, step S140 may comprise steps of:
Step one: build gaussian pyramid iconic model according to image sequence, it is thus achieved that difference of Gaussian pyramid diagram picture；
Step 2: calculate the Local Extremum in difference of Gaussian pyramid diagram picture, by true for calculated Local Extremum
It is set to candidate's extreme point, and calculates corresponding yardstick；
Step 3: matching candidate's extreme point, it is thus achieved that the position of the characteristic point of moving target and yardstick in image sequence；
Step 4: according to position and the yardstick of characteristic point, determine the Feature Descriptor of Feature point correspondence.
For ease of describing, aforementioned four step is combined and illustrates.
Gaussian pyramid iconic model can be built, by 2 adjacent Gaussian image are done according to image sequence data
Subtract each other process, it is possible to obtain difference of Gaussian pyramid diagram picture.Calculate the Local Extremum in difference of Gaussian pyramid diagram picture, will meter
The Local Extremum obtained is defined as candidate's extreme point, and calculates its corresponding yardstick.
Shown in concrete computational methods such as formula (19):
D (x, y, σ)=(G (x, y, k σ)-G (x, y, σ)) * I (x, y)=L (x, y, k σ)-L (x, y, σ) formula (19)
Wherein, (x is y) that ((x y) is L (x, y, σ)=G (x, y, σ) * I current picture point for x, pixel value y) to I
Metric space, * is convolution operator, and σ is scale parameter, and the feature of σ value the biggest expression image is the fuzzyyest, and σ value is the least represents figure
The feature of picture is the most clear.
Matching candidate's extreme point, specifically can carry out matching candidate's extreme point by a three-dimensional quadratic function, it is thus achieved that image
The accurate position of the characteristic point of moving target and yardstick in sequence.In actual applications, the limit of instability can be rejected simultaneously
Edge characteristic point and the characteristic point of low contrast, with the uniqueness of Enhanced feature point, improve matching performance.
Taylor (Taylor) expansion of metric space function D (x, y, σ) represents as shown in formula (20):
Wherein x=(x, y, σ)TIt is the skew relative to sampled point, above formula derivation is obtained formula (21):
When this value on any dimension x, y or σ more than 0.5 time, represent that interpolation off-centring, on its point of proximity, changes and closes
Key point position, interpolation is until restraining again.Extreme point formula (21) is brought in formula (20), available formula (22):
DoG (Difference of Gaussian) pyramid algorith can produce stronger response at edge, can remove
The marginal point that principal curvatures differs greatly.The computational methods of principal curvatures are: calculate 2 × 2 dimension Hash Hessian squares of each characteristic point
Battle array, as shown in formula (23):
The principal curvatures of D is directly proportional to the eigenvalue of this Hessian matrix, if α is the bigger eigenvalue of Hessian matrix, and β
Be the less eigenvalue of Hessian matrix, then the mark of Hessian and determinant can be expressed as formula (24) and formula (25):
Tr (H)=Dxx+Dyy=alpha+beta formula (24)
Det (H)=DxxDyy-(Dxy)2=α β formula (25)
Make α=r β, then obtain formula (26):
Result only depends on the ratio r of two eigenvalues, and along with the increase of r, i.e. minimax eigenvalue ratio increases, and says
It is the biggest that bright two principal curvatures differ, then this point is more likely to be marginal point, so for rejecting marginal point, can be that r arranges one
Individual threshold value, rejects the point higher than this threshold value, concrete, can pass through formula (27) and judge:
In order to make characteristic point have rotational invariance, can be each characteristic point distribution one unification based on topography
The principal direction of attribute.The direction is according to the gradient magnitude of characteristic point each pixel in certain neighborhood and the statistical information pair in direction
In each sampled point L, (x, y), in neighborhood, (x, y) (x y) is expressed as formula to gradient magnitude m of pixel with direction θ
And formula (29) (28):
The Gradient distribution situation of pixel value, histogrammic gradient direction model in key point adjacent area is added up by rectangular histogram
Enclosing is 0-360 degree, has 8 intervals, and each interval 45 degree, histogram peak place is as the principal direction of this feature point, such as Fig. 2 a
Shown in Fig. 2 b.
According to the principal direction of characteristic point, image is rotated, to ensure that it has the invariance of rotation.Then, with this spy
Centered by levying a little, choose the neighborhood window of 16 × 16, as shown in Figure 3.The most each little lattice represent this feature point adjacent area
A pixel value in window, the direction of arrow is the direction of pixel.16 × 16 neighborhood windows are uniformly divided into 16 4 × 4
Subregion, as shown in Figure 4.To calculate in each region 8 directions (0 °, 45 °, 90 °, 135 °, 180 °, 225 °, 270 °,
315 °) gradient accumulated value, use Gaussian Blur method, to each pixel gradient value distribute a weight, further away from feature
The pixel weight of point is the least.Finally the vector information in 8 directions in every sub regions is sorted successively, it is possible to constitute one
The characteristic vector of 4 × 4 × 8=128 dimension, i.e. SIFT feature describe son, as Fig. 5 extracts SIFT feature sample schematic diagram.
S150: clustered by the Feature Descriptor of each two field picture of image sequence, forms visual dictionary, each cluster
A vision word in the correspondence visual dictionary of center.
The Feature Descriptor of every two field picture of image sequence clusters, and forms visual dictionary.Each cluster centre is corresponding
A vision word in visual dictionary.
S160: the frequency occurred according to each vision word in visual dictionary, determines movement of traffic mesh in each two field picture
Target classification.
The frequency that in visual dictionary, each vision word occurs can be added up, using all movement of traffic targets as training sample
This image feature vector is sent into SVM and is gone study, utilizes SVM to adjudicate the generic of this movement of traffic target.
Table 1 show the experiment that movement of traffic target is classified by the technical scheme applying the embodiment of the present invention to be provided
As a result, accuracy rate is higher, has preferable classifying quality.
| Motor vehicles | Bicycle | Pedestrian | |
| Recall ratio (%) | 90.0 | 87.5 | 95.0 |
| Precision ratio (%) | 92.0 | 90.1 | 90.5 |
Table 1
The method that the application embodiment of the present invention is provided, it is thus achieved that after the image sequence in Video Information files, for image
Sequence, sets up background model, it is thus achieved that the target background of each two field picture, according to target background, detects the friendship in each two field picture
Logical moving target, extracts movement of traffic clarification of objective point, it is thus achieved that corresponding Feature Descriptor, is gathered by Feature Descriptor
Class, can form visual dictionary, the frequency occurred according to visual dictionary, it may be determined that movement of traffic target in each two field picture
Classification, relatively accurately can classify to movement of traffic target, have preferable classifying quality.
Corresponding to above method embodiment, the embodiment of the present invention additionally provides a kind of movement of traffic based on video information
Target classification device, a kind of movement of traffic target classification device based on video information described below with above-described based on
The movement of traffic objective classification method of video information can be mutually to should refer to.
Shown in Figure 6, this device includes:
Image sequence obtains module 210, for obtaining the image sequence in Video Information files；
Target background obtains module 220, for for image sequence, sets up background model, it is thus achieved that the mesh of each two field picture
Mark background；
Movement of traffic module of target detection 230, for according to the target background obtained, each frame figure in detection image sequence
The movement of traffic target of picture；
Feature Descriptor obtains module 240, for extracting the spy of the movement of traffic target of each two field picture in image sequence
Levy a little, it is thus achieved that corresponding Feature Descriptor；
Visual dictionary forms module 250, for the Feature Descriptor of each two field picture of image sequence is clustered, and shape
Become visual dictionary, a vision word in each cluster centre correspondence visual dictionary；
Movement of traffic target classification module 260, for the frequency occurred according to each vision word in visual dictionary, determines
The classification of movement of traffic target in each two field picture.
In a kind of detailed description of the invention of the present invention, target background obtains module 220, specifically for:
For each pixel of each two field picture of image sequence, according to the pixel value of this pixel, determine this picture
The current Gauss distribution of vegetarian refreshments；
Each history in the mixed Gauss model corresponding with this pixel pre-build by current Gauss distribution is high
This distribution is mated；
If current Gauss distribution is not mated with any one the history Gauss distribution in mixed Gauss model, and history
The number of Gauss distribution is less than preset first threshold value, then current Gauss distribution increased in mixed Gauss model；
If current Gauss distribution is not mated with any one the history Gauss distribution in mixed Gauss model, and history
The number of Gauss distribution is equal to preset first threshold value, then current Gauss distribution is replaced mixed Gauss model medium priority minimum
History Gauss distribution；
If current Gauss distribution is mated with at least one the history Gauss distribution in mixed Gauss model, the most more new historical
The relevant parameter of Gauss distribution；
According to the priority of Gauss distribution in the mixed Gauss model after updating, right to choose weight values sum is more than presetting second
The Gauss distribution of threshold value, sets up background model, it is thus achieved that target background.
In a kind of detailed description of the invention of the present invention, movement of traffic module of target detection 230, specifically for:
For each two field picture of image sequence, according to the target background of this two field picture obtained, pass through background subtraction
Obtain the sport foreground region of this two field picture；
Movement of traffic target detection is carried out in the sport foreground region of this two field picture.
In a kind of detailed description of the invention of the present invention, movement of traffic module of target detection 230, specifically for:
For continuous print N two field picture in image sequence, two two field pictures being separated by a frame are done difference, obtain multiple frame difference knot
Really, N is preset value；
Each frame difference result is done and computing with the frame difference of intermediate frame and background frames, it is thus achieved that multiple and operation result；
Carry out all of or computing with operation result, incite somebody to action or operation result is defined as movement of traffic target.
In a kind of detailed description of the invention of the present invention, Feature Descriptor obtains module 240, specifically for:
Gaussian pyramid iconic model is built, it is thus achieved that difference of Gaussian pyramid diagram picture according to image sequence；
Calculate the Local Extremum in difference of Gaussian pyramid diagram picture, calculated Local Extremum is defined as candidate
Extreme point, and calculate corresponding yardstick；
Matching candidate's extreme point, it is thus achieved that the position of the characteristic point of moving target and yardstick in image sequence；
Position according to characteristic point and yardstick, determine the Feature Descriptor of Feature point correspondence.
The device that the application embodiment of the present invention is provided, it is thus achieved that after the image sequence in Video Information files, for image
Sequence, sets up background model, it is thus achieved that the target background of each two field picture, according to target background, detects the friendship in each two field picture
Logical moving target, extracts movement of traffic clarification of objective point, it is thus achieved that corresponding Feature Descriptor, is gathered by Feature Descriptor
Class, can form visual dictionary, the frequency occurred according to visual dictionary, it may be determined that movement of traffic target in each two field picture
Classification, relatively accurately can classify to movement of traffic target, have preferable classifying quality.
In this specification, each embodiment uses the mode gone forward one by one to describe, and what each embodiment stressed is and other
The difference of embodiment, between each embodiment, same or similar part sees mutually.For filling disclosed in embodiment
For putting, owing to it corresponds to the method disclosed in Example, so describe is fairly simple, relevant part sees method part
Illustrate.
Professional further appreciates that, in conjunction with the unit of each example that the embodiments described herein describes
And algorithm steps, it is possible to electronic hardware, computer software or the two be implemented in combination in, in order to clearly demonstrate hardware and
The interchangeability of software, the most generally describes composition and the step of each example according to function.These
Function performs with hardware or software mode actually, depends on application-specific and the design constraint of technical scheme.Specialty
Technical staff specifically should can be used for using different methods to realize described function to each, but this realization should not
Think beyond the scope of this invention.
The method described in conjunction with the embodiments described herein or the step of algorithm can direct hardware, processor be held
The software module of row, or the combination of the two implements.Software module can be placed in random access memory (RAM), internal memory, read-only deposit
Reservoir (ROM), electrically programmable ROM, electrically erasable ROM, depositor, hard disk, moveable magnetic disc, CD-ROM or technology
In any other form of storage medium well known in field.
Above a kind of movement of traffic objective classification method based on video information provided by the present invention and device are carried out
It is discussed in detail.Principle and the embodiment of the present invention are set forth by specific case used herein, above example
Explanation be only intended to help to understand method and the core concept thereof of the present invention.It should be pointed out that, for the art is common
For technical staff, under the premise without departing from the principles of the invention, it is also possible to the present invention is carried out some improvement and modification, these
Improve and modify in the protection domain also falling into the claims in the present invention.
Claims (10)
1. a movement of traffic objective classification method based on video information, it is characterised in that including:
Obtain the image sequence in Video Information files；
For described image sequence, set up background model, it is thus achieved that the target background of each two field picture；
According to the target background obtained, detect the movement of traffic target of each two field picture in described image sequence；
Extract the movement of traffic clarification of objective point of each two field picture in described image sequence, it is thus achieved that corresponding Feature Descriptor；
The Feature Descriptor of each two field picture of described image sequence is clustered, forms visual dictionary, each cluster centre
A vision word in corresponding described visual dictionary；
The frequency occurred according to each vision word in described visual dictionary, determines the class of movement of traffic target in each two field picture
Not.
Method the most according to claim 1, it is characterised in that described set up background model for described image sequence, obtains
Obtain the target background of each two field picture, including:
For each pixel of each two field picture of described image sequence, according to the pixel value of this pixel, determine this picture
The current Gauss distribution of vegetarian refreshments；
Each history in the mixed Gauss model corresponding with this pixel pre-build by described current Gauss distribution is high
This distribution is mated；
If described current Gauss distribution is not mated with any one the history Gauss distribution in described mixed Gauss model, and
The number of described history Gauss distribution is less than preset first threshold value, then described current Gauss distribution is increased to described mixed Gaussian
In model；
If described current Gauss distribution is not mated with any one the history Gauss distribution in described mixed Gauss model, and
The number of described history Gauss distribution is equal to described preset first threshold value, then described current Gauss distribution is replaced described mixing height
The history Gauss distribution that this model medium priority is minimum；
If described current Gauss distribution is mated with at least one the history Gauss distribution in described mixed Gauss model, then update
The relevant parameter of history Gauss distribution；
According to the priority of Gauss distribution in the mixed Gauss model after updating, right to choose weight values sum is more than presetting Second Threshold
Gauss distribution, set up background model, it is thus achieved that target background.
Method the most according to claim 1 and 2, it is characterised in that described according to the target background obtained, detects described figure
The movement of traffic target of each two field picture in picture sequence, including:
For each two field picture of described image sequence, according to the target background of this two field picture obtained, pass through background subtraction
Obtain the sport foreground region of this two field picture；
Movement of traffic target detection is carried out in the sport foreground region of this two field picture.
Method the most according to claim 3, it is characterised in that described hand in the sport foreground region of this two field picture
Logical moving object detection, including:
For continuous print N two field picture in described image sequence, two two field pictures being separated by a frame are done difference, obtain multiple frame difference knot
Really, described N is preset value；
Each frame difference result is done and computing with the frame difference of intermediate frame and background frames, it is thus achieved that multiple and operation result；
Carry out all of or computing with operation result, incite somebody to action or operation result is defined as movement of traffic target.
Method the most according to claim 1, it is characterised in that the friendship of each two field picture in the described image sequence of described extraction
The characteristic point of logical moving target, it is thus achieved that corresponding Feature Descriptor, including:
Gaussian pyramid iconic model is built, it is thus achieved that difference of Gaussian pyramid diagram picture according to described image sequence；
Calculate the Local Extremum in described difference of Gaussian pyramid diagram picture, calculated Local Extremum is defined as candidate
Extreme point, and calculate corresponding yardstick；
Candidate's extreme point described in matching, it is thus achieved that the position of the characteristic point of moving target described in described image sequence and yardstick；
Position according to described characteristic point and yardstick, determine the Feature Descriptor of described Feature point correspondence.
6. a movement of traffic target classification device based on video information, it is characterised in that including:
Image sequence obtains module, for obtaining the image sequence in Video Information files；
Target background obtains module, for for described image sequence, sets up background model, it is thus achieved that the target back of the body of each two field picture
Scape；
Movement of traffic module of target detection, for according to the target background obtained, detecting each two field picture in described image sequence
Movement of traffic target；
Feature Descriptor obtains module, for extracting the movement of traffic clarification of objective of each two field picture in described image sequence
Point, it is thus achieved that corresponding Feature Descriptor；
Visual dictionary forms module, for being clustered by the Feature Descriptor of each two field picture of described image sequence, is formed
Visual dictionary, a vision word in the corresponding described visual dictionary of each cluster centre；
Movement of traffic target classification module, for the frequency occurred according to each vision word in described visual dictionary, determines every
The classification of movement of traffic target in one two field picture.
Device the most according to claim 6, it is characterised in that described target background obtains module, specifically for:
For each pixel of each two field picture of described image sequence, according to the pixel value of this pixel, determine this picture
The current Gauss distribution of vegetarian refreshments；
Each history in the mixed Gauss model corresponding with this pixel pre-build by described current Gauss distribution is high
This distribution is mated；
If described current Gauss distribution is not mated with any one the history Gauss distribution in described mixed Gauss model, and
The number of described history Gauss distribution is less than preset first threshold value, then described current Gauss distribution is increased to described mixed Gaussian
In model；
If described current Gauss distribution is not mated with any one the history Gauss distribution in described mixed Gauss model, and
The number of described history Gauss distribution is equal to described preset first threshold value, then described current Gauss distribution is replaced described mixing height
The history Gauss distribution that this model medium priority is minimum；
If described current Gauss distribution is mated with at least one the history Gauss distribution in described mixed Gauss model, then update
The relevant parameter of history Gauss distribution；
According to the priority of Gauss distribution in the mixed Gauss model after updating, right to choose weight values sum is more than presetting Second Threshold
Gauss distribution, set up background model, it is thus achieved that target background.
8., according to the device described in claim 6 or 7, it is characterised in that described movement of traffic module of target detection, specifically use
In:
For each two field picture of described image sequence, according to the target background of this two field picture obtained, pass through background subtraction
Obtain the sport foreground region of this two field picture；
Movement of traffic target detection is carried out in the sport foreground region of this two field picture.
Device the most according to claim 8, it is characterised in that described movement of traffic module of target detection, specifically for:
For continuous print N two field picture in described image sequence, two two field pictures being separated by a frame are done difference, obtain multiple frame difference knot
Really, described N is preset value；
Each frame difference result is done and computing with the frame difference of intermediate frame and background frames, it is thus achieved that multiple and operation result；
Carry out all of or computing with operation result, incite somebody to action or operation result is defined as movement of traffic target.
Device the most according to claim 6, it is characterised in that described Feature Descriptor obtains module, specifically for:
Gaussian pyramid iconic model is built, it is thus achieved that difference of Gaussian pyramid diagram picture according to described image sequence；
Calculate the Local Extremum in described difference of Gaussian pyramid diagram picture, calculated Local Extremum is defined as candidate
Extreme point, and calculate corresponding yardstick；
Candidate's extreme point described in matching, it is thus achieved that the position of the characteristic point of moving target described in described image sequence and yardstick；
Position according to described characteristic point and yardstick, determine the Feature Descriptor of described Feature point correspondence.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610709135.4A CN106295716A (en) | 2016-08-23 | 2016-08-23 | A kind of movement of traffic objective classification method based on video information and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610709135.4A CN106295716A (en) | 2016-08-23 | 2016-08-23 | A kind of movement of traffic objective classification method based on video information and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106295716A true CN106295716A (en) | 2017-01-04 |
Family
ID=57615549
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610709135.4A Pending CN106295716A (en) | 2016-08-23 | 2016-08-23 | A kind of movement of traffic objective classification method based on video information and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106295716A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106897742A (en) * | 2017-02-21 | 2017-06-27 | 北京市商汤科技开发有限公司 | Method, device and electronic equipment for detecting object in video |
| CN109697724A (en) * | 2017-10-24 | 2019-04-30 | 北京京东尚科信息技术有限公司 | Video Image Segmentation method and device, storage medium, electronic equipment |
| CN109977909A (en) * | 2019-04-04 | 2019-07-05 | 山东财经大学 | Finger vein identification method and system based on minutiae point Region Matching |
| CN111612811A (en) * | 2020-06-05 | 2020-09-01 | 中国人民解放军军事科学院国防科技创新研究院 | Video foreground information extraction method and system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102855473A (en) * | 2012-08-21 | 2013-01-02 | 中国科学院信息工程研究所 | Image multi-target detecting method based on similarity measurement |
| CN102945375A (en) * | 2012-11-20 | 2013-02-27 | 天津理工大学 | Multi-view monitoring video behavior detection and recognition method under multiple constraints |
| CN105046683A (en) * | 2014-12-31 | 2015-11-11 | 北京航空航天大学 | Object detection method based on adaptive-parameter-adjustment Gaussian mixture model |
-
2016
- 2016-08-23 CN CN201610709135.4A patent/CN106295716A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102855473A (en) * | 2012-08-21 | 2013-01-02 | 中国科学院信息工程研究所 | Image multi-target detecting method based on similarity measurement |
| CN102945375A (en) * | 2012-11-20 | 2013-02-27 | 天津理工大学 | Multi-view monitoring video behavior detection and recognition method under multiple constraints |
| CN105046683A (en) * | 2014-12-31 | 2015-11-11 | 北京航空航天大学 | Object detection method based on adaptive-parameter-adjustment Gaussian mixture model |
Non-Patent Citations (3)
| Title |
|---|
| 张芯: "监控视频中运动目标分类算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 邓文浩 等: "基于改进的帧差法和Mean-shift结合的运动目标自动检测与跟踪", 《现代电子技术》 * |
| 黄鑫娟 等: "自适应混合高斯背景模型的运动目标检测方法", 《计算机应用》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106897742A (en) * | 2017-02-21 | 2017-06-27 | 北京市商汤科技开发有限公司 | Method, device and electronic equipment for detecting object in video |
| WO2018153323A1 (en) * | 2017-02-21 | 2018-08-30 | 北京市商汤科技开发有限公司 | Method and device for detecting object in video, and electronic device |
| CN106897742B (en) * | 2017-02-21 | 2020-10-27 | 北京市商汤科技开发有限公司 | Method and device for detecting object in video and electronic equipment |
| CN109697724A (en) * | 2017-10-24 | 2019-04-30 | 北京京东尚科信息技术有限公司 | Video Image Segmentation method and device, storage medium, electronic equipment |
| CN109697724B (en) * | 2017-10-24 | 2021-02-26 | 北京京东尚科信息技术有限公司 | Video image segmentation method and device, storage medium and electronic equipment |
| US11227393B2 (en) | 2017-10-24 | 2022-01-18 | Beijing Jingdong Shangke Information Technology Co., Ltd. | Video image segmentation method and apparatus, storage medium and electronic device |
| CN109977909A (en) * | 2019-04-04 | 2019-07-05 | 山东财经大学 | Finger vein identification method and system based on minutiae point Region Matching |
| CN111612811A (en) * | 2020-06-05 | 2020-09-01 | 中国人民解放军军事科学院国防科技创新研究院 | Video foreground information extraction method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10929649B2 (en) | Multi-pose face feature point detection method based on cascade regression | |
| CN110543837B (en) | Visible light airport airplane detection method based on potential target point | |
| CN115861135B (en) | Image enhancement and recognition method applied to panoramic detection of box body | |
| WO2020156361A1 (en) | Training sample obtaining method and apparatus, electronic device and storage medium | |
| CN108304798B (en) | Street level order event video detection method based on deep learning and motion consistency | |
| CN112686812B (en) | Bank card inclination correction detection method and device, readable storage medium and terminal | |
| CN103886589B (en) | Object-oriented automated high-precision edge extracting method | |
| CN110334762B (en) | Feature matching method based on quad tree combined with ORB and SIFT | |
| CN107316031A (en) | The image characteristic extracting method recognized again for pedestrian | |
| CN105989334B (en) | Road detection method based on monocular vision | |
| CN110532946B (en) | Method for identifying axle type of green-traffic vehicle based on convolutional neural network | |
| CN103577815A (en) | Face alignment method and system | |
| CN106446862A (en) | Face detection method and system | |
| CN110706235B (en) | Far infrared pedestrian detection method based on two-stage cascade segmentation | |
| CN106295716A (en) | A kind of movement of traffic objective classification method based on video information and device | |
| CN108108760A (en) | A kind of fast human face recognition | |
| CN108416801B (en) | Har-SURF-RAN characteristic point matching method for stereoscopic vision three-dimensional reconstruction | |
| CN110543848B (en) | Driver action recognition method and device based on three-dimensional convolutional neural network | |
| Huang et al. | Soft-margin mixture of regressions | |
| CN106503748A (en) | A kind of based on S SIFT features and the vehicle targets of SVM training aids | |
| CN109784171A (en) | Car damage identification method for screening images, device, readable storage medium storing program for executing and server | |
| CN105825233A (en) | Pedestrian detection method based on random fern classifier of online learning | |
| CN110889360A (en) | Crowd counting method and system based on switching convolutional network | |
| CN107784263A (en) | Based on the method for improving the Plane Rotation Face datection for accelerating robust features | |
| CN109325434A (en) | A kind of image scene classification method of the probability topic model of multiple features |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170104 |
|
| RJ01 | Rejection of invention patent application after publication |