CN109345564A - A method of it solves not being inconsistent due to self-similarity characteristics generation optical flow field with sports ground - Google Patents
A method of it solves not being inconsistent due to self-similarity characteristics generation optical flow field with sports ground Download PDFInfo
- Publication number
- CN109345564A CN109345564A CN201810851968.3A CN201810851968A CN109345564A CN 109345564 A CN109345564 A CN 109345564A CN 201810851968 A CN201810851968 A CN 201810851968A CN 109345564 A CN109345564 A CN 109345564A
- Authority
- CN
- China
- Prior art keywords
- optical flow
- virtual displacement
- characteristic point
- image
- detection zone
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
It solves to generate the detection method that optical flow field and sports ground are not inconsistent due to self-similarity characteristics the invention discloses a kind of, improves the consistency that self-similarity characteristics generate optical flow field and sports ground, include the following steps: to obtain single channel video flowing, single pass gray level image;Image pyramid processing is carried out to image, does K layers of pyramid;The pyramid first layer of original image is divided into Number=Col × Row detection zone, and detects a characteristic point in each detection zone, calculates virtual displacement sequence V on the original image, and obtain virtual displacement image sequence Iv, include m virtual displacement image;Image pyramid processing is carried out to each virtual displacement image, obtains m K layers of virtual displacement pyramid, each V(i)A corresponding virtual displacement pyramid;Traditional optical flow tracking finally, which is carried out, using the filtered characteristic point of virtual displacement obtains optical flow field.
Description
Technical field
The present invention relates to the detection techniques of optical flow field, more particularly to a kind of solution optical flow field as caused by self-similarity characteristics
The method not being inconsistent with actual motion field.
Background technique
The basic research content that image optical flow field is computer movement vision is calculated, is the weight for carrying out motion analysis and understanding
Tool is wanted, is all played an important role in the applications such as motion detection, estimation, motion tracking, movement identification.Due to optical flow tracking
Be premised on illumination invariant and region consistency are assumed, when exist in image repeat texture when, texture from phase
It can frequently result in the tracking result of optical flow field like feature and the sports ground of actual object be not inconsistent, to influence optical flow tracking method
Accuracy.
Summary of the invention
In view of the deficiencies of the prior art, the present invention intends to provide a kind of solve due to self-similarity characteristics generation
The method that optical flow field and sports ground are not inconsistent is solved above-mentioned with improving the consistency that self-similarity characteristics generate optical flow field and sports ground
The problem of being proposed in background technique.
To achieve the above object, the invention provides the following technical scheme:
A method of it solving the optical flow field due to caused by self-similarity characteristics and not being inconsistent with sports ground, main includes following step
It is rapid:
Step 1: obtaining single channel video flowing, single pass gray level image.
Step 2: image pyramid processing being carried out to image, does K layers of pyramid.
Step 3: the pyramid first layer of original image being divided into Number=Col × Row detection zone, and every
A characteristic point is detected in a detection zone, at most detects Number characteristic point.Characteristic point is indicated with P.
Step 4: calculating virtual displacement sequence V using formula (3)~(6) on the original image, and obtain diastema with formula (1)
Move image sequence Iv, include m virtual displacement image.
Step 5: image pyramid processing is carried out to each virtual displacement image, obtains m K layers of virtual displacement pyramid,
Each V(i)A corresponding virtual displacement pyramid.
Step 6: enabling P=(x, y) is the characteristic point detected in step 3.Using but be not limited to LK optical flow tracking method, will
P point is traced into P '=(x on virtual displacement image pyramid by original image pyramidv,yv) point, light stream vectors be v '=(Δ x,
Δ y), wherein Δ x=xv- x, Δ y=yv-y。
Step 7: light stream vectors (Δ x, Δ y) and the virtual displacement vector V that step 6 is calculated(i)Substitute into confidence level function
The corresponding confidence value of scale i is calculated, m confidence value is obtained.
Step 8: m confidence value being multiplied to obtain final confidence value.The result of optical flow tracking is proved if it is 1
Correctly, the result mistake for otherwise proving optical flow tracking, should be this feature point deletion.
Step 9: characteristic point being updated according to the result of step 8, completes virtual displacement filtering.
Step 10: carrying out traditional optical flow tracking using the filtered characteristic point of virtual displacement and obtain optical flow field.
Further, the method for the screening of characteristic point described in step 3 is the optical flow tracking method based on virtual displacement.
Then the method for the characteristic point screening is pressed it is characterized in that detecting several characteristic points on the original image
Retain a portion characteristic point according to certain algorithm and abandons another part characteristic point.
The method of the present invention for detecting several characteristic points on the original image, it is characterized in that original image
Several small hough transform regions are divided into, each detection zone size is N × N number of pixel, is detected in each detection zone
It is no more than Z characteristic point out, the N, Z are integers, and N > 1, Z >=1.
Preferably, the detection zone size representative value takes N=15, i.e., each detection zone includes 225 pixels.
Preferably, the representative value of the quantity of the characteristic point takes Z=1.
Further, the total quantity of detection zone is jointly true by the size W × H and detection zone size N × N of original image
It is fixed, wherein
The number of X-direction detection zone is denoted as Col, and publicity indicates are as follows:
Col=W/N.
The number of Y direction detection zone is denoted as Row, and publicity indicates are as follows:
Row=H/N.
The total quantity of the detection zone of original image is
Number=Col × Row.
Further, the detection ordering of characteristic point be since be located at the original image upper left corner first detection zone,
Whole detection zones are successively traversed according to sequence from left to right, from top to bottom.
Further, the method that Z characteristic point is detected in each detection zone is, from positioned at the detection zone upper left corner
First pixel starts, and all pixels of detection zone are successively traversed according to sequence from left to right, from top to bottom, to each picture
Plain operation characteristic point detection algorithm, until the quantity of validity feature point has reached or more than Z or single choice, validity feature point
Quantity be less than in Z but detection zone without more pixels for traversal.
Further, the method that Z characteristic point is detected in each detection zone is characterized in, works as validity feature
The quantity of point has reached or is more than Z, and also residual pixel when not traversing in detection zone, and the residual pixel will be by
It skips, starts the processing of next detection zone.
Preferably, in the method for detecting Z characteristic point in each detection zone described, a kind of characteristic point detection calculation
Method is as follows:
Examining or check pixel, nearby diameter is the circular window region of 7 pixels, and edges of regions shares 16 pixels, with the center of circle
P is center pixel.In 16 pixels, meet pixel value all higher than in if there is at least n contiguous pixels
The pixel value of heart pixel p, or equally, there are at least n contiguous pixels and meet the picture that pixel value is entirely below center pixel p
Element value, then the central point p is exactly a satisfactory characteristic point.
In each method according to the present invention, the virtual displacement just refers to the two dimension of a description image translation amount
Vector is indicated, i.e., with symbol v
Further, a virtual displacement vector v is given, is commonly used in calculating to the one and v vector at integral multiple relation
Sequence, defining the sequence vector for convenience of description is V, includes altogether m element, meets following relationship with v:
Wherein, integer i value is 1,2,3 ..., m.
In each method according to the present invention, the virtual displacement image just refers to original image integral translation
The new images obtained after one vector.
Further, original image is integrally translated m times respectively, each translation vector is V(i), then a void can be obtained
It is displaced image sequence Iv, meet following relationship with original image I:
Virtual displacement optical flow tracking method of the present invention carries out the screening of characteristic point, it is characterized in that corresponding according to characteristic point
Virtual displacement vector sum light stream two parameter evaluations of motion vector described in characteristic point confidence level.
Preferably, a kind of method of confidence level that assessing characteristic point is as follows:
A constant parameter θ and binary function P is defined, its satisfaction is made
Wherein v, v ' are two motion vectors, | ... | representation vector modulo operation.
Virtual displacement optical flow tracking method of the present invention, it is characterized in that using optical flow approach to spy on virtual displacement image
Sign clicks through line trace.Specific step is:
1. selecting a detection zone S in I in original image and being subordinated to the characteristic point A of S;
2. determining virtual displacement vector v, virtual displacement image sequence I is obtained according to the vv;
3. i=1 .., m are enabled, it respectively will be in virtual displacement imageAs present image, the original image that former frame is acquired
As history image, characteristic point A is tracked using optical flow method, obtains virtual displacement light stream vectors vi', m diastema is obtained
Move light stream vectors.
4. by v and each vi' confidence level function is substituted into, m confidence value is calculated;
5. m confidence value is multiplied, total confidence value P is obtained;
6. if the characteristic point A is retained P=1, determines the credible result of optical flow tracking.On the contrary, if P=
0, then determine that the result of optical flow tracking is insincere, the characteristic point A is deleted.
Virtual displacement optical flow tracking method of the present invention, it is characterized in that the calculating of the virtual displacement vector v uses fastly
Fast Fourier transformation (FFT) method.
Specific step is:
1. couple original image I (x, y) carries out FFT transform,
Wherein u=0,1,2 ..., W-1, v=0,1,2 ..., H-1.
2. according to one group of constant threshold parameter δ0,δ1A set σ is obtained, i.e.,
σ=(u, v) | δ0<F(u,v)<δ1}
Wherein parameter δ0,δ1It is frequency u, the upper and lower limit of v respectively.
3. calculating a F ' value on the set σ, i.e.,
F '=∑(u,v)∈σF(u,v),
4. two parameter lambdas are calculated according to the F ' valuexAnd λy, i.e.,
5. according to parameter lambdaxAnd λyVirtual displacement vector is generated, i.e.,
6. generating virtual displacement sequence vector according to virtual displacement vector, i.e.,
Wherein, integer i indicates scale parameter, value 1,2,3 ..., m.
Virtual displacement optical flow tracking method of the present invention, it is characterized in that by using dynamic adjustment tracking window size with
And the method for virtual displacement scale parameter realizes best match between the two, so that the matching for promoting optical flow field and sports ground is general
Rate.
Preferably, a kind of method of the selection of optical flow tracking window size is as follows:
1. the size fixation of optical flow tracking window is chosen from following value: 3x3,5x5,7x7,9x9,11x11,13x13,
15x15
2. the preferable width of optical flow tracking window calculates as follows
WinSize_W=2 × vx+1
3. the preferred height of optical flow tracking window calculates as follows:
WinSize_H=2 × vy+1
4. biggish one of access value, is denoted as WS from WinSize_W and WinSize_H
5. from fixed set selection greater than WS all sizes in it is the smallest that.
Optical flow tracking process of the present invention based on virtual displacement, it is characterized in that carrying out light using image pyramid method
Stream tracking.Specific method is:
1. pair original image constructs K tomographic image pyramid.It is detected in the first layer (original image layer) of image pyramid special
Sign point A simultaneously records position coordinates (x of the A in the tomographic image1,y1).According to the aufbauprinciple of image pyramid, characteristic point A is in gold
Corresponding position is in word tower kth tomographic imageWherein k=1 ..., K.
2. virtual displacement vector is calculated with the Fast Fourier Transform (FFT) method when new frame original image arrives
V, using formulaSequence vector V is obtained, and generates virtual displacement image sequence Iv;
3. couple IvEach of scale virtual displacement imageK tomographic image pyramid is constructed, m diastema is obtained
Move image pyramid.
4. respectively willPyramid as present frame, the original image pyramid that is constructed when using former frame as historical frames,
Optical flow tracking is carried out to characteristic point A, obtains virtual displacement light stream vectors vi′;
5. respectively by v and vi' confidence level function calculating confidence value P (v, v ') is substituted into, m confidence value is obtained;
6. m confidence value is multiplied, confidence value of the product P as characteristic point A.
If P=1, the credible result of optical flow tracking is determined, the characteristic point A is retained.On the contrary, if P=0,
Then determine that the result of optical flow tracking is insincere, the characteristic point A is deleted.
In order to explain the structural features and functions of the invention more clearly, come with reference to the accompanying drawing with specific embodiment to this hair
It is bright to be described in detail.
Detailed description of the invention
Fig. 1 is the flow chart of traditional light stream.
Fig. 2 is the flow chart of virtual displacement light stream of the invention.
Fig. 3 is the flow chart of traditional detection.
Fig. 4 is characteristic point circular window of the invention.
Fig. 5 is the flow chart of virtual displacement detection in the present invention.
Fig. 6 is virtual displacement pyramid diagram in the present invention.
Fig. 7 is traditional pyramid diagram.
Fig. 8 is traditional optical flow tracking pyramid.
Fig. 9 is image coordinate in the present invention.
Specific embodiment
The following further describes the technical solution of the present invention in the following with reference to the drawings and specific embodiments.
Referring to Fig. 1-Fig. 7, a method of it solving the optical flow field due to caused by self-similarity characteristics and not being inconsistent with sports ground, mainly
Include the following steps:
Step 1: obtaining single channel video flowing, single pass gray level image.
Step 2: image pyramid processing being carried out to image, does K layers of pyramid.
Step 3: the pyramid first layer of original image being divided into Number=Col × Row detection zone, and every
A characteristic point is detected in a detection zone, at most detects Number characteristic point.Characteristic point is indicated with P.
Step 4: calculating virtual displacement sequence V using formula (3)~(6) on the original image, and obtain diastema with formula (1)
Move image sequence Iv, include m virtual displacement image.
Step 5: image pyramid processing is carried out to each virtual displacement image, obtains m K layers of virtual displacement pyramid,
Each V(i)A corresponding virtual displacement pyramid.
Step 6: enabling P=(x, y) is the characteristic point detected in step 3.Using but be not limited to LK optical flow tracking method, will
P point is traced into P '=(x on virtual displacement image pyramid by original image pyramidv,yv) point, light stream vectors be v '=(Δ x,
Δ y), wherein Δ x=xv- x, Δ y=yv-y。
Step 7: light stream vectors (Δ x, Δ y) and the virtual displacement vector V that step 6 is calculated(i)Substitute into confidence level function
The corresponding confidence value of scale i is calculated, m confidence value is obtained.
Step 8: m confidence value being multiplied to obtain final confidence value.The result of optical flow tracking is proved if it is 1
Correctly, the result mistake for otherwise proving optical flow tracking, should be this feature point deletion.
Step 9: characteristic point being updated according to the result of step 8, completes virtual displacement filtering.
Step 10: carrying out traditional optical flow tracking using the filtered characteristic point of virtual displacement and obtain optical flow field.
The present invention provides a kind of method of characteristic point screening, the method for the characteristic point screening is based on virtual displacement
Optical flow tracking method.
Then the method for the characteristic point screening is pressed it is characterized in that detecting several characteristic points on the original image
Retain a portion characteristic point according to certain algorithm and abandons another part characteristic point.
The method of the present invention for detecting several characteristic points on the original image, it is characterized in that original image
Several small hough transform regions are divided into, each detection zone size is N × N number of pixel, is detected in each detection zone
It is no more than Z characteristic point out, the N, Z are integers, and N > 1, Z >=1.
Preferably, the detection zone size representative value takes N=15, i.e., each detection zone includes 225 pixels.
Preferably, the representative value of the quantity of the characteristic point takes Z=1.
Further, the total quantity of detection zone is jointly true by the size W × H and detection zone size N × N of original image
It is fixed, wherein
The number of X-direction detection zone is denoted as Col, and publicity indicates are as follows:
Col=W/N.
The number of Y direction detection zone is denoted as Row, and publicity indicates are as follows:
Row=H/N.
The total quantity of the detection zone of original image is
Number=Col × Row.
Further, the detection ordering of characteristic point be since be located at the original image upper left corner first detection zone,
Whole detection zones are successively traversed according to sequence from left to right, from top to bottom.
Further, the method that Z characteristic point is detected in each detection zone is, from positioned at the detection zone upper left corner
First pixel starts, and all pixels of detection zone are successively traversed according to sequence from left to right, from top to bottom, to each picture
Plain operation characteristic point detection algorithm, until the quantity of validity feature point has reached or more than Z or single choice, validity feature point
Quantity be less than in Z but detection zone without more pixels for traversal.
Further, the method that Z characteristic point is detected in each detection zone is characterized in, works as validity feature
The quantity of point has reached or is more than Z, and also residual pixel when not traversing in detection zone, and the residual pixel will be by
It skips, starts the processing of next detection zone.
Preferably, in the method for detecting Z characteristic point in each detection zone described, a kind of characteristic point detection calculation
Method is as follows:
Examining or check pixel, nearby diameter is the circular window region of 7 pixels, and edges of regions shares 16 pixels, with the center of circle
P is center pixel.In 16 pixels, meet pixel value all higher than in if there is at least n contiguous pixels
The pixel value of heart pixel p, or equally, there are at least n contiguous pixels and meet the picture that pixel value is entirely below center pixel p
Element value, then the central point p is exactly a satisfactory characteristic point.
In each method according to the present invention, the virtual displacement just refers to the two dimension of a description image translation amount
Vector is indicated, i.e., with symbol v
Further, a virtual displacement vector v is given, is commonly used in calculating to the one and v vector at integral multiple relation
Sequence, defining the sequence vector for convenience of description is V, includes altogether m element, meets following relationship with v:
Wherein, integer i value is 1,2,3 ..., m.
In each method according to the present invention, the virtual displacement image just refers to original image integral translation
The new images obtained after one vector.
Further, original image is integrally translated m times respectively, each translation vector is V(i), then a void can be obtained
It is displaced image sequence Iv, meet following relationship with original image I:
Virtual displacement optical flow tracking method of the present invention carries out the screening of characteristic point, it is characterized in that corresponding according to characteristic point
Virtual displacement vector sum light stream two parameter evaluations of motion vector described in characteristic point confidence level.
Preferably, a kind of method of confidence level that assessing characteristic point is as follows:
A constant parameter θ and binary function P is defined, its satisfaction is made
Wherein v, v ' are two motion vectors, | ... | representation vector modulo operation.
Virtual displacement optical flow tracking method of the present invention, it is characterized in that using optical flow approach to spy on virtual displacement image
Sign clicks through line trace.Specific step is:
7. selecting a detection zone S in I in original image and being subordinated to the characteristic point A of S;
8. determining virtual displacement vector v, virtual displacement image sequence I is obtained according to the vv;
9. i=1 .., m are enabled, it respectively will be in virtual displacement imageAs present image, the original graph that former frame is acquired
As being used as history image, characteristic point A is tracked using optical flow method, obtains virtual displacement light stream vectors vi', m void is obtained
It is displaced light stream vectors.
10. by v and each vi' confidence level function is substituted into, m confidence value is calculated;
11. m confidence value is multiplied, total confidence value P is obtained;
12. if the characteristic point A is retained P=1, determines the credible result of optical flow tracking.On the contrary, if P=
0, then determine that the result of optical flow tracking is insincere, the characteristic point A is deleted.
Virtual displacement optical flow tracking method of the present invention, it is characterized in that the calculating of the virtual displacement vector v uses fastly
Fast Fourier transformation (FFT) method.
Specific step is:
7. couple original image I (x, y) carries out FFT transform,
Wherein u=0,1,2 ..., W-1, v=0,1,2 ..., H-1.
8. according to one group of constant threshold parameter δ0,δ1A set σ is obtained, i.e.,
σ=(u, v) | δ0<F(u,v)<δ1}
Wherein parameter δ0,δ1It is frequency u, the upper and lower limit of v respectively.
9. calculating a F ' value on the set σ, i.e.,
F '=∑(u,v)∈σF(u,v),
10. two parameter lambdas are calculated according to the F ' valuexAnd λy, i.e.,
11. according to parameter lambdaxAnd λyVirtual displacement vector is generated, i.e.,
12. generating virtual displacement sequence vector according to virtual displacement vector, i.e.,
Wherein, integer i indicates scale parameter, value 1,2,3 ..., m.
Virtual displacement optical flow tracking method of the present invention, it is characterized in that by using dynamic adjustment tracking window size with
And the method for virtual displacement scale parameter realizes best match between the two, so that the matching for promoting optical flow field and sports ground is general
Rate.
Preferably, a kind of method of the selection of optical flow tracking window size is as follows:
6. the size fixation of optical flow tracking window is chosen from following value: 3x3,5x5,7x7,9x9,11x11,13x13,
15x15
7. the preferable width of optical flow tracking window calculates as follows
WinSize_W=2 × vx+1
8. the preferred height of optical flow tracking window calculates as follows:
WinSize_H=2 × vy+1
9. biggish one of access value, is denoted as WS from WinSize_W and WinSize_H
10. from fixed set selection greater than WS all sizes in it is the smallest that.
Optical flow tracking process of the present invention based on virtual displacement, it is characterized in that carrying out light using image pyramid method
Stream tracking.Specific method is:
7. pair original image constructs K tomographic image pyramid.It is detected in the first layer (original image layer) of image pyramid special
Sign point A simultaneously records position coordinates (x of the A in the tomographic image1,y1).According to the aufbauprinciple of image pyramid, characteristic point A is in gold
Corresponding position is in word tower kth tomographic imageWherein k=1 ..., K.
8. virtual displacement vector is calculated with the Fast Fourier Transform (FFT) method when new frame original image arrives
V, using formulaSequence vector V is obtained, and generates virtual displacement image sequence Iv;
9. couple IvEach of scale virtual displacement imageK tomographic image pyramid is constructed, m diastema is obtained
Move image pyramid.
10. respectively willPyramid as present frame, the original image pyramid that is constructed when using former frame as historical frames,
Optical flow tracking is carried out to characteristic point A, obtains virtual displacement light stream vectors vi′;
11. respectively by v and vi' confidence level function calculating confidence value P (v, v ') is substituted into, m confidence value is obtained;
12. m confidence value is multiplied, confidence value of the product P as characteristic point A.
If P=1, the credible result of optical flow tracking is determined, the characteristic point A is retained.On the contrary, if P=0,
Then determine that the result of optical flow tracking is insincere, the characteristic point A is deleted.
The technical principle of the invention is described above in combination with a specific embodiment, is only the preferred embodiment of the present invention.This
The protection scope of invention is not limited merely to above-described embodiment, and all technical solutions belonged under thinking of the present invention belong to the present invention
Protection scope.Those skilled in the art, which does not need to pay for creative labor, can associate other specific realities of the invention
Mode is applied, these modes will fall within the scope of protection of the present invention.
Claims (9)
1. a kind of solve to generate the method that optical flow field and sports ground are not inconsistent due to self-similarity characteristics, which is characterized in that including as follows
Step:
Step 1: obtaining single channel video flowing, single pass gray level image;
Step 2: image pyramid processing being carried out to image, does K layers of pyramid;
Step 3: the pyramid first layer of original image being divided into Number=Col × Row detection zone, and in each inspection
It surveys and detects a characteristic point in region, at most detect Number characteristic point;Characteristic point is indicated with P;
Step 4: calculating virtual displacement sequence V on the original image, and obtain virtual displacement image sequence Iv, include m virtual displacement figure
Picture;
Step 5: image pyramid processing being carried out to each virtual displacement image, obtains m K layers of virtual displacement pyramid, each V(i)A corresponding virtual displacement pyramid;
Step 6: enabling P=(x, y) is the characteristic point detected in step 3;Using but be not limited to LK optical flow tracking method, by P point
P '=(x on virtual displacement image pyramid is traced by original image pyramidv, yv) point, light stream vectors are v '=(Δ x, Δ
Y), wherein Δ x=xv- x, Δ y=yv-y;
Step 7: light stream vectors (Δ x, Δ y) and the virtual displacement vector V that step 6 is calculated(i)Confidence level function is substituted into calculate
The corresponding confidence value of scale i, obtains m confidence value;
Step 8: m confidence value being multiplied to obtain final confidence value;The result of optical flow tracking is being proved just if it is 1
Really, the result mistake for otherwise proving optical flow tracking, should be this feature point deletion;
Step 9: characteristic point being updated according to the result of step 8, completes virtual displacement filtering;
Step 10: carrying out traditional optical flow tracking using the filtered characteristic point of virtual displacement and obtain optical flow field.
2. a kind of method for solving not being inconsistent due to self-similarity characteristics generation optical flow field and sports ground according to claim 1,
The method that characteristic point is screened in step 3 is the optical flow tracking method based on virtual displacement, the method for the characteristic point screening, special
Sign is to detect several characteristic points on the original image, then retains a portion characteristic point according to certain algorithm and loses
Abandon another part characteristic point;
Original image is divided into several small squares by the used method for detecting several characteristic points on the original image
Shape detection zone, each detection zone size are N × N number of pixel, detect to be no more than Z characteristic point in each detection zone,
The N, Z are integers, and N > 1, Z >=1;
The total quantity of detection zone is determined jointly by the size W × H and detection zone size N × N of original image, wherein X-axis side
It is denoted as Col to the number of detection zone, publicity indicates are as follows:
Col=W/N;
The number of Y direction detection zone is denoted as Row, and publicity indicates are as follows:
Row=H/N;
The total quantity of the detection zone of original image is
Number=Col × Row;
The detection ordering of characteristic point be since be located at the original image upper left corner first detection zone, according to from left to right,
Sequence from top to bottom successively traverses whole detection zones;
The method that Z characteristic point is detected in each detection zone is opened from first pixel for being located at the detection zone upper left corner
Begin, all pixels of detection zone is successively traversed according to sequence from left to right, from top to bottom, to each pixel operation characteristic point
Detection algorithm, until the quantity of validity feature point has reached or more than Z or single choice, the quantity of validity feature point be less than Z but
It is in detection zone without more pixels for traversal;
The method that Z characteristic point is detected in each detection zone is characterized in, when the quantity of validity feature point has reached or surpasses
When crossing Z, and not traversing in detection zone there are also residual pixel, the residual pixel will be skipped, and start next inspection
Survey the processing in region.
3. a kind of method for solving not being inconsistent due to self-similarity characteristics generation optical flow field and sports ground according to claim 2,
It is characterized in that, detection zone size representative value takes N=15, i.e., each detection zone includes 225 pixels;The quantity of characteristic point
Representative value take Z=1.
4. a kind of method for solving not being inconsistent due to self-similarity characteristics generation optical flow field and sports ground according to claim 2,
It is characterized in that, a kind of feature point detection algorithm is such as in the method for detecting Z characteristic point in each detection zone
Under:
The circular window region that diameter near pixel is 7 pixels is examined or check, edges of regions shares 16 pixels, is with center of circle p
Central pixel point;In 16 pixels, meet pixel value all higher than middle imago if there is at least n contiguous pixels
The pixel value of plain p, or equally, there are at least n contiguous pixels and meet the pixel that pixel value is entirely below center pixel p
Value, then the central point p is exactly a satisfactory characteristic point;
The virtual displacement just refers to the bivector of a description image translation amount, is indicated with symbol v, i.e.,
A virtual displacement vector v is given, is commonly used in calculating to the one and v sequence vector at integral multiple relation, for description side
Just defining the sequence vector is V, includes altogether m element, meets following relationship with v:
Wherein, integer i value is 1,2,3 ..., m;
The virtual displacement image just refers to the new images that will be obtained after one vector of original image integral translation.
5. a kind of method for solving not being inconsistent due to self-similarity characteristics generation optical flow field and sports ground according to claim 4,
The virtual displacement optical flow tracking method carries out the screening of characteristic point, it is characterized in that according to the corresponding virtual displacement vector sum of characteristic point
The confidence level of characteristic point described in two parameter evaluations of light stream motion vector;A kind of method of confidence level that assessing characteristic point is as follows:
A constant parameter θ and binary function P is defined, its satisfaction is made
Wherein v, v ' are two motion vectors, | ... | representation vector modulo operation.
6. a kind of method for solving not being inconsistent due to self-similarity characteristics generation optical flow field and sports ground according to claim 5,
The virtual displacement optical flow tracking method, tracks characteristic point using optical flow approach on virtual displacement image;Specific step
Suddenly it is:
A detection zone S is selected in I in original image and is subordinated to the characteristic point A of S;
It determines virtual displacement vector v, virtual displacement image sequence I is obtained according to the vv;
I=1 .., m are enabled, it respectively will be in virtual displacement imageAs present image, using the original image of former frame acquisition as going through
History image tracks characteristic point A using optical flow method, obtains virtual displacement light stream vectors v 'i, m virtual displacement light stream is obtained
Vector;
By v and each v 'iConfidence level function is substituted into, m confidence value is calculated;
M confidence value is multiplied, total confidence value P is obtained;
If P=1, the credible result of optical flow tracking is determined, the characteristic point A is retained;On the contrary, sentencing if P=0
The result for determining optical flow tracking is insincere, and the characteristic point A is deleted.
7. a kind of method for solving not being inconsistent due to self-similarity characteristics generation optical flow field and sports ground according to claim 6,
The virtual displacement optical flow tracking method, it is characterized in that the calculating of the virtual displacement vector v uses Fast Fourier Transform (FFT)
(FFT) method, specific step is:
FFT transform is carried out to original image I (x, y),
Wherein u=0,1,2 ..., W-1, v=0,1,2 ..., H-1;
According to one group of constant threshold parameter δ0, δ1A set σ is obtained, i.e.,
σ={ (u, v) I δ0< F (u, v) < δ1}
Wherein parameter δ0, δ1It is frequency u, the upper and lower limit of v respectively;,
A F ' value is calculated on the set σ, i.e.,
F '=∑(u, v) ∈ σF (u, v),
Two parameter lambdas are calculated according to the F ' valuexAnd λy, i.e.,
According to parameter lambdaxAnd λyVirtual displacement vector is generated, i.e.,
Virtual displacement sequence vector is generated according to virtual displacement vector, i.e.,
Wherein, integer i indicates scale parameter, value 1,2,3 ..., m.
8. a kind of method for solving not being inconsistent due to self-similarity characteristics generation optical flow field and sports ground according to claim 1,
It is characterized in that, the virtual displacement optical flow tracking method, it is characterized in that by using dynamic adjustment tracking window size and
The method of virtual displacement scale parameter realizes best match between the two, so that the matching probability of optical flow field and sports ground is promoted,
A kind of method of the selection of optical flow tracking window size is as follows:
The size fixation of optical flow tracking window is chosen from following value:
3x3,5x5,7x7,9x9,11x11,13x13,15x15;
The preferable width of optical flow tracking window calculates as follows:
WinSize_W=2 × vx+1
The preferred height of optical flow tracking window calculates as follows:
WinSize_H=2 × vy+1
Biggish one of access value, is denoted as WS from WinSize_W and WinSize_H;
From in fixed set selection greater than WS all sizes in it is the smallest that.
9. a kind of method for solving not being inconsistent due to self-similarity characteristics generation optical flow field and sports ground according to claim 1,
The optical flow tracking process based on virtual displacement, it is characterized in that carrying out optical flow tracking using image pyramid method;Specific side
Method is:
K tomographic image pyramid is constructed to original image;Characteristic point A is detected simultaneously in the first layer (original image layer) of image pyramid
Record position coordinates (x of the A in the tomographic image1, y1);According to the aufbauprinciple of image pyramid, characteristic point A is in pyramid kth
Corresponding position is in tomographic imageWherein k=1 ..., K;
When new frame original image arrives, virtual displacement vector v is calculated with the Fast Fourier Transform (FFT) method, applies
FormulaSequence vector V is obtained, and generates virtual displacement image sequence Iv;
To the IvEach of scale virtual displacement imageK tomographic image pyramid is constructed, m virtual displacement image is obtained
Pyramid;
Respectively willPyramid is as present frame, and the original image pyramid constructed when using former frame is as historical frames, to feature
Point A carries out optical flow tracking, obtains virtual displacement light stream vectors v 'i;
Respectively by v and v 'iIt substitutes into confidence level function and calculates confidence value P (v, v '), m confidence value is obtained;
M confidence value is multiplied, confidence value of the product P as characteristic point A;
If P=1, the credible result of optical flow tracking is determined, the characteristic point A is retained;On the contrary, sentencing if P=0
The result for determining optical flow tracking is insincere, and the characteristic point A is deleted.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810851968.3A CN109345564A (en) | 2018-07-30 | 2018-07-30 | A method of it solves not being inconsistent due to self-similarity characteristics generation optical flow field with sports ground |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810851968.3A CN109345564A (en) | 2018-07-30 | 2018-07-30 | A method of it solves not being inconsistent due to self-similarity characteristics generation optical flow field with sports ground |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109345564A true CN109345564A (en) | 2019-02-15 |
Family
ID=65291265
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810851968.3A Pending CN109345564A (en) | 2018-07-30 | 2018-07-30 | A method of it solves not being inconsistent due to self-similarity characteristics generation optical flow field with sports ground |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109345564A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110322477A (en) * | 2019-06-10 | 2019-10-11 | 广州视源电子科技股份有限公司 | Feature point observation window setting method, tracking method, device, equipment and medium |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103985137A (en) * | 2014-04-25 | 2014-08-13 | 北京大学深圳研究院 | Moving object tracking method and system applied to human-computer interaction |
US20140254882A1 (en) * | 2013-03-11 | 2014-09-11 | Adobe Systems Incorporated | Optical Flow with Nearest Neighbor Field Fusion |
US9305362B1 (en) * | 2014-02-28 | 2016-04-05 | Xilinx, Inc. | Image stabilization |
CN108280831A (en) * | 2018-02-02 | 2018-07-13 | 南昌航空大学 | A kind of acquisition methods and system of image sequence light stream |
-
2018
- 2018-07-30 CN CN201810851968.3A patent/CN109345564A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20140254882A1 (en) * | 2013-03-11 | 2014-09-11 | Adobe Systems Incorporated | Optical Flow with Nearest Neighbor Field Fusion |
US9305362B1 (en) * | 2014-02-28 | 2016-04-05 | Xilinx, Inc. | Image stabilization |
CN103985137A (en) * | 2014-04-25 | 2014-08-13 | 北京大学深圳研究院 | Moving object tracking method and system applied to human-computer interaction |
CN108280831A (en) * | 2018-02-02 | 2018-07-13 | 南昌航空大学 | A kind of acquisition methods and system of image sequence light stream |
Non-Patent Citations (1)
Title |
---|
李成美等: ""一种改进光流法的运动目标检测及跟踪算法"", 《仪器仪表学报》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110322477A (en) * | 2019-06-10 | 2019-10-11 | 广州视源电子科技股份有限公司 | Feature point observation window setting method, tracking method, device, equipment and medium |
CN110322477B (en) * | 2019-06-10 | 2022-01-04 | 广州视源电子科技股份有限公司 | Feature point observation window setting method, tracking method, device, equipment and medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110866924B (en) | Line structured light center line extraction method and storage medium | |
CN113592845A (en) | Defect detection method and device for battery coating and storage medium | |
CN109376631B (en) | Loop detection method and device based on neural network | |
CN104781848B (en) | Image monitoring apparatus for estimating gradient of singleton, and method therefor | |
Wang et al. | Evaluating edge detection through boundary detection | |
US9659390B2 (en) | Tomosynthesis reconstruction with rib suppression | |
US8396285B2 (en) | Estimating vanishing points in images | |
Sheng et al. | Geometric occlusion analysis in depth estimation using integral guided filter for light-field image | |
JP2010536412A (en) | Image forming method for sampling a cross section in a three-dimensional (3D) image data volume | |
JP2006524394A (en) | Delineation of human contours in images | |
CN113066064B (en) | Cone beam CT image biological structure identification and three-dimensional reconstruction system based on artificial intelligence | |
CN109791615A (en) | For detecting and tracking the method, target object tracking equipment and computer program product of target object | |
CN112085675B (en) | Depth image denoising method, foreground segmentation method and human motion monitoring method | |
CN111095075A (en) | Fully automatic, template-free particle pickup for electron microscopes | |
CN106407959A (en) | Low-illumination complicated background license plate positioning method based on wavelet transform and SVM | |
Ghebreab et al. | Combining strings and necklaces for interactive three-dimensional segmentation of spinal images using an integral deformable spine model | |
CN110245660B (en) | Webpage glance path prediction method based on saliency feature fusion | |
CN108710879A (en) | A kind of pedestrian candidate region generation method based on Grid Clustering Algorithm | |
Xing et al. | Self-supervised guided segmentation framework for unsupervised anomaly detection | |
CN109345564A (en) | A method of it solves not being inconsistent due to self-similarity characteristics generation optical flow field with sports ground | |
CN106600615A (en) | Image edge detection algorithm evaluation system and method | |
JP4836065B2 (en) | Edge tracking method and computer program therefor | |
CN117522862A (en) | Image processing method and processing system based on CT image pneumonia recognition | |
CN113409334A (en) | Centroid-based structured light angle point detection method | |
Barba et al. | The use of local entropy measures in edge detection for cytological image analysis |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190215 |