CN105513092B - A kind of template characteristic selection method for target following - Google Patents
A kind of template characteristic selection method for target following Download PDFInfo
- Publication number
- CN105513092B CN105513092B CN201510845774.9A CN201510845774A CN105513092B CN 105513092 B CN105513092 B CN 105513092B CN 201510845774 A CN201510845774 A CN 201510845774A CN 105513092 B CN105513092 B CN 105513092B
- Authority
- CN
- China
- Prior art keywords
- characteristic point
- light stream
- mrow
- template
- stream vector
- 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.)
- Expired - Fee Related
Links
Classifications
-
- 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
- G06T2207/20076—Probabilistic image processing
Landscapes
- Image Analysis (AREA)
Abstract
The present invention discloses a kind of template characteristic selection method for target following, belongs to Visual Tracking field.This method is used to solve the problem of in existing long-range object tracking process due to template tracking drift caused by non-targeted match point in template.Specially：Then the light stream vector of original match characteristic point pair first in calculation template carries out Polar coordinates to the light stream vector calculated, and carry out probability density statistics generation angular distribution histogram to it in polar coordinate space；Medium filtering finally is carried out to angle distribution histogram, target matching characteristics point is obtained, effectively rejects the non-targeted characteristic point in target template.
Description
Technical field
The present invention relates to the light stream field analysis of template characteristic point in a kind of Visual Tracking more particularly to target following,
Field analysis and probability density distribution analysis are moved, belongs to image procossing and area of pattern recognition.
Background technology
Template tracking is image procossing and the important component of area of pattern recognition, is widely used in intelligent monitoring,
In the vision tracking module of the various systems such as human-computer interaction and unmanned platform navigation.Its basic process can be described as：Do not having
In the case of having any priori, some object (i.e. template) is selected in the first two field picture with rectangular window, then rear
To the object into line trace in continuous video sequence.
At present, researcher has carried out substantial amounts of research to template track algorithm, wherein matched based on template characteristic
LK optical flow methods are one of most important technological means.Its techniqueflow is：Feature extraction is carried out to target template first, obtains mould
Characteristic point in plate；Then using LK optical flow methods in next two field picture to each characteristic point into line trace, obtain each characteristic point
Matching characteristic point is consequently formed matching characteristic point pair；Finally by matching characteristic point between two field pictures (matching characteristic point to)
Sports ground acquire geometric transform relation between two field pictures, and finally obtain the position of target and size in the two field picture,
Complete a tracking cycle.Constantly repeating above-mentioned tracking cycle can be in the realization target following in entire video sequence.The skill
The hypothesis of art means is：Although each characteristic point is tracked separately matching by LK optical flow methods in template, it is used as and is tracked mesh
Target is a part of, and all characteristic points follow Movement consistency in template.
However, the shortcomings that existing feature based matched Moulding board method, is：With the continuous progress of tracking cycle,
Template is easily drifted about, and ultimately results in target loss.Itself main reason is that, although most of region is traced in template
Target is covered, but background pixel is inevitably introduced into, and then pixel in background image and background and objective contour
Intersection point also tend to be extracted as characteristic point, and tracked by LK optical flow methods.Therefore, matched spy between the two field pictures obtained
There are non-targeted characteristic points pair for the centering of sign point.But these non-targeted characteristic points and a part for target is not belonging to, do not meet fortune
Uniformity is moved it is assumed that therefore being interfered to the solution of geometric transformation, the estimation of template size in next frame and position is caused
Deviation.With the continuous repetition of tracking cycle, the non-targeted characteristic point in template is constantly introduced into, and template deviation constantly accumulates,
And ultimately cause the drift of template.
The content of the invention
For solve the problems, such as in object tracking process due to non-targeted characteristic point is introduced into and caused by template drift, the present invention
It is proposed a kind of template characteristic selection method for target following, by the matching characteristic point between two field pictures to formation
Optical flow field carry out based on directional spreding probability density analysis, reject template in non-targeted characteristic point, so as to contain target with
Template drift during track.
This method concretely comprises the following steps：
Step 1: in calculation template original match characteristic point pair light stream vector
If there is N number of original match characteristic point pair in template, original match characteristic point is tracked according to LK optical flow methods, for n-th+
Characteristic point in 1 two field pictureIt is corresponded to in n-th frame image
Characteristic pointIt is each original in (1) calculation template according to the following equation
The light stream vector d of matching characteristic point pairi(n|n+1),1≤i≤N:
Wherein, n is image frame number；WithRespectively in n-th frame image InWith the (n+1)th two field picture In+1In it is corresponding
Ith feature point,WithRespectively characteristic pointAnd characteristic pointSeat in image coordinate system
Mark；
Step 2: the light stream vector using original match point pair in polar coordinate representation template
According to following formula (2)
Obtain the polar coordinate representation (ρ of the light stream vector of each original match characteristic point pairi(n|n+1),θi(n|n+1))；Wherein
ρi(n | n+1) it is light stream vector diThe length of (n | n+1), θi(n | n+1) it is light stream vector diThe deflection of (n | n+1)；
Step 3: carrying out probability density statistics in polar coordinate space interior focusing flow vector, light stream vector is obtained in each area
Between probability density, and thus generate angular distribution histogram；The unit width of the angular distribution histogram is Δ θ, highly
For Fb(j), wherein：After the polar coordinate space of light stream vector is is divided into J section by Δ θ, the width in each section；Fb(j) light
Flow vector is in the probability density in j-th of section, j=1,2 ..., J；
Step 4: carrying out medium filtering to angle distribution histogram, meet the light stream arrow in the section of medium filtering condition
The characteristic point of amount is the target feature point tracked, is retained；It is unsatisfactory for the light stream vector in the section of value filtering condition
Characteristic point is non-targeted characteristic point, is deleted from template.
Advantageous effect：
According to target feature point Movement consistency it is assumed that being carried out by the optical flow field to original match characteristic point pair in template
Directional spreding density analysis rejects original match characteristic point centering non-targeted point (i.e. background characteristics point and background and mesh in template
Mark boundary characteristic point), so that the problem of effectively inhibiting template drift in object tracking process, greatly improves target tracking algorism
Accuracy.
Description of the drawings
Fig. 1 is the flow chart of the template characteristic selection method for target following；
Fig. 2 is original match feature-point optical flow polar plot；
Fig. 3 is characterized the polar diagram of a light stream vector；
Fig. 4 is characterized the angular distribution histogram of a light stream vector；
Fig. 5 is target matching characteristics point and its light stream vector figure.
Specific embodiment
With reference to the accompanying drawings and examples, the present invention is described in further detail.
The present embodiment provides a kind of template characteristic selection method for target following, for solving in object tracking process
Due to non-targeted characteristic point is introduced into and caused by template drift the problem of.
The overall flow of this method is as shown in Figure 1：
Step 1: in calculation template original match characteristic point pair light stream vector, obtain optical flow field D (n | n+1)；
If in template the number of original match characteristic point pair be N, the original match characteristic point traced into according to LK optical flow methods,
To the (n+1)th two field picture In+1Interior characteristic pointAnd n-th frame
Image InIn corresponding characteristic point(1) calculates mould according to the following equation
The light stream vector d of each original match characteristic point pair in platei(n|n+1),1≤i≤N:
Wherein, n is image frame number；WithRespectively in n-th frame image InWith the (n+1)th two field picture In+1In it is corresponding
Ith feature point,WithRespectively characteristic pointAnd characteristic pointSeat in image coordinate system
Mark；N is total logarithm of original match characteristic point in template.
Finally obtain optical flow field D (n | n+1)={ d of original match point in templatei(n | n+1) | 1≤i≤N }, such as Fig. 2 institutes
Show.
Step 2: by the optical flow field D of original match point in template (n | n+1) Polar coordinates
According to the light stream vector d of each original match characteristic point pair in templatei(n | n+1) solve below equation：
Obtain the polar coordinate representation (ρ of the light stream vector of each original match characteristic point pairi(n|n+1),θi(n | n+1)), such as scheme
Shown in 3.Wherein ρi(n | n+1) it is light stream vector di(n | n+1) length, θi(n | n+1) it is light stream vector diThe direction of (n | n+1)
Angle.
Step 3: probability density statistics is carried out to it in the polar coordinate space of light stream vector and generates angular distribution Nogata
Figure, concretely comprises the following steps：
(301) angular range of 0~360 ° of polar coordinates is divided into J section, wherein each interval width is Δ θ；
(302) statistics falls into the number S of the light stream vector in j-th of sectionj, j=1,2 ..., J；Then following formula are pressed
(2) probability density F of the light stream vector in j-th of section is obtainedb(j)：
(303) the probability density F in j-th of the section calculated according to step (302)b(j), in 0~360 ° of angle model
Angular distribution histogram is established in the distribution for enclosing interior focusing flow vector, as shown in Figure 4；Wherein histogram constituent parts width is Δ θ, high
It spends for Fb(j)。
Step 4: carrying out medium filtering to angle distribution histogram, target matching characteristics point is obtained, is concretely comprised the following steps：
(401) duplicate removal processing is carried out to angle distribution histogram, obtains single angular histogram amplitude unique (Fb
(j))；
(402) single angular histogram amplitude is ranked up by size, finds in-between value median (unique
(Fb(j)))；
(403) with median median (unique (Fb(j)) it is) thresholding, if the histogram amplitude in j-th of section
unique(Fb(j)) meet：
unique(Fb(j))≥median(unique(Fb(j))), j=1,2 ..., J
The characteristic point for the light stream vector being then considered in j-th of section is the target feature point tracked, is retained；
Rather it is thought that the characteristic point of the light stream vector in the section that condition is not satisfied is non-targeted characteristic point, deleted；Finally protect
The characteristic point stayed is the target feature point in template, as shown in Figure 5.
In conclusion the foregoing is merely a prefered embodiment of the invention, it is not intended to limit the scope of the present invention.
Within the spirit and principles of the invention, any modifications, equivalent replacements and improvements are made should be included in the present invention's
Within protection domain.
Claims (2)
1. a kind of template characteristic selection method for target following, which is characterized in that concretely comprise the following steps：
Step 1: in calculation template original match characteristic point pair light stream vector
If there is N number of original match characteristic point pair in template, original match characteristic point is tracked according to LK optical flow methods, for the (n+1)th frame
Characteristic point in imageIt is corresponding in n-th frame image
Characteristic pointIt is original each in (1) calculation template according to the following equation
Light stream vector d with characteristic point pairi(n | n+1), 1≤i≤N:
<mrow>
<msub>
<mi>d</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>n</mi>
<mo>|</mo>
<mi>n</mi>
<mo>+</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mrow>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mrow>
<mi>n</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
</msubsup>
<mo>-</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mi>n</mi>
</msubsup>
<mo>,</mo>
<msubsup>
<mi>y</mi>
<mi>i</mi>
<mrow>
<mi>n</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
</msubsup>
<mo>-</mo>
<msubsup>
<mi>y</mi>
<mi>i</mi>
<mi>n</mi>
</msubsup>
<mo>)</mo>
</mrow>
<mo>,</mo>
<mn>1</mn>
<mo>&le;</mo>
<mi>i</mi>
<mo>&le;</mo>
<mi>N</mi>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, n is image frame number；WithRespectively in n-th frame image InWith the (n+1)th two field picture In+1In corresponding i-th
Characteristic point,WithRespectively characteristic pointAnd characteristic pointCoordinate in image coordinate system；
Step 2: the light stream vector using original match characteristic point pair in polar coordinate representation template
According to following formula (2)
Obtain the polar coordinate representation (ρ of the light stream vector of each original match characteristic point pairi(n | n+1), θi(n|n+1))；Wherein ρi(n
| n+1) it is light stream vector diThe length of (n | n+1), θi(n | n+1) it is light stream vector diThe deflection of (n | n+1)；
Step 3: carrying out probability density statistics in polar coordinate space interior focusing flow vector, light stream vector is obtained in each section
Probability density, and thus generate angular distribution histogram；The unit width of the angular distribution histogram is Δ θ, is highly Fb
(j), wherein：After the polar coordinate space of light stream vector is is divided into J section by Δ θ, the width in each section；Fb(j) it is light
Flow vector is in the probability density in j-th of section, j=1,2 ..., J；
Step 4: carrying out medium filtering to angle distribution histogram, meet light stream vector in the section of medium filtering condition
Characteristic point is the target feature point tracked, is retained；It is unsatisfactory for the feature of the light stream vector in the section of value filtering condition
Point is non-targeted characteristic point, is deleted from template；
Medium filtering is carried out in the step 4 to angle distribution histogram to concretely comprise the following steps：
(401) duplicate removal processing is carried out to angle distribution histogram, obtains single angular distribution histogram amplitude unique (Fb
(j))；
(402) single angular distribution histogram amplitude is ranked up by size, obtains value median (unique (F among itb
(j)))；
(403) with median median (unique (Fb(j)) it is) thresholding, if the angular distribution histogram amplitude in j-th of section
unique(Fb(j)) meet：
unique(Fb(j))≥median(unique(Fb(j))), j=1,2 ..., J
Then think that it meets medium filtering condition, otherwise it is assumed that it is unsatisfactory for medium filtering condition.
2. the template characteristic selection method of target following to be used for as described in claim 1, which is characterized in that the step three lives
Angled distribution histogram concretely comprises the following steps：
(301) angular range of 0~360 ° of polar coordinates is divided into J section, wherein each interval width is Δ θ；
(302) statistics falls into the number S of the light stream vector in j-th of sectionj, light stream vector is obtained at j-th in j=1,2 ..., J
The probability density F in sectionb(j)：
<mrow>
<msub>
<mi>F</mi>
<mi>b</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>j</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<msub>
<mi>S</mi>
<mi>j</mi>
</msub>
<mi>N</mi>
</mfrac>
<mo>,</mo>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
<mo>,</mo>
<mn>2</mn>
<mo>,</mo>
<mn>...</mn>
<mo>,</mo>
<mi>J</mi>
</mrow>
(303) according to probability density F of the light stream vector that step (302) is calculated in j-th of sectionb(j), at 0~360 ° of angle
Angular distribution histogram is established to the distribution of light stream vector in the range of degree.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510845774.9A CN105513092B (en) | 2015-11-26 | 2015-11-26 | A kind of template characteristic selection method for target following |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510845774.9A CN105513092B (en) | 2015-11-26 | 2015-11-26 | A kind of template characteristic selection method for target following |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105513092A CN105513092A (en) | 2016-04-20 |
| CN105513092B true CN105513092B (en) | 2018-05-22 |
Family
ID=55721049
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510845774.9A Expired - Fee Related CN105513092B (en) | 2015-11-26 | 2015-11-26 | A kind of template characteristic selection method for target following |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105513092B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109685830B (en) * | 2018-12-20 | 2021-06-15 | 浙江大华技术股份有限公司 | Target tracking method, device and equipment and computer storage medium |
| CN109993777B (en) * | 2019-04-04 | 2021-06-29 | 杭州电子科技大学 | Target tracking method and system based on dual-template adaptive threshold |
| CN113012216B (en) * | 2019-12-20 | 2023-07-07 | 舜宇光学（浙江）研究院有限公司 | Feature classification optimization method, SLAM positioning method, system and electronic equipment |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101923717A (en) * | 2009-06-10 | 2010-12-22 | 新奥特（北京）视频技术有限公司 | Method for accurately tracking characteristic points of quick movement target |
-
2015
- 2015-11-26 CN CN201510845774.9A patent/CN105513092B/en not_active Expired - Fee Related
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101923717A (en) * | 2009-06-10 | 2010-12-22 | 新奥特（北京）视频技术有限公司 | Method for accurately tracking characteristic points of quick movement target |
Non-Patent Citations (4)
| Title |
|---|
| A robust template tracking algorithm with weighted active drift correction;Baojie Fan et al.;《Pattern Recognition Letters》;20111231;第1317-1327页 * |
| Robust template tracking with drift correction;David Schreiber;《Pattern Recognition Letters》;20071231;第1484-1491页 * |
| 基于LK光流跟踪法的有效目标点增强追踪;蒋菱 等;《微型机与应用》;20150331;第34卷(第6期);第45-48页 * |
| 基于改进LK光流的目标跟踪算法研究;张忠义;《信息技术》;20151031(第10期);第127-130页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105513092A (en) | 2016-04-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ding et al. | Learning depth-guided convolutions for monocular 3d object detection | |
| Teoh et al. | Symmetry-based monocular vehicle detection system | |
| CN109102522B (en) | Target tracking method and device | |
| Chougule et al. | Reliable multilane detection and classification by utilizing cnn as a regression network | |
| CN110738690A (en) | unmanned aerial vehicle video middle vehicle speed correction method based on multi-target tracking framework | |
| CN101840507B (en) | Target tracking method based on character feature invariant and graph theory clustering | |
| CN108229350A (en) | A kind of face detection of robot device, identification and tracking and system | |
| CN104517275A (en) | Object detection method and system | |
| CN104036524A (en) | Fast target tracking method with improved SIFT algorithm | |
| CN104463851A (en) | Automatic shoe sole edge line tracking method based on robot | |
| CN110060275A (en) | A kind of method and system of microcirculation in human body blood flow velocity detection | |
| CN115049700A (en) | Target detection method and device | |
| CN104036523A (en) | Improved mean shift target tracking method based on surf features | |
| CN105513092B (en) | A kind of template characteristic selection method for target following | |
| CN108009494A (en) | A kind of intersection wireless vehicle tracking based on unmanned plane | |
| CN106534833B (en) | A kind of double vision point three-dimensional video-frequency antihunt means of joint space time shaft | |
| CN111881790A (en) | Automatic extraction method and device for road crosswalk in high-precision map making | |
| CN110705342A (en) | Lane line segmentation detection method and device | |
| Li et al. | Road markings extraction based on threshold segmentation | |
| CN106920253A (en) | It is a kind of based on the multi-object tracking method for blocking layering | |
| Ozgunalp | Robust lane‐detection algorithm based on improved symmetrical local threshold for feature extraction and inverse perspective mapping | |
| CN113516853A (en) | Multi-lane traffic flow detection method for complex monitoring scene | |
| CN115760893A (en) | Single droplet particle size and speed measuring method based on nuclear correlation filtering algorithm | |
| CN105761507A (en) | Vehicle counting method based on three-dimensional trajectory clustering | |
| US20160063716A1 (en) | Line parametric object estimation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20180522 Termination date: 20181126 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |