CN101324958A - Method and apparatus for tracking object - Google Patents

Method and apparatus for tracking object Download PDF

Info

Publication number
CN101324958A
CN101324958A CNA2008101155377A CN200810115537A CN101324958A CN 101324958 A CN101324958 A CN 101324958A CN A2008101155377 A CNA2008101155377 A CN A2008101155377A CN 200810115537 A CN200810115537 A CN 200810115537A CN 101324958 A CN101324958 A CN 101324958A
Authority
CN
China
Prior art keywords
target area
mrow
parameter
msub
candidate
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
Application number
CNA2008101155377A
Other languages
Chinese (zh)
Inventor
卢晓鹏
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Vimicro Corp
Original Assignee
Vimicro Corp
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Vimicro Corp filed Critical Vimicro Corp
Priority to CNA2008101155377A priority Critical patent/CN101324958A/en
Publication of CN101324958A publication Critical patent/CN101324958A/en
Pending legal-status Critical Current

Links

Images

Landscapes

  • Image Analysis (AREA)

Abstract

The invention provides a target tracking method and a target tracking device. The method comprises the following steps: associating a mixture Gaussian model with the similarity of a target based on image features; adopting the image features in target area to describe weighted factors in the mixture Gaussian model; estimating target location parameters and covariance matrix parameters of the shape and the size of the target after the EM iteration processing; and changing target area parameters self-adaptively according to movement states of the target to be tracked, so as to achieve the effect of accurately tracking target on a real-time basis.

Description

Target tracking method and device
Technical Field
The present invention relates to image processing technologies, and in particular, to a target tracking method and apparatus.
Background
Target tracking has wide application in human-computer interaction, automatic surveillance, video retrieval, traffic detection, and vehicle navigation. The task of object tracking is to determine the geometric state of the object in the video stream, including position, shape size, etc. Since the tracked target usually has irregular motion and is interfered by a complex background, the target tracking algorithm faces many challenges, and is one of the research hotspots in the technical field of image processing.
Currently, commonly used target tracking methods generally adopt a detection or manual calibration method to determine an initial position of a tracking target area in a first frame of acquired image, search an area closest to a tracking object in a subsequently acquired image, and use an area matched with the tracking object as a tracking target area.
In the target tracking process, a tracking queue is required to be maintained, the tracking queue includes reference frames, each reference frame represents a target area, and each target area includes information such as position and size. For example, the position of the rectangular reference frame may be the center coordinates of the rectangle, and the size may be the length, width, and height of the rectangle.
When the tracking starts, a tracking target region is usually detected from a first frame acquired image, after each iteration processing, the target elliptical region is increased or decreased by corresponding times along with the approaching or departing of the target region, and the region with the largest similarity measure is taken as the target region of the current frame. However, due to the irregularity of the movement of the tracked target, the movement is unpredictable, and the target elliptical area is increased or decreased by a preset multiple, so that some phenomena of wrong tracking and target loss are inevitably caused.
Yet another tracking method is: and detecting a tracking target area from the first frame acquired image, and continuously performing template matching in a large-range elliptical area when searching the position and the size of the tracking target in the current frame, and taking the area with the maximum similarity measure as the target area of the current frame. Although the method has higher accuracy than the former method, the method greatly increases the amount of calculation due to the large-scale search, and cannot meet the requirement of real-time tracking.
Disclosure of Invention
In view of this, the present invention provides a target tracking method, which can adaptively change a target area parameter according to a motion condition of a tracked target.
The invention also provides a target tracking device which can change the target area parameters in a self-adaptive manner according to the motion condition of the tracked target.
The following is a technical scheme provided by the embodiment of the invention:
a target tracking method, comprising the steps of:
1) detecting a tracking target in a reference image, acquiring an initial value of a target area parameter, and calculating the image characteristics of a target area in the reference image;
2) determining a candidate target area in the current frame by taking the initial value of the target area parameter as a candidate target area parameter, and calculating the image characteristic of the candidate target area;
3) calculating a similarity coefficient between the candidate target area and the target area in the reference image according to the image characteristics of the target area in the reference image and the image characteristics of the candidate target area;
4) calculating new target area parameters according to the similarity coefficient between the target area in the reference image and the target area in the current frame;
5) judging whether the obtained new target area parameter is the true value of the target area parameter in the current frame, if so, taking the new target area parameter as the true value of the target area parameter in the current frame, if not, returning to the step 2) to replace the candidate target area parameter with the new target area parameter, and repeating the steps 2) to 5) until the obtained new target area parameter is the true value of the target area parameter in the current frame.
The initial values of the parameters of the target area comprise an initial value of a parameter of the center position of the target area and an initial value of a parameter of the shape and the size of the target area;
the candidate target area parameters comprise a candidate target area center position parameter and a candidate target area shape size parameter.
The image feature of the target region is the discretization color probability distribution of the target region, and the image feature of the candidate target region is the discretization color probability distribution of the candidate target region.
The discretized color probability distribution is:
<math> <mrow> <msub> <mover> <mi>q</mi> <mo>^</mo> </mover> <mi>u</mi> </msub> <mo>=</mo> <mi>C</mi> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>n</mi> </msubsup> <mi>N</mi> <mrow> <mo>(</mo> <msubsup> <mi>x</mi> <mi>i</mi> <mo>*</mo> </msubsup> <mo>;</mo> <msub> <mi>&mu;</mi> <mi>i</mi> </msub> <mo>;</mo> <msub> <mi>&sigma;</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mi>&delta;</mi> <mrow> <mo>[</mo> <mi>b</mi> <mrow> <mo>(</mo> <msubsup> <mi>x</mi> <mi>i</mi> <mo>*</mo> </msubsup> <mo>)</mo> </mrow> <mo>-</mo> <mi>u</mi> <mo>]</mo> </mrow> <mo>,</mo> </mrow> </math>
wherein muiIs a target region center position parameter, σiA discretization grade is represented by u as a target area shape size parameter; n (-) is a Gaussian kernel function defining a function b (x)i *):R2→ 1.. multidot.m represents pixel point xi *δ is a function containing two variables, which is 1 when the variables have the same value; when the variables have different values, the function is 0, C is a normalization constant, and <math> <mrow> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>u</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>m</mi> </msubsup> <msub> <mover> <mi>q</mi> <mo>^</mo> </mover> <mi>u</mi> </msub> <mo>=</mo> <mn>1</mn> <mo>.</mo> </mrow> </math>
the similarity coefficient between the candidate target region and the target region in the reference image is as follows:
Figure A20081011553700083
wherein,
Figure A20081011553700084
for a discretized color probability distribution of the candidate target region,
Figure A20081011553700085
for discretized color probability distribution of the reference target area, θiFor the candidate target region parameters, m, u represent the discretization level, and u is 1.
Judging whether the obtained new target area parameter is the true value of the target area parameter in the current frame or not, wherein the true value is as follows:
judging whether the absolute value of the difference between the obtained new target area parameter and the candidate target area parameter is smaller than a preset threshold value or not, and if so, considering that the obtained new target area parameter is the true value of the target area parameter in the current frame; if so, the obtained new target area parameter is not considered to be the true value of the target area parameter in the current frame.
The present invention also provides a target tracking apparatus, comprising: the device comprises a detection module and a tracking module.
The detection module is used for detecting a tracking target from a reference image, acquiring a target area parameter initial value, calculating the image characteristics of the target area in the reference image, and sending the target area parameter initial value and the image characteristics to the tracking module;
the tracking module is used for receiving the initial value of the target area parameter from the detection module, taking the initial value of the target area parameter as a candidate target area parameter and calculating the image characteristic of the candidate target area;
the tracking module is further configured to calculate a similarity coefficient between the candidate target region and the candidate target region in the reference image according to the image feature of the target region in the reference image from the detection module and the image feature of the candidate target region, and calculate a true value of a current frame target region parameter according to the similarity coefficient and the candidate target region parameter.
The tracking module includes: a candidate target area updating unit, a feature extracting unit, a calculating unit, a judging unit, wherein
The candidate target area updating unit is used for receiving an image from the outside and a target area parameter initial value from the detection module and determining a candidate target area in the current frame according to the target area parameter initial value,
the candidate target area updating unit is also used for receiving the result obtained by the calculating unit as a candidate target area parameter and sending the candidate target area parameter to the feature extracting unit, the calculating unit and the judging unit;
the feature extraction unit is used for calculating the image features of the candidate target regions according to the received candidate target region parameters;
the calculating unit is used for calculating new target area parameters according to the image features from the feature extracting unit and the target area parameters from the candidate target area updating unit and sending the calculation results to the judging unit and the candidate target area updating unit;
and the judging unit is used for judging whether the new target area parameter from the calculating unit is the real value of the current frame target area parameter.
The calculation unit includes: a similarity coefficient calculating subunit and a target area parameter calculating subunit;
the similarity coefficient calculating subunit is used for receiving the color probability distribution from the feature extraction unit, calculating the similarity coefficient between the candidate target region and the target region in the reference image, and sending the similarity coefficient to the target region parameter calculating subunit;
and the target area parameter calculating subunit is used for calculating a new target area parameter according to the similarity coefficient from the similarity coefficient calculating subunit and the candidate area parameter from the candidate target area updating unit, and sending the new target area parameter to the candidate target area updating unit and the judging unit.
The judging unit includes: a difference value calculating subunit, a threshold value storing subunit and a comparing subunit;
the difference value calculating subunit is configured to calculate an absolute value of a difference between the candidate target region parameter from the candidate target region updating unit and the calculation result from the target region parameter calculating subunit, and output the absolute value to the comparing unit;
the threshold storage subunit is configured to store a threshold used for determining whether the new target area parameter is a true value of the current frame target area parameter;
and the comparison subunit is used for comparing the calculation result from the difference value operator unit with the threshold value in the threshold value storage subunit and outputting the comparison result to the calculation unit.
According to the technical scheme, the target tracking method and the target tracking device provided by the invention have the advantages that the Gaussian mixture model is connected with the target similarity based on the image characteristics, the image characteristics of the target area are adopted to describe the weighting factors in the Gaussian mixture model, and the covariance matrix parameters of the target position parameters and the target shape size are estimated through EM (effective electromagnetic field) iterative processing.
Drawings
FIG. 1 is a schematic diagram of a target tracking method according to the present invention;
FIG. 2 is a schematic diagram of an embodiment of a target tracking method provided in the present invention;
FIG. 3 is a schematic structural diagram of a target tracking apparatus according to the present invention;
FIG. 4 is a schematic diagram of a structural embodiment of a target tracking device provided in the present invention;
FIG. 5 is a schematic diagram of an embodiment of a specific structure of a computing unit of a target tracking apparatus according to the present invention;
fig. 6 is a schematic diagram of an embodiment of a specific structure of a determination unit of a target tracking apparatus according to the present invention.
Detailed Description
The invention provides a target tracking method and a target tracking device, which relate a Gaussian mixture model and target similarity based on image characteristics, describe weighting factors in the Gaussian mixture model by adopting the image characteristics of a target area, estimate the covariance matrix parameters of target position parameters and target shape size through EM (effective magnetic field) iterative processing, and realize the self-adaptive change of the target area parameters according to the motion condition of a tracked target so as to achieve the effect of accurately tracking the target in real time.
The EM algorithm is first introduced here. The basic idea of the EM algorithm is to find the parameter value that maximizes the similarity measure under given observation conditions. The algorithm takes an iterative form. The algorithm starts with a guess of a parameter, alternately implementing two steps: e step and M step. The essence of the step E is to estimate the expected value of the position distribution in the whole observation range through a consistent probability model, observation conditions and the current parameter guess value; and M, taking the expected value (usually a function) obtained in the step E as a likelihood function, and performing maximum likelihood estimation on the parameters.
The focus problem of video tracking is mostly focused on determining parameters such as the center position of a target area and the shape and size of a target in continuous image frames. Research finds that the parameter estimation of incomplete data is very suitable to be realized by adopting an EM algorithm, and the entire video image can not be considered in the tracking process, but only the content contained in the target area to be tracked can be focused.
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to the accompanying drawings and examples.
FIG. 1 is a schematic flow chart diagram of a target tracking method provided by the present invention, which includes the following steps:
step 101, detecting a tracking target in a reference image, acquiring an initial value of a target area parameter, and calculating image characteristics of the target area in the reference image.
In the step, the target detection method in the prior art is adopted for detecting the tracking target in the reference image, such as an inter-frame difference method, a background frame difference method, a time difference method and the like.
The initial values of the parameters of the target area comprise an initial value of a parameter of the center position of the target area and an initial value of a parameter of the shape and the size of the target area. The image feature may be a discretized color distribution probability of the target region, a gradient feature, or the like.
And step 102, determining a candidate target area in the current frame according to the initial value of the target area parameter, obtaining the parameter of the candidate target area, and calculating the image characteristic of the candidate target area.
The candidate target area parameters comprise a candidate target area center position parameter and a candidate target area shape size parameter. The calculated candidate target region image features are of the same type as the image features of the target region in the reference image obtained in step 101.
In this step, determining a candidate target region in the current frame according to the initial value of the target region parameter specifically includes: and enabling the parameter of the center position of the candidate target area to be equal to the initial value of the parameter of the center position of the target area, and enabling the parameter of the shape size of the candidate target area to be equal to the initial value of the parameter of the shape size of the target area.
And 103, calculating a similarity coefficient between the target area in the reference image and the candidate target area in the current frame according to the image characteristics of the target area in the reference image and the image characteristics of the candidate target area.
The similarity coefficient between the target region in the reference image and the candidate target region in the current frame is obtained by taking a square root of the product of the color probability distribution of each discretization level in the target region in the reference image and the color probability distribution of each corresponding discretization level in the candidate target region, and summing the square roots.
And 104, performing EM iteration processing on the candidate target area parameters according to the similarity coefficient of the target area in the reference image and the target area in the current frame, and gradually approaching the real value of the target area parameters of the current frame.
The method comprises the following specific steps: calculating new target area parameters according to the similarity coefficient between the candidate target area and the target area in the reference image;
judging whether the absolute value of the difference between the obtained new target area parameter and the candidate target area parameter is smaller than a preset threshold value or not, if so, determining the obtained target area parameter to be the true value of the target area parameter in the current frame; otherwise, the obtained new target area parameter is used as a candidate target area parameter for the next EM processing.
The following describes the target tracking method according to the present invention in detail with reference to specific embodiments.
Before describing the embodiments of the present invention, the modeling process and principles of the embodiments of the present invention will be described.
The embodiment provided by the invention adopts a mixed Gaussian model, and approximates the image characteristic function p (x | theta) of the image point x in the target area by the weighted sum of a plurality of Gaussian distribution functions:
<math> <mrow> <mi>p</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>|</mo> <mi>&theta;</mi> <mo>)</mo> </mrow> <mo>=</mo> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>M</mi> </msubsup> <msub> <mi>&alpha;</mi> <mi>i</mi> </msub> <msub> <mi>p</mi> <mi>i</mi> </msub> <mrow> <mo>(</mo> <mi>x</mi> <mo>|</mo> <msub> <mi>&theta;</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>1</mn> <mo>)</mo> </mrow> </mrow> </math>
wherein p isiIs a parameter thetaiIs given as a gaussian distribution of (a), θ ═ μ, σ2In which μ is the mean, σ2Is the variance. There are M possibilities for the parameter theta within the target region, alphaiIn the tracking problem, the target area parameter of the current frame can be understood as the sum of the original parameter of the target in the previous frame and the displacement of each image point and the change parameter of the area shape size. Here, theSetting the initial position X of each image point in the target area as { X ═ X1,...,xNDefining the target shape change and displacement change parameters as a subset Y ═ Y } as determined data, N is the number of image points in the target area1,...,yMY is not observable unknown data and has Y for image point i in the target regioniE.g. { 1., M }. I.e. if the change in the parameter of the image point i is generated by the k-th gaussian term in equation (1), then yiK. Suppose given yiIs a random vector, then the parameter likelihood estimate for the entire target region is:
<math> <mrow> <mi>ln</mi> <mrow> <mo>(</mo> <mi>f</mi> <mrow> <mo>(</mo> <mi>X</mi> <mo>,</mo> <mi>Y</mi> <mo>|</mo> <mi>&theta;</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mo>=</mo> <mi>ln</mi> <mrow> <mo>(</mo> <munderover> <mi>&Pi;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </munderover> <mi>p</mi> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>,</mo> <msub> <mi>y</mi> <mi>i</mi> </msub> <mo>|</mo> <mi>&theta;</mi> <mo>)</mo> </mrow> <mo>)</mo> </mrow> </mrow> </math>
<math> <mrow> <mo>=</mo> <munderover> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </munderover> <mi>ln</mi> <mrow> <mo>(</mo> <munderover> <mi>&Sigma;</mi> <mrow> <mi>j</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>M</mi> </munderover> <msub> <mi>&alpha;</mi> <mi>j</mi> </msub> <msub> <mi>p</mi> <mi>j</mi> </msub> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>,</mo> <msub> <mi>y</mi> <mi>i</mi> </msub> <mo>|</mo> <msub> <mi>&theta;</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mo>=</mo> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </msubsup> <mi>ln</mi> <mrow> <mo>(</mo> <msub> <mi>&alpha;</mi> <mi>yi</mi> </msub> <msub> <mi>p</mi> <mi>yi</mi> </msub> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>|</mo> <msub> <mi>&theta;</mi> <mi>yi</mi> </msub> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>2</mn> <mo>)</mo> </mrow> </mrow> </math>
derivation in connection with EM step:
e, step E: given an observation X and a known
Figure A20081011553700133
Under the condition, an expected value of log-likelihood ln (f (X, Y | theta)) is calculated,
<math> <mrow> <mi>Q</mi> <mrow> <mo>(</mo> <mi>&theta;</mi> <mo>,</mo> <msup> <mi>&theta;</mi> <mi>g</mi> </msup> <mo>)</mo> </mrow> <mo>=</mo> <munder> <mi>&Sigma;</mi> <mrow> <mi>y</mi> <mo>&Element;</mo> <mrow> <mo>(</mo> <msub> <mi>y</mi> <mn>1</mn> </msub> <mo>,</mo> <mo>&CenterDot;</mo> <mo>&CenterDot;</mo> <mo>&CenterDot;</mo> <msub> <mrow> <mo>,</mo> <mi>y</mi> </mrow> <mi>m</mi> </msub> <mo>)</mo> </mrow> </mrow> </munder> <mi>ln</mi> <mi>f</mi> <mrow> <mo>(</mo> <mi>&theta;</mi> <mo>|</mo> <mi>X</mi> <mo>,</mo> <mi>Y</mi> <mo>)</mo> </mrow> <mi>p</mi> <mrow> <mo>(</mo> <mi>y</mi> <mo>|</mo> <mi>X</mi> <mo>,</mo> <msup> <mi>&theta;</mi> <mi>g</mi> </msup> <mo>)</mo> </mrow> </mrow> </math>
<math> <mrow> <mo>=</mo> <munderover> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <mi>l</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>M</mi> </munderover> <mi>ln</mi> <mrow> <mo>(</mo> <msub> <mi>&alpha;</mi> <mi>l</mi> </msub> <msub> <mi>p</mi> <mi>l</mi> </msub> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>|</mo> <msub> <mi>&theta;</mi> <mi>l</mi> </msub> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mrow> <mo>(</mo> <munderover> <mi>&Pi;</mi> <mrow> <mi>j</mi> <mo>=</mo> <mn>1</mn> <mo>,</mo> <mi>j</mi> <mo>&NotEqual;</mo> <mi>i</mi> </mrow> <mi>N</mi> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <msub> <mi>y</mi> <mi>j</mi> </msub> <mo>=</mo> <mn>1</mn> </mrow> <mi>M</mi> </munderover> <mi>p</mi> <mrow> <mo>(</mo> <msub> <mi>y</mi> <mi>i</mi> </msub> <mo>|</mo> <msub> <mi>x</mi> <mi>j</mi> </msub> <mo>,</mo> <msup> <mi>&theta;</mi> <mi>g</mi> </msup> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mi>p</mi> <mrow> <mo>(</mo> <mi>l</mi> <mo>|</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>,</mo> <msup> <mi>&theta;</mi> <mi>g</mi> </msup> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>3</mn> <mo>)</mo> </mrow> </mrow> </math>
and M: finding Q (theta )g) The maximized theta is used as a new estimate of this parameter. Because of alphalAnd thetalIndependent, we can maximize the data items in equation (3) that contain the parameters, respectively. For the inclusion parameter alphalThe first term of (a) introduces a Lagrangian factor λ and has a constraint term Σlαl1, solving the following equation:
<math> <mrow> <mfrac> <mo>&PartialD;</mo> <mrow> <mo>&PartialD;</mo> <msub> <mi>&alpha;</mi> <mi>l</mi> </msub> </mrow> </mfrac> <mrow> <mo>(</mo> <munderover> <mi>&Sigma;</mi> <mrow> <mi>l</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>M</mi> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </munderover> <mi>ln</mi> <mrow> <mo>(</mo> <msub> <mi>&alpha;</mi> <mi>l</mi> </msub> <mo>)</mo> </mrow> <mi>p</mi> <mrow> <mo>(</mo> <mi>l</mi> <mo>|</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>,</mo> <msup> <mi>&theta;</mi> <mi>g</mi> </msup> <mo>)</mo> </mrow> <mo>+</mo> <mi>&lambda;</mi> <mrow> <mo>(</mo> <munder> <mi>&Sigma;</mi> <mi>l</mi> </munder> <msub> <mi>&alpha;</mi> <mi>l</mi> </msub> <mo>-</mo> <mn>1</mn> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mo>=</mo> <mn>0</mn> </mrow> </math>
it is possible to obtain:
<math> <mrow> <msub> <mi>&alpha;</mi> <mi>l</mi> </msub> <mo>=</mo> <mfrac> <mn>1</mn> <mi>N</mi> </mfrac> <munderover> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </munderover> <mi>p</mi> <mrow> <mo>(</mo> <mi>l</mi> <mo>|</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>,</mo> <msup> <mi>&theta;</mi> <mi>g</mi> </msup> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>4</mn> <mo>)</mo> </mrow> </mrow> </math>
gaussian distribution pl(x|μl,∑l) Can be expressed as:
<math> <mrow> <msub> <mi>p</mi> <mi>l</mi> </msub> <mrow> <mo>(</mo> <mi>x</mi> <mo>|</mo> <msub> <mi>&mu;</mi> <mi>l</mi> </msub> <mo>,</mo> <msub> <mi>&sigma;</mi> <mi>l</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mn>1</mn> <mrow> <msup> <mrow> <mo>(</mo> <mn>2</mn> <mi>&pi;</mi> <mo>)</mo> </mrow> <mrow> <mi>d</mi> <mo>/</mo> <mn>2</mn> </mrow> </msup> <msup> <mrow> <mo>|</mo> <msub> <mi>&sigma;</mi> <mi>l</mi> </msub> <mo>|</mo> </mrow> <mrow> <mn>1</mn> <mo>/</mo> <mn>2</mn> </mrow> </msup> </mrow> </mfrac> <mi>exp</mi> <mrow> <mo>(</mo> <mo>-</mo> <mfrac> <mn>1</mn> <mn>2</mn> </mfrac> <msup> <mrow> <mo>(</mo> <mi>x</mi> <mo>-</mo> <msub> <mi>&mu;</mi> <mi>l</mi> </msub> <mo>)</mo> </mrow> <mi>T</mi> </msup> <msup> <msub> <mi>&sigma;</mi> <mi>l</mi> </msub> <mrow> <mo>-</mo> <mn>1</mn> </mrow> </msup> <mrow> <mo>(</mo> <mi>x</mi> <mo>-</mo> <msub> <mi>&mu;</mi> <mi>l</mi> </msub> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>5</mn> <mo>)</mo> </mrow> </mrow> </math>
wherein, mulDenotes the mean value, σlThe variance is indicated.
Then the pair contains the parameter thetalThe second term of (2), can be written as:
<math> <mrow> <munderover> <mi>&Sigma;</mi> <mrow> <mi>l</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>M</mi> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </munderover> <mi>ln</mi> <mrow> <mo>(</mo> <msub> <mi>p</mi> <mi>l</mi> </msub> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>|</mo> <msub> <mi>&theta;</mi> <mi>l</mi> </msub> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mi>p</mi> <mrow> <mo>(</mo> <mi>l</mi> <mo>|</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>,</mo> <msup> <mi>&theta;</mi> <mi>g</mi> </msup> <mo>)</mo> </mrow> </mrow> </math>
<math> <mrow> <mo>=</mo> <munderover> <mi>&Sigma;</mi> <mrow> <mi>l</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>M</mi> </munderover> <munderover> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </munderover> <mrow> <mo>(</mo> <mo>-</mo> <mfrac> <mn>1</mn> <mn>2</mn> </mfrac> <mi>ln</mi> <mrow> <mo>(</mo> <mo>|</mo> <msub> <mi>&sigma;</mi> <mi>l</mi> </msub> <mo>|</mo> <mo>-</mo> <mfrac> <mn>1</mn> <mn>2</mn> </mfrac> <msup> <mrow> <mo>(</mo> <mi>x</mi> <mo>-</mo> <msub> <mi>&mu;</mi> <mi>l</mi> </msub> <mo>)</mo> </mrow> <mi>T</mi> </msup> <msup> <msub> <mi>&sigma;</mi> <mi>l</mi> </msub> <mrow> <mo>-</mo> <mn>1</mn> </mrow> </msup> <mrow> <mo>(</mo> <mi>x</mi> <mo>-</mo> <msub> <mi>&mu;</mi> <mi>l</mi> </msub> <mo>)</mo> </mrow> <mo>)</mo> </mrow> <mi>p</mi> <mrow> <mo>(</mo> <mi>l</mi> <mo>|</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>,</mo> <msup> <mi>&theta;</mi> <mi>g</mi> </msup> <mo>)</mo> </mrow> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>6</mn> <mo>)</mo> </mrow> </mrow> </math>
formula (4.26) vs. mulCalculating the partial derivative and making it equal to 0, we can obtain:
<math> <mrow> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </msubsup> <msup> <msub> <mi>&sigma;</mi> <mi>l</mi> </msub> <mrow> <mo>-</mo> <mn>1</mn> </mrow> </msup> <mrow> <mo>(</mo> <mi>x</mi> <mo>-</mo> <msub> <mi>&mu;</mi> <mi>l</mi> </msub> <mo>)</mo> </mrow> <msub> <mi>&alpha;</mi> <mi>l</mi> </msub> <mi>p</mi> <mrow> <mo>(</mo> <mi>l</mi> <mo>|</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>,</mo> <msup> <mi>&theta;</mi> <mi>g</mi> </msup> <mo>)</mo> </mrow> <mo>=</mo> <mn>0</mn> </mrow> </math>
therefore, it can be found that:
<math> <mrow> <msub> <mi>&mu;</mi> <mi>l</mi> </msub> <mo>=</mo> <mfrac> <mrow> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </msubsup> <msub> <mi>x</mi> <mi>i</mi> </msub> <mi>p</mi> <mrow> <mo>(</mo> <mi>l</mi> <mo>|</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>,</mo> <msup> <mi>&theta;</mi> <mi>g</mi> </msup> <mo>)</mo> </mrow> </mrow> <mrow> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </msubsup> <mi>p</mi> <mrow> <mo>(</mo> <mi>l</mi> <mo>|</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>,</mo> <msup> <mi>&theta;</mi> <mi>g</mi> </msup> <mo>)</mo> </mrow> </mrow> </mfrac> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>7</mn> <mo>)</mo> </mrow> </mrow> </math>
equation (4.27) vs. σlCalculating the partial derivative and making it equal to 0, we can obtain:
<math> <mrow> <msub> <mi>&sigma;</mi> <mi>l</mi> </msub> <mo>=</mo> <mfrac> <mrow> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </msubsup> <mi>p</mi> <mrow> <mo>(</mo> <mi>l</mi> <mo>|</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>,</mo> <msup> <mi>&theta;</mi> <mi>g</mi> </msup> <mo>)</mo> </mrow> <mrow> <mo>(</mo> <mi>x</mi> <mo>-</mo> <msub> <mi>&mu;</mi> <mi>l</mi> </msub> <mo>)</mo> </mrow> <msup> <mrow> <mo>(</mo> <mi>x</mi> <mo>-</mo> <msub> <mi>&mu;</mi> <mi>l</mi> </msub> <mo>)</mo> </mrow> <mi>T</mi> </msup> </mrow> <mrow> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </msubsup> <mi>p</mi> <mrow> <mo>(</mo> <mi>l</mi> <mo>|</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>,</mo> <msup> <mi>&theta;</mi> <mi>g</mi> </msup> <mo>)</mo> </mrow> </mrow> </mfrac> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>8</mn> <mo>)</mo> </mrow> </mrow> </math>
fig. 2 is a schematic diagram of a specific embodiment of a target tracking method provided by the present invention, in this embodiment, a method of detecting a tracking target uses a background frame difference method, and an image feature of a target region takes discretization color probability distribution of the target region as an example. The method comprises the following steps:
step 201, detecting a tracking target in a reference image, obtaining a central position parameter and a shape size parameter of a target area in the reference image as initial values of the target area parameter, and calculating a discretization color probability distribution of the area.
In this step, a background frame difference method is used to detect a tracking target in the reference image, specifically:
background modeling is performed first, and the background model used here is a time-averaged image, that is, an average image of the same scene in a period of time is used as the background modeling of the scene.
Then, the image inputted at the time T is used as a reference image, the reference image is subtracted from the background image and the previous frame image respectively to obtain two frame difference images, and the two frame difference images are respectively subjected to binarization processing by using a thresholding method. Then, using mathematical morphology methods, such as expansion operation, erosion operation, on operation, off operation, and the like, to perform filtering processing on the two frames of binary images, filling the holes in the foreground region, and simultaneously removing the isolated regions and non-connected regions with small areas, and only keeping the connected parts of the connected regions with the areas larger than the given threshold.
And finally, performing logical AND operation on the two filtered binary images, performing mathematical form filtering processing on the operated images, and obtaining the contour of the tracking target from the filtered images by adopting a contour tracking method. A target area of a tracking target is represented by a rectangular reference frame, and the target area parameters include a shape size parameter and a center position parameter of the target area. The shape and size parameters of the target area are the length and width of the rectangular reference frame, and the central position parameters of the target area are the number of rows (from top to bottom) and the number of columns (from left to right) of the central point of the rectangular reference frame in the image.
Assuming a target region detected in a reference image <math> <mrow> <mi>S</mi> <mo>=</mo> <mrow> <mo>{</mo> <msubsup> <mi>x</mi> <mi>i</mi> <mo>*</mo> </msubsup> <mo>}</mo> </mrow> <mo>,</mo> <mi>i</mi> <mo>=</mo> <mn>1</mn> <mo>,</mo> <mo>&CenterDot;</mo> <mo>&CenterDot;</mo> <mo>&CenterDot;</mo> <mo>,</mo> <mi>n</mi> <mo>,</mo> </mrow> </math> Target area parameter is theta0=(μ0,σ0) Wherein theta0As initial values of parameters of the target area, mu0Is an initial value of a parameter of the center position of the target area, sigma0Is an initial value of a size parameter of the shape of the target area and has <math> <mrow> <msub> <mi>&sigma;</mi> <mn>0</mn> </msub> <mo>=</mo> <msub> <mi>&Sigma;</mi> <mrow> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>&Element;</mo> <mi>S</mi> </mrow> </msub> <mrow> <mo>(</mo> <msub> <mover> <mi>x</mi> <mo>&RightArrow;</mo> </mover> <mi>i</mi> </msub> <mo>-</mo> <msub> <mover> <mi>&mu;</mi> <mo>&RightArrow;</mo> </mover> <mn>0</mn> </msub> <mo>)</mo> </mrow> <msup> <mrow> <mo>(</mo> <msub> <mover> <mi>x</mi> <mo>&RightArrow;</mo> </mover> <mi>i</mi> </msub> <mo>-</mo> <msub> <mover> <mi>&mu;</mi> <mo>&RightArrow;</mo> </mover> <mn>0</mn> </msub> <mo>)</mo> </mrow> <mi>T</mi> </msup> <mo>.</mo> </mrow> </math> Defining a function b (x)i *):R2→ 1.. m represents a pixel xi *The color value of (a). Then the parameter θ becomes (μ)0,σ0) In a defined target region, pixel point xi *Discretized color probability distribution of <math> <mrow> <msub> <mi>q</mi> <mi>u</mi> </msub> <mo>=</mo> <mrow> <mo>{</mo> <msub> <mover> <mi>q</mi> <mo>^</mo> </mover> <mi>u</mi> </msub> <mrow> <mo>(</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>}</mo> </mrow> <mo>,</mo> <mi>u</mi> <mo>=</mo> <mn>1</mn> <mo>,</mo> <mo>&CenterDot;</mo> <mo>&CenterDot;</mo> <mo>&CenterDot;</mo> <mo>,</mo> <mi>m</mi> <mo>.</mo> </mrow> </math> Can be expressed as:
<math> <mrow> <msub> <mover> <mi>q</mi> <mo>^</mo> </mover> <mi>u</mi> </msub> <mo>=</mo> <mi>C</mi> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>n</mi> </msubsup> <mi>N</mi> <mrow> <mo>(</mo> <msubsup> <mi>x</mi> <mi>i</mi> <mo>*</mo> </msubsup> <mo>;</mo> <msub> <mi>&mu;</mi> <mn>0</mn> </msub> <mo>;</mo> <msub> <mi>&sigma;</mi> <mn>0</mn> </msub> <mo>)</mo> </mrow> <mi>&delta;</mi> <mrow> <mo>[</mo> <mi>b</mi> <mrow> <mo>(</mo> <msubsup> <mi>x</mi> <mi>i</mi> <mo>*</mo> </msubsup> <mo>)</mo> </mrow> <mo>-</mo> <mi>u</mi> <mo>]</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>9</mn> <mo>)</mo> </mrow> </mrow> </math>
wherein, the delta is a kronecker delta function, and the main meaning of the delta is a function of two variables, and when the variables have the same value, the function is 1; when the variables have different values, the function is 0. C is a normalization constant and has <math> <mrow> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>u</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>m</mi> </msubsup> <msub> <mover> <mi>q</mi> <mo>^</mo> </mover> <mi>u</mi> </msub> <mo>=</mo> <mn>1</mn> <mo>.</mo> </mrow> </math> m represents the level of discretization and N (-) is a gaussian kernel function.
Step 202, in the current frame, according to the initial value mu of the target area central position parameter0And an initial value σ of a target region shape size parameter0Candidate target regions are determined.
This step corresponds to the E step in the EM algorithm, i.e. the target region parameter in the reference image is given as the observation condition, and a candidate target region is determined in the current frame using this observation condition.
The method comprises the following steps: finding the initial value mu of the parameter of the center position of the target area in the current frame image0Corresponding to the position, determining the initial value sigma of the size parameter of the target region shape0Equal rectangular areas are candidate target areas. That is, the determined candidate target region parameter θ is assumedi=(μi,σi) Wherein the center position parameter is muiWhen mu is equal to mu0Equal; the center position parameter of the candidate target region is sigmaiThen there is σi=σ0。。
Step 203, according to the candidate target area parameter thetai=(μi,σi) And calculating the color probability distribution of each pixel point in the candidate target region, and calculating the similarity coefficient of the candidate target region and the target region in the reference image according to the discretization color probability distribution of the candidate target region and the discretization color probability distribution of the target region in the reference image.
The discretization color probability distribution of the candidate target region is assumed to beThen there are:
<math> <mrow> <msub> <mover> <mi>p</mi> <mo>^</mo> </mover> <mi>u</mi> </msub> <mrow> <mo>(</mo> <msub> <mi>&theta;</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <msub> <mi>C</mi> <mi>&sigma;i</mi> </msub> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <msub> <mi>n</mi> <mi>&sigma;i</mi> </msub> </msubsup> <mi>N</mi> <mrow> <mo>(</mo> <msubsup> <mi>x</mi> <mi>i</mi> <mo>*</mo> </msubsup> <mo>;</mo> <msub> <mi>&mu;</mi> <mi>i</mi> </msub> <mo>;</mo> <msub> <mi>&sigma;</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mi>&delta;</mi> <mrow> <mo>[</mo> <mi>b</mi> <mrow> <mo>(</mo> <msubsup> <mi>x</mi> <mi>i</mi> <mo>*</mo> </msubsup> <mo>)</mo> </mrow> <mo>-</mo> <msub> <mi>u</mi> <mi>i</mi> </msub> <mo>]</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>10</mn> <mo>)</mo> </mrow> </mrow> </math>
further combining the discretized color probability distribution of the target region in the reference image obtained in step 201The similarity coefficient of the two can be obtained as follows:
Figure A20081011553700164
and step 204, performing EM processing on the candidate target area parameters according to the similarity coefficient obtained in the step 203, and calculating new target area parameters.
The step is equivalent to M steps in an EM algorithm, the expected value of the position distribution in the whole observation range is estimated under the observation condition given by the step E, and the actual value is gradually approximated through iterative processing.
In this step, the similarity coefficient obtained in step 203 is first expanded by taylor's formula, and the high-order terms are truncated, so as to obtain:
Figure A20081011553700165
where c is1,c2Are all constant factors.
Then, by modifying equation (9), it is possible to obtain:
<math> <mrow> <msub> <mi>w</mi> <mi>i</mi> </msub> <mo>=</mo> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>m</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>M</mi> </msubsup> <msqrt> <msub> <mi>q</mi> <mi>u</mi> </msub> <mo>/</mo> <msub> <mover> <mi>p</mi> <mo>^</mo> </mover> <mi>u</mi> </msub> <mrow> <mo>(</mo> <msup> <mi>&theta;</mi> <mi>k</mi> </msup> <mo>)</mo> </mrow> </msqrt> <mi>&delta;</mi> <mrow> <mo>[</mo> <mi>b</mi> <mrow> <mo>(</mo> <msubsup> <mi>x</mi> <mi>i</mi> <mo>*</mo> </msubsup> <mo>)</mo> </mrow> <mo>-</mo> <mi>u</mi> <mo>]</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>13</mn> <mo>)</mo> </mrow> </mrow> </math>
it is necessary to introduce a normalization constant, the parameter α in equation (1)lThe relationship to the similarity coefficient of the color distribution is: alpha is alphalγ · ρ. From equation (4) and the above relation, it can be obtained:
<math> <mrow> <mi>p</mi> <mrow> <mo>(</mo> <mi>l</mi> <mo>|</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>,</mo> <msup> <mi>&theta;</mi> <mi>g</mi> </msup> <mo>)</mo> </mrow> <mo>=</mo> <mo>=</mo> <msub> <mi>w</mi> <mi>i</mi> </msub> <mi>N</mi> <mrow> <mo>(</mo> <msubsup> <mi>x</mi> <mi>i</mi> <mo>*</mo> </msubsup> <mo>;</mo> <msub> <mi>&theta;</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>14</mn> <mo>)</mo> </mrow> </mrow> </math>
substituting the formula (14) into the formulas (7) and (8) to obtain new target area parameters as follows:
<math> <mrow> <msub> <mi>&mu;</mi> <mrow> <mi>i</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> <mo>=</mo> <mfrac> <mrow> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>K</mi> </msubsup> <msubsup> <mi>x</mi> <mi>i</mi> <mo>*</mo> </msubsup> <msub> <mi>w</mi> <mi>i</mi> </msub> <mi>N</mi> <mrow> <mo>(</mo> <msubsup> <mi>x</mi> <mi>i</mi> <mo>*</mo> </msubsup> <mo>;</mo> <msub> <mi>&theta;</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> </mrow> <mrow> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>K</mi> </msubsup> <msub> <mi>w</mi> <mi>i</mi> </msub> <mi>N</mi> <mrow> <mo>(</mo> <msubsup> <mi>x</mi> <mi>i</mi> <mo>*</mo> </msubsup> <mo>;</mo> <msub> <mi>&theta;</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> </mrow> </mfrac> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>15</mn> <mo>)</mo> </mrow> </mrow> </math>
<math> <mrow> <msub> <mi>&sigma;</mi> <mrow> <mi>i</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> <mo>=</mo> <mfrac> <mrow> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>K</mi> </msubsup> <msub> <mi>w</mi> <mi>i</mi> </msub> <mi>N</mi> <mrow> <mo>(</mo> <msubsup> <mi>x</mi> <mi>i</mi> <mo>*</mo> </msubsup> <mo>;</mo> <msub> <mi>&theta;</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mrow> <mo>(</mo> <mi>x</mi> <mo>-</mo> <msub> <mi>&mu;</mi> <mrow> <mi>i</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> <mo>)</mo> </mrow> <msup> <mrow> <mo>(</mo> <mi>x</mi> <mo>-</mo> <msub> <mi>&mu;</mi> <mrow> <mi>i</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> <mo>)</mo> </mrow> <mi>T</mi> </msup> </mrow> <mrow> <mrow> <mo>(</mo> <mn>1</mn> <mo>-</mo> <mi>&gamma;</mi> <mo>)</mo> </mrow> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>K</mi> </msubsup> <msub> <mi>w</mi> <mi>i</mi> </msub> <mi>N</mi> <mrow> <mo>(</mo> <msubsup> <mi>x</mi> <mi>i</mi> <mo>*</mo> </msubsup> <mo>;</mo> <msub> <mi>&theta;</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> </mrow> </mfrac> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>16</mn> <mo>)</mo> </mrow> </mrow> </math>
wherein mui+1As a new central position parameter, σ, of the target regioni+1The shape and size parameters of the new target area are shown, and K is the number of pixel points in the new target area.
Step 205, according to the new target area parameter obtained in step 204, determine whether it is the real value of the target area parameter in the current frame.
The method comprises the following specific steps: comparing the new target region parameter θi+1=(μi+1,σi+1) And candidate target region parameter thetai=(μi,σi) If the absolute value of the difference between the two is less than the preset threshold value, then eta is determinedi+1=(μi+1,σi+1) The real value of the target area parameter in the current frame is calculated, otherwise theta is calculatedi+1=(μi+1,σi+1) As candidate target area parameters and returns to step 203.
The threshold value is a value small enough to ensure that the new target area parameter is obtained sufficientlyThe true value of the parameter of the target area in the current frame is approximated, and the value range of the threshold value can be 0 to 10e-6
The embodiment of the invention also provides a target tracking device. Fig. 3 is a schematic structural diagram of an object detection apparatus provided in the present invention. The device includes:
the detection module 301 is configured to detect a tracking target from a reference image, obtain an initial value of a target area parameter, calculate an image feature in the target area in the reference image, and send the initial value of the target area parameter and the image feature to the tracking module 302;
a tracking module 302, configured to receive the initial value of the target area parameter from the detection module 301, and calculate an image feature of the candidate target area according to the initial value of the target area parameter as a candidate target area parameter;
the tracking module 302 is further configured to calculate, by combining with the image features in the target region in the reference image from the detecting module 301, a similarity coefficient between the candidate target region and the candidate target region in the reference image, and calculate a true value of the target region parameter of the current frame according to the similarity coefficient and the candidate target region parameter.
FIG. 4 is a diagram of an embodiment of a specific structure of a tracking module 302, which includes:
the candidate target region updating unit 401 receives an image from the outside and an initial value of a target region parameter from the detection module 301, determines a candidate target region in the current frame based on the initial value of the target region parameter, and sends the initial value of the target region parameter as a candidate target region parameter to the feature extraction unit 402.
A candidate target area updating unit 401, configured to receive the result obtained by the calculating unit 403 and send the result to the feature extracting unit 402, the calculating unit 403, and the determining unit 404 as a candidate target area parameter;
a feature extraction unit 402, which calculates the image features of the candidate target region according to the received candidate target region parameters, and sends the obtained candidate target region image features to a calculation unit;
a calculating unit 403 for calculating a similarity coefficient between the candidate target region and the target region in the reference image according to the initial value of the image feature from the detecting module 301 and the image feature of the candidate target region from the feature extracting unit 402; is also used for calculating new target area parameters according to the image features from the feature extraction unit 402 and the target area parameters from the candidate target area updating unit 401, and sending the calculation results to the judgment unit 404 and the candidate target area updating unit 401;
a judging unit 404, configured to judge whether an absolute value of a difference between the new target area parameter from the calculating unit 403 and the candidate target area parameter from the candidate target area updating unit 401 is smaller than a threshold, and if so, judge that the new target area parameter is a true value of the current frame target area parameter;
otherwise, the new target area parameter will be returned to the candidate target area updating unit 401 as the candidate target area parameter for the new round of calculation.
Fig. 5 is a schematic diagram of an embodiment of a specific structure of a computing unit, which includes:
a similarity coefficient calculating subunit 501, which receives the image features from the feature extracting unit 402, calculates a similarity coefficient between the candidate target region and the target region in the reference image, and sends the similarity coefficient to the target region parameter calculating subunit 502;
the target area parameter calculating subunit 502 calculates a new target area parameter from the similarity coefficient calculating subunit 501 and the candidate area parameter from the candidate target area updating unit 401, and sends the obtained new target area parameter to the candidate target area updating unit 401 and the judging unit 404.
Fig. 6 is a schematic diagram of an embodiment of a specific structure of the determination unit 404, which includes:
a difference value calculating subunit 601 that calculates an absolute value of a difference between the candidate target region parameter from the candidate target region updating unit 401 and the new target region parameter from the target region calculating subunit 502, and outputs the absolute value to the comparing subunit 603;
a threshold storage subunit 602 that stores a threshold used to determine whether the new target region parameter is the true value of the current frame target region parameter;
a comparing subunit 603, configured to compare the calculation result from the difference calculating subunit 601 with the threshold in the threshold storing subunit 602, and if the comparison result is smaller than the threshold, determine that the new target region parameter is the true value of the current frame target region parameter;
otherwise, the new target area parameter will be returned to the candidate target area updating unit 401 as the candidate target area parameter for the new round of calculation.
It can be seen from the foregoing embodiments that, in the target tracking method and apparatus provided by the present invention, the gaussian mixture model and the similarity of the target based on the image characteristics are associated, the image characteristics of the target region are used to describe the weighting factors in the gaussian mixture model, and the covariance matrix parameters of the target position parameter and the target shape size are estimated through the EM iteration process, so that the target region parameters are adaptively changed according to the motion condition of the tracked target, thereby achieving the effect of accurately tracking the target in real time.
In summary, the above description is only a preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (10)

1. A target tracking method, comprising the steps of:
1) detecting a tracking target in a reference image, acquiring an initial value of a target area parameter, and calculating the image characteristics of a target area in the reference image;
2) determining a candidate target area in the current frame by taking the initial value of the target area parameter as a candidate target area parameter, and calculating the image characteristic of the candidate target area;
3) calculating a similarity coefficient between the candidate target area and the target area in the reference image according to the image characteristics of the target area in the reference image and the image characteristics of the candidate target area;
4) calculating new target area parameters according to the similarity coefficient between the target area in the reference image and the target area in the current frame;
5) judging whether the obtained new target area parameter is the true value of the target area parameter in the current frame, if so, taking the new target area parameter as the true value of the target area parameter in the current frame, if not, returning to the step 2) to replace the candidate target area parameter with the new target area parameter, and repeating the steps 2) to 5) until the obtained new target area parameter is the true value of the target area parameter in the current frame.
2. The method of claim 1, wherein: the initial values of the parameters of the target area comprise an initial value of a parameter of the center position of the target area and an initial value of a parameter of the shape and the size of the target area;
the candidate target area parameters comprise a candidate target area center position parameter and a candidate target area shape size parameter.
3. The method of claim 1, wherein: the image feature of the target region is the discretization color probability distribution of the target region, and the image feature of the candidate target region is the discretization color probability distribution of the candidate target region.
4. The method of claim 3, wherein: the discretized color probability distribution is:
<math> <mrow> <msub> <mover> <mi>q</mi> <mo>^</mo> </mover> <mi>u</mi> </msub> <mo>=</mo> <mi>C</mi> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>n</mi> </msubsup> <mi>N</mi> <mrow> <mo>(</mo> <msubsup> <mi>x</mi> <mi>i</mi> <mo>*</mo> </msubsup> <mo>;</mo> <msub> <mi>&mu;</mi> <mi>i</mi> </msub> <mo>;</mo> <msub> <mi>&sigma;</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mi>&delta;</mi> <mo>[</mo> <mi>b</mi> <mrow> <mo>(</mo> <msubsup> <mi>x</mi> <mi>i</mi> <mo>*</mo> </msubsup> <mo>)</mo> </mrow> <mo>-</mo> <mi>u</mi> <mo>]</mo> <mo>,</mo> </mrow> </math>
wherein muiIs a target region center position parameter, σiA discretization grade is represented by u as a target area shape size parameter; n (-) is a Gaussian kernel function defining a function b (x)i *):R2→ 1.. multidot.m represents pixel point xi *δ is a function containing two variables, which is 1 when the variables have the same value; when the variables have different values, the function is 0, C is a normalization constant, and <math> <mrow> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>u</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>m</mi> </msubsup> <msub> <mover> <mi>q</mi> <mo>^</mo> </mover> <mi>u</mi> </msub> <mo>=</mo> <mn>1</mn> <mo>.</mo> </mrow> </math>
5. the method of claim 1, wherein the similarity coefficient between the candidate target region and the target region in the reference image is:
Figure A2008101155370003C2
wherein,
Figure A2008101155370003C3
for a discretized color probability distribution of the candidate target region,for discretized color probability distribution of the reference target area, θiFor the candidate target region parameters, m and u represent discretization levels, and u is 1, …, m.
6. The method of claim 1, wherein: judging whether the obtained new target area parameter is the true value of the target area parameter in the current frame or not, wherein the true value is as follows:
judging whether the absolute value of the difference between the obtained new target area parameter and the candidate target area parameter is smaller than a preset threshold value or not, and if so, considering that the obtained new target area parameter is the true value of the target area parameter in the current frame; if so, the obtained new target area parameter is not considered to be the true value of the target area parameter in the current frame.
7. An object tracking apparatus, characterized in that the apparatus comprises: the device comprises a detection module and a tracking module.
The detection module is used for detecting a tracking target from a reference image, acquiring a target area parameter initial value, calculating the image characteristics of the target area in the reference image, and sending the target area parameter initial value and the image characteristics to the tracking module;
the tracking module is used for receiving the initial value of the target area parameter from the detection module, taking the initial value of the target area parameter as a candidate target area parameter and calculating the image characteristic of the candidate target area;
the tracking module is further configured to calculate a similarity coefficient between the candidate target region and the candidate target region in the reference image according to the image feature of the target region in the reference image from the detection module and the image feature of the candidate target region, and calculate a true value of a current frame target region parameter according to the similarity coefficient and the candidate target region parameter.
8. The apparatus of claim 7, wherein the tracking module comprises: a candidate target area updating unit, a feature extracting unit, a calculating unit, a judging unit, wherein
The candidate target area updating unit is used for receiving an image from the outside and a target area parameter initial value from the detection module and determining a candidate target area in the current frame according to the target area parameter initial value,
the candidate target area updating unit is also used for receiving the result obtained by the calculating unit as a candidate target area parameter and sending the candidate target area parameter to the feature extracting unit, the calculating unit and the judging unit;
the feature extraction unit is used for calculating the image features of the candidate target regions according to the received candidate target region parameters;
the calculating unit is used for calculating new target area parameters according to the image features from the feature extracting unit and the target area parameters from the candidate target area updating unit and sending the calculation results to the judging unit and the candidate target area updating unit;
and the judging unit is used for judging whether the new target area parameter from the calculating unit is the real value of the current frame target area parameter.
9. The apparatus of claim 8, wherein the computing unit comprises: a similarity coefficient calculating subunit and a target area parameter calculating subunit;
the similarity coefficient calculating subunit is used for receiving the color probability distribution from the feature extraction unit, calculating the similarity coefficient between the candidate target region and the target region in the reference image, and sending the similarity coefficient to the target region parameter calculating subunit;
and the target area parameter calculating subunit is used for calculating a new target area parameter according to the similarity coefficient from the similarity coefficient calculating subunit and the candidate area parameter from the candidate target area updating unit, and sending the new target area parameter to the candidate target area updating unit and the judging unit.
10. The apparatus of claim 9, wherein the determining unit comprises: a difference value calculating subunit, a threshold value storing subunit and a comparing subunit;
the difference value calculating subunit is configured to calculate an absolute value of a difference between the candidate target region parameter from the candidate target region updating unit and the calculation result from the target region parameter calculating subunit, and output the absolute value to the comparing unit;
the threshold storage subunit is configured to store a threshold used for determining whether the new target area parameter is a true value of the current frame target area parameter;
and the comparison subunit is used for comparing the calculation result from the difference value operator unit with the threshold value in the threshold value storage subunit and outputting the comparison result to the calculation unit.
CNA2008101155377A 2008-06-25 2008-06-25 Method and apparatus for tracking object Pending CN101324958A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CNA2008101155377A CN101324958A (en) 2008-06-25 2008-06-25 Method and apparatus for tracking object

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CNA2008101155377A CN101324958A (en) 2008-06-25 2008-06-25 Method and apparatus for tracking object

Publications (1)

Publication Number Publication Date
CN101324958A true CN101324958A (en) 2008-12-17

Family

ID=40188484

Family Applications (1)

Application Number Title Priority Date Filing Date
CNA2008101155377A Pending CN101324958A (en) 2008-06-25 2008-06-25 Method and apparatus for tracking object

Country Status (1)

Country Link
CN (1) CN101324958A (en)

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101719278B (en) * 2009-12-21 2012-01-04 西安电子科技大学 Automatic tracking method for video microimage cells based on KHM algorithm
CN102867193A (en) * 2012-09-14 2013-01-09 成都国科海博计算机系统有限公司 Biological detection method and device and biological detector
CN103339655A (en) * 2011-02-03 2013-10-02 株式会社理光 Image capturing apparatus, image capturing method, and computer program product
CN104700431A (en) * 2015-02-10 2015-06-10 浙江工业大学 Saliency based flexible object nature contour tracking method
CN105740751A (en) * 2014-12-11 2016-07-06 深圳市赛为智能股份有限公司 Object detection and identification method and system
CN106228805A (en) * 2016-06-30 2016-12-14 北京航空航天大学 A kind of interactive many wireless vehicle trackings and device
CN106485231A (en) * 2016-10-21 2017-03-08 上海未来伙伴机器人有限公司 A kind of target color lump extracting method and device
CN107169436A (en) * 2017-05-11 2017-09-15 南宁市正祥科技有限公司 Adaptive motion vehicle checking method
CN108629935A (en) * 2018-05-17 2018-10-09 山东深图智能科技有限公司 A kind of method and system for climbing building pivot frame larceny based on video monitoring detection
CN109461174A (en) * 2018-10-25 2019-03-12 北京陌上花科技有限公司 Video object area tracking method and video plane advertisement method for implantation and system

Cited By (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101719278B (en) * 2009-12-21 2012-01-04 西安电子科技大学 Automatic tracking method for video microimage cells based on KHM algorithm
CN103339655A (en) * 2011-02-03 2013-10-02 株式会社理光 Image capturing apparatus, image capturing method, and computer program product
CN103339655B (en) * 2011-02-03 2016-03-30 株式会社理光 Image capture device, image capture method and computer program
US9336606B2 (en) 2011-02-03 2016-05-10 Ricoh Company, Ltd. Image capturing apparatus, image capturing method, and computer program product
CN102867193A (en) * 2012-09-14 2013-01-09 成都国科海博计算机系统有限公司 Biological detection method and device and biological detector
CN102867193B (en) * 2012-09-14 2015-06-17 成都国科海博信息技术股份有限公司 Biological detection method and device and biological detector
CN105740751A (en) * 2014-12-11 2016-07-06 深圳市赛为智能股份有限公司 Object detection and identification method and system
CN104700431A (en) * 2015-02-10 2015-06-10 浙江工业大学 Saliency based flexible object nature contour tracking method
CN104700431B (en) * 2015-02-10 2017-08-11 浙江工业大学 A kind of natural contour tracing method of the flexible object based on significance
CN106228805A (en) * 2016-06-30 2016-12-14 北京航空航天大学 A kind of interactive many wireless vehicle trackings and device
CN106485231A (en) * 2016-10-21 2017-03-08 上海未来伙伴机器人有限公司 A kind of target color lump extracting method and device
CN107169436A (en) * 2017-05-11 2017-09-15 南宁市正祥科技有限公司 Adaptive motion vehicle checking method
CN108629935A (en) * 2018-05-17 2018-10-09 山东深图智能科技有限公司 A kind of method and system for climbing building pivot frame larceny based on video monitoring detection
CN109461174A (en) * 2018-10-25 2019-03-12 北京陌上花科技有限公司 Video object area tracking method and video plane advertisement method for implantation and system
CN109461174B (en) * 2018-10-25 2021-01-29 北京陌上花科技有限公司 Video target area tracking method and video plane advertisement implanting method and system

Similar Documents

Publication Publication Date Title
CN101324958A (en) Method and apparatus for tracking object
CN106408594B (en) Video multi-target tracking based on more Bernoulli Jacob&#39;s Eigen Covariances
CN107527009B (en) Remnant detection method based on YOLO target detection
CN105279484B (en) Method for checking object and object test equipment
EP1975879B1 (en) Computer implemented method for tracking object in sequence of frames of video
CN108647649B (en) Method for detecting abnormal behaviors in video
CN102831409B (en) Motion pedestrian&#39;s video automatic tracking method based on particle filter and system
CN111861925B (en) Image rain removing method based on attention mechanism and door control circulation unit
CN107203781A (en) A kind of object detection method Weakly supervised end to end instructed based on conspicuousness
CN110175649B (en) Rapid multi-scale estimation target tracking method for re-detection
CN103927764B (en) A kind of wireless vehicle tracking of combining target information and estimation
CN112052802B (en) Machine vision-based front vehicle behavior recognition method
CN104050685B (en) Moving target detecting method based on particle filter visual attention model
CN103324951B (en) A kind of polynary isomery monitoring water environment data evaluation and method for early warning
CN110163103B (en) Live pig behavior identification method and device based on video image
CN105740751A (en) Object detection and identification method and system
CN108596032B (en) Detection method, device, equipment and medium for fighting behavior in video
CN109978916B (en) Vibe moving target detection method based on gray level image feature matching
Firouznia et al. Adaptive chaotic sampling particle filter to handle occlusion and fast motion in visual object tracking
CN103870837B (en) It is a kind of to detect the method and apparatus that whether there is vehicle in region to be detected
CN106023256A (en) State observation method for planar target particle filter tracking of augmented reality auxiliary maintenance system
CN101877135B (en) Moving target detecting method based on background reconstruction
CN102663419A (en) Pan-tilt tracking method based on representation model and classification model
CN110570450B (en) Target tracking method based on cascade context-aware framework
CN104616323A (en) Space-time significance detecting method based on slow characteristic analysis

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C02 Deemed withdrawal of patent application after publication (patent law 2001)
WD01 Invention patent application deemed withdrawn after publication

Open date: 20081217