CN105139424B - Method for tracking target based on signal filtering - Google Patents
Method for tracking target based on signal filtering Download PDFInfo
- Publication number
- CN105139424B CN105139424B CN201510529167.1A CN201510529167A CN105139424B CN 105139424 B CN105139424 B CN 105139424B CN 201510529167 A CN201510529167 A CN 201510529167A CN 105139424 B CN105139424 B CN 105139424B
- Authority
- CN
- China
- Prior art keywords
- target
- belief
- signal
- frame
- tracking
- 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.)
- Active
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
- G06T2207/20024—Filtering details
Landscapes
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses a kind of method for tracking target based on signal filtering comprising following steps: input signal and output target degree of belief mapping being established functional relation by filter model；Estimated matrix is mapped using target degree of belief and initial filter model is calculated in current goal candidate region, and initial filter model and previous moment filter model weighted sum are obtained current more new model；It is filtered using signal of the filter model to current goal candidate region and obtains the mapping of target degree of belief, using the maximum in the mapping of target degree of belief as current goal tracking result.The present invention solves the problems, such as that tracking accuracy is undesirable in original method for tracking target, real-time is poor, tracking poor robustness.
Description
Technical field
The present invention relates to signal trace fields, specifically, being to be related to a kind of method for tracking target based on signal filtering.
Background technique
Target following refers to the position for obtaining target in time domain continuous signal, is the core content of signal processing research, quilt
It is widely used in analyzing the signal source that the multiple sensors such as radar imagery, visual image, one-dimensional radar return generate, such as Fig. 1 institute
Show.Vehicle violation form can be judged by tracking vehicle target；AT STATION, the public places such as square pass through tracking
Pedestrian target can detecte incident of violence and illegal activities；Target following technology is combined with Radar Technology, it can be to Hai Lu
It vacates existing some Unknown Subjects and carries out monitoring and behavioural analysis, prevent anomalous event.
Track algorithm can generally carve the position for providing target in signal at the beginning, and estimation analysis on this basis is subsequent
The kinematic parameter (including position, motion profile, speed etc.) of moment target.Target following result can effectively analyze the fortune of target
Dynamic state predicts the movement tendency of target, and provides foundation for high-level semantic identification.Existing method is dry in object variations, noise
Speed, precision and the robustness disturb, tracked when high-speed motion, partial occlusion is not met by the demand of practical application.
In order to solve Target Tracking Problem, researcher proposes big quantity algorithm or method, including template matching, mean-
Shift, particle filter, Kalman filtering etc..But existing method is not met by the demand of practical application, is primarily present with next
A little insufficient: (1) tracking may drift about when target Self-variation, noise jamming, high-speed motion, partial occlusion, and
Tracking is difficult to restore the tracking to specified target after generating drift；(2) target following real-time is poor；(3) target tracking accuracy
Fail to reach practical application request.
Summary of the invention
Technical problem to be solved by the invention is to provide a kind of method for tracking target based on signal filtering, to solve original
The problem of tracking accuracy is undesirable in some target tracking domains, real-time is poor, tracking poor robustness.
In order to solve the above technical problems, the present invention provides a kind of method for tracking target based on signal filtering, including such as
Lower step:
Input signal and output target degree of belief mapping are established into functional relation by filter model；
Estimated matrix is mapped using target degree of belief and initial filter model is calculated in current goal candidate region, and
Initial filter model and previous moment filter model weighted sum are obtained current more new model；
It is filtered using signal of the filter model to current goal candidate region and obtains the mapping of target degree of belief, with mesh
The maximum in degree of belief mapping is marked as current goal tracking result.
Further, calculating the initial filter model includes: the target's center position for obtaining measured signal t frame,
The current object candidate area is characterized using the rectangular window of fixed size, degree of belief mapping estimated matrix is established, using quick
The initial filter model of measured signal t frame is calculated in Fourier transformation and inverse transformation.
Further, degree of belief mapping estimated matrix is established using Gaussian function or Bessel function.
Further, it includes: current according to the interception of previous moment target's center position for obtaining the candidate region of current goal
The window at moment estimates the region at current time as target candidate as object candidate area, or according to target trajectory
Region.
Further, before being filtered to current goal candidate region signal, pre-treatment step, institute also are carried out to signal
Stating pretreatment includes that signal space transformation and/or signal denoising processing and/or signal characteristic abstraction are carried out to signal.
Further, after obtaining current initial filter model, then the model is smoothed, it is described smooth
Processing uses following calculation method:Wherein,
And ft model(x) t-1 frame and the final filter model of t frame, f are respectively indicatedtIt (x) is the initial model of t frame, and fixed
JusticeIt is weighted factor for 0, ρ, the value of ρ is greater than 0 less than 1.
Further, the degree of belief mapping estimated matrix remains unchanged during entire tracking.
Further, when t+1 frame signal arrives, the mesh of the filter model that is updated using t frame to t+1 frame
Mark candidate region is filtered and can map in the hope of the target degree of belief of t+1 frame, it may be assumed that
Wherein,Indicate the target of t+1 frame
Candidate region, t+1 frame signal intercept rectangular window as object candidate area using the target position of t frame,For
The target degree of belief of t+1 frame maps.
Compared with prior art, a kind of method for tracking target based on signal filtering of the present invention, has reached as follows
Effect:
1) present invention uses filter model online updating mode, and filter model is made to can adapt to the variation of target, solves
The certainly relatively low problem of target tracking accuracy；
2) present invention simplifies the calculating of filter model update and the mapping of target degree of belief using Fast Fourier Transform (FFT)
Journey solves the problems, such as that target following real-time is poor；
3) present invention updates filter model by smoother mode, prevents model variation from excessively acutely generating target
Drift, solves the problems, such as target following poor robustness.
4) the new method for tracking target of one kind provided by the invention, can effectively improve the performance of target following various aspects,
Closer to practical application request, there is important theoretical research and practical application value.
Detailed description of the invention
Fig. 1 is the schematic diagram of the practical application of the method for tracking target disclosed in this invention based on signal filtering.
The step of Fig. 2 is a kind of method for tracking target based on signal filtering disclosed in this invention figure.
Fig. 3 is a kind of method for tracking target based on signal filtering disclosed in this invention in t frame filter model
The flow diagram of update.
Fig. 4 is a kind of trusting in t+1 frame target for method for tracking target based on signal filtering disclosed in this invention
Spend the flow diagram of mapping calculation.
Specific embodiment
Below in conjunction with attached drawing, invention is further described in detail, but not as a limitation of the invention.
Fig. 1 is the schematic diagram of the practical application of the method for tracking target disclosed in this invention based on signal filtering.
As shown in Fig. 2, a kind of method for tracking target based on signal filtering of the present invention, includes the following steps:
Step S1, input signal and output target degree of belief mapping are established by functional relation by filter model；
Step S2, estimated matrix is mapped using target degree of belief and initial filter is calculated in current goal candidate region
Model, and initial filter model and previous moment filter model weighted sum are obtained current more new model；
Step S3, acquisition target degree of belief is filtered using signal of the filter model to current goal candidate region to reflect
It penetrates, using the maximum in the mapping of target degree of belief as current goal tracking result.
The present invention can be applied to the tracking of target in one-dimensional signal (such as radar echo signal), and also can be applied to can
The tracking of target in the 2D signals such as visible image, radar image.Below by taking target following in visual image as an example, this is described in detail
The embodiment of invention.
In step 1, the present invention is using the candidate region of target as input signal, and each position occurs in candidate region
Target degree of belief mapping be used as output signal, then it is assumed that there are a filter models to be obtained by being filtered to input signal
Output signal is obtained, as shown in formula (1).
O in formulaconf(x) mapping of target degree of belief, i.e., the likelihood occurred in x position target are indicated.I (x) is the mesh of input
Candidate region is marked, f (x) is the filter model for needing to find.It can be seen that the present invention is by filter model input signal
Connection is established with output target degree of belief mapping, so that Target Tracking Problem is just converted into a signal filtering problem.
In step 2, estimated matrix is mapped using target degree of belief and initial filter is calculated in current goal candidate region
Wave device model, and initial filter model and previous moment filter model weighted sum are obtained current more new model, it is real
The update of existing filter model.
Specifically, referring to shown in Fig. 3, the update of filter model mainly includes the following steps:
1) (first frame image provides target by detection or artificial mode for the target's center position of acquisition t frame image
Center), and target area is indicated using the rectangular window of fixed size.Intercept target rectangle region It(x) it is used as filtering
Device model modification.Here why using target area rather than whole image is inputted as filter, be because of being removed in image
Target area also includes a large amount of environmental informations, these information are to Target Modeling and track useless, or even can generate interference.
2) degree of belief mapping estimated matrix O is established using Gaussian functionconf(x).Specifically, foundation and It(x) sizes such as
Matrix Oconf(x), and according to Gaussian function to Oconf(x) each position assignment in, it may be assumed that
Oconf(x)=a × exp (- (x-xcenter)2/b2) (2)
Wherein, xcenterRepresenting matrix Oconf(x) center position coordinates, a, b are the coefficient of Gaussian function.It requires emphasis
Be degree of belief mapping estimated matrix Oconf(x) it is remained unchanged in entire tracking process.It should be noted that the present invention uses height
This function establishes degree of belief mapping estimated matrix, and degree of belief mapping estimation can also be established by other functions such as Bessel function
Matrix is not limited in any way herein.
3) product calculation that can be converted into due to the convolution algorithm in airspace in frequency domain, available according to formula (1):
F{Oconf(x) }=F { ft(x)}·F{It(x)} (3)
F { } indicates the Fast Fourier Transform (FFT) of two-dimensional discrete signal, representing matrix dot-product operation in above formula.Exchange equation
(3) every available following expression in position:
Wherein, F-1The Fast Fourier Transform Inverse of { } expression 2D signal.It(x) and Oconf(x) matrix brings formula into
It (4) can be in the hope of the initial model f of t framet(x)。
4) for the target that is tracked in actual scene there are deformation, block, visual angle change situations such as.In order to improve target
The robustness of tracking prevents tracking from drifting about, the present invention more new model in such a way that one kind is smoother, it may be assumed that
Wherein,And ft model(x) t-1 frame and the final filter model of t frame are respectively indicated, and is definedIt is 0.ρ is weighted factor, and value is greater than 0 less than 1.It is emphasized that initial model ft(x) it does not apply directly
In target following, ft model(x) it is only the filter model that t+1 frame image is used.
The present invention is updated using filter model and signal filtering recurrence executive mode, realization 2D signal (visual image,
Radar imagery) or one-dimensional signal (radar echo signal) in arbitrary target tracking, use signal filtering technique and quick Fu
In leaf transformation simplify the calculating process of former track algorithm, to guarantee the real-time of algorithm.Using filter model online updating
With smooth update mode, guarantee that model can adapt to the variation of target, but is unlikely to excessive variation and generates tracking drift.
In step 3, the main calculating for realizing target degree of belief, i.e., using filter model to current goal candidate regions
The signal in domain is filtered realization to target degree of belief mapping calculation, using the maximum in the mapping of target degree of belief as current mesh
Mark tracking result.
Specifically, the detailed process of target degree of belief mapping calculation and acquisition target's center position is as shown in Figure 4.
When t+1 frame image arrives, the object candidate area of the filter model that is updated using t frame to t+1 frame
Being filtered can map in the hope of the target degree of belief of t+1 frame, it may be assumed that
Wherein,Indicate the object candidate area of t+1 frame.Assuming that adjacent two field pictures target will not generate it is larger
Displacement, so t+1 frame image intercepts rectangular window as object candidate area using the target position of t frame.For
The target degree of belief of t+1 frame maps.In addition, the present invention intercepts the window at current time as target according to previous moment position
Candidate region such as can also estimate current target candidate region according to target trajectory using other strategies.
The present invention defines t+1 frame target's center position as the mapping of target degree of beliefThe corresponding seat of maximum
Mark, it may be assumed that
The target following of t+1 frame image is can be obtained as a result, can count according to above-mentioned steps S2 by above formula (7)
Calculate the filter model of t+1 frameAbove-mentioned step S2 and step S3 be alternately performed be achieved that in consecutive image appoint
The tracking for target of anticipating.
The method for tracking target based on signal filtering proposed through the invention, can be applied individually to any the auxiliary of imaging radar
The work such as analysis, video clipping can be used for forming more complicated system, such as non-cooperating formula target recognition and tracking, video
Monitoring, Activity recognition, automobile assistant driving, human-computer interaction, recognition of face etc..
Filter model proposed by the present invention can characterize target signature well, and be adapted to the variation of target, from
And effectively raise the precision and robustness of target following.In addition, the invention is simplified using Fast Fourier Transform (FFT) and was calculated
Journey greatly improves the processing speed of track algorithm.Compared to existing method, method proposed by the present invention is needed closer to practical application
It asks, has widened the application range of target tracking algorism.
Several preferred embodiments of the invention have shown and described in above description, but as previously described, it should be understood that the present invention
Be not limited to forms disclosed herein, should not be regarded as an exclusion of other examples, and can be used for various other combinations,
Modification and environment, and the above teachings or related fields of technology or knowledge can be passed through within that scope of the inventive concept describe herein
It is modified.And changes and modifications made by those skilled in the art do not depart from the spirit and scope of the present invention, then it all should be in this hair
In the protection scope of bright appended claims.
Claims (7)
1. a kind of method for tracking target based on signal filtering, which comprises the steps of:
Input signal and output target degree of belief mapping are established into functional relation by filter model, the input signal is mesh
Mark candidate region signal；
Estimated matrix is mapped using target degree of belief and initial filter model is calculated in current goal candidate region, and at the beginning of
Beginning filter model and previous moment filter model weighted sum obtain current more new model；
The signal of current goal candidate region is filtered using the more new model and obtains the mapping of target degree of belief, and with mesh
The corresponding coordinate of maximum in degree of belief mapping is marked as current goal tracking result；
Wherein, the calculating process of the target degree of belief mapping includes:
When t+1 frame signal arrives, the filter model that is updated using t frame to the object candidate area of t+1 frame into
Row filtering acquires the target degree of belief mapping of t+1 frame.
2. method for tracking target as described in claim 1, which is characterized in that calculate the initial filter model and further wrap
It includes: obtaining the target's center position of measured signal t frame, current goal candidate regions are characterized using the rectangular window of fixed size
Domain establishes target degree of belief mapping estimated matrix, measured signal t frame is calculated using Fast Fourier Transform (FFT) and inverse transformation
Initial filter model；
The Fast Fourier Transform (FFT) is calculated using following:
F{Oconf(x) }=F { ft(x)}·F{It(x)}；
The Fast Fourier Transform Inverse is calculated using following:
Wherein, Oconf(x) indicate that degree of belief maps estimated matrix, It(x) target rectangle region, f are indicatedt(x) t frame is indicated
Initial model.
3. method for tracking target as claimed in claim 2, which is characterized in that establish degree of belief mapping estimated matrix using Gauss
Function or Bessel function, and degree of belief mapping estimated matrix remains unchanged during entire tracking.
4. method for tracking target as described in claim 1, which is characterized in that further wrap the candidate region for obtaining current goal
It includes: being transported according to the window that previous moment target's center position intercepts current time as object candidate area, or according to target
Dynamic rail mark estimates the region at current time as object candidate area.
5. method for tracking target as claimed in claim 2, which is characterized in that after obtaining current initial filter model, then
The model is smoothed, the smoothing processing uses following calculation method:
Wherein,xThe position coordinates of target area are expressed as,
And ft model(x) t-1 frame and the final filter model of t frame, f are respectively indicatedtIt (x) is the initial filter mould of t frame
Type, and defineIt is weighted factor for 0, ρ, the value of ρ is greater than 0 less than 1.
6. method for tracking target as described in claim 1, which is characterized in that the target degree of belief mapping of the t+1 frame
Calculation formula are as follows:Wherein,Indicate t+1 frame
Object candidate area signal, t+1 frame signal intercepts rectangular window as object candidate area using the target position of t frame,It is mapped for the target degree of belief of t+1 frame.
7. method for tracking target as described in claim 1, which is characterized in that filtered to current goal candidate region signal
Wavefront further includes that pretreated step is carried out to signal, the pretreatment include signal space transformation is carried out to signal, and/or
Signal denoising processing and/or signal characteristic abstraction.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510529167.1A CN105139424B (en) | 2015-08-25 | 2015-08-25 | Method for tracking target based on signal filtering |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510529167.1A CN105139424B (en) | 2015-08-25 | 2015-08-25 | Method for tracking target based on signal filtering |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105139424A CN105139424A (en) | 2015-12-09 |
| CN105139424B true CN105139424B (en) | 2019-01-18 |
Family
ID=54724756
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510529167.1A Active CN105139424B (en) | 2015-08-25 | 2015-08-25 | Method for tracking target based on signal filtering |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105139424B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180275265A1 (en) * | 2017-03-23 | 2018-09-27 | GM Global Technology Operations LLC | Target tracking using region covariance |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1741069A (en) * | 2005-09-22 | 2006-03-01 | 上海交通大学 | Probability video tracing method based on adaptive surface model |
| CN103106667A (en) * | 2013-02-01 | 2013-05-15 | 山东科技大学 | Motion target tracing method towards shielding and scene change |
| CN104200485A (en) * | 2014-07-10 | 2014-12-10 | 浙江工业大学 | Video-monitoring-oriented human body tracking method |
-
2015
- 2015-08-25 CN CN201510529167.1A patent/CN105139424B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1741069A (en) * | 2005-09-22 | 2006-03-01 | 上海交通大学 | Probability video tracing method based on adaptive surface model |
| CN103106667A (en) * | 2013-02-01 | 2013-05-15 | 山东科技大学 | Motion target tracing method towards shielding and scene change |
| CN104200485A (en) * | 2014-07-10 | 2014-12-10 | 浙江工业大学 | Video-monitoring-oriented human body tracking method |
Non-Patent Citations (3)
| Title |
|---|
| Current Statistical Model Probability Hypothesis Density Filter for Multiple Maneuvering Targets Tracking;Mengjun Jin,et al.;《International Conference on Wireless Communications & Signal Processing》;20091231;1-5 |
| 一种基于粒子滤波的被动多传感器多目标跟踪算法;赵 欣，姬红兵，杨柏胜;《系统仿真学报》;20091031;第21卷(第20期);6382-6387 |
| 基于概率假设密度滤波方法的多目标跟踪技术综述;杨峰，王永齐，梁彦，潘泉;《自动化学报》;20131130;第39卷(第11期);1944-1956 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105139424A (en) | 2015-12-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhao et al. | Vehicle-logo recognition based on modified HU invariant moments and SVM | |
| CN102855459A (en) | Method and system for detecting and verifying specific foreground objects | |
| CN104484890A (en) | Video target tracking method based on compound sparse model | |
| Attari et al. | Target tracking formulation of the SVSF with data association techniques | |
| WO2022099528A1 (en) | Method and apparatus for calculating normal vector of point cloud, computer device, and storage medium | |
| Chen et al. | A novel spatial–temporal detection method of dim infrared moving small target | |
| CN102521811B (en) | The SAR image method for reducing speckle estimated based on anisotropy parameter and mutual information homogeneity | |
| CN105139424B (en) | Method for tracking target based on signal filtering | |
| Thomas et al. | Fast approach for moving vehicle localization and bounding box estimation in highway traffic videos | |
| Balasundaram et al. | Vision based motion tracking in real time videos | |
| CN111739047A (en) | Tongue image segmentation method and system based on bispectrum reconstruction | |
| Lin et al. | Infrared moving small target detection and tracking algorithm based on feature point matching | |
| Ferrer et al. | Two-steps perceptual important points estimator in 8-connected curves from handwritten signature | |
| Kumar et al. | Log Gabor filter based feature detection in image verification application | |
| Masood et al. | A novel technique for recognition and tracking of moving objects based on E-MACH and proximate gradient (PG) filters | |
| Yanik et al. | Computationally efficient FBP-type direct segmentation of synthetic aperture radar images | |
| Hongyin et al. | ISAR imaging based on improved phase retrieval algorithm | |
| Lucena et al. | Optical flow-based observation models for particle filter tracking | |
| CN110033474B (en) | Target detection method, target detection device, computer equipment and storage medium | |
| CN104751448A (en) | Online video tracking method based on PCA (Principal Component Analysis) and noise separation | |
| Heideklang et al. | A Novel Flexible Model for the Extraction of Features from Brain Signals in the Time‐Frequency Domain | |
| CN109359663A (en) | A kind of video tracing method returned based on color cluster and space-time canonical | |
| Wu et al. | Research on vehicle tracking algorithm using Contourlet transform | |
| CN108257156A (en) | A kind of method of the automatic tracing target object from video | |
| Zhang et al. | An Iris Location Algorithm Based on Gray Projection and Hough Transform |
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 |