CN102706345B - Maneuvering target tracking method based on fading memory sequential detector - Google Patents
Maneuvering target tracking method based on fading memory sequential detector Download PDFInfo
- Publication number
- CN102706345B CN102706345B CN201210190065.8A CN201210190065A CN102706345B CN 102706345 B CN102706345 B CN 102706345B CN 201210190065 A CN201210190065 A CN 201210190065A CN 102706345 B CN102706345 B CN 102706345B
- Authority
- CN
- China
- Prior art keywords
- motor
- driven
- state
- target
- model
- 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
Landscapes
- Feedback Control In General (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses a maneuvering target tracking method based on a fading memory sequential detector. Aiming at advantages and disadvantages of a decision-based single-model tracking method and a multi-model tracking method, the invention provides the maneuvering target tracking method based on the fading memory sequential detector. The method comprises the following steps of: performing Kalman filter-unscented Kalman filter (KF-UKF) combined filtration, wherein Kalman filtration based on a uniform model is adopted for a non-maneuvering target; when a maneuvering target is detected, expanding the dimension of a coefficient of variation (CV) model, and performing state estimation by adopting an unscented Kalman filter (UKF); detecting maneuvering by using the fading memory sequential detector; and after the maneuvering is detected, correcting a dimension expanded state vector and a covariance matrix, so that the matched state is matched with the actual moving mode. By the method, the state of the target can be corrected in real time, and matched filters can be selected according to different moving modes, so that the tracking precision is improved.
Description
Technical field
The invention belongs to motor-driven detection and maneuvering target tracking field, relate to a kind of maneuvering target tracking method based on Attenuation Memory Recursive sequential detector.
Background technology
In recent years, all have a wide range of applications in the military-civil fields such as maneuvering target tracking traffic control aloft (ATC), ground investigation supervision, homing.The difficulty of maneuvering target tracking mainly comes from the uncertainty of target travel and state observation.The uncertain uncertainty referring to target identities of observation, belongs to the category of multiple target tracking; The uncertainty of target travel mainly refers to the uncertainty of target movement model, mainly comprise: unknown motor-driven input and process noise statistical nature, these factors all can cause trace model and target travel mode mismatch, cause tracking performance to decline, and can cause filtering divergence time serious.
Current, the research of various maneuvering target tracking is all that the motor-driven uncertainty of surrounding target is launched, and becomes the main maneuvering target tracking method of two classes based on the single model method of decision-making and multi-model process.
Because maneuvering target motor pattern is represented by model usually, in fact the target travel mode decision thus in maneuvering target tracking is exactly target movement model decision-making.It is one of technology relevant to model decision in current maneuvering target tracking that target maneuver detects.In tracing process, after target travel pattern changes, usually can cause the mismatch of Filtering Model and motor pattern, cause filtering error to increase, must to be made Filtering Model used by corresponding mode decision process and adjusting timely.Motor-driven detection is exactly by detecting motor-driven generation, then adjusts a kind of decision-making technic that trace model mates corresponding mode of motion.
Although multi-model process enjoys favor with its good tracking performance and robustness, its design process is complicated and need great computational resource when realizing.In many resource-constrained occasions, the single model method based on decision-making has to be applied widely.Nearest research shows, the motor-driven detection method of no-float, and the single model method based on decision-making can obtain the tracking performance close with multi-model process.Thus timely, correct motor-driven detection becomes the key point of this kind of method.
For the single model tracking based on decision-making, the experts and scholars of domestic and international target tracking domain and engineering technical personnel have carried out large quantifier elimination.But their detecting device used is mostly that existence common issue of this kind of detecting device is the selection of motor-driven detector window length based on the fixing sliding window detecting device of length of window.Under normal circumstances, detection probability increases with length of window and increases, but length of window increase can cause the increase of motor-driven testing lag, and this is huge on the impact of performance of target tracking; Although and length of window reduces to reduce motor-driven testing lag, false-alarm probability can increase thereupon, thus how selection window length is the difficult point of this kind of detecting device.
Summary of the invention
The object of the invention is to the deficiency for existing detecting device, devise a kind of Attenuation Memory Recursive sequential detector.And detecting device is merged in target tracking algorism, for motor-driven detection, target tracking domain research provide a kind of maneuvering target tracking method based on Attenuation Memory Recursive sequential detector.
Based on a maneuvering target tracking method for Attenuation Memory Recursive sequential detector, comprise the following steps:
Step 1, state covariance put initial value and Kalman filtering: when target is at linear uniform motion, and moving target adopts two-dimension Kalman filtering device, and system model adopts CV model, carries out Kalman filtering according to state equation and measurement equation.
Step 2, detector design: according to the likelihood ratio function of definition, design a kind of Attenuation Memory Recursive sequential detector based on Generalized Likelihood Ratio, described Attenuation Memory Recursive sequential detector is realized by Attenuation Memory Recursive Sequential Detection Algorithm.
Step 3, input are estimated and state compensation: utilize Attenuation Memory Recursive sequential detector to detect the time of motor-driven generation, estimate afterwards to motor-driven amplitude.According to the latitude estimated, new breath and covariance are revised.
Step 4, be augmented and revise in real time with state: when
kmoment is when detecting motor-driven, to [
n,k] state in the time interval compensates, be then augmented state vector and covariance matrix, and revise in real time CV model according to the size of the latitude estimated.
Step 5, exit motor-driven differentiation: under given level of significance, if discriminant is set up, then think that the generation of acceleration is not by the motor-driven generation of target, can think that target is in non-maneuver state.
Step 6, Unscented Kalman Filter are estimated: when target occurs motor-driven, and adopt Unscented Kalman Filter device to carry out filtering to dbjective state, first Unscented Kalman Filter process carries out Unscented transform, then carry out filtering by the state variable after conversion.
Beneficial effect of the present invention: the present invention utilizes observation data to be the condition that order obtains, propose the decay sequential detector based on Generalized Likelihood Ratio, the benefit of Attenuation Memory Recursive make use of not data in the same time different this is theoretical on the impact of motor-driven differentiation, make to close on the impact of time data on motor-driven differentiation larger, moment far away is less on motor-driven differentiation impact, detecting device precision can be improved, shorten motor-driven testing lag.The motor-driven detecting device of design is integrated in tracker simultaneously, devise the maneuvering target tracking method based on Attenuation Memory Recursive sequential detector, while method adopts the method for Kalman filter and Unscented Kalman Filter device Federated filter to carry out filtering, target adopts Kalman filter when being in non-maneuver state, when target is in maneuvering condition, adopt Unscented Kalman Filter device to carry out filtering, improve filtering accuracy.This method is for process maneuvering target track question, particularly for the motor-driven motor pattern alternate with non-maneuver, this method has good tracking performance, the maneuvering target tracking method of simultaneously comparing based on multi-model, The method reduces calculated amount, improve the practicality of method.The present invention has stronger motor-driven detection perform and practicality.
Accompanying drawing explanation
Fig. 1 is the inventive method process flow diagram.
Embodiment
Below in conjunction with accompanying drawing, the invention will be further described.
As shown in Figure 1, the present invention includes following steps:
Step 1, state covariance put initial value and Kalman filtering: when target is at linear uniform motion, and moving target adopts two-dimension Kalman filtering device, and system model adopts CV model, carries out Kalman filtering according to state equation and measurement equation.
Step 2, detector design: according to the likelihood ratio function of definition, design a kind of Attenuation Memory Recursive sequential detector based on Generalized Likelihood Ratio.Described Attenuation Memory Recursive sequential detector is realized by Attenuation Memory Recursive Sequential Detection Algorithm.In target maneuver detects, observation data sequentially obtains often, thus compared to the sliding window detecting device of batch processing, Sequential detect is applicable to application more with motor-driven detection, and Sequential detect is without the need to presetting detection window length, thus eliminate an important problem in the detection of sliding window.
Step 3, input are estimated and state compensation: utilize Attenuation Memory Recursive sequential detector to detect the time of motor-driven generation, estimate afterwards to motor-driven amplitude.According to the latitude estimated, new breath and covariance are revised.
Step 4, be augmented and revise in real time with state: when detecting motor-driven in the k moment, to [
n,k] state in the time interval compensates, be then augmented state vector and covariance matrix, and revise in real time CV model according to the size of the latitude estimated.
Step 5, exit motor-driven differentiation: under given level of significance, if discriminant is set up, then think that the generation of acceleration is not by the motor-driven generation of target, can think that target is in non-maneuver state.
Step 6, Unscented Kalman Filter are estimated: when target occurs motor-driven, and adopt Unscented Kalman Filter device to carry out filtering to dbjective state, first Unscented Kalman Filter process carries out Unscented transform, then carry out filtering by the state variable after conversion.
In step 1, to put the concrete steps of initial value and Kalman filtering as follows for state covariance:
When target is at linear uniform motion, moving target adopts two-dimension Kalman filtering device, and system model adopts CV model.Kalman filtering is carried out according to state equation and measurement equation.
Initial covariance is:
(1)
In formula,
for measuring noise square difference,
for the sampling period.Because
for the covariance of location components, size equals the covariance of measurement noise.
,
for the covariance of speed component, can be obtained by the relationship of speed and position,
=
=
.
for the covariance of component of acceleration, by the relationship of position and acceleration,
=
, in like manner can obtain,
.Now just can obtain initial covariance matrix.
In step 2, the concrete steps of detector design are as follows:
When hypothetical target is motor-driven, state equation is:
(2)
Wherein, c is input control matrix,
for the latitude size in k moment,
for process noise.
with
all there is Gaussian characteristics.
First dualism hypothesis is done:
：
：
(3)
Wherein suppose
indicate without motor-driven, suppose
represent that appearance is motor-driven,
represent motor-driven amplitude.
Definition likelihood ratio function
(4)
Wherein,
for new breath, there is white Gaussian noise characteristic,
for the joint probability density function of breath new under motor-driven hypothesis.
The Attenuation Memory Recursive Sequential Detection Algorithm of recursive form is as follows:
，
(5)
Wherein,
for decay factor,
value can rule of thumb draw.
According to formula (5) and 0 relation, detection threshold can be set to 0, probability density function Gaussian distributed, motor-driven decision rule is as follows:
(1) if
, then accept
suppose namely exist motor-driven, so motor-driven initial time is:
, carry out state compensation afterwards;
(2) if
then continue checkout procedure
.
In step 3, input is estimated with the concrete steps of state compensation as follows:
Utilize Attenuation Memory Recursive sequential detector to detect the time of motor-driven generation, estimate afterwards to latitude, suppose that the n moment occurs motor-driven, the k moment detects motor-driven generation, being then estimated as of latitude:
(6)
Wherein,
for the k moment state-transition matrix,
for motor-driven error co-variance matrix,
for new breath.
The state estimation in a sampling period of Kalman filter filtering and covariance is adopted to be estimated as:
(7)
Wherein,
,
for filter gain,
for state one-step prediction covariance matrix,
for newly ceasing covariance matrix.
Therefore [
n,k] state revision in the time interval is as follows:
(8)
In formula (8), [
n,k] correction of state in the time interval, contain the time interval analytic relationship corresponding to latitude size, because dbjective state is sextuple, comprise, position, speed, acceleration aliquot, and the size of latitude has been exactly the size of acceleration.Therefore latitude and the time interval meet the analytic relationship of formula (8).
New breath is revised:
(9)
Covariance correction:
(10)
Wherein
for the error co-variance matrix of motor-driven amplitude.
,
,
be with [
n ... k] between the relevant matrix of coefficients of state.
The concrete steps revised in real time with state are augmented as follows in step 4:
When detecting motor-driven in the k moment, and to [
n,k] after state in the time interval compensates, finally state vector and covariance matrix are augmented, and according to the size of the latitude estimated, model are revised in real time.
(11)
(12)
Wherein,
for the estimation of latitude.
The concrete steps exiting motor-driven differentiation in step 5 are as follows:
The normalization square of input amplitude is
(13)
[
n,k] in the time interval, the normalization quadratic sum of amplitude is
(14)
Discriminant:
(15)
In given level of significance
under, if meet above formula, then think acceleration not by the motor-driven generation of target, do non-maneuver process under this kind of situation, can think that target is in non-maneuver state.
The concrete steps that in step 6, Unscented Kalman Filter device is estimated are as follows:
In the present invention, namely UKF wave filter is Unscented Kalman Filter device.
First UKF carries out tasteless conversion (UT) when treatment state equation, then uses the state variable after tasteless conversion to carry out filtering estimation, to reduce evaluated error.
Calculate 2n+1
point
, i=1,2 ..., 2n, wherein n is state vector dimension.
(16)
In formula
,
determine
spread of points degree, usually get one less on the occasion of, as 0.01,
usually 0 is got.
Calculate
point
, i=1,2 ..., 2n, the propagation by state equation:
(17)
In formula
.
Calculate
point
,
propagation by measurement equation:
(18)
Calculate the step look-ahead exported:
(19)
In new measurement amount
filtering renewal is carried out after arrival:
(20)
Wherein
it is filter gain matrix.
The main flow of the inventive method is: hypothetical target is initial to move with uniform velocity, first according to measuring value, initial value is put to state vector and covariance matrix, adopt and based on the two-dimension Kalman filtering device of CV model, state is estimated, utilize Attenuation Memory Recursive sequential detector to carry out motor-driven detection simultaneously.If do not have motor-driven, return the state estimation proceeding next cycle; If detect motor-driven, then the time of motor-driven amplitude and motor-driven generation is estimated.Consider the time delay of motor-driven detection, now filter state and covariance are compensated and corrected.Finally according to revised state and motor-driven amplitude, CV model is augmented, obtains a model revised in real time, carry out UKF filtering according to real-time revised model afterwards.Repeat motor-driven testing process, if detect motor-driven again, then carry out input and estimate and state compensation, real-time update is carried out to model simultaneously, if do not detect motor-driven, then carry out exiting motor-driven judgement, if motor-driven amplitude is greater than thresholding, then represents and do not exit motor-driven, repeat UKF filtering, if be less than thresholding, then think that target is in non-maneuver state, now, adopt the Kalman filtering of two dimension, repeat above process, constitute the maneuvering target tracking method based on Attenuation Memory Recursive sequential detector.
Claims (1)
1., based on a maneuvering target tracking method for Attenuation Memory Recursive sequential detector, it is characterized in that the method comprises the following steps:
Step 1, state covariance put initial value and Kalman filtering: when target is at linear uniform motion, and moving target adopts two-dimension Kalman filtering device, and system model adopts CV model, carries out Kalman filtering according to state equation and measurement equation;
Step 2, detector design: according to the likelihood ratio function of definition, design the Attenuation Memory Recursive sequential detector based on Generalized Likelihood Ratio, described Attenuation Memory Recursive sequential detector is realized by Attenuation Memory Recursive Sequential Detection Algorithm;
Step 3, input are estimated and state compensation: utilize Attenuation Memory Recursive sequential detector to detect the time of motor-driven generation, estimate afterwards to motor-driven amplitude; According to the latitude estimated, new breath and covariance are revised;
Step 4, be augmented and revise in real time with state: when detecting motor-driven in the k moment, to [n, k] state in the time interval compensates, is then augmented state vector and covariance matrix, and revises in real time CV model according to the size of the latitude estimated;
Step 5, exit motor-driven differentiation: under given level of significance, if discriminant is set up, then think that the generation of acceleration is not by the motor-driven generation of target, can think that target is in non-maneuver state;
Step 6, Unscented Kalman Filter are estimated: when target occurs motor-driven, and adopt Unscented Kalman Filter device to carry out filtering to dbjective state, first Unscented Kalman Filter process carries out Unscented transform, then carry out filtering by the state variable after conversion;
Step 2 is specifically:
First dualism hypothesis is done:
H
0：u
m＝0 m∈[1,k]
Wherein suppose H
0indicate without motor-driven, suppose H
1represent that appearance is motor-driven, u
mrepresent motor-driven amplitude, m represents the moment;
Definition likelihood ratio function L
k
Wherein,
for new breath, there is white Gaussian noise characteristic,
for the joint probability density function of breath new under motor-driven hypothesis;
The Attenuation Memory Recursive Sequential Detection Algorithm of recursive form is as follows:
Wherein, α is decay factor,
According to formula (3) and 0 relation, detection threshold is set to 0, probability density function Gaussian distributed, motor-driven decision rule is as follows:
(1) if L
k>0, then accept H
1suppose namely exist motor-driven, so motor-driven initial time is:
Carry out state compensation afterwards;
(2) if L
k≤ 0 is continued checkout procedure k → k+1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210190065.8A CN102706345B (en) | 2012-06-11 | 2012-06-11 | Maneuvering target tracking method based on fading memory sequential detector |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210190065.8A CN102706345B (en) | 2012-06-11 | 2012-06-11 | Maneuvering target tracking method based on fading memory sequential detector |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102706345A CN102706345A (en) | 2012-10-03 |
| CN102706345B true CN102706345B (en) | 2015-01-28 |
Family
ID=46899297
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210190065.8A Active CN102706345B (en) | 2012-06-11 | 2012-06-11 | Maneuvering target tracking method based on fading memory sequential detector |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102706345B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103901426A (en) * | 2012-12-28 | 2014-07-02 | 重庆凯泽科技有限公司 | Single-maneuvering target tracking system |
| CN103744058A (en) * | 2013-12-24 | 2014-04-23 | 河海大学 | Ballistic trajectory formation method based on exponential weighting attenuated memory filtering |
| CN104180801B (en) * | 2014-08-20 | 2017-06-06 | 广州海格通信集团股份有限公司 | Forecasting Methodology and system based on ADS B system track points |
| CN106910211A (en) * | 2015-12-21 | 2017-06-30 | 中国石油天然气股份有限公司 | Multiple maneuver target tracking methods under complex environment |
| CN106772351B (en) * | 2016-11-16 | 2019-04-23 | 西安电子科技大学 | Kalman filter method based on the memory of limited step |
| CN108332939B (en) * | 2017-12-31 | 2020-04-07 | 中国科学院声学研究所 | Cross-medium time-varying fluid parameter online estimation method |
| CN108629935B (en) * | 2018-05-17 | 2020-03-24 | 山东深图智能科技有限公司 | Method and system for detecting burglary of climbing stairs and turning windows based on video monitoring |
| CN113155122A (en) * | 2021-04-01 | 2021-07-23 | 广州大学 | Maneuvering target tracking method based on adaptive filtering |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2128644A1 (en) * | 2007-03-22 | 2009-12-02 | Furuno Electric Co., Ltd. | Gps composite navigation apparatus |
| CN102043906A (en) * | 2010-12-15 | 2011-05-04 | 北京航空航天大学 | Diagonal-form fusion method for maneuvering target tracking |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE102007018190B3 (en) * | 2007-04-18 | 2009-01-22 | Lfk-Lenkflugkörpersysteme Gmbh | Method for ascertaining motion of target object involves utilizing semi-martingale algorithm based on model equations that represented by smooth semi-martingales for estimating motion |
-
2012
- 2012-06-11 CN CN201210190065.8A patent/CN102706345B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2128644A1 (en) * | 2007-03-22 | 2009-12-02 | Furuno Electric Co., Ltd. | Gps composite navigation apparatus |
| CN102043906A (en) * | 2010-12-15 | 2011-05-04 | 北京航空航天大学 | Diagonal-form fusion method for maneuvering target tracking |
Non-Patent Citations (2)
| Title |
|---|
| Sequential Detection of Target Maneuvers;Jifeng Ru;《2005 7th International Conference on Information Fusion(FUSION)》;20050728;第1卷;第346页 * |
| 基于新息序列修正的输入估计算法;盛琥等;《系统工程与电子技术》;20090930;第31卷(第9期);第2123-2124页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102706345A (en) | 2012-10-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102706345B (en) | Maneuvering target tracking method based on fading memory sequential detector | |
| CN103472445B (en) | Detecting tracking integrated method for multi-target scene | |
| US20230297642A1 (en) | Bearings-only target tracking method based on pseudo-linear maximum correlation entropy kalman filtering | |
| CN106407677B (en) | A kind of multi-object tracking method in the case of missing measurement | |
| CN105549049A (en) | Adaptive Kalman filtering algorithm applied to GPS navigation | |
| CN103383261B (en) | A kind of modified can't harm the indoor moving targets location method of Kalman filtering | |
| CN102322861B (en) | Flight path fusion method | |
| CN102568004A (en) | Tracking algorithm for high maneuvering targets | |
| CN104035083B (en) | A kind of radar target tracking method based on measurement conversion | |
| Deng et al. | Adaptive unscented Kalman filter for parameter and state estimation of nonlinear high-speed objects | |
| CN104155650A (en) | Object tracking method based on trace point quality evaluation by entropy weight method | |
| CN104237879A (en) | Multi-target tracking method in radar system | |
| Ma et al. | Target tracking system for multi-sensor data fusion | |
| CN107064865A (en) | The passive co-located method of polar coordinates Dynamic Programming clustered based on depth | |
| CN114090718B (en) | Bi-LSTM prediction and fuzzy analysis based interrupted track correlation method | |
| CN111711432B (en) | Target tracking algorithm based on UKF and PF hybrid filtering | |
| Straka et al. | Performance evaluation of local state estimation methods in bearings-only tracking problems | |
| CN101299271A (en) | Polynomial forecast model of maneuvering target state equation and tracking method | |
| Choi et al. | State estimation with delayed measurements considering uncertainty of time delay | |
| CN104021285B (en) | A kind of interactive multi-model method for tracking target with optimal motion pattern switching parameter | |
| CN110501686A (en) | A kind of method for estimating state based on NEW ADAPTIVE high-order Unscented kalman filtering | |
| CN110532517A (en) | Gas pipeline method for parameter estimation based on improved ARUKF | |
| CN107329131A (en) | A kind of radar dim target detection tracking of utilization particle filter | |
| CN105549003A (en) | Automobile radar target tracking method | |
| Wang et al. | An adaptive outlier-robust Kalman filter based on sliding window and Pearson type VII distribution modeling |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| EE01 | Entry into force of recordation of patent licensing contract | ||
| EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20121003 Assignee: HANGZHOU DAZHI ELECTROMECHANICAL Co.,Ltd. Assignor: HANGZHOU DIANZI University Contract record no.: X2020330000073 Denomination of invention: A maneuvering target tracking method based on attenuated memory sequential detector Granted publication date: 20150128 License type: Common License Record date: 20201010 |