Presentation is loading. Please wait.

Presentation is loading. Please wait.

Adaptive Signal Processing Class Project Adaptive Interacting Multiple Model Technique for Tracking Maneuvering Targets Viji Paul, Sahay Shishir Brijendra,

Published byNeil Burke Modified over 8 years ago

Similar presentations

Presentation on theme: "Adaptive Signal Processing Class Project Adaptive Interacting Multiple Model Technique for Tracking Maneuvering Targets Viji Paul, Sahay Shishir Brijendra,"— Presentation transcript:

1 Adaptive Signal Processing Class Project Adaptive Interacting Multiple Model Technique for Tracking Maneuvering Targets Viji Paul, Sahay Shishir Brijendra, Krishnamoorthy Iyer, Meles Gebreyesus

2 Adaptive Signal Processing Class Project Outline of the Presentation Introduction to the problem Multiple Model Technique IMM and the models used Multiple Target Scenario Simulation results Adaptive Cancellation of Oscillation Effect Simulation results

3 Adaptive Signal Processing Class Project Outline of the Presentation Introduction to the problem Multiple Model Technique IMM and the models used Multiple Target Scenario Simulation results Adaptive Cancellation of Oscillation Effect Simulation results

4 Adaptive Signal Processing Class Project Radar site Observed Position Predicted Position Error Target Tracking Using Surveillance radar

5 Adaptive Signal Processing Class Project Target Tracking Using Surveillance radar

6 Adaptive Signal Processing Class Project In the MM estimation, it is assumed that the possible system behavior patterns, called system modes, can be represented by a set of models. A bank of filters runs in parallel at every time, each based on a particular model, to obtain the model-conditional estimates. Overall state estimate is a certain combination of these model-conditional estimates. Multiple Model Estimation

7 Adaptive Signal Processing Class Project Multiple Model Estimation A Bayesian framework starting with prior probabilities of each model being correct (i.e. the system is in a particular mode) is used. The model, assumed to be in effect throughout the process, is one of r possible models (the system is in one of r modes)

8 Adaptive Signal Processing Class Project Multiple Model Estimation The prior probability that is correct (i.e. the system is in mode j) is j = 1…….r where is the prior information; and

9 Adaptive Signal Processing Class Project Multiple Model Estimation The output of each mode-matched filter: Mode-conditioned State Estimate Associated State Error Covariance Matrix Mode Likelihood Function

10 Adaptive Signal Processing Class Project Multiple Model Structure

11 Adaptive Signal Processing Class Project Output Estimate After the filters are initialized, they run recursively on their own estimates. Their likelihood functions are used to update the mode probabilities. The latest mode probabilities are used to combine the mode-conditioned estimates and covariances.

12 Adaptive Signal Processing Class Project Output Estimate The combination of mode-conditioned estimates is therefore done as follows And the covariance of the combined estimate is

13 Adaptive Signal Processing Class Project Multiple Model Approach for Switching Models Consider an example with two models, and at time, k =2 one has possible histories. 111 212 321 422

14 Adaptive Signal Processing Class Project Multiple Model Approach for Switching Models The mode history – or sequence of models – through time k is denoted as where is the model index at time k from history l. It is important to note that number of histories increases exponentially with time.

15 Adaptive Signal Processing Class Project Interacting Multiple Model Algorithm

16 Adaptive Signal Processing Class Project Steps in IMM One cycle of the algorithm consists of the following: Step 1: Calculation of the mixing probabilities. Step2: Mixing- Calculation of mixed initial conditions Step3: Mode matched filtering. Step4: Mode probability update. Step5: Estimate and covariance combination

17 Adaptive Signal Processing Class Project Block Diagram of the Tracking Routine Model j Kalman Filter j = 1…r IMM Block State estimates for established target State estimate covariance for established target Model probabilities for established target Sensor measurements

18 Adaptive Signal Processing Class Project Tracking Maneuvering target A weaving target track constructed of linked coordinated turns

19 Adaptive Signal Processing Class Project IMM One cycle of the algorithm consists of the following: Step 1: Calculation of the mixing probabilities. The probability that mode was in effect at time k-1 given that is in effect at k, conditioned on is:

20 Adaptive Signal Processing Class Project IMM Algorithm The above are the mixing probabilities, which can be written as Where the normalizing constants are j = 1,…,r.

21 Adaptive Signal Processing Class Project IMM Algorithm Step 2: Mixing. Starting with one computes the mixed initial condition for the filter matched to j = 1,…,r

22 Adaptive Signal Processing Class Project IMM Algorithm The covariance corresponding to the above is

23 Adaptive Signal Processing Class Project IMM Algorithm Step 3: Mode-matched filtering. The estimate and covariance are used as input to the filter matched to which uses to yield and The likelihood functions corresponding to the r filters are computed using the mixed initial condition and the associated covariance

24 Adaptive Signal Processing Class Project IMM Algorithm Step 4: Mode probability update. This is done as follows

25 Adaptive Signal Processing Class Project IMM Algorithm Step 5: Estimate and covariance combination. Combination of the model-conditioned estimates and covariances is done according to the mixture equations

26 Adaptive Signal Processing Class Project There are total four targets moving with different kinematics. Initial Range : Range of the target at time zero. Initial Theta : Azimuth of the target at time zero. Split no : Number of splits into which time is partitioned. No of Scans : Number of scans in each time split. Start Scan : Starting scan number of each partition of time. End Scan : Ending scan number of each partition of time. IMM Estimator for Tracking Multiple Targets: Parameters used for scenario generation

27 Adaptive Signal Processing Class Project Turn Rate : Amount of course change in degree per second. Velocity : Velocity in each partition of time. Acceleration : Acceleration in each partition of time. Heading : Heading in each partition of time. IMM Estimator for Tracking Multiple Targets: Parameters used for scenario generation

28 Adaptive Signal Processing Class Project --- Target data ---1 Initial Range : 10000.000000. Initial Theta : -0.785398. Split no : 1.0000002.0000003.0000004.0000005.0000006.0000007.000000. Scan no : 20.00000031.00000017.00000031.00000020.00000031.00000014.000000. Start Scan : 0.00000021.00000053.00000071.000000103.000000124.000000156.000000. End Scan : 20.00000052.00000070.000000102.000000123.000000155.000000170.000000. Turn Rate : 0.0000009.0000000.0000009.0000000.0000009.0000000.000000. Velocity : 150.000000150.000000150.000000150.000000150.000000150.000000150.000000. Accelaration : 0.00000023.5241140.00000023.5241140.00000023.5241140.000000. Heading : -0.7853981.374447-0.7853981.374447-0.7853981.374447-0.785398. IMM Estimator for Tracking Multiple Targets: Parameters used for scenario generation

29 Adaptive Signal Processing Class Project --- Target data ---2 Initial Range : 10000.000000. Initial Theta : 0.785398. Split no : 1.000000. Scan no : 170.000000. Start Scan : 0.000000. End Scan : 170.000000. Turn Rate : 0.000000. Velocity : 100.000000. Accelaration : 0.000000. Heading : 0.785398. IMM Estimator for Tracking Multiple Targets: Parameters used for scenario generation

30 Adaptive Signal Processing Class Project --- Target data ---3 Initial Range : 10000.000000. Initial Theta : -2.356194. Split no : 1.0000002.0000003.0000004.000000. Scan no : 30.00000034.00000031.00000072.000000. Start Scan : 0.00000031.00000066.00000098.000000. End Scan : 30.00000065.00000097.000000170.000000. Turn Rate : 0.000000-9.0000009.0000000.000000. Velocity : 100.000000100.000000100.000000100.000000. Accelaration : 0.00000015.68274215.68274215.682742. Heading : -2.3561943.141593-2.552544-0.792379. IMM Estimator for Tracking Multiple Targets: Parameters used for scenario generation

31 Adaptive Signal Processing Class Project --- Target data ---4 Initial Range : 10000.000000. Initial Theta : 2.356194. Split no : 1.0000002.0000003.0000004.000000. Scan no : 30.00000034.00000031.00000072.000000. Start Scan : 0.00000031.00000066.00000098.000000. End Scan : 30.00000065.00000097.000000170.000000. Turn Rate : 0.000000-9.0000009.0000000.000000. Velocity : 100.000000100.000000100.000000100.000000. Accelaration : 0.00000015.68274215.68274215.682742. Heading : 2.3561941.5707962.159845-2.363176. IMM Estimator for Tracking Multiple Targets: Parameters used for scenario generation

32 Adaptive Signal Processing Class Project Maneuvering Models Constant Velocity Model For small sample intervals T, the following model is commonly used (Blackman & Popoli, Sec. 4.2.2):

33 Adaptive Signal Processing Class Project Maneuvering Models Constant Acceleration Model

34 Adaptive Signal Processing Class Project Maneuvering Models Coordinated Turn Model

35 Adaptive Signal Processing Class Project Multi Target Scenario

36 Adaptive Signal Processing Class Project Variation of Model weights For Target 1

37 Adaptive Signal Processing Class Project Variation of Model weights For Target 2

38 Adaptive Signal Processing Class Project Variation of Model weights For Target 3

39 Adaptive Signal Processing Class Project Variation of Model weights For Target 4

40 Adaptive Signal Processing Class Project Tracking From Unstable Platform The environment strongly impacts radar performance

41 Adaptive Signal Processing Class Project Platform Oscillations Roll, Yaw and Pitch Only Roll has been considered in this simulation. All the three motions are sinusoidal or DC shifted sinusoidal. At max, frequency of the sinusoid is about 1/10 Hz.

42 Adaptive Signal Processing Class Project Physically stabilized beam

43 Adaptive Signal Processing Class Project Tracking Maneuvering target A weaving target track constructed of linked coordinated turns. Perturbations are seen because of platform motion.

44 Adaptive Signal Processing Class Project Target Tracking Data Flow Estimated State Target Sensor (Obsvn Device) Signal / Data Pre- Processor Tracker (State Estimation / Data Association) Electro Magnetic or Acoustic Energy Channel Signal / Raw Data Tracking Algorithm Data Conversion Decoupling Detection- Subsystem Typical Target tracking system

45 Adaptive Signal Processing Class Project Measurement corrupted by Oscillations Increased deterioration at larger ranges.

46 Adaptive Signal Processing Class Project For Target 1 Model weight variations due to Platform Oscillations

47 Adaptive Signal Processing Class Project For Target 2 Model weight variations due to Platform Oscillations

48 Adaptive Signal Processing Class Project For Target 3 Model weight variations due to Platform Oscillations

49 Adaptive Signal Processing Class Project For Target 4 Model weight variations due to Platform Oscillations

50 Adaptive Signal Processing Class Project Target Tracking Data Flow with Adaptive Compensation Estimated State Target Sensor (Obsvn Device) Signal / Data Pre- Processor Tracker (State Estimation / Data Association) Electro Magnetic or Acoustic Energy Channel Signal / Raw Data Tracking Algorithm Data Conversion Decoupling Detection- Subsystem Typical Target tracking system Motion Sensor LMS based Algo

51 Adaptive Signal Processing Class Project Measurements corrupted by a proportional multiplication of oscillation

52 Adaptive Signal Processing Class Project Reminds You of Something ??? Output from the Gyro Modified form of Gyro output Signal from the radar

53 Adaptive Signal Processing Class Project Compensated for Platform Oscillation

54 Adaptive Signal Processing Class Project Its OK in Theory but is the Target a Sitting Duck ? Operational Solution: Sea state does not change drastically. Ships are always in formation during an operation. During the pre-detection phase, i.e. while approaching the Theatre of Operation, the weights of the Adaptive Filter can be “set” using the LMS Algorithm. The same weights can then be used during the Target Detection phase. Ship with Surv Radar Friendly Ship in Company

55 Adaptive Signal Processing Class Project Compensated for Platform Oscillation For Target 1

56 Adaptive Signal Processing Class Project Compensated for Platform Oscillation For Target 2

57 Adaptive Signal Processing Class Project Compensated for Platform Oscillation For Target 3

58 Adaptive Signal Processing Class Project Compensated for Platform Oscillation For Target 4

59 Adaptive Signal Processing Class Project Conclusion Analyzed Multiple Model Technique IMM based estimation is implemented Generated a Multi Target Scenario Applied IMM Verified the algorithm Introduced Platform Oscillations Added LMS based adaptive compensation

60 Adaptive Signal Processing Class Project Thank you

Download ppt "Adaptive Signal Processing Class Project Adaptive Interacting Multiple Model Technique for Tracking Maneuvering Targets Viji Paul, Sahay Shishir Brijendra,"

Similar presentations

TARGET DETECTION AND TRACKING IN A WIRELESS SENSOR NETWORK Clement Kam, William Hodgkiss, Dept. of Electrical and Computer Engineering, University of California,

TARGET DETECTION AND TRACKING IN A WIRELESS SENSOR NETWORK Clement Kam, William Hodgkiss, Dept. of Electrical and Computer Engineering, University of California,
Bayesian Belief Propagation

Bayesian Belief Propagation
ECE 8443 – Pattern Recognition ECE 8423 – Adaptive Signal Processing Objectives: The Linear Prediction Model The Autocorrelation Method Levinson and Durbin.

ECE 8443 – Pattern Recognition ECE 8423 – Adaptive Signal Processing Objectives: The Linear Prediction Model The Autocorrelation Method Levinson and Durbin.
VSMC MIMO: A Spectral Efficient Scheme for Cooperative Relay in Cognitive Radio Networks 1.

VSMC MIMO: A Spectral Efficient Scheme for Cooperative Relay in Cognitive Radio Networks 1.
Change Detection C. Stauffer and W.E.L. Grimson, “Learning patterns of activity using real time tracking,” IEEE Trans. On PAMI, 22(8):747-757, Aug 2000.

Change Detection C. Stauffer and W.E.L. Grimson, “Learning patterns of activity using real time tracking,” IEEE Trans. On PAMI, 22(8): , Aug 2000.
ECE 8443 – Pattern Recognition ECE 8423 – Adaptive Signal Processing Objectives: The FIR Adaptive Filter The LMS Adaptive Filter Stability and Convergence.

ECE 8443 – Pattern Recognition ECE 8423 – Adaptive Signal Processing Objectives: The FIR Adaptive Filter The LMS Adaptive Filter Stability and Convergence.
Reducing Drift in Parametric Motion Tracking

Reducing Drift in Parametric Motion Tracking
(Includes references to Brian Clipp

(Includes references to Brian Clipp
Adaptive Sampling for Sensor Networks Ankur Jain ٭ and Edward Y. Chang University of California, Santa Barbara DMSN 2004.

Adaptive Sampling for Sensor Networks Ankur Jain ٭ and Edward Y. Chang University of California, Santa Barbara DMSN 2004.
Probabilistic video stabilization using Kalman filtering and mosaicking.

Probabilistic video stabilization using Kalman filtering and mosaicking.
Adaptive Rao-Blackwellized Particle Filter and It’s Evaluation for Tracking in Surveillance Xinyu Xu and Baoxin Li, Senior Member, IEEE.

Adaptive Rao-Blackwellized Particle Filter and It’s Evaluation for Tracking in Surveillance Xinyu Xu and Baoxin Li, Senior Member, IEEE.
October 28, 2008Final Work Presentation1. October 28, 2008Final Work Presentation2.

October 28, 2008Final Work Presentation1. October 28, 2008Final Work Presentation2.
Tracking using the Kalman Filter. Point Tracking Estimate the location of a given point along a sequence of images. (x 0,y 0 ) (x n,y n )

Tracking using the Kalman Filter. Point Tracking Estimate the location of a given point along a sequence of images. (x 0,y 0 ) (x n,y n )
Prepared By: Kevin Meier Alok Desai

Prepared By: Kevin Meier Alok Desai
A Probabilistic Approach to Collaborative Multi-robot Localization Dieter Fox, Wolfram Burgard, Hannes Kruppa, Sebastin Thrun Presented by Rajkumar Parthasarathy.

A Probabilistic Approach to Collaborative Multi-robot Localization Dieter Fox, Wolfram Burgard, Hannes Kruppa, Sebastin Thrun Presented by Rajkumar Parthasarathy.
1 Integration of Background Modeling and Object Tracking Yu-Ting Chen, Chu-Song Chen, Yi-Ping Hung IEEE ICME, 2006.

1 Integration of Background Modeling and Object Tracking Yu-Ting Chen, Chu-Song Chen, Yi-Ping Hung IEEE ICME, 2006.
Tracking a maneuvering object in a noisy environment using IMMPDAF By: Igor Tolchinsky Alexander Levin Supervisor: Daniel Sigalov Spring 2006.

Tracking a maneuvering object in a noisy environment using IMMPDAF By: Igor Tolchinsky Alexander Levin Supervisor: Daniel Sigalov Spring 2006.
© 2003 by Davi GeigerComputer Vision November 2003 L1.1 Tracking We are given a contour   with coordinates   ={x 1, x 2, …, x N } at the initial frame.

© 2003 by Davi GeigerComputer Vision November 2003 L1.1 Tracking We are given a contour   with coordinates   ={x 1, x 2, …, x N } at the initial frame.
April 2004Dartmouth College 1 A Hybrid Tracker and Smoother for Highly Maneuvering Targets Stephen Linder This material is based on work supported by Dr.

April 2004Dartmouth College 1 A Hybrid Tracker and Smoother for Highly Maneuvering Targets Stephen Linder This material is based on work supported by Dr.
February 2001SUNY Plattsburgh Concise Track Characterization of Maneuvering Targets Stephen Linder Matthew Ryan Richard Quintin This material is based.

February 2001SUNY Plattsburgh Concise Track Characterization of Maneuvering Targets Stephen Linder Matthew Ryan Richard Quintin This material is based.

Similar presentations

Ads by Google