Title: Stanford CS223B Computer Vision, Winter 2006 Lecture 12 Filters / Motion Tracking 2
1Stanford CS223B Computer Vision, Winter
2006Lecture 12 Filters / Motion Tracking 2
- Professor Sebastian Thrun
- CAs Dan Maynes-Aminzade, Mitul Saha, Greg
Corrado
2Team Names
- noname
- Skynet
- AED
- WeeDrivers
- Fear Factor
- Dangerous Driver
- First Star on the Left
- Team Lee
- The Five-Seconds Rule
- Colorado
- Pien
- Triangulated
- Making Drivers Obsolete
- LyricalAssasins
- Gal
- Brad Evan
- Three Blind Mice
- Optical Optima
- The Visionaries
3- Online data
- cs223b.stanford.edu/competition
4USB Frash Drive
- team.txt -replace with you team name, number
- /data
- /data/clip1
- /data/clip2
- /data/clip3
- /results
- Will not be available
- /submit
- clip1.png
- clip2.png
- clip3.png
5Competition
- 5 seconds of driving (150 frames)
- 3 seconds of blackout
6Important Information
- Memorize your team number
- Pick up flash drive today (SCPD will post on
Web) - Add files in submit directory
- Change team.txt
- Return on Monday
7Moving Objects
8Kalman Filter Tracking
9Particle Filter Tracking
10Mixture of KF / PF (Unscented PF)
11Review Kalman Filters
prior
12Summary Kalman Filter
- Estimates state of a system
- Position
- Velocity
- Many other continuous state variables possible
- KF maintains
- Mean vector for the state
- Covariance matrix of state uncertainty
- Implements
- Time update prediction
- Measurement update
- Standard Kalman filter is linear-Gaussian
- Linear system dynamics, linear sensor model
- Additive Gaussian noise (independent)
- Nonlinear extensions extended KF, unscented KF
linearize - More info
- CS226
- Probabilistic Robotics (Thrun/Burgard/Fox, MIT
Press)
13Particle Filters
- An alternative technique for tracking
- Easier to implement
- Nonlinear
- Better for data association
- In CV, known as Condensation Algorithm
14Particle Filter
15Particle Filters Basic Idea
See e.g., Doucet 98, deFreitas 98
16Basic Particle Filter Algorithm
Initialization X0 ? n particles x0 i
p(x0) particleFilters(Xt -1 ) for i1 to n
xt i p(xt xt -1i)
(prediction) wti p(zt
xti)
(importance weights) endfor for i1 to
n include xt i in Xt with probability ?
wti (resampling)
17Particle Filters Illustration With Wolfram
Burgard, Dieter Fox, Frank Dellaert
18Particle Filter Explained
19Monte Carlo Localization (1)
20Monte Carlo Localization (2)
21Particles Robustness
22Tracking People from Moving Platform
- ? robot location (particles)
- ? people location (particles)
- ? laser measurements (wall)
With Michael Montemerlo
23Tracking People from Moving Platform
- ? robot location (particles)
- ? people location (particles)
- ? laser measurements (wall)
With Michael Montemerlo
24Examples
Siu Chi Chan McGill University
25Examples
D. Stavens. D. Lieb. A. Lookingbill
Particle filter
Optical flow
26Another Example
Mike Isard and Andrew Blake
27Tracking Fast moving Objects
K. Toyama, A.Blake
28More Particle Filter Tracking
David Stavens, Andrew Lookingbill, David Lieb,
CS223b Winter 2004
29Nonlinearity in the Particle Filter
30Data Association in Particle Filters
- Suppose k features in image, k state variables.
Which ones to marry? - Particle filter Each particle selects its own
data association - Probabilistic interpretation Particles are
posteriors over continuous state and discrete
data associations. (KF can only to continuous
state)
31Summary Kalman Filter
Particle
- Estimates state of a system
- Position
- Velocity
- Many other continuous state variables possible
- KF maintains
- Mean vector for the state
- Covariance matrix of state uncertainty
- Implements
- Time update prediction
- Measurement update
- Standard Kalman filter is linear-Gaussian
- Linear system dynamics, linear sensor model
- Additive Gaussian noise (independent)
- Nonlinear extensions extended KF, unscented KF
linearize - More info
- CS226
- Probabilistic Robotics (Thrun/Burgard/Fox, MIT
Press)
and discrete
set of particles (example states)
predictive sampling
resampling, importance weights
fully nonlinear
easy to implement