Contour Tracking With Energy Minimization and HMM

1
Contour Tracking With Energy Minimization and HMM

• Min Li
• CISC849 presentation
• Feb, 27, 2002

2
Contour Tracking
3
Active Contour model
Active Contour Model
Energy Minimization
Probability Maximization (HMM)
4
Energy Minimization Method

• Contour representation
• Contour energies

5
Energy definitions

• Internal Energy
• External Energy

6
HIDDEN MARKOV MODEL(HMM)

• If equal opportunities
• If the third one prefers head
• If probability from 1st or 2nd to 3rd is 0,
  start from 1
• HMM Initial state Observation probability
• Transition probability between states
  

7
HIDDEN MARKOV MODEL(HMM) cont.

• N number of states
• M number of distinct observations
• T length of observation sequence
• V the set of possible
  observations
• the probability of being in
  state i at the beginning
• A where
• An HMM can be noted as
• Problem Given the model
  ,how to choose a state sequence
  , so that is maximized.

8
Viterbi Algorithm

• Define
• equals
• Viterbi Algorithm Finding a path with minimum
  weight.
  

9
Viterbi Algorithm(cont.)
10
Contour tracking using HMM

• Restrict the searching to the contour normal line
• 2N pixels on a normal line, 2N states

11
Observation likelihood( by chen_at_uiuc)

• Color Information

12
Transition probabilities(by chen_at_uiuc)

• Smoothness
• JPDAF based

Matching distance of background( i1,N on line
and j1,N on line )