Implementation of Hidden Markov Modells- Detail information

[xandi1987]

Hello,

 

I am quite new to HMM but I know the general approach of the system. I am using matlab for calculating a transition/emission matrix for a observation- sequences. I want to use Baumwelch alg. for training and the viterbi- alg. for determining the most like- full sequence. My problem is the following (lets say the system is for gesture recognition and we have 2 gestures A,B):

 

By training I get for every sequence the parameter of the HMM for classification. Lets assume I have 30 samples for every gesture. By the Baumwelch alg. I get now 60 HMM (for every sample one). And here stops my understanding. Do I have to "sum up" these 60 HMM to 2 one HMM_A,HMM_B? And if I do, how can I do it?

 

 

 

I hope somebody can give me a hint and that my English was good enough.

 

Kind regards

 

Alex