I am working on a project that uses the leap motion controller. I am using a Hidden Markov Model to model whether a hand is moving left, or, whether a hand is moving right. For this, I do the following:
1) Take the live data from the leap motion controller (80 values containing 2 values for each set)
2) Extract features "Velocity vector"
I have plotted the live data and it seems to be extracting the features ok. Now my problem is generating the HMM's. For this, I store the first 80 values inside a 2-Dimentional Vector and initialise and train each of the models with these values, for example:
-> Live data -> Left-> Capture first 80 values -> initialise HMM -> train HMM
-> Live data -> Right -> Capture first 80 values -> initialise HMM -> train HMM
I then save the output of each model into different text files:
1
2
p0:
1.000000
a:
1.000000
mu:
20.578226 -11.572160
cov:
2012.605072 0.000000 0.000000 19224.648008
1
2
p0:
1.000000
a:
1.000000
mu:
20.721460 6.474166
cov:
3576.301595 0.000000 0.000000 26737.119734
For the recognition part of this problem, I read the values in to form the models of the HMM, for the live data again I extract the features and pass this to a viterbi_scoring algorithm. For this, I pass in: raw extracted features and each of the models.
I then output each of the scores, but none of them make any sense. I can't seem to figure out where I am going wrong. The methods seems to be right.
Any help or suggestions would be greatly appreciated.