I am new to statistical pattern recognition and trying to learn.To begin with I am trying to work with two class problems and trying to classify motion activities as mentioned in the paper "Object Trajectory-Based Activity Classification and Recognition using Hidden Markov Models". In this paper, the authors have used GMM for estimating the PDF for each motion pattern class. But I have several doubts and shall be grateful for the following answers
- If there are 5 experimental data with 4 feature vectors(4 columns of data) of length 1000 samples/rows for each of the two classes(say,running and walking) then is the pdf from Gaussian mixture model obtained for each of the 5 experimental data or only for each class? If it is for each class, then how is it done? I really do not know and please pardon if this sounds too trivial.
- Is it always necessary to find the pdf of the data before classification task in general? Does the pdf control in deciding which model based approach to choose for classification purpose?
- Are there any such model based approaches for classification?
- As far as my understanding goes, GMM is a clustering algorithm like k-means. So, are there other ways other than using Hidden Markov Model (I want to avoid HMM due to further complexities in understanding it) for classification with GMM?
- Can GMM be combined with regression models like AR,MA,ARMA for model based classification? If so, then pointers to resources which explain this shall be helpful.
- Is there a Matlab implementation of multivariate GMM for the said application of classification?
Thank you in advance.