Hello. I have not allied HMM techniques to EMG data but I have been involved in HMMs for cell membrane kinetics. In this application, there are only 2 basic states (open and closed) of a single channel and a membrane is assumed to consist of a large number of identical channels. If channels are also assumed independent (not true in practice!) then a simple Markov model can be used to describe the probabilistic dynamics. Perhaps you could start with a simple discrete-time EMG model that has a small number of states. The HMM for this would hopefully easy to identify with standard HMM techniques (Baum Welch). The key for any HMM modelling is to use as much prior information about the system to constrain the HMM in terms of number of states, transition probability structure and observation probability model. Links to some previous work I was involved in are attached. There are some references in these papers that you might find useful.

Article Identification of individual channel kinetics from recording...

Article Evaluation and Estimation of Various Markov Models with Appl...

Dear Maxim,

  I appreciate for your graet paper, it has a very comprehensible method and was useful for my project. But they studied on posture detection( sit to stand or walking or running positions) . I'm studing just on upright standing equilibrium and the data is EMG derived from TA and MG muscles...

Regards,

Hello dear Mr.Pulford,

   Thank you very much for your great idea about discrete-time EMG model. As I understood we can derive any discrete data to a HMM for estimating hidden states. In your second article (application to membrane channel kinetics) we have just 2 states, open and closed, as you said. But if we have more states in a data sequence the HMM still could be useful? For example in upright standing for a muscle, activation levels are important. Because for instance in level 1 and 2 the muscle being activate and in level 3 it can't be activate. And these measurements are different from one subject to another. Moreover there is an intermittent control in human posture. So identification of appropriate sequence is difficult. Do you have any idea? Please let me know.

Hello.

Maybe you could have a look into the following articles, which address the used of Concurrent Hidden Markov Model (CHHM) for behaviour analysis, which involves pose estimation. We used different features from movement, including the analysis in the frequency domain, which I assume it could be used also in EMG signals.

If you have trouble accessing the publications let me know.

Cheers,

Conference Paper Using Concurrent Hidden Markov Models to Analyze Human Behav...

Conference Paper Probabilistic LMA-based Human Motion Analysis by Conjugating...

Conference Paper Probabilistic LMA-based classification of human behaviour un...

Hi,

i basically agree with what Graham Pulford reported.

I have no experience in EMG signals processing but i use HMM for electrical characterization. I enclose two links to publications in which we use HMM to characterize random telegraph signals in emerging memories, if you want to compare. 

HMM is very useful if your system can be described as a memoryless process with a given number of states. Even more than two, of course. The HMM tries to find the best description of the system, so how and when the system switches from one state to another, based on the experimental observation. So it will work also with more than two levels. As a drawback you should estimate the number of levels beforehand as this is a required input for the algorithm.

There is a great guide to HMM by Rabiner. It shows application to speech recognition but if you have a look at it you may find it useful to understand whether HMM is a suitable solution for you or not.

Hope this is helpful.

Regards.

Article RTS noise characterization of HfOx RRAM in high resistive state

Conference Paper Random Telegraph Signal noise properties of HfOx RRAM in hig...

Dear Dr. Tayebeh Jahani

I am searching for references about postural control and technical development in football. So, how I started in few days ago, I'm not able to help you like the those researchers that wrote above.

This question was very important to me also, I'll try to know more about this method.

Greetings. 

Dear João Quintas,

  Thank you for your great papers. The method of your first paper was interesting for my approach. About  second and third one actually in posture( special upright standing) we can not use acceleration signals. Because person have no specified movement...but Bayesian network and LMA system was interesting too.

Dear Francesco M. Puglisi,

 As João Quintas point to the " USING CONCURRENT HIDDEN MARKOV MODELS TO ANALYSE HUMAN BEHAVIOURS

IN A SMART HOME ENVIRONMENT " paper, we can use HMM even for multiple states , and I reviewd your proposed papers in electrical engineering. It was interesting. Thank you so much.

Dear Rodrigo Almeida,

 As I said before I research on upright posture. Any physiologic charactristics and muscle activation or torque at ankle muscles and intermittent or continious models with considering muscle synergy should be investigated. So if this subject are interesting for you please follow me and I'll try to put you in my progress.

Good luck to you.

Like Dr. Pulford, my exposure to HMMs was in ion channel kinetics.  Check out the free software QUB from SUNY Buffalo.  I don't know about EMGs, but I've found a way of building HMMs from EEGs, namely sleep cycles.  See my patent #7912717, "Method for building hidden Markov models," 2011.  I don't know of any EMG cycle, though, nor even any observed sleep-like EMG states.  Apparently, "HMM" in this context refers to "Hill-based muscle model" so I apologize for any inappropriateness..