when one of your hand lifts up or grasp anything and the other is at rest, the EMGs will differ on basis of signal wave length, power,spectral magnitude,etc...

Why don't you use a classifier on the basis on one of the features. You also need to see which feature is the most optimal for the use.

If data is separable you may use any linear classifier

i agree with Traian, i think most significantly in this case will be bicep, tricep and flexor digitorum

Hello Lijo. of course using classifier is more elegant method but yet I am thinking, is it really necessary? I would calculate the RMS of signals and try to estimate which one is active by using a threshold basically. since it is easy to implement. however I am assuming one of the hand in relax position while other is active.

It rather depends on what you want to know. Is it just a matter of wanting to know which hand moves? - in which case the video approach or even just an accelerometer on the hand might tell you what you want to know - or are you interested in the particular pattern of muscle activation. I suppose the other relevant issue is what else is the subject doing at the same time? If they are doing something with the other hand then EMG may well have a hard time discriminating which hand is following your instruction.

i guess Lijo is more keen on using EMG and i agree with Jeremy that it will become very non discriminating if you use the other hand for performing some other function. However, at rest it will be simple

I agree with Jeremy, what's the motive of this experiment. If we're looking at rehabilitation training, I think there is a great need! Since the action will be controlled and the outcome measurement is to ensure the subject has strictly followed the instruction, especially in remote monitoring rehab efforts. I am currently working in such applications.