Three learning systems in the reconnaissance of basic movements in contemporary dance

Although action and gesture recognition by human beings, for example dance movements, is a high level process executed in the human brain there is quite a bit of structure imposed on the task so that it be can be analyzed by a computer by dividing it into a chain of more simple steps. One technique that has been successfully used in developing automatic systems to identify this type of activities, is the hidden Markov models (HMM). In this work, an automatic recognition system of basic movements in the contemporary dance that uses three learning techniques is presented. The first two of these techniques are used in order to generate the discrete observations needed in the Markov model associated to each of these movements. In the first one, the principal components analysis is used to extract the most relevant information coming from sensors tied to dancers who execute the movements. In the second one, the vector quantization is used in order to convert the output of the first data pretreatment to discrete symbols and so to feed the HMM to identify a given movement or for adjusting model parameters so as to best account for the observations.