Authors: M. Boukoutsou, D. F. Kavelidis, D. Natsidou, N. Papageorgiou, I. Roboli and L. J. Hatjileontiadis
Abstract: This paper is a comprehensive study on classification of motion capture data based on features extracted from wavelet analysis (using Discrete Wavelet Transform), Higher-Order Spectral Analysis (HOSA) and Cepstral Analysis. More specifically, the analysis presented is a continuation of the open research on the influence of auditory stimuli on human micromotion, following the MICRO project of the RITMO Centre of the University of Oslo. The idea of the classification is to confirm that this micromotion is affected by the music genre, using new approaches on the specific problem. The emphasis on this paper is given more to the time-series representing the motion of the head and not the direct correlation with the auditory stimuli, but rather the labels of each of the music genre. From the analysis mentioned above, features were extracted and a classifier was used; HOS-Cepstrum Classifier. The results show that the best accuracy (holdout- 25%) the classifier could achieve is equal to 64.81%.
As of the completion of the project, it will probably be modified in case of more research in the field, but in any case it will not be maintained for now.
The code for this paper is private. If interested, reach out to me: