URL:

files/proceedings/2010/28.pdf

Abstract:

In this paper we describe a method to detect patterns in dance movements. Such patterns can be used in the context of interactive dance systems to allow dancers to influence computational systems with their body movements. For the detection of motion patterns, dynamic time warping is used to compute the distance between two given move- ments. A custom threshold clustering algorithm is used for subsequent unsupervised classification of movements. For the evaluation of the presented method, a wearable sensor system was built. To quantify the accuracy of the classifica- tion, a custom label space mapping was designed to allow comparison of sequences with disparate label sets.