Feature Comparison and Feature Fusion for Traditional Dances Recognition

Abstract

AbstractTraditional dances constitute a significant part of the cultural heritage around the world. The great variety of traditional dances along with the complexity of some dances increases the difficulty of identifying such dances, thus making the traditional dance recognition a challenging subset within the general field of activity recognition. In this paper, three types of features are extracted to represent traditional dance video sequences and a bag of words approach is used to perform activity recognition in a dataset that consist of Greek traditional dances. Each type of features is compared in a stand alone manner in terms of recognition accuracy whereas a fusion approach is also investigated. Features extracted through the training of a neural network as well as fusion of all three types of features achieved the highest classification rate.KeywordsDance recognitionDense TrajectoriesSpatio-temporal interest pointsSubspace AnalysisNeural networks