Feature construction for Latin couple dance

Motivation

Human history is written not only by material objects or written text but also through traditions. One of the elements of the tradition is how we dance. To understand how the traditions connect and evolves, it is necessary to be able to deconstruct systematically and implement a large scale comparison of different types of dance.

Goals

The goal of this work is to develop a handcrafted dance motion descriptor and use it to find similarities between different dances or repeating patterns within one dance.

Tasks

• Define the implementation of pose descriptors
• Implementation of a bag of words approach
• Classification of parts of the dance

Necesary skills

• Programming in Python
• Basic knowledge of machine learning

Contact

Johannes Stork and Tomasz Kucner