In recent years there has been a proliferation of online digital archives of the performing arts, and in particular Dance. These digital repositories promise new opportunities to search and discover historically significant performance. However current dance archives are searchable only using a text-based index of keywords, identified by the archive curators. These keywords typically focus upon authorship or time-location metadata rather than the choreographic details of the content. We report work in progress towards a system that enables choreographic search, through a sketch based interface that matches sketched human stick figures to human poses in lowfidelity dance video. Pose is the essential element of dance, and users are adept at drawing stick-figure representations of pose regardless of their artistic apptitude. Our sketch retrieval system offers an intuitive and accessible interface for choreographic search.
In this paper we outline the front-end of our sketch based pose search system. Our front-end accepts a free-hand sketch as input and transforms this into a parameterised stick-man; a set of joint angles and related skeleton limb lengths. This is a significant challenge given the intra-class variation exhibited by sketched stick-men. The joint angles of the stick-man form the representation for the subsequent pose search.