Within Terpsichore framework new algorithms have been developed focusing on (i) dance summarization of choreographic data that include point joints, (ii) dance classification and (iii) posture identification algorithms including deep machine learning, (iv) deployment of a new game that can be used for dance educational purposes enhancing the social impact of the project, (v) new calibration tools for depth sensors, (vii) a semantic metadata description framework for dances, (viii) Labanotation of the dance movements, (ix) and the creation of three innovative dance dataset.
Terpsichore consortium during the lifecycle of the project carried out successfully three Summer Schools and many dissemination events. Terpsichore presence at workshops and conferences is really important since each partner receive training on keys aspects regarding the Terpsichore objectives. The transdisciplinary environment and the diversity of the participant’s research backgrounds is a source of scientific inspiration for new approaches, collaborations and research activities. At this point is important to mention that the presence of the Terpsichore consortium at various workshops and conferences disseminating the importance and the rehabilitation of the intangible cultural heritage. Moreover, the transdisciplinary and inter-sectorial knowledge via the presence at conferences and workshops providing Terpsichore partners the high-quality learning.
To overcome the challenges of the innovative project, we carried out research in the area of 3D capturing and imaging, computer vision, machine learning, 3D modeling, symbolic representation and finally virtual scene generation. In particular, the main research objectives in 3D computer vision research focused on 3D modeling process, while maintain high resolution accuracy. Additionally, the research is focused on complex background environments and of moving objects. To address these difficulties in the Terpsichore project, we have introduced a scalable capturing framework by incorporating state of the art devices able to acquire depth information in real-time constraints. Moreover, the current 3D modeling algorithms are not appropriate of complex human movements and complex background regions. Additionally, in case where multiple dancers interact with each other and with the environment several research challenges are emerged. To address these aspects, we need, on the one hand, computer vision tools, able to detect the foreground/background content under a highly dynamic framework, to track geometrically enriched points of interest through time and finally to estimate 3D skeletons from the 3D voxels. In addition, we need technologies able to fit the 3D skeletons into pre-defined deformable models. Furthermore, the general advantage of the project is that a complete pipeline from capturing to digital visualization.
