Motion capturing devices have become widely available, which resulted in large volumes of 3D human motion data produced in a variety of application domains, ranging from entertainment to medicine. However, automatized processing of such data is a challenging problem because their inherent spatio-temporal nature implies that the same action can be performed in a number of alternatives that vary in speed, timing, or location in space. Moreover, the captured data are imprecise and voluminous, as hundreds of megabytes per hour are obtained during tracking only 3D positions of body joints. Therefore, the employment of basic data-processing paradigms is much more intriguing, when compared to the traditional domains such as text or images. In the proposed project, we aim at developing new theories and technologies for three interconnected open problems of content-based searching, annotating, and mining in motion data streams. Taking into account the fast growth of motion data volumes, a lot of attention will be given to the scalability of proposed solutions.

Publikace