Classification of trampoline jumps using inertial sensors

The automatic segmentation and classification of an unknown motion data stream according to given motion categories constitute an important research problem with applications in computer animation, medicine and sports sciences. In this paper, the scenario of trampoline motions is considered, where an athlete performs a routine consisting of sequence of jumps that belong to predefined motion categories such as pike jumps or somersaults. As main contribution, a fully automated approach for capturing, segmenting, and classifying trampoline routines according to these categories is introduced. Since trampoline motions are highly dynamic and spacious, optical motion capturing is problematic. Instead, it is reverted to a small number of inertial sensors attached to the athlete’s body. To cope with measurement noise and performance differences, suitable feature and class representations are introduced that are robust to spatial and temporal variations while capturing the characteristics of each motion category. The experiments show that the approach reliably classifies trampoline jumps across different athletes even in the presence of significant style variations.