IMU-based Trick Classification in Skateboarding Benjamin H. Groh Thomas Kautz Dominik Schuldhaus
[email protected] [email protected] [email protected] Bjoern M. Eskofier bjoern.eskofi
[email protected] Digital Sports Group, Pattern Recognition Lab Friedrich-Alexander University Erlangen-Nürnberg (FAU), Germany ABSTRACT can find its application for non-professional skateboarders The popularity of skateboarding continuously grows for ath- by providing feedback about their trick performance. letes performing the sport and for spectators following com- For the development of a trick-detection and classification petitions. The presentation and the assessment of the ath- system, sophisticated pattern recognition procedures are nec- letes' performance can be supported by state-of-the-art mo- essary. These methods are applied to the motion data which tion analysis and pattern recognition methods. In this pa- are often acquired by inertial measurement units (IMU). In per, we present a trick classification analysis based on mo- contrast to video-based methods, IMUs do not require any tion data of inertial measurement units. Six tricks were per- external equipment and the acquisition is not confined to a formed by seven skateboarders. A trick event detection algo- specified area. rithm and four different classification methods were applied The first known approach to apply pattern recognition meth- to the collected data. A sensitivity of the event detection ods to skateboarding was proposed by Anlauff et al. [2]. of 94.2 % was achieved. The classification of correctly de- They developed a real-time skateboarding game Tilt 'n' Roll tected trick events provides an accuracy of 97.8 % for the for a Nokia N900 smartphone.