Embedded surface classification in digital sports

In this presentation, we give a detailed analysis of the considerations needed for mapping the complete pattern classification chain to the restricted embedded system hardware environment. We describe the methodology of the design, realization and testing process that takes these hardware limitations into account. For this purpose, we consider a particular embedded application from the field of digital sports: a novel running shoe that is capable of sensing run-specific parameters and adapting the cushioning setting accordingly. Of utmost importance in this context is the classification of the current surface condition in order to enable optimal adaptation to the prevailing situation. Following our design approach, we provide a classification system with a runner-independent surface classification rate of more than 80%. This system is implemented in the current version of the aforementioned running shoe. The presented methodology is quite general as it makes no system-dependent assumptions and can thus be transferred to many other embedded classification applications.