This paper presents methods for classifying speed and track inclination groups during recreational runs using input data from the "adidas_1" running shoe. Running speed, altitude and shoe heel compression were recorded continuously while athletes ran freely outdoors. A total of 84 one-hour-runs were collected in order to have sufficient ground truth as well as sensor data for classification. The data was analyzed using features computed for each step of the athlete. The goal of this work was to distinguish three speed and three surface inclination classes, respectively. The speed and inclination classes were established using the collected ground truth data. The results showed that surface inclination classification was only possible with an accuracy of 67.2% due to measurement restrictions. However, it is also demonstrated that speed classification was feasible with up to 89.2% accuracy. The developed classification system for speed classification was implemented and verified on the embedded microprocessor of the "adidas_1". Such a system can be used to support sportsmen, for example by adapting their equipment to the specific running speed. The employed pattern recognition methods are general in nature and can thus be applied to other embedded classification applications as well.
© Copyright 2010 International Journal of Computer Science in Sport. Sciendo. All rights reserved.