INTRODUCTION: Requiring complex motion sequences, snowboarding is difficult to learn and improving one's skills often requires lessons given by an experienced teacher. Yet even this approach has its limitations, as instructors might not notice all mistakes. We have therefore been developing a wearable trainer capable of supporting snowboarders in improving their riding style [1,2]. Such trainers are built on sensor systems recognizing the activities performed by its users. In snowboarding, traditional activity recognition approaches make use of insoles with force-sensitive resistors (FSRs), which, however, are particularly uncomfortable to wear. In this paper, we show how FSRs can be integrated into snowboard bindings and how basic snowboarding moves such as turns and the riding edge can be recognized with the algorithm Nearest Centroid Classifier (NCC) from the sensor data. METHOD: On ski slopes, FSRs are exposed to extreme temperatures, forces and humidity. Thus, we designed a protective sealing made of PVC. The packaged sensors are integrated at different locations: Two below the ball of the foot, one below the heel, one inside the lower strap and two more in the upper strap of each binding (Figure 1). A netbook stored in a backpack logs the sensor data. We evaluated our sensor bindings on an intermediate ski slope in the Matterhorn Glacier Paradise (3369 m at 4°C) with one experienced goofy snowboarder on slushy snow. The lest rider descended the slope performing carved backside and frontside turns, which was recorded by a video camera. The sensor data were later labeled as either backside or frontside riding based on the video ground truth. The data then underwent low pass filtering and were scaled with respect to the body weight of the snowboarder. To assess the accuracy of NCC, we performed a cross-validation of the classifier on the sensor data. RESULTS AND DISCUSSION: Feature selection revealed that snowboarding moves are recognized best when using the pressure data of the sensor placed under the inner ball of the leading foot. In this case, NCC detected 49 of the 51 (96.1 %) turns correctly. This accuracy is competitive with other approaches based on insoles (96.7 %) [3], gyroscopes (90.5 %) [1] and socks (84.3 %) [2]. The riding edge was detected with an accuracy of 89.3 % at any time (Table 1). CONCLUSION: Besides the recognition of basic snowboarding moves, our bindings can be applied for various training purposes such as the determination the weight distribution between the front and the back foot. This is especially interesting for novices struggling to control the board due to too much weight on the back foot. Allowing for the monitoring of advanced carving parameters such as the rider-initiated torsion of the snowboard and the regularity of turns, our bindings can also assist experienced snowboarders in improving their carving skills.
© Copyright 2010 Book of Abstracts. 5th International Congress on Science and Skiing, Dec. 14 - 19, 2010, St. Christoph am Arlberg. Veröffentlicht von University of Salzburg, Interfakultärer Fachbereich Sport- und Bewegungswissenschaft/USI. Alle Rechte vorbehalten.