Connected Skiing: Motion Quality Quantification in Alpine Skiing

Abstract

Recent developments in sensing technology have made wearable computing smaller and cheaper. While many wearable technologies aim to quantify motion, there are few which aim to qualify motion. (2) To develop a wearable system to quantify motion quality during alpine skiing, IMUs were affixed to the ski boots of nineteen expert alpine skiers while they completed a set protocol of skiing styles, included carving and drifting in long, medium, and short radii. The IMU data were processed according to the previously published skiing activity recognition chain algorithms for turn segmentation, enrichment, and turn style classification Principal component models were learned on the time series variables edge angle, symmetry, radial force, and speed to identify the sources of variability in a subset of reference skiers. The remaining data were scored by comparing the PC score distributions of variables to the reference dataset. (3) The algorithm was able to differentiate between an expert and beginner skier, but not between an expert and a ski instructor, or a ski instructor and a beginner. (4) The scoring algorithm is a novel concept to quantify motion quality but is limited by the accuracy and relevance of the input data.

Keywords: IMU; carving; principal component analysis; scoring; wearable.

Figure 1

Measurement system and axis orientation. The X axis (red) points to the right, the Y axis (green) points vertically, and the Z axis (blue) points posteriorly.

Figure 2

Cumulative explained variability of the first six principal components from the reference dataset for edge angle (solid line), edging symmetry (dashed line), radial force (dotted line), and speed (dot–dash line), in (a) carving long radius, (b) carving medium radius, (c) carving short radius, (d) drifting long radius, (e) drifting medium radius, and (f) drifting short radius.

Figure 3

Responses for first three principal components for edge angle (a–c), edge angle symmetry (d–f), radial force (g–i), and speed (j–l) for carving short from the reference dataset. Blue lines represent the mean load score −1SD and red lines represent mean load score +1SD.

Figure 4

Distribution of algorithm-assigned scores from the test datasetfor the skiing styles carving and drifting in long (a,d), medium (b,e), and short (c,f) radius.

Figure 5

Mean +/− 1 standard deviation Z-transformed PC scores for PC 1–3 (top to bottom) across all runs (long, short, self-selected) for a beginner skier (left), a ski instructor (middle), and an expert skier (right).