Development of an Automatic Alpine Skiing Turn Detection Algorithm Based on a Simple Sensor Setup

Abstract

In order to gain insight into skiing performance, it is necessary to determine the point where each turn begins. Recent developments in sensor technology have made it possible to develop simpler automatic turn detection methodologies, however they are not feasible for regular use. The aim of this study was to develop a sensor set up and an algorithm to precisely detect turns during alpine ski, which is feasible for a daily use. An IMU was attached to the posterior upper cuff of each ski boot. Turn movements were reproduced on a ski-ergometer at different turn durations and slopes. Algorithms were developed to analyze vertical, medio-lateral, anterior-posterior axes, and resultant accelerometer and gyroscope signals. Raw signals, and signals filtered with 3, 6, 9, and 12 Hz cut-offs were used to identify turn switch points. Video recordings were assessed to establish a reference turn-switch and precision (mean bias = 5.2, LoA = 51.4 ms). Precision was adjusted based on reference and the best signals were selected. The z-axis and resultant gyroscope signals, filtered at 3Hz are the most precise signals (0.056 and 0.063 s, respectively) to automatically detect turn switches during alpine skiing using this simple system.

Development and Validation of a Gyroscope-Based Turn Detection Algorithm for Alpine Skiing in the Field

“…The performance metrics and video recordings were used to evaluate the performance of the current algorithm iteration (see section Evaluation design) and detect weaknesses. The starting point was the evaluation of the algorithm developed under laboratory conditions by Martínez et al (2019). Matlab (R2017B, MathWorks, Natick, MA, USA) and R (3.5.1, R Core Team, Auckland, New Zealand) were used to test and develop the algorithm.…”

“…To restrict high rates to parallel turns, the arithmetic mean of both sides ω roll smooths unilateral rotations. In the next step, a fourth-order zero-lag 0.5 Hz lowpass Butterworth (BW) (Martínez et al, 2019) was applied to produce a decision signal (ω ds ) that describes the rotation rate in the roll axis (Figure 3). Figure 3 shows the local extrema in the two exemplary decision signals ω ds = BW 0.5 (−ω roll ).…”

“…The degree of smoothing between raw and decision signal can be observed in Figures 4, 6A. To determine the turn switches more accurately, a higher cutoff frequency seems to be more advisable. As shown in the study of Martínez et al (2019), f c = 3.0 Hz is a proper cut-off frequency to perform such a "fine tuning." Therefore, a new fine tuning signal is defined by ω ft = BW 3.0 (−ω roll ) (Figure 6A).…”

Application of Inertial Motion Unit-Based Kinematics to Assess the Effect of Boot Modifications on Ski Jump Landings—A Methodological Study

“…A possible circumvention of this difficulty is the use of inertial motion unit (IMU) -based kinematic measurements in conjunction with differential global navigation satellite systems (GNSS), which provide a spatial resolution of a few millimeters [10]. Similar approaches have been utilized previously for slightly different applications [11][12][13][14][15][16][17][18]. We propose that a combination of 2D video recordings with IMU-based kinematics will provide a feasible solution for the assessment of full body motion, for example, in the evaluation of ski jump biomechanics in terrain parks.…”

Computationally Efficient 3D Orientation Tracking Using Gyroscope Measurements

“…3DO is computed for each consecutive measurement step 2. All consecutive measurements are combined into a single composite, at which point only the final 3DO is computed 3.…”