In this study, several exercises were analysed. The air squat (AS, Figure 2a) is a squat without any external weight. In the overhead squat (OHS, Figure 2b), an athlete balances a barbell above his head with straight arms while performing a squat. During the front squat (FS, Figure 2c), the athlete has the barbell on his shoulders. The deadlift (DL, Figure 2d) is not a squat exercise since the barbell starts from the ground and finishes at hip height. However, in this study, an aim was also to understand the differences between warm-up and stretching routines on CoP features during different lifting exercises.
Subjects
Thirteen athletes (mean age 28.2 ± 5.9 yrs) from a local functional training facility volunteered to be analysed in this study and underwent either stretching or warm-up exercises. In addition, 5 subjects were tested without stretching or warm-up exercises in order to provide a control. Inclusion criteria were an age between 18 and 75, no medical issues affecting their physical performance, and no prior training on the day of testing. The study was approved by the Ethics Committee of the ETH Zurich, Switzerland, for which all participants signed a form of consent in order to participate. All athletes were also required to have non-pathological RoM for the tested exercises; the facility performs functional movement screen (FMS) for each athlete at the date of the sign-up to identify deficiencies in RoM [18]. The subjects’ age, gender, and experience level (“novice”, “proficient”, “expert”) were all recorded. Male athletes used a 20 kg barbell, while female athletes were provided with a 15 kg barbell.
Pressure sensor measurement system
A wearable, non-obtrusive sensor system was used that is able to capture dynamic movement and plantar pressure data (Figure 3) [19]. The system was validated in prior work where it was shown to provide valid estimations on subjects’ balance performance [20, 21].
The system was comprised of a thin and flexible foot-shaped plastic foil containing 1260 force-sensitive resistors (FSR) (Figure 4). A raw sensor sample featured 21 sensing points in the x-direction and 60 sensing points in y-direction (Figure 5). It has been validated against commercial plantar-pressure sensing systems [20]. Although the system is able to detect small differences between different shoe models [19], in this study, the sensor foil was glued onto a flat plywood surface for measurements (Figure 4b). In so doing, the impact of differences in shoes or feet sizes of the various subjects could be removed and the data could be acquired without individual bias. For this study, “zero-drop” shoes i.e. shoes with no drop from heels to toe, or only wearing socks was required from all subjects. During the tests, the athletes stood on the foil in socks or in their own (“zero-rise”) shoes. The sensor system sampled FSR values and concurrently recorded motion data from an inertial measurement unit (IMU). IMU data consisted of three-dimensional acceleration, rotation rate, and compass values. Accelerometer readings from the IMU were used to segment the data during analysis. At 100Hz, the system calculated the CoP of each foot and stored it locally.In advance of the testing, the pressure measurement system was presented to each subject and all procedures were explained. Firstly, subjects were asked to perform 10 squats without additional weight (air squat, AS), followed by 10 overhead squats (OHS) with the assigned weight, and 10 front squats (FS) with the same weight. Finally, every subject performed 10 deadlifts (DL) (Figure 2). A coach of the facility supervised the correct execution of the exercises.
After a 10 minute rest, each subject was then randomly assigned to perform the warm-up routines (WR), the stretching routines (SR) (Figure 6) or simply to further wait (control, CTR). Stretching exercises combined dynamic stretching routines [11], self-myofascial release (SMR) techniques [12], and proprioceptive neuromuscular facilitation [13] (Figure 6b). The warm-up routines consisted of a combination of exercises commonly used in the functional training facility (Figure 2). The subjects were asked not to go into exhaustion during warm-up. The CTR group was asked to wait 10 minutes sitting or standing. After the 10 minutes, each exercise was recorded once again. Between exercises, the subjects performed several steps to pick-up the barbell or to readjust stance etc. The movements were visible in the force data and also in accelerometer data.
All data were analysed using the Matlab software (R213b, MathWorks, Natick, MA). IMU and pressure data were segmented into episodes of AS, OHS, FS, and DL. Data intervals were labelled with the appropriate exercises. If an athlete performed unexpected movements during data acquisition, the sample was annotated to avoid false labels. For each interval of AS, OHS, FS, and DL, the algorithms extracted the features from pressure data. Since every subject had a different baseline level of mobility, and because feet sizes were different, all data were normalised to feet sizes. Here, in a pre-processing step, the region of interest for each data sample was extracted, i.e. the parts of the insole where the feet were standing and CoP coordinates were mapped to a common range ([0,1]).
For both feet, the centre of pressure was extracted: CoPL and CoPR. The left and the right CoP were then combined into a single CoP for the whole body. We calculated the following features from CoP: mean, and the coefficient of variance (CV). All features were calculated on both dimensions, i.e. x and y (Figure 6), in a 300 ms sliding window with 50% overlap. CV was calculated as the fraction of the standard deviation from the mean value of the sample, i.e. CV = σ/μ. This feature reflected how dispersed or scattered the CoP was for a given exercise, as a proxy for stability in balancing exercises. For exercises that required limited balancing (e.g. DL), the CV feature was used as a surrogate for the regions of each subject’s feet that were used to generate a reaction force. These features have been demonstrated to be valid indicators for balance, stability and body-weight distribution [21].
Prior to addressing the hypotheses, the samples from all groups (pre-stretching/warm-up/control) were tested by ANOVA to examine whether statistically significant differences in the data sets were present. Based on all features, a three-factor analysis of variance (ANOVA, (alpha = 0.05, factors = [age, gender, exp.])) was performed to work out if age, experience or gender had a measurable effect on the data.
To answer the three questions listed in the first section, repeated two-factor ANOVA (α = 0.05, factors = [group{WR,SR,CTR} condition{pre,post}]) was used to detect statistically significant differences between pre- and post-routine data and to answer the three questions.