Wearable sensors were developed to compensate the lack of portability of multi-camera systems (both multi-video cameras and optokinetic systems), as they are wireless, small, and light. A second limitation of multi-camera systems relates to the lack of control of the ambient environment in which data are collected (in terms of light, calibration volume, and/or area, etc.) which led scientists to develop new devices to track the behavioural dynamics over long periods of time, involving time-series analysis. Indeed, getting 10%-20% of an event cannot represent the performance during a whole event; therefore, new technologies are requested to track athletes’ behaviours over a longer period of time, covering all the area or volume of performance. To overcome the previous limitations, devices such as smartwatches, smartphones, global positioning system (GPS), inertial measurement unit (IMU), or eye tracking system (glasses) have been developed. Those devices are often of low cost, light, user-centric, portable (i.e., wearable), and hence easy to use in field conditions, which could favour in situ and representative design of the task and procedures. In comparison to 3D multi-cameras systems, smartwatches, GPS, and IMUs can record a high volume of continuous data over an entire competitive or training event.

Smartwatch, Smartphone, and Global Positioning System (GPS)

In regard to smartwatches, Mooney et al. (2017) have tested the reliability of two commercially swimming activity monitors (Finis Swimsense and Garmin Swim™) against a multi-video camera system used for notational analysis. Ten swimmers had to swim 1,500 m in the four swimming techniques, then the reliability of the smartwatches was assessed for five features (i.e., stroke swimming technique, swim distance, lap time, stroke count, and average speed). High reliability was found to detect the swimming technique, the swim distance, and the number of swimming laps performed in the middle of a swimming interval; however, the number of laps performed at the beginning and end of an interval were not as accurately timed. In addition, a statistical difference was found for stroke count measurements, which affect the accuracy of stroke rate, stroke length, and average speed measurements. Mooney et al. (2017) concluded that these smartwatches appear suited for recreational use, but further development to improve accuracy of the lap time and stroke count detection would be required for competitive settings. In running, Pobiruchin, Suleder, Zowalla, and Wiesner (2017) used a pre-race and post-race survey to investigate the accuracy of tracked distances recorded by smart devices (e.g., smartwatches and smartphones). The mean of the track distances recorded by mobile phones with combined application (mean absolute error of 0.35 km) was significantly different to GPS-enabled sport watches (mean absolute error of 0.12 km) for the half-marathon event. Again, smart devices appear suitable for recreational use but not for scientific purpose.

The GPS integrated in a smart suit or in a bra is regularly used to quantify the external load in team sports (Cummins, Orr, O’Connor, & West, 2013; Hausler, Halaki, & Orr, 2016; Jennings, Cormack, Coutts, Boyd, & Aughey, 2010; Johnston et al., 2012; Scott, Scott, & Kelly, 2016; Willmott, James, Bliss, Left- wich, & Maxwell, 2019). Scott et al. (2016) investigated the reliability of GPS in a team sport setting, with a particular focus on measurements of distance, speed, and accelerations across sampling rates of 1, 5, 10, and 15 Hz GPS. Low sampling rate GPS (1 and 5 Hz) exhibited limitations in measuring distance during high-intensity running, speed and short linear running (particularly those involving changes of direction), whereas high sampling rate GPS (10 and 15 Hz) appeared more reliable across linear and team sport simulated running (Scott et al., 2016). Several studies confirmed the higher validity and reliability of GPS with high sample rate for distance and speed measurements in team sports, especially for low-to-moderate speeds (below 20 knrh-1) (Gray, Jenkins, & Andrews, 2010; Jennings et al., 2010; Johnston et al., 2012; Waldron, Worsfold, Twist, & Lamb, 2011). For instance, Jennings et al. (2010) observed a lower coefficient of variation (1.4%-2.6%) during walking for a 5 Hz GPS than during sprinting over a 20-m distance (19.7%—30%). Similarly, these authors observed higher coefficients of variation for the same activity (30.8% during walking and 77.2% during sprinting) over a 10-m distance with a 1 Hz GPS (Jennings et al., 2010). Finally, as already mentioned by Scott et al. (2016), the reliability of GPS is also affected by tight change of direction as Jennings et al. (2010) observed a lower coefficient of variation for gradual (11.5%) than for tight (15.2%) change of direction during walking. Therefore, assuming that speed measurement is prone to error at high speed (above 20 kmh ') and during non-linear run with tight changes of direction, cautions might be considered when practitioners break players’ activity into speed zones between 0 and 36 km h 1 (Cummins et al., 2013). When such breaks are made, possible differences in the number of sprints at high intensity or in the time spent in high-speed zone due to gender, player’s position in the field (attackers/defenders), or age may hide lack of reliability. In fact, few studies assessed the reliability of GPS against gold standard motion capture system (Randers et al., 2010; Waldron et al., 2011).

Randers et al. (2010) compared the activity of 20 football players during a match using four different motion capture systems (a video-based time-motion analysis system - VTM, a semi-automatic multiple-camera system - MCS, and two commercially available GPS systems: 5 Hz GPS and 1 Hz GPS). Although those four systems were able to track the players, significant differences in the total distance covered during the match were observed between the GPS (10.72 ± 0.70 km for the 5 Hz GPS and 9.52 ± 0.89 km for the 1 Hz GPS) and the other two systems (10.83 ± 0.77 km for the MCS and 9.51 ± 0.74 km for the VTM). The distance covered by high-intensity running was significantly different between 5 Hz GPS (2.03 ± 0.60 km) and MCS (2.65 ± 0.53 km), and between 1 Hz GPS (1.66 ± 0.44 km) and both MCS (2.65 ± 0.53 km) and VTM (1.61 ± 0.37 km). The distance covered by sprinting was significantly different between 1 Hz GPS (0.23 ± 0.16 km) and both VTM (0.38 + 0.18 km) and MCS (0.42 ± 0.17 km), whereas 5 Hz GPS (0.37 ± 0.19 km) did not significantly differ with the VTM and MCS. Randers et al. (2010) concluded that there were large between-system differences in the identification of the absolute distances covered, implying that any comparison of results using different match analysis systems should be done with caution.

In another study, Waldron et al. (2011) investigated the validity and reliability of a 5 Hz GPS and timing gates (Brower Timing Systems, Draper, UT) to measure sprinting speed and distance in rugby players, and the reliability of proper accelerations recorded via GPS-accelerometer integration. Results of validity indicated that the GPS measurements systematically underestimated both distance and timing gate speed (coefficient of variation ranged between 4.81% and 9.81%). When the GPS measurements were compared between two tests, a high reliability was observed for all variables of distance and speed (coefficient of variation ranged between 1.62% and 2.3%). However, the timing gates were more reliable (coefficient of variation ranged between 1% and 1.54%) than equivalent GPS measurements. Finally, acceleration measurements (via GPS-accelerometer integration) were less reliable (coefficient of variation ranged between 4.69% and 14.12% when peak acceleration and frequency are compared between two tests). Waldron et al. (2011) concluded that the timing gates and the GPS were reliable systems to assess speed and distance, although the validity of the GPS remains questionable. The error found in acceleration measurements (via GPS-accelerometer integration) indicates the limits of this device for detecting changes in performance. One way to improve the accuracy of the GPS could be by fuzzing data from GPS, IMU, camera, and digital map (Baranski & Strumillo, 2012). Baranski and Strumillo (2012) did so in pedestrian navigation, for 90% of the navigation time, and the algorithm was able to estimate a pedestrian location with an error smaller than 2 m, compared to an error of 6.5 m for a navigation based solely on GPS (Baranski & Strumillo, 2012). In conclusion, advanced sensor fuzzing GPS, accelerometer, and local radio system might offer promising perspectives to improve the measurement reliability for distance and speed. According to the reliability of GPS measurement, multi-camera motion capture systems seem preferred to track players’ location in the field and then assess advanced metrics such as interpersonal coordination within and between teams.

< Prev   CONTENTS   Source   Next >