Table of Contents
Background and Introduction
The overarching goal of the Proximity Project is to provide information about physical distance between players that may be used to better understand the possibility of viral transmission during the course of a game or training session.
For this report, a radius of specified distance around each athlete is referred to as the close proximity zone. Any instance in which another athlete enters the close proximity zone is referred to as an incursion. Incursions may vary in duration, and overlapping incursions are possible when more than 2 athletes are positioned together. If Athlete A breaches the close proximity zone of Athlete B, the inverse is also true. Therefore, each incursion is essentially doubled in order to achieve an accurate count for each athlete.
The overall objective is to assess the quality of the detection of any incursion and the related accumulated total duration within a 2m radius around an athlete.
For this investigation, a simplified environment consisting of only 2 athletes is considered. Real-world results will likely be more complex, involving interactions between many pairs of athletes, but the protocol followed for this report allows for a focused understanding unencumbered by the complexities of a typical training session.
The Validation Protocol
A primary objective of the Proximity Testing Protocol was to assess the effect of the close proximity zone radius on incursion detection and total duration of incursions detected. The effect of dwell-time is also assessed, although it should be noted that dwell-time considerations were not a primary concern when designing the protocol, and insights concerning dwell-time likely require further analysis.
The Proximity Testing Protocol consisted of 8 drills (listed below). Two trials of the Proximity Testing Protocol were completed for a total of 4 athletes and two locations. In this document, it is implied that a single incursion involves 2 athletes.
- Ai:
- Athletes start 10m apart.
- They jog facing toward each other until they are 1m apart, pivot to face the same direction (standing shoulder to shoulder), and remain in this position for 5 seconds.
- The athletes then jog back facing away from each other to their original starting locations and pause for another 5 seconds facing in the same direction.
- Repeat 10 times.
- Aii:
- Athletes start 10m apart.
- They jog facing toward each other until they are 2m apart, pivot to face the same direction (standing shoulder to shoulder), and remain in this position for 5 seconds.
- The athletes then jog back facing away from each other to their original starting locations and pause for another 5 seconds facing in the same direction.
- Repeat 10 times.
- Athletes start 10m apart.
- Aiii:
- Athletes start 10m apart.
- They jog facing toward each other until they are 5m apart, pivot to face the same direction (standing shoulder to shoulder), and remain in this position for 5 seconds.
- The athletes then jog back facing away from each other to their original starting locations and pause for another 5 seconds facing in the same direction.
- Repeat 10 times.
- Athletes start 10m apart.
- Bi:
- Same as Ai, but no stopping in the middle (1m distance)
- Bii:
- Same as Ai, but stop for only 2 seconds in the middle (1m distance)
- Biii:
- Same as Ai, but stop for 10 seconds in the middle (1m distance)
- Ci:
- For 60s in a 10m by 10m area, athletes make random movements, coming together occasionally.
- Repeat 3 times.
- Cii:
- For 60s in a 10m by 10m area, athletes make random movements with one athlete shadowing the other.
- Repeat 3 times.
A schematic diagram of the validation protocol is included below. 
Data Collection and Brief Error Discussion
Wearable Device and Satellite Positional Information
The data for this investigation was measured using Catapult Vector S7 devices which rely on GPS and GLONASS satellite constellations for 10 Hz positional information. As stated in the introduction, the objective of detecting all incursions within a 2m threshold around an athlete is used as a baseline for which to assess the dwell-time and close proximity zone radius parameters. Due to measurement error associated with satellite-derived positional information, close proximity is regarded as 2m (±1m).
The following table contains positional quality information for each portion of the protocol.
Video Verification of Expected Results
Expected incursion duration data was verified after completing the protocol using Vision video analysis software.
Participants followed strict adherence to the protocol described above, but some of the expected durations drifted slightly from those described in the protocol. The video-verified durations were used in the following analysis.
Even after manual video verification of expected results by an experienced video analyst, accuracy of these expected results is still subject to human error.
Notes:
- The protocol was designed so that phase Aii could be used to test the limit of the 2m threshold. It was decided that, ideally, no incursions would occur at this exact limit, and expected incursion duration was set to zero.
- The expected incursion count for phase Cii is 1 as the athletes are constantly engaged in a 2m shadowing exercise.
Summary Tables and Graphs
Data Summary By Period
The table below contains the complete set of data collected for each trial, drill, close proximity zone radius, and dwell-time. The table illustrates the effect of manipulating dwell-time and close proximity zone radius for each drill. Expected incursion durations and counts can be compared with measured incursion duration and count for each period, radius, and dwell-time combination. Durations have been rounded to the nearest second. Percentage errors have been selected as an evaluation metric*.
High-Level Overview by Close Proximity Zone Radius and Dwell-Time
To simplify the table above, all periods can be combined and grouped by dwell-time and Close Proximity Zone radius only. This provides a high-level overview of overall incursion count and duration that might be more suitable for quick comparisons.
A plot of the data in the table above may be easier to interpret. The data in the table directly above is summarized in the following plot.
The plots above show results for each individual trial and for the combined trials. While only two trials are shown, the similarity between the plots for each trial suggests strong reliability. The plots also illustrate the trade-off between sensitivity (the proportion of incursions that are correctly identified) and specificity (the proportion of non-incursions that are correctly identified). Ideally, parameters would be chosen such that both are maximized, but the lighter areas of the graph indicate a more optimal trade-off. In practical terms, percentage difference would be zero, but arguments could be made to err on the side of over-detection for the use-case of tracking possible viral transmission between athletes.
Another representation of the combined data from each trial is visualized below.
From the scatter plot, it is apparent that a dwell-time of 1s combined with a Close Proximity Zone radius of 2.5m provides results with the lowest deviation from expected (closest to the origin) for this protocol. However, incursion count is slightly higher (+3%) and duration is slightly lower (-2%) than expected with this combination.
Conclusions
Based on the Proximity Testing Protocol results, which were obtained in a controlled setting with only two athletes, a Close Proximity Zone Radius of 2.5m and a dwell-time of 1s seems to be most favorable when the objective is to report time spent within 2m of other athletes. While other internal investigations also support this general finding, it should be noted that the results may vary depending on a range of technical and environmental factors.
* Percentage errors were selected for readibility, but the reader should be aware that this implies infinite error for any nonzero measurement when zero is the expected result. A side-by-side analysis was completed using relative differences scaled by the mean of the expected and measured results. The differences were minimal when comparing the analyses, and the authors opted to use percentage error for ease of understanding.