The Triton unit is well equipped to recognize and handle all swimming behaviours, however, there are some activities that will produce odd-looking data in your workouts
TritonWear is great at recognizing swimming but has a few dislikes. It will still capture data for all activities, and will automatically disregard outlier data during calculations. However, recognizing unsupported activities will help you understand why you see certain results in TritonWear when tracking live data, or reviewing historical swims. The most commonly encountered unsupported activities that will affect data accuracy are:
- Snorkels. Because TritonWear depends on head movement to recognize the stroke, snorkels make it harder to classify stroke types. This is because the head does not move for breathing, which is one of the indicators used to identify the stroke being performed.
- Weak Starts. In order to recognize the beginning of swimming the unit looks for a clear start swim signal. If a swimmer lazily pushes off the signal might not change fast enough for the unit to consider it the start of swimming, so the length would not be captured.
- Partial Lengths. The unit uses the start and end swim, along with the pool distance (set at the beginning of the workout) to calculate the metrics. If an athlete swims any less than the full pool distance, the system will still calculate the metrics as though they swam a full 25m. This will result in some very unexpected metrics values.
- Mid-length Stroke Switch. The unit will only classify stroke type once per length, so if an athlete switches strokes mid-length the unit will continue to calculate their metrics based on the initial stroke type, but will no longer find strokes as the movements entirely change. This will cause the data for that lap to be lost.
- Ending on a Turn. For TritonWear, a turn signifies another lap is about to be executed. If no lap is swum after a turn the unit will continue to calculate any movement into a lap until it thinks the lap should have ended (time passes), producing false metrics.
- Drills. Most drills will be recognized, as they appear mostly like the stroke they are teaching, however drills that completely change the look of a stroke (breast with dolphin kick, one arm fly, no breath free, treading, etc.) will cause the unit to not know what stroke type to assign, and therefore produce no metric. Drills using equipment (fins, paddles, pull buoys) will produce metrics normally but will be considered outliers in calculating Focus comparisons. The only caveat being if the equipment is used more often than not, it can impact the Focus comparison calculations, as the machine learning will not be able to tell which is typical and which is an outlier.
- One-arm Fly. As mentioned above, 1 arm fly looks very confusing for our system. It has characteristics consistent with both freestyle and butterfly and is therefore often classified incorrectly.
For a more complete look at all of the elements that go into good data collection, check out this book.