BlackBerry has launched the KEYone aka "Mercury", a few hours ago, to kick off this…
Wearables: Algorithm reacts in real-time
Researchers at the University of Sussex have developed a new algorithm that allows Wearables to record every movement in real time without prior data input and to assign a specific activity more precisely. Machine learning thus increases the hit rate rapidly.
“Current activity detection systems usually fail because they can only detect a previously defined collection of activities while human movements are not limited and change over time,” explains Hristijan Gjoreski. “We have access here with machine learning, which can recognize new human activities in real time – and rival thinking approaches.”
Traditional models first collect information about the movement to finally assess what a user has done. When a system steps, it assumes that the person is running. If there is a pause, the activity of running is completed. As soon as the person goes back, the system starts a new run. The new algorithm can prevent this, because it analyzes the person in real time. If a run is interrupted briefly, the system waits.
The system can also use the recorded data of the movements to determine whether the run is to be continued or not – and it keeps the data already recorded in a kind of waiting position. If the person actually continues to run, the previous run is also continued at the stopped position. The algorithm also reliably detects whether the user is sitting or standing.
“Smartwatches of the future will be able to better analyze and understand activities by automatically detecting when we are dealing with a new activity. Just as for fitness and lifestyle tracker, the system can be used in health care or research on consumer behavior, “says researcher Daniel Roggen.
The new system will be presented this month at the International Symposium on Wearable Computers 2017.