Smartphones have become essential devices in our everyday lives, therefore they are the perfect basis for pervasive movement monitoring, if they are carried close to our body, in a pocket or belt strap. Nevertheless, in some situations, for example at home, when bathing or when the smartphone cannot be placed close to the body, wearable devices are a better choice. Fraunhofer AICOS hardware team has been developing a wearable device containing an inertial measurement unit with accelerometer, gyroscope, magnetometer and barometer sensors, the Pandlets. The Pandlets are a flexible platform created to support Fraunhofer AICOS research in various domains. It can include diverse sensor modules and allows a seamless integration with Android operative system. In the scope of the partnership between Fraunhofer AICOS and the Dutch company Gociety the wearable Pandlets was designed and shaped into the GoLive Wear. This wearable presents enhanced features such as waterproof casing with an innovative clip design and wireless charging. The GoLive Wear is in final tests and will be introduced in the market in the second half of 2016.
Fraunhofer AICOS has been developing algorithms for pervasive daily physical activities monitoring and falls detection having as input signals from inertial sensors built-in smartphones. These algorithms have now been adapted to process the signals from GoLive Wear inertial sensors taking into account the fact that the wearable may be placed in the body differently from the smartphone.
Activity monitoring algorithms use accelerometer signals, which were acquired from a set of 18 volunteers. Daily living activities such as walk, run, sit and stand were performed, with the GoLive Wear sensors placed in different positions (as chest and belt). In addition, samples of not using the wearable sensor and random movements (called tilt) were also collected. Overall, the dataset considered has more than 21 hours of acquisitions. Temporal and statistical features were extracted from signals, which were then used to train a classification algorithm based on decision trees. As well as in the fall detector algorithm, the overall offline results achieved the same level of accuracy as for the algorithm developed to the smartphone, where all the activities (walk, run, stand, sit, not using and tilt) were recognized with an accuracy above 90%. Real time tests are still in progress. Moreover, the algorithms include also estimation the number of steps, the walked distance and the expended kilocalories.
Data from the GoLive Wear sensors were collected yielding 1000 samples of simulated falls and 1000 samples of activities that are not considered a fall event, as sit on chair, run or put the sensor in the table. This dataset was used to adapt the state machine for fall detection that is based on thresholds of signal amplitude and angle variations. Testing the algorithm with a different group of samples, the overall results revealed to be in the same level of accuracy as for the algorithm developed to the smartphone. The capability to detect falls and the ability to do not detect a non-fall events achieved rates above 90%. When using the algorithm continuously during the day, the number of false alarms was also considerable small. The algorithm was adjusted to three levels of sensibility: low, medium and high sensitivity.
Pervasive daily physical activities monitoring is a key technology to enable seniors to live autonomously for longer periods. Adequate physical activity has been shown to correlate with the prevention of several pathological conditions such as obesity, cardiovascular diseases, osteoporosis, cancer and depression. Moreover, by constantly monitoring the user’s movements it is possible to automatically detect when a fall occurs. Automatic fall detection has the potential to accelerate the assistance response time after a fall event, which has been shown to improve the prognosis of the fall victims. Fraunhofer AICOS has been working since 2009 in solutions for pervasive daily physical activities monitoring and falls detection based on the inertial sensors built-in smartphones and, more recently, wearables.