Analyzing Multiple Accelerometer Configurations to Detect Falls and Motion

The use of wearable devices with accelerometers developed to detect falls and motion has been continuously growing because of their small size, low weight, energy efficiency, and low price. However, the number of sensors, their position in the body and estimation methodologies are still open issues when tested in uncontrolled conditions. In this paper, we perform a discriminant analysis to determine which combinations of feature extraction characteristics and accelerometer positions are best fitted to estimate falls and particular movements. A dataset of 33 activities recorded with eight accelerometers distributed along the body of six participants is released as part of this work. Our analysis concludes that a waist-arm combination with a statistical feature is best fitted for most cases. But this work, rather than a single conclusion, is intended to provide a benchmark to other authors. The specific tests explain for example that the foot is not a good location for the accelerometer, or that static features are less relevant than dynamic ones.