An Approach for Determining Linear Velocities of Athletes from Acceleration Measurements using a Neural Network

Peter Christ, Felix Werner, Ulrich Rückert, and Jörg Mielebacher


gait analysis, velocity estimation, accelerometer, wireless body sensor, neural network


With recent technology advancements, miniaturized wireless body sensor (WBS) systems equipped with accelerometers allow the recording of an athlete’s actions without impact on their execution. This paper addresses the problem of determining linear velocities of athletes using a single tri-axial accelerometer located in a WBS on a chest strap. We propose using a neural network to associate features extracted from acceleration signals in time and frequency domain with particular velocities. This method neither requires a kinematic model nor information on characteristics of an athlete (e.g. height or weight). In a treadmill experiment with 20 subjects we obtain more than 97% correct classifications for velocities from 3 to 9 km/h. We demonstrate that using a set of simple features (variance, amplitude, RMS) a classification rate of more than 95% can be obtained. Misclassifications are mainly found at higher velocities (9 and 11 km/h) due to severe inter-subject differences in the acceleration signals.

Important Links:

Go Back