Development of a step counter based on artificial neural networks

Abstract

The research field of indoor navigation focuses on the position estimate in GNSS-shaded (Global Navigation Satellite System) areas. Accelerometer and gyroscope as Micro Electro Mechanical Systems are used for a position estimate based on Pedestrian Dead Reckoning, where the accelerometer is used for step detection. To realize a usefull position estimate for pedestrian navigation, the distance accuracy of the step length has to be less than a few centimeters to reach this accuracy, because the distance errors add up themselves time-dependent. This step length is derived from the sensor data of the integrated accelerometer of the smartphone. In this article, a step counter with step length estimate based on a artificial neural network (ANN) is described. The Matlab toolbox Neural Network is used to generate the structure of ANN. After leveling the three axis accelerometer the z-axis acceleration will be used to realize a ANN based on data from more than forty persons. Besides that, the results will be compared to an alternative approach, while two conditions are used which successively must be fulfilled. The results of this investigation reveal a step recognition rate of 99.5% as well as an average distance error of 9% of the respective distance.