Enhanced User Motion Detection using Multiple Smartphone Sensors

Abstract: Number of research papers have been published in recent years which demonstrate how data received from sensors can be used for context aware computing. Also due the fact that nowadays smartphones come equipped with many motion sensors, people have begun developing interesting mobile applications that do automatic context aware computing. In this paper we discuss how data sensed from smartphone‟s motion sensors like accelerometer and gyroscope, can be used to predict user‟s motion type like cycling, walking, driving car and travelling in train, and furthermore do automatic motion profiling like putting the phone on vibrate mode or opening an application etc. First of all, we compared performance of various algorithms like Random Forest, J48, REPTree, Naïve Bayes, Rotation Forest (ensemble method) etc. while classifying motion types. Secondly, we pointed out the challenges faced when using only two sensors (i.e. accelerometer & gyroscope) to predict motion type and also when an identical motion type exists in data set like walking and cycling. Thirdly, we proposed a technique that makes use of an additional sensor GPS & google maps along with accelerometer & gyroscope to correct wrong predictions made by classifying algorithm to further improve the existing model with aim to deploy it for practical situation. This technique is based on usage of speed and location parameters that logically corrects wrong class. This is a promising approach when data set contains similar motion types.

Keywords: Context Awareness; Motion Type Prediction; Algorithm Comparison; Error Correction.