Gait Related Activity Based Person Authentication with Smartphone Sensors

Abstract

Smartphones are recently becoming more and more sophisticated with numerous applications and a large number of people are becoming habituated with their use in everyday life. With the vast use of smartphones in various routine everyday transactions, the need of secured access control is increasing as people tend to store their personal and important information in the mobile devices. The existing popular methods of securing mobile devices, pincodes or patterns, can be vulnerable if gets lost or stolen. In this work, a novel framework for user authentication technique based on human gait related activities analyzed from smartphone sensors data has been studied. Being non-intrusive and continuously available, human gait behaviour analyzed from smartphone sensors data provides an opportunity of developing convenient and user friendly means of user authentication. Benchmark data sets from smartphone sensors are used for simulation experiments. It is found that activity dependent authentication method produces better accuracy than activity independent authentication. It is also found that convolutional neural networks based classification is promising compared to traditional machine learning classifiers.