GAIT BASED PEDESTRIAN IDENTIFICATION WITH REDUCING DEPENDENCY OF ACCELEROMETER POSITION

摘要

Identification of a user with accelerometer is a technology receiving high attention in pervasive computing, as it allows areas such as homes or vehicles to provide personalized service to the user. Conventional methods in this field are often based on machine learning that train on gait data obtained from accelerometers. These methods share a common issue in being sensitive to the location of where the sensor is attached; specifically, the sensor is intended to be located at the same position during the training phase, making it less versatile. In this paper, we propose an approach to incorporate multiple sensors that extract Reducing Dependency of accelerometer Position (RDP) features from their walking acceleration data to create a classifier with Support Vector Machine (SVM). Our experiments show that in comparison to a method that does not extract RDP, the accuracy is improved by 14.0-17.2% where the sensors are placed in different positions.