Robust Personal Identification Using Wearable Devices Based on LSTM and CNN

Abstract

Various studies exist to identify individuals. Personal identification research based on inertial data, that is, acceleration and angular velocity acquired with an inertial sensor, is also one of these efforts. In fact, when learning inertial data with convolutional neural network (CNN), it is known to be able to identify individuals with high accuracy. However, we found that the individual identification model using inertial data significantly lowers the performance of recognition from 99% to 81% when the shoes worn by the individual change. This paper deals with solving this problem by using a gait cycle extracted from inertial data. First, we study a method to detect the gait cycle using long short-term memory, a representative recurrent neural network model. Second, the CNN model using the gait cycle is implemented, and then the model is evaluated with the typical performance evaluation indicators such as accuracy, precision, recall, and F1-score. As a result, it is confirmed that the proposed model can identify individuals with more than 90% accuracy even when the shoes worn are different.