One-Class Subject Identification From Smartphone-Acquired Walking Data

In this work, a novel type of human identification system is proposed, which has the aim to recognize a user from his biometric traits of his way of walk. A smartphone is utilized to acquire motion data from the built-in sensors. Data from accelerometer and gyroscope are processed through a cycle extraction phase, a Convolutional Neural Network for feature extraction and a One-Class SVM classifier for identification. From quantitave results the system achieves an Equal Error Rate close to 1%