Spatial footstep recognition by convolutional neural networks for biometric applications

We propose a Convolutional Neural Network model to learn spatial footstep features end-to-end from a floor sensor system for biometric applications. Our model's generalization performance is assessed by independent validation and evaluation datasets from the largest footstep database to date, containing nearly 20,000 footstep signals from 127 users. We report footstep recognition performance as Equal Error Rate (EER) in the range of 9% to 13% depending on the test set. This improves previously reported footstep recognition rates in the spatial domain up to 4% EER.