An LSTM Recurrent Network for Step Counting

TL;DR

This paper introduces an LSTM recurrent neural network model designed to accurately count steps using smartphone sensor data, benefiting both blind and sighted users with a 5% error rate.

Contribution

It presents a novel application of LSTM networks for step counting tailored to different user groups using a new annotated dataset.

Findings

Achieved 5% overcount and undercount rate in step counting

Demonstrated effectiveness for both blind and sighted users

Utilized Leave-One-Out training modality for model evaluation

Abstract

Smartphones with sensors such as accelerometer and gyroscope can be used as pedometers and navigators. In this paper, we propose to use an LSTM recurrent network for counting the number of steps taken by both blind and sighted users, based on an annotated smartphone sensor dataset, WeAllWork. The models were trained separately for sighted people, blind people with a long cane or a guide dog for Leave-One-Out training modality. It achieved 5% overcount and undercount rate.