Deep Learning Methods for Speed Estimation of Bipedal Motion from Wearable IMU Sensors

Abstract

The estimation of the speed of human motion from wearable IMU sensors is required in applications such as pedestrian dead reckoning. In this paper, we test deep learning methods for the prediction of the motion speed from raw readings of a low-cost IMU sensor. Each subject was observed using three sensors at the shoe, shin, and thigh. We show that existing general-purpose architectures outperform classical feature-based approaches and propose a novel architecture tailored for this task. The proposed architecture is based on a semi-supervised variational auto-encoder structure with innovated decoder in the form of a dense layer with a sinusoidal activation function. The proposed architecture achieved the lowest average error on the test data. Analysis of sensor placement reveals that the best location for the sensor is the shoe. Significant accuracy gain was observed when all three sensors were available. All data acquired in this experiment and the code of the estimation methods are available for download.

Research on Deep Learning-Based Sensor Technology for Falls in Elderly Communities

“…In order to be able to describe various variables during the elderly's fall process conveniently, a spatial coordinate system must be established. Therefore, in elderly falls, accelerometers and gyroscopes in three-dimensional space are mainly used through inertial sensors to detect acceleration values and body tilt angles and analyze and judge the occurrence of body falls [5]. The fall alarm can use a six-axis motion sensor to detect the fall of the human body and use a combination of GPS and base station positioning (LBS) to track the movement trajectory of the elderly in real-time, realize fall alarm, trajectory query, and other functions, and also has a one-key SOS emergency call function, with the advantages of stability, reliability, and convenient portability [6].…”

A Deep Learning Approach for Biped Robot Locomotion Interface Using a Single Inertial Sensor

“…[ 18 , 19 ] proved that having only a few independent high-level user commands is enough to generate a completely new motion set from an animal-motion-trained network. However, unlike [ 18 , 19 ], such parameters can be computed from only a few wearable sensors without the need for industry-level motion capture systems [ 20 , 21 , 22 ]. These studies implemented estimation gait parameters (such as gait speed and specific gait events) utilizing deep learning methods from a few inertial sensors (≤3).…”