Online Human Gesture Recognition using Recurrent Neural Networks and Wearable Sensors

TL;DR

This paper introduces SLOTH, an online gesture recognition system using wearable accelerometers and RNNs, achieving reliable, real-time interpretation of human gestures for improved human-robot interaction.

Contribution

The paper presents a novel architecture combining wearable sensors and RNNs for continuous, online gesture recognition with high accuracy and responsiveness.

Findings

High precision and recall in gesture recognition

Immediate system reactivity achieved

Effective modeling of gesture probabilities

Abstract

Gestures are a natural communication modality for humans. The ability to interpret gestures is fundamental for robots aiming to naturally interact with humans. Wearable sensors are promising to monitor human activity, in particular the usage of triaxial accelerometers for gesture recognition have been explored. Despite this, the state of the art presents lack of systems for reliable online gesture recognition using accelerometer data. The article proposes SLOTH, an architecture for online gesture recognition, based on a wearable triaxial accelerometer, a Recurrent Neural Network (RNN) probabilistic classifier and a procedure for continuous gesture detection, relying on modelling gesture probabilities, that guarantees (i) good recognition results in terms of precision and recall, (ii) immediate system reactivity.