Energy Efficient Personalized Hand-Gesture Recognition with Neuromorphic Computing

TL;DR

This paper presents a novel neuromorphic approach for energy-efficient personalized hand-gesture recognition, enabling natural human-robot interaction with low power consumption.

Contribution

It introduces a new methodology for training spiking convolutional neural networks tailored for gesture recognition on neuromorphic hardware.

Findings

Preliminary results demonstrate effective gesture recognition.

Potential for personalized human-robot interaction.

Discussion of methods to improve model performance.

Abstract

Hand gestures are a form of non-verbal communication that is used in social interaction and it is therefore required for more natural human-robot interaction. Neuromorphic (brain-inspired) computing offers a low-power solution for Spiking neural networks (SNNs) that can be used for the classification and recognition of gestures. This article introduces the preliminary results of a novel methodology for training spiking convolutional neural networks for hand-gesture recognition so that a humanoid robot with integrated neuromorphic hardware will be able to personalise the interaction with a user according to the shown hand gesture. It also describes other approaches that could improve the overall performance of the model.