Watch a 21-minute conference talk from tinyML EMEA exploring an innovative ultra-low power gesture recognition system that utilizes piezoelectric micromachined ultrasonic transducer (pMUT) arrays and spike-based beamforming. Learn how Research Engineer Emmanuel Hardy from CEA Leti demonstrates a novel approach to reducing energy consumption in battery-powered smart sensors by processing information in the spiking domain. Discover the system's components including pMUT arrays, driving/sensing electronics, and a unique spike-based beamforming strategy that extracts distance and angle information without conventional analog-to-digital converters. Follow along as Hardy explains how their Spiking Recurrent Neural Network achieves 86% classification accuracy across five 3D gestures, and explore practical applications, experimental results, and state-of-the-art comparisons in this cutting-edge research presentation.
Ultra-Low Power Gesture Recognition with pMUT Arrays and Spike-based Beamforming
EDGE AI FOUNDATION via YouTube
Overview
Syllabus
Intro
System Overview
Key Points
The team
Acoustic setup
Spike-based Beamforming
Feature Vector
Example of Gestures
Gesture Classifier
Spiking Recurrent Unit
Right-Left example
Experimental Setup
Gesture Dataset
Classification Results
State-of-the-Art
Applications and Perspectives
Takeaway Points
Taught by
EDGE AI FOUNDATION
