Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Resonate-and-Fire Spiking Neurons for Target Detection and Hand Gesture Recognition: A Hybrid Approach

Ahmed Shaaban, Zeineb Chaabouni, Maximilian Strobel, Wolfgang Furtner, Robert Weigel, Fabian Lurz

TL;DR

This work addresses hand gesture recognition (HGR) with FMCW radar by bypassing computationally heavy FFTs. It introduces resonate-and-fire (RAF) neurons for direct time-domain hand detection, followed by a Goertzel-based extraction of five features and a GRU classifier to distinguish five gestures, achieving 98.21% mean accuracy. The approach demonstrates comparable performance to FFT-based methods while reducing pipeline complexity, with potential benefits for neuromorphic hardware implementations. Overall, the paper presents a practical, FFT-free HGR solution that leverages RAF dynamics and light feature extraction for efficient, accurate gesture recognition in radar data.

Abstract

Hand gesture recognition using radar often relies on computationally expensive fast Fourier transforms. This paper proposes an alternative approach that bypasses fast Fourier transforms using resonate-and-fire neurons. These neurons directly detect the hand in the time-domain signal, eliminating the need for fast Fourier transforms to retrieve range information. Following detection, a simple Goertzel algorithm is employed to extract five key features, eliminating the need for a second fast Fourier transform. These features are then fed into a recurrent neural network, achieving an accuracy of 98.21% for classifying five gestures. The proposed approach demonstrates competitive performance with reduced complexity compared to traditional methods

Resonate-and-Fire Spiking Neurons for Target Detection and Hand Gesture Recognition: A Hybrid Approach

TL;DR

This work addresses hand gesture recognition (HGR) with FMCW radar by bypassing computationally heavy FFTs. It introduces resonate-and-fire (RAF) neurons for direct time-domain hand detection, followed by a Goertzel-based extraction of five features and a GRU classifier to distinguish five gestures, achieving 98.21% mean accuracy. The approach demonstrates comparable performance to FFT-based methods while reducing pipeline complexity, with potential benefits for neuromorphic hardware implementations. Overall, the paper presents a practical, FFT-free HGR solution that leverages RAF dynamics and light feature extraction for efficient, accurate gesture recognition in radar data.

Abstract

Hand gesture recognition using radar often relies on computationally expensive fast Fourier transforms. This paper proposes an alternative approach that bypasses fast Fourier transforms using resonate-and-fire neurons. These neurons directly detect the hand in the time-domain signal, eliminating the need for fast Fourier transforms to retrieve range information. Following detection, a simple Goertzel algorithm is employed to extract five key features, eliminating the need for a second fast Fourier transform. These features are then fed into a recurrent neural network, achieving an accuracy of 98.21% for classifying five gestures. The proposed approach demonstrates competitive performance with reduced complexity compared to traditional methods
Paper Structure (19 sections, 1 figure, 3 tables)

This paper contains 19 sections, 1 figure, 3 tables.

Table of Contents

  1. Introduction
  2. Radar System and Dataset
  3. Radar System
  4. Dataset
  5. Proposed Recognition System
  6. Resonate-and-Fire Neurons: Frequency-Selective Processing without Training
  7. Refined Gesture Labeling and Efficient Target Detection using RAF Neurons
  8. Goertzel Algorithm
  9. Features Extraction
  10. Architecture
  11. Experimental Setup
  12. Dataset Preparation
  13. Network Training
  14. Testing and Evaluation
  15. Results and Discussion
  16. ...and 4 more sections

Figures (1)

  • Figure 1: Hand gesture recognition approach: The approach utilizes a layer of 32 resonate-and-fire (RAF) neurons to effectively detect the hand's resonance frequency. Subsequently, the Goertzel algorithm is employed at this specific frequency to calculate the discrete Fourier transform coefficient. This coefficient, along with simple angular and phase estimations, allows for extracting five essential features. These features are then fed into a recurrent neural network, enabling the system to classify between five distinct hand gestures.