Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Exploring FMCW Radars and Feature Maps for Activity Recognition: A Benchmark Study

Ali Samimi Fard, Mohammadreza Mashhadigholamali, Samaneh Zolfaghari, Hajar Abedi, Mainak Chakraborty, Luigi Borzì, Masoud Daneshtalab, George Shaker

TL;DR

This study introduces a privacy-preserving HAR framework using a 60 GHz FMCW radar, treating multi-dimensional feature maps (Range-Doppler, Range-Azimuth, Range-Elevation) as 3D data vectors instead of images. A novel dataset collected in home-like environments supports seven activity classes, with two validation schemes (CSV and LOPO-CV) to assess generalization. ConvLSTM emerges as the top model, leveraging spatiotemporal cues across all three feature maps to achieve up to 90.51% accuracy in CSV and 89.56% in LOPO-CV for seven activities, with even higher performance for four activities. The findings demonstrate that combining RD, RA, and RE maps preserves structural information and improves recognition, enabling scalable, non-intrusive activity monitoring suitable for real-world deployment, particularly in healthcare and smart homes.

Abstract

Human Activity Recognition has gained significant attention due to its diverse applications, including ambient assisted living and remote sensing. Wearable sensor-based solutions often suffer from user discomfort and reliability issues, while video-based methods raise privacy concerns and perform poorly in low-light conditions or long ranges. This study introduces a Frequency-Modulated Continuous Wave radar-based framework for human activity recognition, leveraging a 60 GHz radar and multi-dimensional feature maps. Unlike conventional approaches that process feature maps as images, this study feeds multi-dimensional feature maps -- Range-Doppler, Range-Azimuth, and Range-Elevation -- as data vectors directly into the machine learning (SVM, MLP) and deep learning (CNN, LSTM, ConvLSTM) models, preserving the spatial and temporal structures of the data. These features were extracted from a novel dataset with seven activity classes and validated using two different validation approaches. The ConvLSTM model outperformed conventional machine learning and deep learning models, achieving an accuracy of 90.51% and an F1-score of 87.31% on cross-scene validation and an accuracy of 89.56% and an F1-score of 87.15% on leave-one-person-out cross-validation. The results highlight the approach's potential for scalable, non-intrusive, and privacy-preserving activity monitoring in real-world scenarios.

Exploring FMCW Radars and Feature Maps for Activity Recognition: A Benchmark Study

TL;DR

This study introduces a privacy-preserving HAR framework using a 60 GHz FMCW radar, treating multi-dimensional feature maps (Range-Doppler, Range-Azimuth, Range-Elevation) as 3D data vectors instead of images. A novel dataset collected in home-like environments supports seven activity classes, with two validation schemes (CSV and LOPO-CV) to assess generalization. ConvLSTM emerges as the top model, leveraging spatiotemporal cues across all three feature maps to achieve up to 90.51% accuracy in CSV and 89.56% in LOPO-CV for seven activities, with even higher performance for four activities. The findings demonstrate that combining RD, RA, and RE maps preserves structural information and improves recognition, enabling scalable, non-intrusive activity monitoring suitable for real-world deployment, particularly in healthcare and smart homes.

Abstract

Human Activity Recognition has gained significant attention due to its diverse applications, including ambient assisted living and remote sensing. Wearable sensor-based solutions often suffer from user discomfort and reliability issues, while video-based methods raise privacy concerns and perform poorly in low-light conditions or long ranges. This study introduces a Frequency-Modulated Continuous Wave radar-based framework for human activity recognition, leveraging a 60 GHz radar and multi-dimensional feature maps. Unlike conventional approaches that process feature maps as images, this study feeds multi-dimensional feature maps -- Range-Doppler, Range-Azimuth, and Range-Elevation -- as data vectors directly into the machine learning (SVM, MLP) and deep learning (CNN, LSTM, ConvLSTM) models, preserving the spatial and temporal structures of the data. These features were extracted from a novel dataset with seven activity classes and validated using two different validation approaches. The ConvLSTM model outperformed conventional machine learning and deep learning models, achieving an accuracy of 90.51% and an F1-score of 87.31% on cross-scene validation and an accuracy of 89.56% and an F1-score of 87.15% on leave-one-person-out cross-validation. The results highlight the approach's potential for scalable, non-intrusive, and privacy-preserving activity monitoring in real-world scenarios.

Paper Structure

This paper contains 18 sections, 2 equations, 4 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Methodology
  4. Radar Setup and Data Collection
  5. Data Preprocessing and Feature Extraction
  6. Blackman-Harris Window
  7. Range-FFT Map
  8. Moving Target Indicator (MTI)
  9. Range-Doppler Map
  10. Capon Algorithm
  11. Activity Recognition Models
  12. Experimental Evaluation
  13. Experimental Setup
  14. Experimental Results
  15. CSV
  16. ...and 3 more sections

Figures (4)

  • Figure 1: Overview of the proposed framework for FMCW radar-based HAR.
  • Figure 2: Examples of a participant performing the activities with corresponding feature maps. ($A_1$) walking, ($A_2$) sitting on the bed, ($A_3$) sitting on the chair, ($A_4$) lying down on the bed, ($A_5$) lying down on the floor, ($A_6$) empty room, ($A_7$) transition.
  • Figure 3: Confusion matrices for the ConvLSTM model on various activity sets under CSV (RD+RA+RE inputs).
  • Figure 4: Training and validation loss curves for the ConvLSTM model under CSV (RD+RA+RE inputs).