Attention-based deep learning networks for identification of human gait using radar micro-Doppler spectrograms

Hannah Garcia Doherty,Ronny I A Harmanny,Roeland P Trommel,Vasileios Papanastasiou,Roberto Arnaiz Burgueño

doi:10.1017/s1759078721000830

Abstract

AbstractIdentification of human individuals within a group of 39 persons using micro-Doppler (μ-D) features has been investigated. Deep convolutional neural networks with two different training procedures have been used to perform classification. Visualization of the inner network layers revealed the sections of the input image most relevant when determining the class label of the target. A convolutional block attention module is added to provide a weighted feature vector in the channel and feature dimension, highlighting the relevant μ-D feature-filled areas in the image and improving classification performance.

Highlights

For all sorts of military and civilian safety and security applications it is interesting to use radar to track individuals moving over ground
Non-machine learning approaches were used in [1, 2] proposing a particle filter method and handcrafted features, respectively
In [5, 6] deep convolutional neural networks (DCNNs) were used to deal with challenging problems such as personnel recognition based on multistatic μ-D and multi-target human gait classification revealing the potential of such networks

Summary

Introduction

For all sorts of military and civilian safety and security applications it is interesting to use radar to track individuals moving over ground. By adapting the time-on-target and measuring the μ-D signal of ground moving targets it is possible to classify moving personnel based on the most relevant and unique form of human motion, their human gait. Human gait classification based on μ-D has attracted the interest of many researchers. In [3, 4] machine learning approaches were used to solve gait classification problems such as activity and walking style classification. Deep learning-based methods have gained popularity in several fields, including radar. In [5, 6] deep convolutional neural networks (DCNNs) were used to deal with challenging problems such as personnel recognition based on multistatic μ-D and multi-target human gait classification revealing the potential of such networks. Identification accuracy of above 89% was achieved for less than 10 subjects in [7] and 98% in [8] considering that the subjects were walking on a treadmill

Methods

Results

Conclusion