Feature diversity for fall detection and human indoor activities classification using radar systems

Abstract

This paper presents preliminary analysis of radar signatures for fall detection and classification of human indoor activities, to monitor the daily behaviour of individuals at risk of deteriorating physical or cognitive health. Two datasets of signatures in different environments have been collected, one of which included signatures generated from signals simultaneously collected from a radar and an RGB-D Kinect sensor, on a couple of older individuals. This preliminary analysis shows the potential effectiveness of different features and classifiers, and highlights the need of additional investigation to characterise and exploit the diversity of features and classification methods, in different experimental scenarios with different subjects.

A Non-Stationary Channel Model for the Development of Non-Wearable Radio Fall Detection Systems

“…It is noteworthy that some daily-life activities can imprint the time-frequency distribution of the channel with similar signatures to those of fall incidents. Lying in bed, sitting on chair or ground, jumping over a couple of stairs, and chaotic movements of pets are examples of activities that either have similar trajectory and/or temporal features to that of a falling incident 14 . To reduce the number of potential false alarms caused by such activities, the utilization of machine learning models has been proposed in the literature (see the systems based on the learning method in [26]).…”

“…The use of a multiple range-Doppler radar for increasing the reliability of the fall detection system was proposed very recently in [13]. In order to monitor the daily behaviour of individuals at risk of deteriorating physical or cognitive health, analysis of radar signatures for fall detection and classification of human indoor activities has been presented in [14]. The system proposed in [15] applies the frequency distribution trajectories corresponding to the velocities of the movements while falling to a hidden Markov model.…”

A Multisensory Approach for Remote Health Monitoring of Older People

“…For radar, the micro-Doppler effect [21] is visible within the spectrograms. These are Doppler versus time plots where the movements of torso, limbs, and other body parts generate a distinctive pattern; some examples are shown in [22]. To generate the spectrogram, Short-Time Fourier Transform of the radar data with a window of 0.2s and 95% overlap were taken to characterise the time variant Doppler shifts associated with the movement of different body parts.…”

“…Different features have been suggested in the literature for classification with radar [22,26] and these features, listed in Table II, can be grouped into three categories: Physical, Transform domain, and Textural.…”

Performance Analysis of Classification Algorithms for Activity Recognition Using Micro-Doppler Feature

“…Motion capture (Mocap) [13,14] is the process of recording the movement of objects or people. Complex movement can be recreated in a physically accurate manner, such as secondary motions, weight, and exchange of forces.…”