Home // SENSORDEVICES 2023, The Fourteenth International Conference on Sensor Device Technologies and Applications // View article

Multi Human Posture Classification Using MIMO FMCW Radar Point Cloud and Deep Learning

Authors:
Sohaib Abdullah
Shahzad Ahmed
Junbyung Park
Chanwoo Choi
Sung Ho Cho

Keywords: Human Posture Recognition, FMCW Radar, Deep Learning, CNN

Abstract:
Human action and pose recognition in context of health and safety has lately attracted a huge amount of attraction. Human pose recognition using radar is a challenging task since the human under consideration is static. This paper uses Multi Input Multi Output (MIMO) Frequency Modulated Continuous Wave (FMCW) radar to recognize postures of two co-located humans using Convolutional Neural Network (CNN). Two humans at different angles and arbitrary distance (in living room) are simultaneously considered for data collection. Radar-extracted spherical coordinates of posture are acquired using Fast Fourier transform (FFT) and afterwards, spatial transformation is used to convert these points into Cartesian coordinate system. The resultant image shows the posture of two persons in a single image. A clustering approach is used to classify the two postures and CNN is trained to classify each posture. Promising accuracy is achieved for one and two persons posture recognition.

Pages: 24 to 29

Copyright: Copyright (c) IARIA, 2023

Publication date: September 25, 2023

Published in: conference

ISSN: 2308-3514

ISBN: 978-1-68558-091-9

Location: Porto, Portugal

Dates: from September 25, 2023 to September 29, 2023