Time Domain Integration in Range Doppler/Angle Images for Human Gesture Recognition with FMCW Radar

Maiwald T, Schmidt P, Fischer G, Lurz F (2025)

Publication Language: English

Publication Status: Accepted

Publication Type: Unpublished / Preprint

Future Publication Type: Conference contribution

Publication year: 2025

Event location: Bali, Indonesia ID

Abstract

This paper presents a radar data labeling method and discusses results obtained by training a CNN for human gesture recognition with FMCW radar. Superimposed range-Doppler and range-angle images are used to integrate time dependent information into single images, which reduces the amount of data the CNN has to deal with. Obtained results indicate 92% accuracy in terms of the automated label assignment. An average classification accuracy of 85% is achieved, whereas this value drops by 2% only if in addition the CNN also differentiates between the left and right hand resulting in 23 categories.

Authors with CRIS profile

Timo Maiwald Lehrstuhl für Technische Elektronik Georg Fischer Professur für Technische Elektronik

Related research project(s)

Embedded Machine Learning (eMiL) Oct. 1, 2021 - Sept. 30, 2024

Involved external institutions

Otto-von-Guericke-Universität Magdeburg DE Germany (DE)

How to cite

APA:

Maiwald, T., Schmidt, P., Fischer, G., & Lurz, F. (2025). Time Domain Integration in Range Doppler/Angle Images for Human Gesture Recognition with FMCW Radar. (Unpublished, Accepted).

MLA:

Maiwald, Timo, et al. Time Domain Integration in Range Doppler/Angle Images for Human Gesture Recognition with FMCW Radar. Unpublished, Accepted. 2025.

BibTeX: Download