Radar-Based Human Activity Recognition in Restrooms and Beds

Background

In elderly monitoring and nursing-care support, early detection of falls, long periods of inactivity, bed-leaving events, and abnormal behavior in restrooms is important. However, restrooms and bed areas are highly privacy-sensitive, and continuous camera monitoring is often unacceptable.

Radar can measure human range, velocity, motion, and reflection distributions without capturing visual images. Therefore, it is a promising sensing modality for activity recognition and anomaly detection in privacy-sensitive environments such as restrooms and beds.

Furthermore, FMCW radar and MIMO radar can provide radar images that include range and angle information, in addition to simple velocity waveforms. Radar-based human imaging may capture human position and posture changes in more detail and can be applied to healthcare monitoring.

Research Objectives

This research aims to realize radar-based activity recognition and human imaging in daily living environments where privacy must be protected.

Main topics include:

• Recognition and posture estimation of standing, sitting, and falling behaviors in restrooms

• Recognition and posture estimation of sitting up, turning over, and bed-leaving motions around beds

• Activity classification using Doppler radar

• Human imaging using FMCW radar and MIMO radar

• Machine learning and deep learning using radar images and time-velocity distributions

• Privacy-preserving monitoring without cameras

• Development of fundamental radar motion imaging technologies

Related Work

Our previous work has investigated classification of human behaviors and falls using multiple Doppler radars installed in a restroom. By using time-velocity distributions and radar signal features with machine learning, the research evaluates the possibility of camera-free monitoring systems for privacy-sensitive spaces.

In addition, our research explores the use of FMCW radar and MIMO radar for measuring range-angle information, reflection distributions, and radar-based human images around beds and indoor environments. The aim is to capture human position, posture, and motion changes as radar images and apply them to monitoring and healthcare applications.

Selected Publications

• Xiangbo Kong, Kenshi Saho, and Akari Takebayashi , "MDSCNet: A Lightweight Radar Image-Based Model for Multi-Action Classification in Elderly Healthcare," Inventions, vol. 10, article no. 98, October 2025.

• Kenshi Saho, Sora Hayashi, Mutsuki Tsuyama, Lin Meng, and Masao Masugi, "Machine Learning-Based Classification of Human Behaviors and Falls in Restroom via Dual Doppler Radar Measurements," Sensors, vol. 22, article no. 1721, March 2022. 

• ZhiChen Wang, Zelin Meng, Kenshi Saho, Kazuki Uemura, Naoto Nojiri, and Lin Meng, "Deep Learning based Elderly Gender Classification using Doppler Radar," Personal and Ubiquitous Computing, vol. 26, pp. 1067-1079, August 2022.  

• Shuhei Hashimoto, Xiangbo Kong, Kosuke Manabe, Hiroshi Minematsu, and Kenshi Saho, "Classification of Behaviors Related to Bed-Leaving and Bed-Lying Using Millimeter-Wave FMCW Radar,"  The 2023 International Symposium on Advanced Technologies and Applications in the Internet of Things (ATAIT 2023), Kusatsu, Japan, August 2023. 

Keywords

restroom activity recognition, bed activity recognition, fall detection, bed-leaving detection, radar-based human imaging, FMCW radar, MIMO radar, Doppler radar, dual Doppler radar, time-velocity distribution, radar image, human sensing, privacy-preserving monitoring, healthcare monitoring, nursing-care monitoring, machine learning, radar signal processing