Development of Motion Detection Algorithm Using 3d Sensors for Patient Monitoring Support Service System

• Published in: Studies in health technology and informatics Vol. 329; p. 371
• Main Authors: Mukai, Masami, Yoshida, Yukihiro, Yotsukura, Masaya, Machida, Mieko, Kanemitsu, Miyuki, Miura, Yoshiaki, Nagase, Katsuya, Ozeki, Yota, Saito, Tomohisa, Kataoka, Masato, Watanabe, Shun-Ichi
• Format: Journal Article
• Language: English
• Published: Netherlands 07.08.2025
• Subjects: Algorithms; Detection Algorithms; Humans; Imaging, Three-Dimensional - instrumentation; Imaging, Three-Dimensional - methods; Monitoring, Physiologic - instrumentation; Monitoring, Physiologic - methods; Movement; Video Recognition; Patient Privacy; 3D Sensor; Patient Safety; SlowFast
• ISSN: 1879-8365
• DOI: 10.3233/SHTI250864

With the aging of the overall patient population, the incidence of patients developing delirium during hospitalization is increasing. This study aims to improve post-operative safety management and reduce the workload of nurses related to patient care. We have developed a monitoring system that uses 3D sensors to detect specific behaviors and motions that require attention in cases where patients exhibit abnormal behaviors, such as falls and self-removal of IV lines, and trigger alerts. In this paper, we present an algorithm for detecting dangerous motions. We use the point cloud data generated by the 3D sensors that collect 3D information. We analyze the motions of subjects based on changes in the point cloud data and tag specific human body motions and behaviors. When there are no obstacles in the imaging direction of the 3D sensors, we detect human body movements (supine position on the bed, half sitting up, and separated from the bed) with an F-measure of 98.33% and motions (thrashing limbs, touching mouth/neck/arms, no action) with an F-measure of 98.23%. We detect the basic motions that trigger alert notifications. However, the detection accuracy decreases depending on the imaging conditions and subject movements. We use invisible and safe near-infrared light for motion detection and recognition to perform imaging even after lights are turned off, without disturbing patients' sleep. Motion recognition using point cloud data is a privacy-friendly monitoring method with a low risk of acquiring personally identifiable information. In the future, we plan to verify the algorithm using actual patients and investigate the detection of motions in addition to those considered in this study.