Embedded Bed-exit Monitoring System Using Deep Learning

In this study, we present a practical bed-exit monitoring system designed specifically for healthcare settings, distinguished by its focus on privacy, accuracy, cost, and ease of use.The NVIDIA Jetson Xavier, a compact System-on-Module, powered the system, using a camera serial interface for video data acquisition.Specifically, the presented system aims to recognize the bed-exit behavior from a series of images with narrow fields of view for privacy preservation.Our approach encompasses a three-stage process for detecting, tracking, and classifying human body trunk movements relative to a bed.First, the You Only Look Once (YOLO) algorithm detects the human body trunk within the scene.Following detection, the Simple Online and Real-time Tracking (SORT) algorithm tracks the detected body trunk objects across frames.Finally, deep learning techniques such as long short-term memory (LSTM) or gated recurrent unit (GRU) networks classify the tracked objects' actions into three categories: getting off the bed, being on the bed, and returning to the bed.Our experimental findings demonstrate that this system achieves a high accuracy rate of 97.97% and operates at a processing speed of 7.1 frames per second.This methodology offers precise bed-exit monitoring and represents a significant step forward in improving patient safety and the efficiency of care in healthcare environments, underscoring its importance in minimizing fall hazards and enhancing patient care quality.