AI Tech for Nursing: Real Time Bed-Exit Prediction System

Prototyping
Professor Name：Rong-Shue Hsiao
Cooperation Partner: Tian‑Xiang Chen and Chun‑Hao Kao

The system can predict the intention of the patients/elderly to leave bed, and reduce the false alarms. Two prerequisites of hospitals build the system more difficult:
1. Ultra-low resolution depth clip (32 * 24 pixels) to avoid privacy concerns.
2. The installation positions of the image sensor can only be installed on head of a patient’s bed. Not all limbs could be observed after sitting.
We apply deep leaning-based 3D convolutional neural network (3D CNN) framework to overcome the above two difficulties. Experiments demonstrated the online and real-time inference of the activity in bed and recorded live on videos synchronously (YouTube link: http://youtu.be/R-vH-PPLbqc ).The experimental video showed that detection of difficult scenarios can be easily accomplished, and the design can be applied to home and hospital bed structure. 
Market potential in international market: 
Researches shows that 60% of the elderly fall in the patient room, 80% around the bed, 30% fall will cause injuries; in United States, each fall of the elderly cost average $50,534, which have great market demand.