A dynamic evidential network for fall detection

Area of Science:

• Computer Science
• Artificial Intelligence
• Biomedical Engineering

Background:

• Remote home healthcare monitoring systems utilize multisensor fusion for enhanced data accuracy and reliability.
• Heterogeneous sensor data in these systems often exhibit imperfections and varying trust levels.
• Dempster-Shafer theory (DST) is a suitable framework for managing imperfect information in multisensor fusion.

Purpose of the Study:

• To develop a robust remote home healthcare monitoring application for detecting distress situations, specifically falls.
• To improve the accuracy and reliability of fall detection by employing multisensor fusion techniques.
• To address challenges posed by sensor noise, signal variability, and non-stationarity in real-world monitoring environments.

Main Methods:

• Utilized Dempster-Shafer theory (DST) and its graphical representation, evidential networks, for heterogeneous data fusion.
• Implemented evidential networks within a remote medical monitoring platform to enhance automatic fall detection performance.
• Introduced dynamic evidential networks to compensate for the non-stationary nature of sensor signals by considering time evolution.

Main Results:

• The proposed evidential network structure demonstrated improved performance in automatic fall detection.
• Dynamic evidential networks effectively compensated for non-stationary sensor effects in simulated fall scenarios.
• The multisensor fusion approach provided more accurate and reliable information compared to individual sensors.

Conclusions:

• Evidential networks, particularly dynamic variants, offer a reliable method for multisensor data fusion in remote healthcare monitoring.
• The developed system shows promise for increasing the reliability and performance of automatic fall detection in home healthcare settings.
• Addressing sensor imperfections and signal non-stationarity is crucial for effective remote patient monitoring systems.