OCR Text |
Show COLLEGE OF ENGINEERING UNDERGRADUATE RESEARCH ABSTRACTS USING RF SENSOR NETWORKS TO DETECT FALLING Brad Mager (Neal Patwari) Department of Electrical and Computer Engineering University of Utah BACKGROUND A major concern for the elderly is that of falling, which affects nearly one-third of all adults over 65 each year and is a leading cause of injury-related deaths. While most existing fall detection systems require the user to wear or carry a device, a better approach is an environment monitoring system that avoids this necessity. One candidate for real-time monitoring employs arrays of wireless radio-frequency (RF) sensor nodes to detect where people are in a room or building, based on h o w they affect the RF field within the sensor array. This technology does not require the person to wear a device, which makes it suitable for assisted-living applications. OBJECTIVE The goal of this research was to demonstrate that an RF sensor network could successfully detect falls and distinguish them from other types of motions. METHODS W e deploy a two-level array of RF sensor nodes around the perimeter of a room and use the shadowing losses in the signals relayed between sensors to detect a person's horizontal and vertical position. A hidden Markov model then determines the subject's vertical pose, and the amount of time between a standing pose and a lying d o w n pose determines if a fall has occurred. Using a 3 m x 3 m sensor array, w e conducted nine carefully-timed sets of motions (e.g., walking, lying down, sitting, falling) three times each with two subjects, for a total of 54 experiments. The Markov model was trained with one subject's data, and that model was used to process the second subject's data. RESULTS A simple experiment in an uncluttered room achieved 1 0 0 % reliability in fall detection. Finding the proper amount of time between standing and lying d o w n was crucial in reaching a balance between detecting every fall and minimizing false alarms. CONCLUSION A two-level array of RF sensor nodes can accurately detect falls when the data are processed with the proper algorithm, suggesting the viability of such a system for use in assisted-living facilities or other elder care situations. Neal Patwari |