Printable Version

Before the Fall


Michael P. Rigney, Ph.D., is a principal engineer in the Manufacturing Systems Department of SwRIís Automation and Data Systems Division. His technical expertise includes developing machine vision inspection, pattern recognition and quality control systems; integrating visible and infrared images, spectrometric sensors and other sensing technologies; and using artificial intelligence techniques for pattern recognition and sensor fusion.


An SwRI-developed monitoring system helps nursing homes prevent patients from attempting to exit their beds without assistance

 By Michael P. Rigney, Ph.D.

Falls account for the highest number of incidents and injuries reported in nursing homes. Most falls occur when residents are in their rooms and unattended by staff. According to the National Center for Injury Prevention and Control at the Centers for Disease Control, 10 to 20 percent of falls in nursing homes cause serious injury, and as many as 75 percent of nursing home residents fall each year. In 2000, the total direct cost of fall injuries for people age 65 and older exceeded $19 billion, and costs may reach $54 billion by 2020.

With a vision to improve the standard of care in nursing homes and critical care facilities, a commercial client asked engineers at Southwest Research Institute (SwRI) to develop an advanced technology fall-prevention solution. The resulting SwRI-developed software application is designed to alert caregivers to impending bed exits by residents so that intervention and assistance can be provided, reducing the risk of a fall-related injury.

Modeling bed-exit behavior

Bed-exiting behavior monitoring software builds on previous SwRI work in automated video surveillance applications such as highway infrastructure protection, intelligent transportation system incident detection, intrusion detection and animal-vehicle collision hazard warning. The common thread in these automated video surveillance applications is that it is impractical for a human observer to monitor continuous video feeds from a large number of cameras in order to detect an occasional incident.

The monitoring system consists of a computer, a video camera and an infrared (IR) illuminator within each residentís room, linked to an intelligent alerting system at the facility. Video from a camera positioned over a monitored bed is sent to an in-room computer. Invisible near-infrared illumination is automatically turned on under low-light conditions, allowing the system to operate even when normal room lighting is turned off. In addition, the camera automatically switches between daytime and nighttime modes based on room illumination levels. SwRI assisted the client with specification, evaluation and testing of system cameras, IR illuminators and the in-room computer.

The in-room computer processes the video feed using SwRI-developed image analysis software that provides bed-exiting behavior detection. Two primary software modules provide automatic bed location and bed-exiting behavior monitoring. These modules are based on video image analysis and pattern recognition algorithms. A separate module running at the facility level integrates utilities for system configuration, camera calibration, behavior classifier training and performance testing.

Automatic bed location allows the system to accommodate the dynamics of a nursing home environment, where beds are frequently moved about during medical or housekeeping procedures. Beds are often mounted on casters, making even inadvertent movement likely. Knowledge of the bedís location is important for subsequent motion analysis, because the bed-exit software is based on recognizing specific motions of the patient relative to the bed.


Reference marks (fiducials) serve as targets for one available bed location method so that patient motions can be analyzed properly in relation to the bed. Beds without fiducials are located by matching the observed bed boundary to a geometric model.


A bed model and a pattern recognition process determine the bedís location each time the system is enabled for bed-exit monitoring. The bed model describes physical size and shape of the bed. Using camera calibration parameters, the bed model can be projected into the video image. A pattern recognition algorithm matches the bed model projection with the image data, thus determining the bedís location.

After locating the bed, the bed-exiting behavior monitor continuously analyzes video from the overhead camera. The analysis begins by updating an adaptive model of the scene. Motion detection and analysis operations compare live video data to the adaptive scene model, and in conjunction with the bed location and bed model, compute specific attributes that characterize the residentís location and activity. These attributes feed into a behavior classifier, which identifies the residentís activity as either normal motion or one of a number of bed-exiting behaviors.


Ceiling-mounted video cameras and infrared illuminators produce visible-light images of patients during daylight hours and infrared images when the room is darkened at night.


The classifier has been trained to detect what most people would consider a ďnormalĒ bed exit: sitting up, pulling back covers, swinging legs over the side of the bed, scooting to the edge and transitioning to a standing pose. It also recognizes other exiting behaviors that residents may exhibit, such as rolling over the side rail or exiting over the footboard. Detected changes in behavior are signaled to a supervisory process running on a computer at the nursing home facility level.

In-room computers, cameras, nursing station consoles and the facility computer all are linked by a local area network. When bed-exiting behavior occurs, an alert is sent immediately to the nursing station. The nurse can see the resident on the nursing station console, which displays the video feed from the over-bed camera or video from a secondary in-room camera. Meanwhile, video of the nurse is displayed to the resident on the in-room monitor. Two-way audio communication complements the video to support interaction between the nurse at her station and the resident. At the same time, caregivers throughout the facility are alerted with a text message via a wireless network to quickly intervene directly with the resident, with the closest caregiver generally being notified. The caregiver typically can respond before the resident actually exits the bed.

Monitoring for security as well as safety

While the base technology was designed initially to detect bed-exiting behavior, the behavior classifier has been extended to recognize bed-entry behavior, characterized by the approach of a person toward the bed. Related to this function is a wireless identification (ID) system that tracks residents, staff and visitors as they move about the facility. But if someone removes their wireless identifier, their location and interaction with residents cannot be tracked by the system. Bed-entry detection, however, ensures that an interaction between an in-bed resident and an un-tagged person would be detected, resulting in the event being recorded and an alert issued to the nursing station.


When a patient exhibits movement characterized as bed-exiting behavior (Top), a machine-vision algorithm detects and analyzes image motion (Bottom) and generates an alert message.


In addition, SwRI has been asked to develop a related video detection capability to further strengthen resident security and abuse prevention. Although bed-entry detection is responsive if someone approaches the bed, a person without a wireless ID could still move freely about the room. This could allow an un-tagged person to circumvent a system design feature that passively monitors all interactions between residents, staff and visitors through video recording. The new capability, room occupancy detection, will analyze video from a second camera that views the entire room and count the number of people in the room. Room occupancy computed from video analysis will be compared to the occupancy count determined by the wireless tracking system. If someone enters the room without their wireless ID, the occupancy count discrepancy will trigger an alert at the nursing station and a recording of the incident.

SwRI-developed video incident detection software will help mitigate the risk of fall-related injury and promote an improved standard of care and quality of life for nursing home residents. Not all falls can be avoided, but a reduction of fall-related injuries helps decrease nursing care cost and reduces liability exposure for nursing homes.

Questions about this article? Contact Rigney at 210-522-5104 or michael.rigney@swri.org.
 

Published in the Summer 2009 issue of Technology Todayģ, published by Southwest Research Institute. For more information, contact Joe Fohn.

Summer 2009 Technology Today
SwRI Publications SwRI Home