Alex Mihailidis, PhD, PEng; Tracy Lee, MASc, Intelligent Assistive Technology and Systems Lab, Department of Occupational Therapy, University of Toronto, Toronto, ON.
It is estimated that one in three older adults will experience a fall over a one-year span, with one-third of these falls occurring in the home.1 Providing immediate response and care when a fall occurs is a key concern and is becoming increasingly difficult to convey as more older adults choose to remain in their own homes, often alone. Situations involving the person left lying on the floor after a fall for an extended period of time before receiving assistance have been reported. This often drastically reduces their probability of recovery, and survival.2
Worn fall detectors (e.g., Tunstall Group Ltd., www.tunstall.co.uk) and emergency response systems (ERS) (e.g., Lifeline Systems Inc., www.lifelinesys.com) are some examples of currently available commercial technologies that attempt to address the problem of fall detection in the older population. Worn fall detectors, mechanical sensors that are worn on the hip, are triggered when both the orientation and acceleration forces of the person reach a pre-set threshold. A common form of ERS is a telephone-based personal system consisting of the person wearing a small help button as a necklace or wristband, which the person pushes manually when an accident has occurred. A two-way telephone system connects the user to emergency services. A primary limitation of these devices is that they require effort from the user in order to be effective. For example, the user must remember to wear the device, which may become less reliable as people age and/or develop cognitive impairments. Furthermore, if a fall causing serious injury occurs, the user may be incapable of pushing the button, thus rendering the device ineffective.
To overcome some of the difficulties associated with aging-in-place and with currently available devices, we have been developing an intelligent home environment that can monitor and assist older adults during activities of daily living. As part of this work, we are developing an intelligent ERS that can automatically and confidently detect if an emergency situation has occurred, such as the person becoming ill or falling, and that subsequently calls for appropriate assistance. Here we focus on the fall detection component of this system.
Fall Detection System
Currently, the fall detection component of this new ERS uses computer vision consisting of a ceiling-mounted digital camera to locate and track the occupant when he or she enters the room. Using simple background subtraction algorithms3 combined with a connective-component labelling technique,4 the image is processed and the shape of the person is determined and extracted as a silhouette. Various features and geometric properties of this silhouette are then calculated by the system, which are used to characterize the person’s posture—i.e., depending on whether the person is standing, sitting or lying down, the silhouette will adopt different shapes and sizes. Once these data have been determined the collected images are then discarded by the system, thus preserving privacy. Combining information on the change in properties such as the area or perimeter of the silhouette with reduced motion of the person, and then comparing these values with pre-set thresholds, the sensing agent is able to detect a fall. Once a fall has been detected, the system prompts the user to check if he or she is okay. It uses voice recognition software to “listen” for the user’s response, or lack thereof, upon which it decides the appropriate actions/responses to execute. For example, if the person has fallen and does not respond, the system will automatically contact the closest emergency facility. However, if the person responds that he or she is fine, no action is necessary and the system learns that that incident was a false alarm.
Initial pilot studies conducted with the system have shown that it can reliably detect falls with an 85% accuracy rate based on a sample of 100 varied actions and postures. However, it should be noted that these preliminary results were obtained in ideal conditions and are constrained to the limitations of the system, which will be further discussed.
Future Work and Implicationsto the Health Care Community
Although the current system has shown promising results, there are issues that need to be considered. For example, the shape parameters identified to characterize a fall are user-dependent (specifically on the height of the user) and may be affected if the person uses assistive devices such as a walker. The system also is currently constrained to track a single person. If another person or a large pet were introduced to the environment, the system may be confused and sound a false alarm.
Future work will focus on addressing these and other limitations, as well as expanding the current system to include other required features of the ERS. For example, new algorithms will be developed to improve the intelligence of the system so that it can intrinsically recognize areas of acceptable inactivity (e.g., the bed or sofa) and eventually learn the living patterns of the person. Learning such patterns and detecting deviations may be used by the system to indicate the onset of health problems.
Work to date has provided some evidence that using intelligent computer systems to ensure the safety of seniors in their homes and to monitor their daily activities offers both a practical and feasible solution. Environments and homes that can intelligently aid in caregiving could play a very significant role in enhancing aging-in-place and thus help to reduce the burden of care on the health care industry.
Johnson M, Cusick A, Chang S. Home-Screen: A short scale to measure fall risk in the home. Public Health Nursing 2001;18:169-77.
Tinetti ME, Liu W, Claus EB. Predictors and prognosis of inability to get up after falls among elderly persons. JAMA 1993;269:65.
Wren C, Azarbayejani A, Darrell T, et al. Pfinder: Real-time tracking of the human body. IEEE transactions on pattern analysis and machine intelligence. 1997;19:780-5.
Ronse C, Denijver PA. Connected components in binary images: the detection problem. Hertfordshire, England: Research Studies Press, 1984.