The Rise of AI Fall Detection Systems

Each year, millions of older adults in the United States experience falls, resulting in serious injuries and even death. The issue is expected to become more prevalent as the population ages, creating new challenges for the healthcare industry and society as a whole.

The scope of the problem is significant, with the Centers for Disease Control and Prevention estimating that one out of every four Americans aged 65 and older will experience a fall each year. In 2019, older adults accounted for over 36 million fall-related visits to emergency departments, and over three million were hospitalized due to fall-related injuries. Additionally, over 39,000 older adults died due to fall-related injuries in 2019.

The physical and mental impacts of falls can be severe. For many older adults, a fall can lead to a loss of mobility and independence, as well as chronic pain and disability. Falls can also have a significant impact on mental health, leading to feelings of isolation, depression, and anxiety.

The medical system burden and costs associated with falls are also significant. In 2015, the total medical costs of falls among older adults exceeded $50 billion. Medicare and Medicaid, and by extension, US taxpayers, were responsible for about 75% of these costs. These costs include hospitalizations, emergency department visits, rehabilitation services, and ongoing medical care for long-term injuries.

To address the problem, healthcare providers and policymakers must take proactive steps to prevent falls among older adults. These steps include identifying individuals at risk for falls and implementing fall prevention programs. Fortunately, monitoring and preventing falls is becoming easier with advances in artificial intelligence and edge computing systems. These advances, coupled with reductions in the costs of the underlying technologies, enabled a group of engineering students at Rice University to build a fairly sophisticated sensor system for monitoring fall risk.

Where the students’ system differs from most other presently available options is that it not only detects falls that have occurred, but it also seeks to better understand why someone fell to prevent it from happening again in the future. Towards these goals, two primary components were developed.

The first component is a wearable device that contains a processing unit running a machine learning algorithm that was trained to recognize when the wearer falls down. Complementing this is an ultrawideband sensor and accelerometer, which provide information about the wearer’s location and movements, respectively. A real-time clock module is included to timestamp all of this information, which is stored on an SD card, for later analysis.

The other component of the system is a lidar sensor on a tripod that would be located in the same room as the wearer of the fall detector. The tripod can be raised and lowered so that the lidar sensor can capture measurements of the room at different heights. This helps it to distinguish features like furniture and walls.

When a fall occurs, the location and movement of the individual at that time can be correlated with nearby objects in the room to determine what they were doing. This is important, because it is common that people do not recall exactly what they were doing immediately before a fall. Using this information, it might be determined, for example, that a person was trying to sit down on a couch right before falling down. That knowledge could be used by a medical professional to determine that the height of the couch is either too high or too low, which presents a risk to that individual.

Recommendations made based on information gleaned from this system could serve to prevent future incidents and help the fall victim to maintain their autonomy for a longer period of time.