Can we measure mobility and other activity levels at home?

Human activity recognition is an exciting and challenging research area. Understanding the daily behaviour of an elderly person living independently can be valuable for clinicians. Relying on people’s own self-assessment of daily activities has proven inaccurate. This is why there is a great interest in developing automated systems that can perform a number of tasks, like measuring people’s activity level over a period of time or detecting a fall. First, they must collect data from the user, who could be elderly or someone with limited mobility and receiving therapy. Second, the system must analyse the data accurately and in real time. Third, the system must generate reports on the user’s activity level that are understandable to a clinician.

Generally, to collect data for indoor activity recognition,

there are two popular approaches: environmental sensors and wearable sensors. Wearables may include fall detection pendants, inertial sensor-based watches, and personal alert fobs. Wearable devices are mostly battery powered and may have little or no installation cost. However, their effectiveness in collecting data is heavily reliant on the person wearing them all the time. Environmental sensors include technologies such as microphones, microwave sensors, passive infrared sensors and vibration sensors to collect data about movement. Data collected from such sensors may not be enough to detect crucial events like falls and may incur some initial installation cost. As an alternative to these specialised sensors, video-based data offers rich information about the person in the scene, what they are doing and if they require immediate attention.

To ensure that the person is always visible, multiple cameras or specialised wide-angle lenses are needed. Setting up such a system requires an expert installer. Depending on the type of camera used and what it measures, there may be some privacy concerns, which may affect video coverage in a home environment. Human activity happens fast. To capture an event like a fall requires recording video at a rate of 50 to 100 frames per second. When combined with the output from all our other sensors, this generates a massive amount of data. If it can be processed in real-time, it enables live monitoring and response to emergencies. It also avoids storing all this data to process later offline.

On the other hand, building a system that can process all this data in realtime is expensive, requires powerful computers and is technologically challenging.