Analysis: new research shows how behavioural biometrics such as walking provide valuable data about who we are and our health and well-being 

Image analysis of faces, fingerprints and retinal scans to identify people is well established at this stage. But "behavioural biometrics" such as walking provide valuable biomechanical information since every human has to walk or interact with gravity to get about.

Walking generates periodic motion affected by body characteristics such as age, fitness, sex, health and identity and can uniquely identify a person. Humans can recognise family members by the sound of their footsteps. We can spot a loved one walking at a distance or in a crowd by the sounds and visual signals their body makes. Clinicians can identify diseases and conditions from observing how a patient walks, again from their experiential training from many patients. 

Current technology for identifying humans from motion uses cameras, but the information is affected by ambient lighting, viewing angle and interference by other moving objects in the field of view. Since every camera pixel must be scrutinised, analysis is data heavy. Clinical measurement of a person’s gait requires them to walk on an expensive pressure sensitive mat, several metres long, in a gym or sports lab. 

A fixed, ambient floor sensor is preferable to wearable gait sensors or mobile phone apps because every person needs to walk on a floor

This collaborative project with Krikor Ozanyan and Omar Costilla Reyes at the University of Manchester and Ruben Vera-Rodriguez at Universidad Autonoma de Madrid monitored forces exerted on the floor during a footstep cycle. It analysed signals using computational techniques, in which walking patterns of known individuals are used to train artificial neural networks to identify unknown individuals, using specific signal features automatically extracted by the computer.

We analysed 20,000 footstep signals from more than 120 users and evaluated performance in scenarios where person identification is required. These included airport security checkpoints (the smallest training set, as an airport would include many unknown new individuals), work environments (medium training set, since all legitimate individuals are known, but access of individuals to workspace areas needs to be controlled for safety or security reasons) and home environments.

The latter was the largest training set, as legitimate users are small and well defined, but identification of an intruder or change in physical condition of a known individual requires rapid identification. This achieved a performance that was 3.7 times better than the state-of-the-art. 

"Human walking characteristics can be assessed in everyday life by using a wide-area floor sensor deployable in smart housing developed for older people to enable them to live at home safely"

This technology can be applied to health checks because human gait is affected by physical and mental well-being. Mental processes that control planning, attention and remembering and juggling multiple tasks affect physical coordination. Ageing and dementia-related decline in mental performance affects balance and limb co-ordination, causing changes in gait.

Early signs of degradation in cognitive and physical function is evaluated by measuring the change in walking pattern of a person performing a controlled distraction, such as mental arithmetic or spelling tests. It is well known that the distraction caused by speaking or texting on a mobile phone, can cause the user to bump into things or crash their car. The effect of a controlled distraction on physical co-ordination can be related to early on-set of degenerative diseases. 

For monitoring well-being, human walking characteristics can be assessed in everyday life by using a wide-area floor sensor deployable in smart housing developed for older people to enable them to live at home safely. We have developed floor sensors embedded into carpets or flooring and made unobtrusive and affordable over large areas, so that a person’s walking patterns can be monitored during their everyday motion including getting up, going to the kitchen to fill a kettle and going to the bathroom. Thus, decline in habitual movement, that would usually only be noticed by a relative or carer in close contact with the patient, can be quantified and used to trigger improved diet, exercise and medication. If done at an early stage, these measure could reverse the disease process.

This method of non-intrusive motion measurement is highly valuable with the potential to catch early, subtle changes 

A fixed, ambient floor sensor is preferable to wearable gait sensors or mobile phone apps because motivating old and ill people to put on, switch on and maintain wearable sensors is challenging, but every person needs to walk on a floor. Our floor sensor provides integrated pressure measurements and is data conservative compared with vision or pixelated systems, so the data can be cloud stored and analysed in realtime or retrospectively. We have tested a demo mat, two metres long by a metre wide, and costs for the research materials came to around £100 (€115).  

While walking patterns used to distinguish individuals or spot changes in health and well being are difficult to quantify, we have shown that applying Artificial Intelligence and Machine Learning technologies to sensor mats can analyse walking patterns and automatically extract significant features. This method of non-intrusive motion measurement is highly valuable, with the potential to catch early, subtle changes in an individual’s walking pattern and initial changes in molecular biomarkers confirmed by body fluid sampling and related changes in brain structure shown in brain scans.

Currently, biomarker and brain changes are confirmed retrospectively when degradation in a patient’s physical and mental function is disrupting everyday life and is picked up by family and carers. But this is too late and irreversible damage has already occurred. Identifying and relating early physical decline to initial molecular and brain structure changes enables unobtrusive monitoring of large patient numbers, purely from their habitual movement in their own environments. 

"This technology can be applied to health checks because human gait is affected by physical and mental well-being"

It could also provide low cost evaluation of drugs and other treatments (diet, exercise) to treat dementia on a range of patients as an alternative to time-consuming and expensive brain scans, clinical sampling and analysis of body tissue. Early intervention and appropriate treatment will enable patients to stay in their own homes, reducing healthcare costs, and prevent falling, hip fractures and hip replacements, which further degrade the health, mobility, independence and mortality of elderly people.

Currently, there is no drug cure for Alzheimers, and hence a massive need to develop new drugs. Methods of unobtrusively monitoring large numbers of older people in their own homes allows both alerting of early signs of dementia and monitoring of drug regimes enabling cost effective drug discovery and patient trials.  At NUIG, we plan to investigate photonic and laser inscription methods of functionalising smart sensing surfaces for measuring strain and deformation to detect walking and balance for healthcare applications.

The views expressed here are those of the author and do not represent or reflect the views of RTÉ