Here Comes the Hotstepper

With the rise in popularity of smart home technologies, there has been a corresponding increase in interest in transparent user identification methods. Such methods can tell not only if a person has entered a room, but also specifically who that person is. This technology enables homes to personalize experiences for each user as they move around the house, allowing for the customization of lighting, temperature, and other preferences.

Another potential benefit of transparent user identification methods is improved home security. By recognizing each individual in the home, the smart home can detect when an unauthorized person enters the house and alert the homeowner or authorities accordingly. Additionally, the system can help prevent accidents by identifying children or pets in the home and adjusting settings accordingly.

Despite the many potential benefits of transparent user identification methods, the technology remains relatively rare. This is primarily due to the complexity and expense of implementing such a system. The technology relies on a combination of sensors and advanced algorithms to accurately identify individuals, which can be difficult to implement and maintain. Additionally, the cost of such a system may be prohibitive for many homeowners.

Clever thinking by a group of researchers led by a team at the University of Duisburg-Essen in Germany has led to the development of a new, more practical approach to this problem. Rather than relying on cameras, fingerprint readers, or any other traditional biometrics, this new system called HotFoot leverages the fact that our feet all have unique characteristics. With the help of a thermal camera, those unique characteristics can be used for user identification, even if those users are wearing shoes.

As a person walks into a room, HotFoot captures thermal images of the floor where they recently stepped. The heat signature of their feet remains on the flooring long enough to capture it on all flooring types tested — carpet, laminate, and linoleum — and for all tested types of footwear — two types of shoes or socks. This data is then fed into a random forest classifier to differentiate between the users that it has been trained to recognize.

A trial was conducted with 21 participants walking on three types of flooring, and with three types of footwear. While assessing the performance of the model at accurately identifying individuals, an AUC of between 91.1% and 98.9% was observed. This result shows that thermal imaging may be a feasible method for the continuous and unobtrusive identification of individuals.

There are, however, some caveats to this study, so further validations are needed to confirm that the results would translate into good real-world performance. The number of flooring and footwear types was limited, and not representative of the numerous varieties that would be encountered during actual use. How HotFoot would adapt to a wider range of conditions is presently unknown. Moreover, the tests were all conducted during a single session. What impact, if any, there would be on the predictions if the same individuals returned the next day is not yet understood. Finally, the study only involved 21 individuals, which is quite a small cohort. This is not necessarily a big problem, though, because HotFoot is primarily targeted at residential use cases, where the pool of individuals to identify is generally quite small.

If some of these open questions can be answered, HotFoot might prove to be a very useful, and convenient means of identifying people for smart home applications.